Skip to contentSkip to conversation

sched: subport level configuration of pipe nodes

PATCH Superseded Jasvinder Singh 3 participants 121 messages Aug 23, 2019 – Oct 25, 2019 Page 1 of 2

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover sched: subport level configuration of pipe nodes

This patchset refactors the dpdk qos sched library to allow subport level configuration flexibility of the pipe nodes.

Currently, all parameters for the pipe nodes (subscribers) configuration are part of the port level structure which forces all groups of subscribers (pipes) in different subports to have similar configurations in terms of their number, queue sizes, traffic-classes, etc.

The new implementation moves pipe nodes configuration parameters from port level to subport level structure. This allows different subports of the same port to have different configuration for the pipe nodes, for examples- number of pipes, queue sizes, pipe profiles, etc.

In order to keep the implementation complexity under control, all pipes within the same subport share the same configuration for queue sizes.

Jasvinder Singh (15):
sched: add pipe config params to subport struct
sched: modify internal structs for config flexibility
sched: remove pipe params config from port level
shced: add pipe config to subport level
sched: modify pipe functions for config flexibility
sched: modify pkt enqueue for config flexibility
sched: update memory compute to support flexiblity
sched: update grinder functions for config flexibility
sched: update pkt dequeue for flexible config
sched: update queue stats read for config flexibility test/sched: modify tests for subport config flexibility net/softnic: add subport config flexibility to TM function
ip_pipeline: add subport config flexibility to TM function examples/qos_sched: add subport configuration flexibility
sched: remove redundant code

 app/test/test_sched.c                    |   35 +-
 doc/guides/rel_notes/release_19_11.rst   |    6 +-
 drivers/net/softnic/rte_eth_softnic_tm.c |   51 +-
 examples/ip_pipeline/cli.c               |   71 +-
 examples/ip_pipeline/tmgr.c              |   23 +-
 examples/ip_pipeline/tmgr.h              |    7 +-
 examples/qos_sched/app_thread.c          |    4 +-
 examples/qos_sched/cfg_file.c            |  226 ++--
 examples/qos_sched/init.c                |   54 +-
 examples/qos_sched/main.h                |    1 +
 examples/qos_sched/profile.cfg           |    5 +-
 examples/qos_sched/profile_ov.cfg        |    5 +-
 examples/qos_sched/stats.c               |   36 +-
 lib/librte_sched/Makefile                |    2 +-
 lib/librte_sched/meson.build             |    2 +-
 lib/librte_sched/rte_sched.c             | 1400 +++++++++++++---------
 lib/librte_sched/rte_sched.h             |  114 +-
 17 files changed, 1180 insertions(+), 862 deletions(-)
Signature · 1 line
2.21.0
#1 lore raw

PATCH 1/15 sched: add pipe config params to subport struct

Add pipe configuration parameters to subport level structure to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/Makefile    |  2 +-
 lib/librte_sched/meson.build |  2 +-
 lib/librte_sched/rte_sched.h | 81 +++++++++++++++++++++++-------------
 3 files changed, 55 insertions(+), 30 deletions(-)
lib/librte_sched/Makefile +1 -1
index 3d7f410e1..6e4a72d89 100644
@@ -18,7 +18,7 @@ LDLIBS += -lrte_timer
 
 EXPORT_MAP := rte_sched_version.map
 
-LIBABIVER := 3
+LIBABIVER := 4
 
 #
 # all source are stored in SRCS-y
lib/librte_sched/meson.build +1 -1
index 59d43c6d8..9f40a2368 100644
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2017 Intel Corporation
 
-version = 3
+version = 4
 sources = files('rte_sched.c', 'rte_red.c', 'rte_approx.c')
 headers = files('rte_sched.h', 'rte_sched_common.h',
 		'rte_red.h', 'rte_approx.h')
lib/librte_sched/rte_sched.h +53 -28

126 lines not on this page · raw message

Signature · 1 line
2.21.0
#2 lore raw

PATCH 2/15 sched: modify internal structs for subport config

In reply to Jasvinder Singh

Update internal structures related to port and subport to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 106 ++++++++++++++++++++++++++---------
 1 file changed, 79 insertions(+), 27 deletions(-)
lib/librte_sched/rte_sched.c +79 -27
index d8ab21d64..6a9990e5e 100644
@@ -47,33 +47,6 @@
  */
 #define RTE_SCHED_TIME_SHIFT		      8
 
-struct rte_sched_subport {
-	/* Token bucket (TB) */
-	uint64_t tb_time; /* time of last update */
-	uint32_t tb_period;
-	uint32_t tb_credits_per_period;
-	uint32_t tb_size;
-	uint32_t tb_credits;
-
-	/* Traffic classes (TCs) */
-	uint64_t tc_time; /* time of next update */
-	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_period;
-
-	/* TC oversubscription */
-	uint32_t tc_ov_wm;
-	uint32_t tc_ov_wm_min;
-	uint32_t tc_ov_wm_max;
-	uint8_t tc_ov_period_id;
-	uint8_t tc_ov;
-	uint32_t tc_ov_n;
-	double tc_ov_rate;
-
-	/* Statistics */
-	struct rte_sched_subport_stats stats;
-};
-
 struct rte_sched_pipe_profile {
 	/* Token bucket (TB) */
 	uint32_t tb_period;
@@ -166,10 +139,77 @@ struct rte_sched_grinder {
 	uint8_t wrr_cost[RTE_SCHED_BE_QUEUES_PER_PIPE];
 };
 
+struct rte_sched_subport {
+	/* Token bucket (TB) */
+	uint64_t tb_time; /* time of last update */
+	uint32_t tb_period;
+	uint32_t tb_credits_per_period;
+	uint32_t tb_size;
+	uint32_t tb_credits;
+
+	/* Traffic classes (TCs) */
+	uint64_t tc_time; /* time of next update */
+	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_period;
+
+	/* TC oversubscription */
+	uint32_t tc_ov_wm;
+	uint32_t tc_ov_wm_min;
+	uint32_t tc_ov_wm_max;
+	uint8_t tc_ov_period_id;
+	uint8_t tc_ov;
+	uint32_t tc_ov_n;
+	double tc_ov_rate;
+
+	/* Statistics */
+	struct rte_sched_subport_stats stats;
+
+	/* Subport pipes */
+	uint32_t n_pipes_per_subport;
+	uint32_t n_pipe_profiles;
+
+	/* Pipe best-effort TC rate */
+	uint32_t pipe_tc_be_rate_max;
+
+	/* Pipe queues size */
+	uint16_t qsize[RTE_SCHED_QUEUES_PER_PIPE];
+
+#ifdef RTE_SCHED_RED
+	struct rte_red_config red_config[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][RTE_COLORS];
+#endif
+
+	/* Scheduling loop detection */
+	uint32_t pipe_loop;
+	uint32_t pipe_exhaustion;
+
+	/* Bitmap */
+	struct rte_bitmap *bmp;
+	uint32_t grinder_base_bmp_pos[RTE_SCHED_PORT_N_GRINDERS] __rte_aligned_16;
+
+	/* Grinders */
+	struct rte_sched_grinder grinder[RTE_SCHED_PORT_N_GRINDERS];
+	uint32_t busy_grinders;
+
+	/* Queue base calculation */
+	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
+	uint32_t qsize_sum;
+
+	struct rte_sched_pipe *pipe;
+	struct rte_sched_queue *queue;
+	struct rte_sched_queue_extra *queue_extra;
+	struct rte_sched_pipe_profile *pipe_profiles;
+	uint8_t *bmp_array;
+	struct rte_mbuf **queue_array;
+	uint8_t memory[0] __rte_cache_aligned;
+} __rte_cache_aligned;
+
 struct rte_sched_port {
 	/* User parameters */
 	uint32_t n_subports_per_port;
 	uint32_t n_pipes_per_subport;
+	uint32_t n_max_pipes_per_subport;
+	uint32_t n_max_pipes_per_subport_log2;
 	uint32_t n_pipes_per_subport_log2;
 	uint16_t pipe_queue[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint8_t pipe_tc[RTE_SCHED_QUEUES_PER_PIPE];
@@ -177,6 +217,7 @@ struct rte_sched_port {
 	uint32_t rate;
 	uint32_t mtu;
 	uint32_t frame_overhead;
+	int socket;
 	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t n_pipe_profiles;
 	uint32_t n_max_pipe_profiles;
@@ -210,6 +251,7 @@ struct rte_sched_port {
 	uint32_t qsize_sum;
 
 	/* Large data structures */
+	struct rte_sched_subport *subports[0];
 	struct rte_sched_subport *subport;
 	struct rte_sched_pipe *pipe;
 	struct rte_sched_queue *queue;
@@ -231,6 +273,16 @@ enum rte_sched_port_array {
 	e_RTE_SCHED_PORT_ARRAY_TOTAL,
 };
 
+enum rte_sched_subport_array {
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE = 0,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA,
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES,
+	e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
+};
+
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline uint32_t
Signature · 1 line
2.21.0
#3 lore raw

PATCH 3/15 sched: remove pipe params config from port level

In reply to Jasvinder Singh

Remove pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 266 +++++++++++------------------------
 1 file changed, 81 insertions(+), 185 deletions(-)
lib/librte_sched/rte_sched.c +81 -185
index 6a9990e5e..db8b0485a 100644
@@ -293,6 +293,31 @@ rte_sched_port_queues_per_subport(struct rte_sched_port *port)
 
 #endif
 
+static inline uint32_t
+rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
+{
+	return RTE_SCHED_QUEUES_PER_PIPE * subport->n_pipes_per_subport;
+}
+
+static inline struct rte_mbuf **
+rte_sched_subport_pipe_qbase(struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t pindex = qindex >> 4;
+	uint32_t qpos = qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1);
+
+	return (subport->queue_array + pindex *
+		subport->qsize_sum + subport->qsize_add[qpos]);
+}
+
+static inline uint16_t
+rte_sched_subport_pipe_qsize(struct rte_sched_port *port,
+struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
+
+	return subport->qsize[tc];
+}
+
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
@@ -416,8 +441,6 @@ pipe_profile_check(struct rte_sched_pipe_params *params,
 static int
 rte_sched_port_check_params(struct rte_sched_port_params *params)
 {
-	uint32_t i;
-
 	if (params == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter params\n", __func__);
@@ -454,49 +477,21 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 		return -EINVAL;
 	}
 
-	/* n_pipes_per_subport: non-zero, power of 2 */
-	if (params->n_pipes_per_subport == 0 ||
-	    !rte_is_power_of_2(params->n_pipes_per_subport)) {
+	/* n_max_pipes_per_subport: non-zero, power of 2 */
+	if (params->n_max_pipes_per_subport == 0 ||
+	    !rte_is_power_of_2(params->n_max_pipes_per_subport)) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for pipes number\n", __func__);
+			"%s: Incorrect value for maximum pipes number\n", __func__);
 		return -EINVAL;
 	}
 
-	/* qsize: if non-zero, power of 2,
-	 * no bigger than 32K (due to 16-bit read/write pointers)
-	 */
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint16_t qsize = params->qsize[i];
-
-		if ((qsize != 0 && !rte_is_power_of_2(qsize)) ||
-			((i == RTE_SCHED_TRAFFIC_CLASS_BE) && (qsize == 0))) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	/* pipe_profiles and n_pipe_profiles */
-	if (params->pipe_profiles == NULL ||
-	    params->n_pipe_profiles == 0 ||
-		 params->n_pipe_profiles > params->n_max_pipe_profiles) {
+	/* n_max_pipe_profiles: non-zero */
+	if (params->n_max_pipe_profiles == 0) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for number of pipe profiles\n", __func__);
+			"%s: Incorrect value for maximum pipe profiles\n", __func__);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < params->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
-		int status;
-
-		status = pipe_profile_check(p, params->rate, &params->qsize[0]);
-		if (status != 0) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Pipe profile check failed(%d)\n", __func__, status);
-			return -EINVAL;
-		}
-	}
-
 	return 0;
 }
 
@@ -584,32 +579,6 @@ rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
 	return size0 + size1;
 }
 
-static void
-rte_sched_port_config_qsize(struct rte_sched_port *port)
-{
-	uint32_t i;
-
-	port->qsize_add[0] = 0;
-
-	/* Strict prority traffic class */
-	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		port->qsize_add[i] = port->qsize_add[i-1] + port->qsize[i-1];
-
-	/* Best-effort traffic class */
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-	port->qsize_sum = port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-}
-
 static void
 rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
 {
@@ -719,56 +688,42 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
-static void
-rte_sched_port_config_pipe_profile_table(struct rte_sched_port *port,
-	struct rte_sched_port_params *params)
-{
-	uint32_t i;
-
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		struct rte_sched_pipe_profile *dst = port->pipe_profiles + i;
-
-		rte_sched_pipe_profile_convert(port, src, dst, params->rate);
-		rte_sched_port_log_pipe_profile(port, i);
-	}
-
-	port->pipe_tc_be_rate_max = 0;
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-		if (port->pipe_tc_be_rate_max < pipe_tc_be_rate)
-			port->pipe_tc_be_rate_max = pipe_tc_be_rate;
-	}
-}
-
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
 	struct rte_sched_port *port = NULL;
-	uint32_t mem_size, bmp_mem_size, n_queues_per_port, i, j, cycles_per_byte;
+	uint32_t size0, size1;
+	uint32_t cycles_per_byte;
+	uint32_t i, j;
+	int status;
 
-	/* Check user parameters. Determine the amount of memory to allocate */
-	mem_size = rte_sched_port_get_memory_footprint(params);
-	if (mem_size == 0)
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 		return NULL;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = params->n_subports_per_port * sizeof(struct rte_sched_subport *);
 
 	/* Allocate memory to store the data structures */
-	port = rte_zmalloc_socket("qos_params", mem_size, RTE_CACHE_LINE_SIZE,
+	port = rte_zmalloc_socket("qos_params", size0 + size1, RTE_CACHE_LINE_SIZE,
 		params->socket);
-	if (port == NULL)
-		return NULL;
+	if (port == NULL) {
+		RTE_LOG(ERR, SCHED, "%s: Memory allocation fails\n", __func__);
 
-	/* compile time checks */
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS & (RTE_SCHED_PORT_N_GRINDERS - 1));
+		return NULL;
+	}
 
 	/* User parameters */
 	port->n_subports_per_port = params->n_subports_per_port;
-	port->n_pipes_per_subport = params->n_pipes_per_subport;
-	port->n_pipes_per_subport_log2 =
-			__builtin_ctz(params->n_pipes_per_subport);
+	port->n_max_pipes_per_subport = params->n_max_pipes_per_subport;
+	port->n_max_pipes_per_subport_log2 =
+			__builtin_ctz(params->n_max_pipes_per_subport);
+	port->n_max_pipe_profiles = params->n_max_pipe_profiles;
+	port->socket = params->socket;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 		port->pipe_queue[i] = i;
@@ -789,32 +744,6 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	port->rate = params->rate;
 	port->mtu = params->mtu + params->frame_overhead;
 	port->frame_overhead = params->frame_overhead;
-	memcpy(port->qsize, params->qsize, sizeof(params->qsize));
-	port->n_pipe_profiles = params->n_pipe_profiles;
-	port->n_max_pipe_profiles = params->n_max_pipe_profiles;
-
-#ifdef RTE_SCHED_RED
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t j;
-
-		for (j = 0; j < RTE_COLORS; j++) {
-			/* if min/max are both zero, then RED is disabled */
-			if ((params->red_params[i][j].min_th |
-			     params->red_params[i][j].max_th) == 0) {
-				continue;
-			}
-
-			if (rte_red_config_init(&port->red_config[i][j],
-				params->red_params[i][j].wq_log2,
-				params->red_params[i][j].min_th,
-				params->red_params[i][j].max_th,
-				params->red_params[i][j].maxp_inv) != 0) {
-				rte_free(port);
-				return NULL;
-			}
-		}
-	}
-#endif
 
 	/* Timing */
 	port->time_cpu_cycles = rte_get_tsc_cycles();
@@ -825,79 +754,32 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 		/ params->rate;
 	port->inv_cycles_per_byte = rte_reciprocal_value(cycles_per_byte);
 
-	/* Scheduling loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
-	port->pipe_exhaustion = 0;
-
 	/* Grinders */
-	port->busy_grinders = 0;
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
 
-	/* Queue base calculation */
-	rte_sched_port_config_qsize(port);
-
-	/* Large data structures */
-	port->subport = (struct rte_sched_subport *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_SUBPORT));
-	port->pipe = (struct rte_sched_pipe *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE));
-	port->queue = (struct rte_sched_queue *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE));
-	port->queue_extra = (struct rte_sched_queue_extra *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA));
-	port->pipe_profiles = (struct rte_sched_pipe_profile *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES));
-	port->bmp_array =  port->memory
-		+ rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY);
-	port->queue_array = (struct rte_mbuf **)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY));
-
-	/* Pipe profile table */
-	rte_sched_port_config_pipe_profile_table(port, params);
-
-	/* Bitmap */
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
-	bmp_mem_size = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	port->bmp = rte_bitmap_init(n_queues_per_port, port->bmp_array,
-				    bmp_mem_size);
-	if (port->bmp == NULL) {
-		RTE_LOG(ERR, SCHED, "Bitmap init error\n");
-		rte_free(port);
-		return NULL;
-	}
-
-	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
-		port->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
-
-
 	return port;
 }
 
-void
-rte_sched_port_free(struct rte_sched_port *port)
+static inline void
+rte_sched_subport_free(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
+	uint32_t n_subport_pipe_queues;
 	uint32_t qindex;
-	uint32_t n_queues_per_port;
 
-	/* Check user parameters */
-	if (port == NULL)
+	if (subport == NULL)
 		return;
 
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(subport);
 
 	/* Free enqueued mbufs */
-	for (qindex = 0; qindex < n_queues_per_port; qindex++) {
-		struct rte_mbuf **mbufs = rte_sched_port_qbase(port, qindex);
-		uint16_t qsize = rte_sched_port_qsize(port, qindex);
+	for (qindex = 0; qindex < n_subport_pipe_queues; qindex++) {
+		struct rte_mbuf **mbufs =
+			rte_sched_subport_pipe_qbase(subport, qindex);
+		uint16_t qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 		if (qsize != 0) {
-			struct rte_sched_queue *queue = port->queue + qindex;
+			struct rte_sched_queue *queue = subport->queue + qindex;
 			uint16_t qr = queue->qr & (qsize - 1);
 			uint16_t qw = queue->qw & (qsize - 1);
 
@@ -906,7 +788,21 @@ rte_sched_port_free(struct rte_sched_port *port)
 		}
 	}
 
-	rte_bitmap_free(port->bmp);
+	rte_bitmap_free(subport->bmp);
+}
+
+void
+rte_sched_port_free(struct rte_sched_port *port)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (port == NULL)
+		return;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		rte_sched_subport_free(port, port->subports[i]);
+
 	rte_free(port);
 }
 
Signature · 1 line
2.21.0
#4 lore raw

PATCH 4/15 shced: add pipe config to subport level

In reply to Jasvinder Singh

Add pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 390 ++++++++++++++++++++++++++++++-----
 1 file changed, 334 insertions(+), 56 deletions(-)
lib/librte_sched/rte_sched.c +334 -56
index db8b0485a..1491cb695 100644
@@ -559,24 +559,92 @@ rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sch
 	return base;
 }
 
-uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+static uint32_t
+rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
+	uint32_t n_max_pipe_profiles,
+	enum rte_sched_subport_array array)
 {
-	uint32_t size0, size1;
-	int status;
+	uint32_t n_pipes_per_subport = params->n_pipes_per_subport;
+	uint32_t n_subport_pipe_queues =
+		RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport;
 
-	status = rte_sched_port_check_params(params);
-	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+	uint32_t size_pipe = n_pipes_per_subport * sizeof(struct rte_sched_pipe);
+	uint32_t size_queue =
+		n_subport_pipe_queues * sizeof(struct rte_sched_queue);
+	uint32_t size_queue_extra
+		= n_subport_pipe_queues * sizeof(struct rte_sched_queue_extra);
+	uint32_t size_pipe_profiles = n_max_pipe_profiles *
+		sizeof(struct rte_sched_pipe_profile);
+	uint32_t size_bmp_array =
+		rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	uint32_t size_per_pipe_queue_array, size_queue_array;
 
-		return 0;
+	uint32_t base, i;
+
+	size_per_pipe_queue_array = 0;
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
+			size_per_pipe_queue_array +=
+				params->qsize[i] * sizeof(struct rte_mbuf *);
+		else
+			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
+				params->qsize[i] * sizeof(struct rte_mbuf *);
 	}
+	size_queue_array = n_pipes_per_subport * size_per_pipe_queue_array;
 
-	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_TOTAL);
+	base = 0;
 
-	return size0 + size1;
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
+
+	return base;
+}
+
+static void
+rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
+{
+	uint32_t i;
+
+	subport->qsize_add[0] = 0;
+
+	/* Strict prority traffic class */
+	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		subport->qsize_add[i] = subport->qsize_add[i-1] + subport->qsize[i-1];
+
+	/* Best-effort traffic class */
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+	subport->qsize_sum = subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
 }
 
 static void
@@ -688,6 +756,126 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static int
+rte_sched_subport_check_params(struct rte_sched_subport_params *params,
+	uint32_t n_max_pipes_per_subport,
+	uint32_t n_max_pipe_profiles,
+	uint32_t rate)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (params == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for parameter params\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_rate == 0 || params->tb_rate > rate) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb rate\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_size == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb size\n", __func__);
+		return -EINVAL;
+	}
+
+	/* qsize: if non-zero, power of 2,
+	 * no bigger than 32K (due to 16-bit read/write pointers)
+	 */
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint16_t qsize = params->qsize[i];
+
+		if (qsize != 0 && !rte_is_power_of_2(qsize)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for qsize\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t tc_rate = params->tc_rate[i];
+		uint16_t qsize = params->qsize[i];
+
+		if ((qsize == 0 && tc_rate != 0) ||
+			(qsize != 0 && tc_rate == 0) ||
+			(tc_rate > params->tb_rate)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for tc rate\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	if (params->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
+		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect qsize or tc rate(best effort)\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tc_period == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tc period\n", __func__);
+		return -EINVAL;
+	}
+
+	/* n_pipes_per_subport: non-zero, power of 2 */
+	if (params->n_pipes_per_subport == 0 ||
+		params->n_pipes_per_subport > n_max_pipes_per_subport ||
+	    !rte_is_power_of_2(params->n_pipes_per_subport)) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipes number\n", __func__);
+		return -EINVAL;
+	}
+
+	/* pipe_profiles and n_pipe_profiles */
+	if (params->pipe_profiles == NULL ||
+	    params->n_pipe_profiles == 0 ||
+		params->n_pipe_profiles > n_max_pipe_profiles) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipe profiles\n", __func__);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < params->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
+		int status;
+
+		status = pipe_profile_check(p, rate, &params->qsize[0]);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Pipe profile check failed(%d)\n", __func__, status);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+uint32_t
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+{
+	uint32_t size0, size1;
+	int status;
+
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"Port scheduler params check failed (%d)\n", status);
+
+		return 0;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = rte_sched_port_get_array_base(params,
+			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	return size0 + size1;
+}
+
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
@@ -809,7 +997,7 @@ rte_sched_port_free(struct rte_sched_port *port)
 static void
 rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 {
-	struct rte_sched_subport *s = port->subport + i;
+	struct rte_sched_subport *s = port->subports[i];
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for subport %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -844,72 +1032,80 @@ rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 		s->tc_ov_wm_max);
 }
 
+static void
+rte_sched_free_memory(struct rte_sched_port *port, uint32_t n_subports)
+{
+	uint32_t i;
+
+	for (i = 0; i < n_subports; i++) {
+		struct rte_sched_subport *subport = port->subports[i];
+
+		rte_sched_subport_free(port, subport);
+	}
+
+	rte_free(port);
+}
+
 int
 rte_sched_subport_config(struct rte_sched_port *port,
 	uint32_t subport_id,
 	struct rte_sched_subport_params *params)
 {
-	struct rte_sched_subport *s;
-	uint32_t i;
+	struct rte_sched_subport *s = NULL;
+	uint32_t n_subports = subport_id;
+	uint32_t n_subport_pipe_queues, i;
+	uint32_t size0, size1, bmp_mem_size;
+	int status;
 
 	/* Check user parameters */
 	if (port == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter port\n", __func__);
-		return -EINVAL;
+		return 0;
 	}
 
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for subport id\n", __func__);
-		return -EINVAL;
-	}
 
-	if (params == NULL) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for parameter params\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (params->tb_rate == 0 || params->tb_rate > port->rate) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb rate\n", __func__);
-		return -EINVAL;
-	}
+	status = rte_sched_subport_check_params(params,
+		port->n_max_pipes_per_subport,
+		port->n_max_pipe_profiles,
+		port->rate);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 
-	if (params->tb_size == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb size\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t tc_rate = params->tc_rate[i];
-		uint16_t qsize = port->qsize[i];
+	/* Determine the amount of memory to allocate */
+	size0 = sizeof(struct rte_sched_subport);
+	size1 = rte_sched_subport_get_array_base(params,
+			port->n_max_pipe_profiles,
+			e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
 
-		if ((qsize == 0 && tc_rate != 0) ||
-			(qsize != 0 && tc_rate == 0) ||
-			(tc_rate > params->tb_rate)) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	if (port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
-		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+	/* Allocate memory to store the data structures */
+	s = rte_zmalloc_socket("subport_params", size0 + size1,
+		RTE_CACHE_LINE_SIZE, port->socket);
+	if (s == NULL) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc rate(best effort)\n", __func__);
-		return -EINVAL;
-	}
+			"%s: Memory allocation fails\n", __func__);
 
-	if (params->tc_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc period\n", __func__);
-		return -EINVAL;
+		rte_sched_free_memory(port, n_subports);
+		return -ENOMEM;
 	}
 
-	s = port->subport + subport_id;
+	n_subports++;
+
+	/* Port */
+	port->subports[subport_id] = s;
 
 	/* Token Bucket (TB) */
 	if (params->tb_rate == port->rate) {
@@ -929,26 +1125,108 @@ rte_sched_subport_config(struct rte_sched_port *port,
 	/* Traffic Classes (TCs) */
 	s->tc_period = rte_sched_time_ms_to_bytes(params->tc_period, port->rate);
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(params->tc_period,
-								 params->tc_rate[i]);
-
+					params->tc_rate[i]);
 	}
 	s->tc_time = port->time + s->tc_period;
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits[i] = s->tc_credits_per_period[i];
 
+	/* compile time checks */
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS &
+		(RTE_SCHED_PORT_N_GRINDERS - 1));
+
+	/* User parameters */
+	s->n_pipes_per_subport = params->n_pipes_per_subport;
+	memcpy(s->qsize, params->qsize, sizeof(params->qsize));
+	s->n_pipe_profiles = params->n_pipe_profiles;
+
+#ifdef RTE_SCHED_RED
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t j;
+
+		for (j = 0; j < RTE_COLORS; j++) {
+			/* if min/max are both zero, then RED is disabled */
+			if ((params->red_params[i][j].min_th |
+			     params->red_params[i][j].max_th) == 0) {
+				continue;
+			}
+
+			if (rte_red_config_init(&s->red_config[i][j],
+				params->red_params[i][j].wq_log2,
+				params->red_params[i][j].min_th,
+				params->red_params[i][j].max_th,
+				params->red_params[i][j].maxp_inv) != 0) {
+				rte_sched_free_memory(port, n_subports);
+
+				RTE_LOG(NOTICE, SCHED,
+				"%s: RED configuration init fails\n", __func__);
+				return -EINVAL;
+			}
+		}
+	}
+#endif
+
+	/* Scheduling loop detection */
+	s->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	s->pipe_exhaustion = 0;
+
+	/* Grinders */
+	s->busy_grinders = 0;
+
+	/* Queue base calculation */
+	rte_sched_subport_config_qsize(s);
+
+	/* Large data structures */
+	s->pipe = (struct rte_sched_pipe *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_PIPE));
+	s->queue = (struct rte_sched_queue *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE));
+	s->queue_extra = (struct rte_sched_queue_extra *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA));
+	s->pipe_profiles = (struct rte_sched_pipe_profile *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES));
+	s->bmp_array =  s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY);
+	s->queue_array = (struct rte_mbuf **)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
+
+	/* Bitmap */
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
+	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	s->bmp = rte_bitmap_init(n_subport_pipe_queues, s->bmp_array,
+				bmp_mem_size);
+	if (s->bmp == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Subport bitmap init error\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
+		s->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
+
+#ifdef RTE_SCHED_SUBPORT_TC_OV
 	/* TC oversubscription */
 	s->tc_ov_wm_min = port->mtu;
 	s->tc_ov_wm_max = rte_sched_time_ms_to_bytes(params->tc_period,
-						     port->pipe_tc_be_rate_max);
+						     s->pipe_tc_be_rate_max);
 	s->tc_ov_wm = s->tc_ov_wm_max;
 	s->tc_ov_period_id = 0;
 	s->tc_ov = 0;
 	s->tc_ov_n = 0;
 	s->tc_ov_rate = 0;
+#endif
 
 	rte_sched_port_log_subport_config(port, subport_id);
 
Signature · 1 line
2.21.0
#5 lore raw

PATCH 5/15 sched: modify pipe functions for config flexibility

In reply to Jasvinder Singh

Modify pipe level functions to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 95 +++++++++++++++++++++++++-----------
 lib/librte_sched/rte_sched.h |  5 +-
 2 files changed, 70 insertions(+), 30 deletions(-)
lib/librte_sched/rte_sched.c +66 -29
index 1491cb695..6bafbd8fe 100644
@@ -648,9 +648,9 @@ rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
 }
 
 static void
-rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
+rte_sched_port_log_pipe_profile(struct rte_sched_subport *subport, uint32_t i)
 {
-	struct rte_sched_pipe_profile *p = port->pipe_profiles + i;
+	struct rte_sched_pipe_profile *p = subport->pipe_profiles + i;
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for pipe profile %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -699,7 +699,7 @@ rte_sched_time_ms_to_bytes(uint32_t time_ms, uint32_t rate)
 }
 
 static void
-rte_sched_pipe_profile_convert(struct rte_sched_port *port,
+rte_sched_pipe_profile_convert(struct rte_sched_subport *subport,
 	struct rte_sched_pipe_params *src,
 	struct rte_sched_pipe_profile *dst,
 	uint32_t rate)
@@ -728,7 +728,7 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 						rate);
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (subport->qsize[i])
 			dst->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(src->tc_period,
 					src->tc_rate[i]);
@@ -756,6 +756,30 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static void
+rte_sched_subport_config_pipe_profile_table(struct rte_sched_subport *subport,
+	struct rte_sched_subport_params *params, uint32_t rate)
+{
+	uint32_t i;
+
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		struct rte_sched_pipe_profile *dst = subport->pipe_profiles + i;
+
+		rte_sched_pipe_profile_convert(subport, src, dst, rate);
+		rte_sched_port_log_pipe_profile(subport, i);
+	}
+
+	subport->pipe_tc_be_rate_max = 0;
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+		if (subport->pipe_tc_be_rate_max < pipe_tc_be_rate)
+			subport->pipe_tc_be_rate_max = pipe_tc_be_rate;
+	}
+}
+
 static int
 rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 	uint32_t n_max_pipes_per_subport,
@@ -1200,6 +1224,9 @@ rte_sched_subport_config(struct rte_sched_port *port,
 		(s->memory + rte_sched_subport_get_array_base(params,
 		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
 
+	/* Pipe profile table */
+	rte_sched_subport_config_pipe_profile_table(s, params, port->rate);
+
 	/* Bitmap */
 	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
 	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
@@ -1242,6 +1269,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	struct rte_sched_subport *s;
 	struct rte_sched_pipe *p;
 	struct rte_sched_pipe_profile *params;
+	uint32_t n_subports = subport_id + 1;
 	uint32_t deactivate, profile, i;
 
 	/* Check user parameters */
@@ -1257,34 +1285,32 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter subport id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (pipe_id >= port->n_pipes_per_subport) {
+	s = port->subports[subport_id];
+	if (pipe_id >= s->n_pipes_per_subport) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (!deactivate && profile >= port->n_pipe_profiles) {
+	if (!deactivate && profile >= s->n_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe profile\n", __func__);
-		return -EINVAL;
-	}
 
-	/* Check that subport configuration is valid */
-	s = port->subport + subport_id;
-	if (s->tb_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Subport configuration invalid\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	p = port->pipe + (subport_id * port->n_pipes_per_subport + pipe_id);
-
 	/* Handle the case when pipe already has a valid configuration */
+	p = s->pipe + pipe_id;
 	if (p->tb_time) {
-		params = port->pipe_profiles + p->profile;
+		params = s->pipe_profiles + p->profile;
 
 		double subport_tc_be_rate =
 			(double) s->tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASS_BE]
@@ -1314,7 +1340,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 
 	/* Apply the new pipe configuration */
 	p->profile = profile;
-	params = port->pipe_profiles + p->profile;
+	params = s->pipe_profiles + p->profile;
 
 	/* Token Bucket (TB) */
 	p->tb_time = port->time;
@@ -1324,7 +1350,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	p->tc_time = port->time + params->tc_period;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (s->qsize[i])
 			p->tc_credits[i] = params->tc_credits_per_period[i];
 
 	{
@@ -1355,9 +1381,11 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 
 int
 rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_pipe_profile *pp;
 	uint32_t i;
 	int status;
@@ -1369,40 +1397,49 @@ rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
+	/* Subport id not exceeds the max limit */
+	if (subport_id > port->n_subports_per_port) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for subport id\n", __func__);
+		return -EINVAL;
+	}
+
+	s = port->subports[subport_id];
+
 	/* Pipe profiles not exceeds the max limit */
-	if (port->n_pipe_profiles >= port->n_max_pipe_profiles) {
+	if (s->n_pipe_profiles >= port->n_max_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Number of pipe profiles exceeds the max limit\n", __func__);
 		return -EINVAL;
 	}
 
 	/* Pipe params */
-	status = pipe_profile_check(params, port->rate, &port->qsize[0]);
+	status = pipe_profile_check(params, port->rate, &s->qsize[0]);
 	if (status != 0) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Pipe profile check failed(%d)\n", __func__, status);
 		return -EINVAL;
 	}
 
-	pp = &port->pipe_profiles[port->n_pipe_profiles];
-	rte_sched_pipe_profile_convert(port, params, pp, port->rate);
+	pp = &s->pipe_profiles[s->n_pipe_profiles];
+	rte_sched_pipe_profile_convert(s, params, pp, port->rate);
 
 	/* Pipe profile not exists */
-	for (i = 0; i < port->n_pipe_profiles; i++)
-		if (memcmp(port->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
+	for (i = 0; i < s->n_pipe_profiles; i++)
+		if (memcmp(s->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
 			RTE_LOG(ERR, SCHED,
 				"%s: Pipe profile doesn't exist\n", __func__);
 			return -EINVAL;
 		}
 
 	/* Pipe profile commit */
-	*pipe_profile_id = port->n_pipe_profiles;
-	port->n_pipe_profiles++;
+	*pipe_profile_id = s->n_pipe_profiles;
+	s->n_pipe_profiles++;
 
-	if (port->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
-		port->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+	if (s->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
+		s->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
 
-	rte_sched_port_log_pipe_profile(port, *pipe_profile_id);
+	rte_sched_port_log_pipe_profile(s, *pipe_profile_id);
 
 	return 0;
 }
lib/librte_sched/rte_sched.h +4 -1
index 48be702bc..56e9fbfa7 100644
@@ -308,6 +308,8 @@ rte_sched_port_free(struct rte_sched_port *port);
  *
  * @param port
  *   Handle to port scheduler instance
+ * @param subport_id
+ *   Subport ID
  * @param params
  *   Pipe profile parameters
  * @param pipe_profile_id
@@ -318,6 +320,7 @@ rte_sched_port_free(struct rte_sched_port *port);
 __rte_experimental
 int
 rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id);
 
@@ -348,7 +351,7 @@ rte_sched_subport_config(struct rte_sched_port *port,
  * @param pipe_id
  *   Pipe ID within subport
  * @param pipe_profile
- *   ID of port-level pre-configured pipe profile
+ *   ID of subport-level pre-configured pipe profile
  * @return
  *   0 upon success, error code otherwise
  */
Signature · 1 line
2.21.0
#6 lore raw

PATCH 6/15 sched: modify pkt enqueue for subport config flexibility

In reply to Jasvinder Singh

Modify scheduler packet enqueue operation of the scheduler to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 278 ++++++++++++++++++++++-------------
 1 file changed, 178 insertions(+), 100 deletions(-)
lib/librte_sched/rte_sched.c +178 -100
index 6bafbd8fe..a5e4c45b4 100644
@@ -1452,10 +1452,10 @@ rte_sched_port_qindex(struct rte_sched_port *port,
 	uint32_t queue)
 {
 	return ((subport & (port->n_subports_per_port - 1)) <<
-			(port->n_pipes_per_subport_log2 + 4)) |
-			((pipe & (port->n_pipes_per_subport - 1)) << 4) |
-			((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
-			(RTE_SCHED_QUEUES_PER_PIPE - 1));
+		(port->n_max_pipes_per_subport_log2 + 4)) |
+		((pipe & (port->subports[subport]->n_pipes_per_subport - 1)) << 4) |
+		((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
+		(RTE_SCHED_QUEUES_PER_PIPE - 1));
 }
 
 void
@@ -1479,8 +1479,9 @@ rte_sched_port_pkt_read_tree_path(struct rte_sched_port *port,
 {
 	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
 
-	*subport = queue_id >> (port->n_pipes_per_subport_log2 + 4);
-	*pipe = (queue_id >> 4) & (port->n_pipes_per_subport - 1);
+	*subport = queue_id >> (port->n_max_pipes_per_subport_log2 + 4);
+	*pipe = (queue_id >> 4) &
+		(port->subports[*subport]->n_pipes_per_subport - 1);
 	*traffic_class = rte_sched_port_pipe_tc(port, queue_id);
 	*queue = rte_sched_port_tc_queue(port, queue_id);
 }
@@ -1524,7 +1525,7 @@ rte_sched_subport_read_stats(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
-	s = port->subport + subport_id;
+	s = port->subports[subport_id];
 
 	/* Copy subport stats and clear */
 	memcpy(stats, &s->stats, sizeof(struct rte_sched_subport_stats));
@@ -1597,43 +1598,50 @@ rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline void
-rte_sched_port_update_subport_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_subport_stats(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc[tc_index] += 1;
-	s->stats.n_bytes_tc[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc[tc_index] += 1;
+	subport->stats.n_bytes_tc[tc_index] += pkt_len;
 }
 
 #ifdef RTE_SCHED_RED
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc_dropped[tc_index] += 1;
-	s->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc_dropped[tc_index] += 1;
+	subport->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
 #ifdef RTE_SCHED_RED
-	s->stats.n_pkts_red_dropped[tc_index] += red;
+	subport->stats.n_pkts_red_dropped[tc_index] += red;
 #endif
 }
 
 static inline void
-rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_queue_stats(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts += 1;
@@ -1642,17 +1650,19 @@ rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex,
 
 #ifdef RTE_SCHED_RED
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts_dropped += 1;
@@ -1667,7 +1677,11 @@ rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
 #ifdef RTE_SCHED_RED
 
 static inline int
-rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint32_t qindex, uint16_t qlen)
+rte_sched_port_red_drop(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt,
+	uint32_t qindex,
+	uint16_t qlen)
 {
 	struct rte_sched_queue_extra *qe;
 	struct rte_red_config *red_cfg;
@@ -1677,12 +1691,12 @@ rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint3
 
 	tc_index = rte_sched_port_pipe_tc(port, qindex);
 	color = rte_sched_port_pkt_read_color(pkt);
-	red_cfg = &port->red_config[tc_index][color];
+	red_cfg = &subport->red_config[tc_index][color];
 
 	if ((red_cfg->min_th | red_cfg->max_th) == 0)
 		return 0;
 
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + qindex;
 	red = &qe->red;
 
 	return rte_red_enqueue(red_cfg, red, qlen, port->time);
@@ -1699,7 +1713,14 @@ rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t q
 
 #else
 
-#define rte_sched_port_red_drop(port, pkt, qindex, qlen)             0
+static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unused,
+	struct rte_sched_subport *subport __rte_unused,
+	struct rte_mbuf *pkt __rte_unused,
+	uint32_t qindex __rte_unused,
+	uint16_t qlen __rte_unused)
+{
+	return 0;
+}
 
 #define rte_sched_port_set_queue_empty_timestamp(port, qindex)
 
@@ -1734,63 +1755,79 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 
 #endif /* RTE_SCHED_DEBUG */
 
+static inline struct rte_sched_subport *
+rte_sched_port_subport(struct rte_sched_port *port,
+	struct rte_mbuf *pkt)
+{
+	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_id = queue_id >> (port->n_max_pipes_per_subport_log2 + 4);
+
+	return port->subports[subport_id];
+}
+
 static inline uint32_t
-rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_port *port,
-				       struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt, uint32_t subport_qmask)
 {
 	struct rte_sched_queue *q;
 #ifdef RTE_SCHED_COLLECT_STATS
 	struct rte_sched_queue_extra *qe;
 #endif
 	uint32_t qindex = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_queue_id = subport_qmask & qindex;
 
-	q = port->queue + qindex;
+	q = subport->queue + subport_queue_id;
 	rte_prefetch0(q);
 #ifdef RTE_SCHED_COLLECT_STATS
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + subport_queue_id;
 	rte_prefetch0(qe);
 #endif
 
-	return qindex;
+	return subport_queue_id;
 }
 
 static inline void
 rte_sched_port_enqueue_qwa_prefetch0(struct rte_sched_port *port,
-				     uint32_t qindex, struct rte_mbuf **qbase)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase)
 {
 	struct rte_sched_queue *q;
 	struct rte_mbuf **q_qw;
 	uint16_t qsize;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	q_qw = qbase + (q->qw & (qsize - 1));
 
 	rte_prefetch0(q_qw);
-	rte_bitmap_prefetch0(port->bmp, qindex);
+	rte_bitmap_prefetch0(subport->bmp, qindex);
 }
 
 static inline int
-rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
-			   struct rte_mbuf **qbase, struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qwa(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase,
+	struct rte_mbuf *pkt)
 {
 	struct rte_sched_queue *q;
 	uint16_t qsize;
 	uint16_t qlen;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	qlen = q->qw - q->qr;
 
 	/* Drop the packet (and update drop stats) when queue is full */
-	if (unlikely(rte_sched_port_red_drop(port, pkt, qindex, qlen) ||
+	if (unlikely(rte_sched_port_red_drop(port, subport, pkt, qindex, qlen) ||
 		     (qlen >= qsize))) {
 		rte_pktmbuf_free(pkt);
 #ifdef RTE_SCHED_COLLECT_STATS
-		rte_sched_port_update_subport_stats_on_drop(port, qindex, pkt,
-							    qlen < qsize);
-		rte_sched_port_update_queue_stats_on_drop(port, qindex, pkt,
-							  qlen < qsize);
+		rte_sched_port_update_subport_stats_on_drop(port, subport,
+			qindex, pkt, qlen < qsize);
+		rte_sched_port_update_queue_stats_on_drop(subport, qindex, pkt,
+			qlen < qsize);
 #endif
 		return 0;
 	}
@@ -1799,13 +1836,13 @@ rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
 	qbase[q->qw & (qsize - 1)] = pkt;
 	q->qw++;
 
-	/* Activate queue in the port bitmap */
-	rte_bitmap_set(port->bmp, qindex);
+	/* Activate queue in the subport bitmap */
+	rte_bitmap_set(subport->bmp, qindex);
 
 	/* Statistics */
 #ifdef RTE_SCHED_COLLECT_STATS
-	rte_sched_port_update_subport_stats(port, qindex, pkt);
-	rte_sched_port_update_queue_stats(port, qindex, pkt);
+	rte_sched_port_update_subport_stats(port, subport, qindex, pkt);
+	rte_sched_port_update_queue_stats(subport, qindex, pkt);
 #endif
 
 	return 1;
@@ -1833,17 +1870,22 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		*pkt30, *pkt31, *pkt_last;
 	struct rte_mbuf **q00_base, **q01_base, **q10_base, **q11_base,
 		**q20_base, **q21_base, **q30_base, **q31_base, **q_last_base;
+	struct rte_sched_subport *subport00, *subport01, *subport10, *subport11,
+		*subport20, *subport21, *subport30, *subport31, *subport_last;
 	uint32_t q00, q01, q10, q11, q20, q21, q30, q31, q_last;
 	uint32_t r00, r01, r10, r11, r20, r21, r30, r31, r_last;
+	uint32_t subport_qmask;
 	uint32_t result, i;
 
 	result = 0;
+	subport_qmask = (1 << (port->n_max_pipes_per_subport_log2 + 4)) - 1;
 
 	/*
 	 * Less then 6 input packets available, which is not enough to
 	 * feed the pipeline
 	 */
 	if (unlikely(n_pkts < 6)) {
+		struct rte_sched_subport *subports[5];
 		struct rte_mbuf **q_base[5];
 		uint32_t q[5];
 
@@ -1851,22 +1893,26 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		for (i = 0; i < n_pkts; i++)
 			rte_prefetch0(pkts[i]);
 
+		/* Prefetch the subport structure for each packet */
+		for (i = 0; i < n_pkts; i++)
+			subports[i] = rte_sched_port_subport(port, pkts[i]);
+
 		/* Prefetch the queue structure for each queue */
 		for (i = 0; i < n_pkts; i++)
-			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(port,
-								      pkts[i]);
+			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(subports[i],
+					pkts[i], subport_qmask);
 
 		/* Prefetch the write pointer location of each queue */
 		for (i = 0; i < n_pkts; i++) {
-			q_base[i] = rte_sched_port_qbase(port, q[i]);
-			rte_sched_port_enqueue_qwa_prefetch0(port, q[i],
-							     q_base[i]);
+			q_base[i] = rte_sched_subport_pipe_qbase(subports[i], q[i]);
+			rte_sched_port_enqueue_qwa_prefetch0(port, subports[i],
+				q[i], q_base[i]);
 		}
 
 		/* Write each packet to its queue */
 		for (i = 0; i < n_pkts; i++)
-			result += rte_sched_port_enqueue_qwa(port, q[i],
-							     q_base[i], pkts[i]);
+			result += rte_sched_port_enqueue_qwa(port, subports[i],
+						q[i], q_base[i], pkts[i]);
 
 		return result;
 	}
@@ -1882,21 +1928,29 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	rte_prefetch0(pkt10);
 	rte_prefetch0(pkt11);
 
-	q20 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt20);
-	q21 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt21);
+	subport20 = rte_sched_port_subport(port, pkt20);
+	subport21 = rte_sched_port_subport(port, pkt21);
+	q20 = rte_sched_port_enqueue_qptrs_prefetch0(subport20,
+			pkt20, subport_qmask);
+	q21 = rte_sched_port_enqueue_qptrs_prefetch0(subport21,
+			pkt21, subport_qmask);
 
 	pkt00 = pkts[4];
 	pkt01 = pkts[5];
 	rte_prefetch0(pkt00);
 	rte_prefetch0(pkt01);
 
-	q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-	q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+	subport10 = rte_sched_port_subport(port, pkt10);
+	subport11 = rte_sched_port_subport(port, pkt11);
+	q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+			pkt10, subport_qmask);
+	q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+			pkt11, subport_qmask);
 
-	q20_base = rte_sched_port_qbase(port, q20);
-	q21_base = rte_sched_port_qbase(port, q21);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+	q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+	q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 	/* Run the pipeline */
 	for (i = 6; i < (n_pkts & (~1)); i += 2) {
@@ -1911,6 +1965,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		q31 = q21;
 		q20 = q10;
 		q21 = q11;
+		subport30 = subport20;
+		subport31 = subport21;
+		subport20 = subport10;
+		subport21 = subport11;
 		q30_base = q20_base;
 		q31_base = q21_base;
 
@@ -1920,19 +1978,25 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		rte_prefetch0(pkt00);
 		rte_prefetch0(pkt01);
 
-		/* Stage 1: Prefetch queue structure storing queue pointers */
-		q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-		q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+		/* Stage 1: Prefetch subport and queue structure storing queue pointers */
+		subport10 = rte_sched_port_subport(port, pkt10);
+		subport11 = rte_sched_port_subport(port, pkt11);
+		q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+				pkt10, subport_qmask);
+		q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+				pkt11, subport_qmask);
 
 		/* Stage 2: Prefetch queue write location */
-		q20_base = rte_sched_port_qbase(port, q20);
-		q21_base = rte_sched_port_qbase(port, q21);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+		q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+		q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 		/* Stage 3: Write packet to queue and activate queue */
-		r30 = rte_sched_port_enqueue_qwa(port, q30, q30_base, pkt30);
-		r31 = rte_sched_port_enqueue_qwa(port, q31, q31_base, pkt31);
+		r30 = rte_sched_port_enqueue_qwa(port, subport30,
+				q30, q30_base, pkt30);
+		r31 = rte_sched_port_enqueue_qwa(port, subport31,
+				q31, q31_base, pkt31);
 		result += r30 + r31;
 	}
 
@@ -1944,38 +2008,52 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	pkt_last = pkts[n_pkts - 1];
 	rte_prefetch0(pkt_last);
 
-	q00 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt00);
-	q01 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt01);
-
-	q10_base = rte_sched_port_qbase(port, q10);
-	q11_base = rte_sched_port_qbase(port, q11);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q10, q10_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q11, q11_base);
-
-	r20 = rte_sched_port_enqueue_qwa(port, q20, q20_base, pkt20);
-	r21 = rte_sched_port_enqueue_qwa(port, q21, q21_base, pkt21);
+	subport00 = rte_sched_port_subport(port, pkt00);
+	subport01 = rte_sched_port_subport(port, pkt01);
+	q00 = rte_sched_port_enqueue_qptrs_prefetch0(subport00,
+			pkt00, subport_qmask);
+	q01 = rte_sched_port_enqueue_qptrs_prefetch0(subport01,
+			pkt01, subport_qmask);
+
+	q10_base = rte_sched_subport_pipe_qbase(subport10, q10);
+	q11_base = rte_sched_subport_pipe_qbase(subport11, q11);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport10, q10, q10_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport11, q11, q11_base);
+
+	r20 = rte_sched_port_enqueue_qwa(port, subport20,
+			q20, q20_base, pkt20);
+	r21 = rte_sched_port_enqueue_qwa(port, subport21,
+			q21, q21_base, pkt21);
 	result += r20 + r21;
 
-	q_last = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt_last);
+	subport_last = rte_sched_port_subport(port, pkt_last);
+	q_last = rte_sched_port_enqueue_qptrs_prefetch0(subport_last,
+				pkt_last, subport_qmask);
 
-	q00_base = rte_sched_port_qbase(port, q00);
-	q01_base = rte_sched_port_qbase(port, q01);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q00, q00_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q01, q01_base);
+	q00_base = rte_sched_subport_pipe_qbase(subport00, q00);
+	q01_base = rte_sched_subport_pipe_qbase(subport01, q01);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport00, q00, q00_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport01, q01, q01_base);
 
-	r10 = rte_sched_port_enqueue_qwa(port, q10, q10_base, pkt10);
-	r11 = rte_sched_port_enqueue_qwa(port, q11, q11_base, pkt11);
+	r10 = rte_sched_port_enqueue_qwa(port, subport10, q10,
+			q10_base, pkt10);
+	r11 = rte_sched_port_enqueue_qwa(port, subport11, q11,
+			q11_base, pkt11);
 	result += r10 + r11;
 
-	q_last_base = rte_sched_port_qbase(port, q_last);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q_last, q_last_base);
+	q_last_base = rte_sched_subport_pipe_qbase(subport_last, q_last);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport_last,
+		q_last, q_last_base);
 
-	r00 = rte_sched_port_enqueue_qwa(port, q00, q00_base, pkt00);
-	r01 = rte_sched_port_enqueue_qwa(port, q01, q01_base, pkt01);
+	r00 = rte_sched_port_enqueue_qwa(port, subport00, q00,
+			q00_base, pkt00);
+	r01 = rte_sched_port_enqueue_qwa(port, subport01, q01,
+			q01_base, pkt01);
 	result += r00 + r01;
 
 	if (n_pkts & 1) {
-		r_last = rte_sched_port_enqueue_qwa(port, q_last, q_last_base, pkt_last);
+		r_last = rte_sched_port_enqueue_qwa(port, subport_last,
+					q_last,	q_last_base, pkt_last);
 		result += r_last;
 	}
 
Signature · 1 line
2.21.0
#7 lore raw

PATCH 7/15 sched: update memory compute to support flexiblity

In reply to Jasvinder Singh

Update memory footprint compute function for allowing subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 102 +++++++++++------------------------
 lib/librte_sched/rte_sched.h |   8 +--
 2 files changed, 36 insertions(+), 74 deletions(-)
lib/librte_sched/rte_sched.c +31 -71
index a5e4c45b4..f8d13c66d 100644
@@ -495,70 +495,6 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 	return 0;
 }
 
-static uint32_t
-rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sched_port_array array)
-{
-	uint32_t n_subports_per_port = params->n_subports_per_port;
-	uint32_t n_pipes_per_subport = params->n_pipes_per_subport;
-	uint32_t n_pipes_per_port = n_pipes_per_subport * n_subports_per_port;
-	uint32_t n_queues_per_port = RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport * n_subports_per_port;
-
-	uint32_t size_subport = n_subports_per_port * sizeof(struct rte_sched_subport);
-	uint32_t size_pipe = n_pipes_per_port * sizeof(struct rte_sched_pipe);
-	uint32_t size_queue = n_queues_per_port * sizeof(struct rte_sched_queue);
-	uint32_t size_queue_extra
-		= n_queues_per_port * sizeof(struct rte_sched_queue_extra);
-	uint32_t size_pipe_profiles
-		= params->n_max_pipe_profiles * sizeof(struct rte_sched_pipe_profile);
-	uint32_t size_bmp_array = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	uint32_t size_per_pipe_queue_array, size_queue_array;
-
-	uint32_t base, i;
-
-	size_per_pipe_queue_array = 0;
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
-			size_per_pipe_queue_array +=
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-		else
-			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-	}
-	size_queue_array = n_pipes_per_port * size_per_pipe_queue_array;
-
-	base = 0;
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_SUBPORT)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_subport);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
-
-	return base;
-}
-
 static uint32_t
 rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
 	uint32_t n_max_pipe_profiles,
@@ -880,22 +816,46 @@ rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 }
 
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params)
 {
-	uint32_t size0, size1;
+	uint32_t size0 = 0, size1 = 0, i;
 	int status;
 
-	status = rte_sched_port_check_params(params);
+	status = rte_sched_port_check_params(port_params);
 	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler port params check failed (%d)\n",
+			__func__, status);
 
 		return 0;
 	}
 
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		status = rte_sched_subport_check_params(sp,
+				port_params->n_max_pipes_per_subport,
+				port_params->n_max_pipe_profiles,
+				port_params->rate);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Port scheduler subport params check failed (%d)\n",
+				__func__, status);
+
+			return 0;
+		}
+	}
+
 	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params,
-			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		size1 += rte_sched_subport_get_array_base(sp,
+					port_params->n_max_pipe_profiles,
+					e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
+	}
 
 	return size0 + size1;
 }
lib/librte_sched/rte_sched.h +5 -3
index 56e9fbfa7..ea2b07448 100644
@@ -364,14 +364,16 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 /**
  * Hierarchical scheduler memory footprint size per port
  *
- * @param params
+ * @param port_params
  *   Port scheduler configuration parameter structure
+ * @param subport_params
+ *   Array of subport parameter structures
  * @return
  *   Memory footprint size in bytes upon success, 0 otherwise
  */
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params);
-
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params);
 /*
  * Statistics
  *
Signature · 1 line
2.21.0
#8 lore raw

PATCH 8/15 sched: update grinder functions for config flexibility

In reply to Jasvinder Singh

Modify packet grinder functions of the schedule to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 223 +++++++++++++++++------------------
 1 file changed, 106 insertions(+), 117 deletions(-)
lib/librte_sched/rte_sched.c +106 -117
index f8d13c66d..0451e10ea 100644
@@ -324,24 +324,6 @@ rte_sched_port_queues_per_port(struct rte_sched_port *port)
 	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
 }
 
-static inline struct rte_mbuf **
-rte_sched_port_qbase(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t pindex = qindex >> 4;
-	uint32_t qpos = qindex & 0xF;
-
-	return (port->queue_array + pindex *
-		port->qsize_sum + port->qsize_add[qpos]);
-}
-
-static inline uint16_t
-rte_sched_port_qsize(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
-
-	return port->qsize[tc];
-}
-
 static inline uint16_t
 rte_sched_port_pipe_queue(struct rte_sched_port *port, uint32_t traffic_class)
 {
@@ -1546,9 +1528,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 #ifdef RTE_SCHED_DEBUG
 
 static inline int
-rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_queue_is_empty(struct rte_sched_subport *subport,
+	uint32_t qindex)
 {
-	struct rte_sched_queue *queue = port->queue + qindex;
+	struct rte_sched_queue *queue = subport->queue + qindex;
 
 	return queue->qr == queue->qw;
 }
@@ -1663,9 +1646,10 @@ rte_sched_port_red_drop(struct rte_sched_port *port,
 }
 
 static inline void
-rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t qindex)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	struct rte_red *red = &qe->red;
 
 	rte_red_mark_queue_empty(red, port->time);
@@ -1682,14 +1666,14 @@ static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unus
 	return 0;
 }
 
-#define rte_sched_port_set_queue_empty_timestamp(port, qindex)
+#define rte_sched_port_set_queue_empty_timestamp(port, subport, qindex)
 
 #endif /* RTE_SCHED_RED */
 
 #ifdef RTE_SCHED_DEBUG
 
 static inline void
-debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
+debug_check_queue_slab(struct rte_sched_subport *subport, uint32_t bmp_pos,
 		       uint64_t bmp_slab)
 {
 	uint64_t mask;
@@ -1701,7 +1685,7 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 	panic = 0;
 	for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 		if (mask & bmp_slab) {
-			if (rte_sched_port_queue_is_empty(port, bmp_pos + i)) {
+			if (rte_sched_port_queue_is_empty(subport, bmp_pos + i)) {
 				printf("Queue %u (slab offset %u) is empty\n", bmp_pos + i, i);
 				panic = 1;
 			}
@@ -2023,10 +2007,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2064,10 +2048,9 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline uint32_t
-grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_tc_ov_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
 	uint32_t tc_ov_consumption[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t tc_consumption = 0, tc_ov_consumption_max;
 	uint32_t tc_ov_wm = subport->tc_ov_wm;
@@ -2107,10 +2090,10 @@ grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2130,7 +2113,7 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 
 	/* Subport TCs */
 	if (unlikely(port->time >= subport->tc_time)) {
-		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, pos);
+		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, subport);
 
 		for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 			subport->tc_credits[i] = subport->tc_credits_per_period[i];
@@ -2160,10 +2143,10 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2195,10 +2178,10 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2243,15 +2226,16 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline int
-grinder_schedule(struct rte_sched_port *port, uint32_t pos)
+grinder_schedule(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_queue *queue = grinder->queue[grinder->qpos];
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t pkt_len = pkt->pkt_len + port->frame_overhead;
 	uint32_t be_tc_active;
 
-	if (!grinder_credits_check(port, pos))
+	if (!grinder_credits_check(port, subport, pos))
 		return 0;
 
 	/* Advance port time */
@@ -2268,15 +2252,15 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 	if (queue->qr == queue->qw) {
 		uint32_t qindex = grinder->qindex[grinder->qpos];
 
-		rte_bitmap_clear(port->bmp, qindex);
+		rte_bitmap_clear(subport->bmp, qindex);
 		grinder->qmask &= ~(1 << grinder->qpos);
 		if (be_tc_active)
 			grinder->wrr_mask[grinder->qpos] = 0;
-		rte_sched_port_set_queue_empty_timestamp(port, qindex);
+		rte_sched_port_set_queue_empty_timestamp(port, subport, qindex);
 	}
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	grinder->productive = 1;
 
 	return 1;
@@ -2285,13 +2269,13 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 #ifdef SCHED_VECTOR_SSE4
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	__m128i index = _mm_set1_epi32(base_pipe);
-	__m128i pipes = _mm_load_si128((__m128i *)port->grinder_base_bmp_pos);
+	__m128i pipes = _mm_load_si128((__m128i *)subport->grinder_base_bmp_pos);
 	__m128i res = _mm_cmpeq_epi32(pipes, index);
 
-	pipes = _mm_load_si128((__m128i *)(port->grinder_base_bmp_pos + 4));
+	pipes = _mm_load_si128((__m128i *)(subport->grinder_base_bmp_pos + 4));
 	pipes = _mm_cmpeq_epi32(pipes, index);
 	res = _mm_or_si128(res, pipes);
 
@@ -2304,10 +2288,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #elif defined(SCHED_VECTOR_NEON)
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32x4_t index, pipes;
-	uint32_t *pos = (uint32_t *)port->grinder_base_bmp_pos;
+	uint32_t *pos = (uint32_t *)subport->grinder_base_bmp_pos;
 
 	index = vmovq_n_u32(base_pipe);
 	pipes = vld1q_u32(pos);
@@ -2324,12 +2308,12 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #else
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32_t i;
 
 	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++) {
-		if (port->grinder_base_bmp_pos[i] == base_pipe)
+		if (subport->grinder_base_bmp_pos[i] == base_pipe)
 			return 1;
 	}
 
@@ -2339,9 +2323,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #endif /* RTE_SCHED_OPTIMIZATIONS */
 
 static inline void
-grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
+grinder_pcache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t w[4];
 
 	grinder->pcache_w = 0;
@@ -2370,9 +2355,10 @@ grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_
 }
 
 static inline void
-grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qindex, uint16_t qmask)
+grinder_tccache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t qindex, uint16_t qmask)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint8_t b, i;
 
 	grinder->tccache_w = 0;
@@ -2393,9 +2379,10 @@ grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qin
 }
 
 static inline int
-grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
+grinder_next_tc(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_mbuf **qbase;
 	uint32_t qindex;
 	uint16_t qsize;
@@ -2404,15 +2391,15 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 0;
 
 	qindex = grinder->tccache_qindex[grinder->tccache_r];
-	qbase = rte_sched_port_qbase(port, qindex);
-	qsize = rte_sched_port_qsize(port, qindex);
+	qbase = rte_sched_subport_pipe_qbase(subport, qindex);
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 
 	grinder->tc_index = rte_sched_port_pipe_tc(port, qindex);
 	grinder->qmask = grinder->tccache_qmask[grinder->tccache_r];
 	grinder->qsize = qsize;
 
 	if (grinder->tc_index < RTE_SCHED_TRAFFIC_CLASS_BE) {
-		grinder->queue[0] = port->queue + qindex;
+		grinder->queue[0] = subport->queue + qindex;
 		grinder->qbase[0] = qbase;
 		grinder->qindex[0] = qindex;
 		grinder->tccache_r++;
@@ -2420,10 +2407,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 1;
 	}
 
-	grinder->queue[0] = port->queue + qindex;
-	grinder->queue[1] = port->queue + qindex + 1;
-	grinder->queue[2] = port->queue + qindex + 2;
-	grinder->queue[3] = port->queue + qindex + 3;
+	grinder->queue[0] = subport->queue + qindex;
+	grinder->queue[1] = subport->queue + qindex + 1;
+	grinder->queue[2] = subport->queue + qindex + 2;
+	grinder->queue[3] = subport->queue + qindex + 3;
 
 	grinder->qbase[0] = qbase;
 	grinder->qbase[1] = qbase + qsize;
@@ -2440,9 +2427,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline int
-grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_next_pipe(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t pipe_qindex;
 	uint16_t pipe_qmask;
 
@@ -2455,22 +2443,22 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 		uint32_t bmp_pos = 0;
 
 		/* Get another non-empty pipe group */
-		if (unlikely(rte_bitmap_scan(port->bmp, &bmp_pos, &bmp_slab) <= 0))
+		if (unlikely(rte_bitmap_scan(subport->bmp, &bmp_pos, &bmp_slab) <= 0))
 			return 0;
 
 #ifdef RTE_SCHED_DEBUG
-		debug_check_queue_slab(port, bmp_pos, bmp_slab);
+		debug_check_queue_slab(subport, bmp_pos, bmp_slab);
 #endif
 
 		/* Return if pipe group already in one of the other grinders */
-		port->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
-		if (unlikely(grinder_pipe_exists(port, bmp_pos)))
+		subport->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
+		if (unlikely(grinder_pipe_exists(subport, bmp_pos)))
 			return 0;
 
-		port->grinder_base_bmp_pos[pos] = bmp_pos;
+		subport->grinder_base_bmp_pos[pos] = bmp_pos;
 
 		/* Install new pipe group into grinder's pipe cache */
-		grinder_pcache_populate(port, pos, bmp_pos, bmp_slab);
+		grinder_pcache_populate(subport, pos, bmp_pos, bmp_slab);
 
 		pipe_qmask = grinder->pcache_qmask[0];
 		pipe_qindex = grinder->pcache_qindex[0];
@@ -2479,18 +2467,18 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 	/* Install new pipe in the grinder */
 	grinder->pindex = pipe_qindex >> 4;
-	grinder->subport = port->subport + (grinder->pindex / port->n_pipes_per_subport);
-	grinder->pipe = port->pipe + grinder->pindex;
+	grinder->subport = subport;
+	grinder->pipe = subport->pipe + grinder->pindex;
 	grinder->pipe_params = NULL; /* to be set after the pipe structure is prefetched */
 	grinder->productive = 0;
 
-	grinder_tccache_populate(port, pos, pipe_qindex, pipe_qmask);
-	grinder_next_tc(port, pos);
+	grinder_tccache_populate(subport, pos, pipe_qindex, pipe_qmask);
+	grinder_next_tc(port, subport, pos);
 
 	/* Check for pipe exhaustion */
-	if (grinder->pindex == port->pipe_loop) {
-		port->pipe_exhaustion = 1;
-		port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	if (grinder->pindex == subport->pipe_loop) {
+		subport->pipe_exhaustion = 1;
+		subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	}
 
 	return 1;
@@ -2498,9 +2486,9 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline void
-grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_load(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *pipe_params = grinder->pipe_params;
 	uint32_t qmask = grinder->qmask;
@@ -2526,9 +2514,9 @@ grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_store(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 
 	pipe->wrr_tokens[0] =
@@ -2546,9 +2534,9 @@ grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t wrr_tokens_min;
 
 	grinder->wrr_tokens[0] |= ~grinder->wrr_mask[0];
@@ -2566,21 +2554,21 @@ grinder_wrr(struct rte_sched_port *port, uint32_t pos)
 }
 
 
-#define grinder_evict(port, pos)
+#define grinder_evict(subport, pos)
 
 static inline void
-grinder_prefetch_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_pipe(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	rte_prefetch0(grinder->pipe);
 	rte_prefetch0(grinder->queue[0]);
 }
 
 static inline void
-grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_tc_queue_arrays(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t qsize, qr[RTE_SCHED_MAX_QUEUES_PER_TC];
 
 	qsize = grinder->qsize;
@@ -2601,17 +2589,17 @@ grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
 	rte_prefetch0(grinder->qbase[0] + qr[0]);
 	rte_prefetch0(grinder->qbase[1] + qr[1]);
 
-	grinder_wrr_load(port, pos);
-	grinder_wrr(port, pos);
+	grinder_wrr_load(subport, pos);
+	grinder_wrr(subport, pos);
 
 	rte_prefetch0(grinder->qbase[2] + qr[2]);
 	rte_prefetch0(grinder->qbase[3] + qr[3]);
 }
 
 static inline void
-grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t qpos = grinder->qpos;
 	struct rte_mbuf **qbase = grinder->qbase[qpos];
 	uint16_t qsize = grinder->qsize;
@@ -2630,14 +2618,15 @@ grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
 static inline uint32_t
 grinder_handle(struct rte_sched_port *port, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_subport *subport = port->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
 	case e_GRINDER_PREFETCH_PIPE:
 	{
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
-			port->busy_grinders++;
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
+			subport->busy_grinders++;
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return 0;
@@ -2650,9 +2639,9 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		struct rte_sched_pipe *pipe = grinder->pipe;
 
-		grinder->pipe_params = port->pipe_profiles + pipe->profile;
-		grinder_prefetch_tc_queue_arrays(port, pos);
-		grinder_credits_update(port, pos);
+		grinder->pipe_params = subport->pipe_profiles + pipe->profile;
+		grinder_prefetch_tc_queue_arrays(subport, pos);
+		grinder_credits_update(port, subport, pos);
 
 		grinder->state = e_GRINDER_PREFETCH_MBUF;
 		return 0;
@@ -2660,7 +2649,7 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 
 	case e_GRINDER_PREFETCH_MBUF:
 	{
-		grinder_prefetch_mbuf(port, pos);
+		grinder_prefetch_mbuf(subport, pos);
 
 		grinder->state = e_GRINDER_READ_MBUF;
 		return 0;
@@ -2670,47 +2659,47 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		uint32_t wrr_active, result = 0;
 
-		result = grinder_schedule(port, pos);
+		result = grinder_schedule(port, subport, pos);
 
 		wrr_active = (grinder->tc_index == RTE_SCHED_TRAFFIC_CLASS_BE);
 
 		/* Look for next packet within the same TC */
 		if (result && grinder->qmask) {
 			if (wrr_active)
-				grinder_wrr(port, pos);
+				grinder_wrr(subport, pos);
 
-			grinder_prefetch_mbuf(port, pos);
+			grinder_prefetch_mbuf(subport, pos);
 
 			return 1;
 		}
 
 		if (wrr_active)
-			grinder_wrr_store(port, pos);
+			grinder_wrr_store(subport, pos);
 
 		/* Look for another active TC within same pipe */
-		if (grinder_next_tc(port, pos)) {
-			grinder_prefetch_tc_queue_arrays(port, pos);
+		if (grinder_next_tc(port, subport, pos)) {
+			grinder_prefetch_tc_queue_arrays(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_MBUF;
 			return result;
 		}
 
 		if (grinder->productive == 0 &&
-		    port->pipe_loop == RTE_SCHED_PIPE_INVALID)
-			port->pipe_loop = grinder->pindex;
+		    subport->pipe_loop == RTE_SCHED_PIPE_INVALID)
+			subport->pipe_loop = grinder->pindex;
 
-		grinder_evict(port, pos);
+		grinder_evict(subport, pos);
 
 		/* Look for another active pipe */
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return result;
 		}
 
 		/* No active pipe found */
-		port->busy_grinders--;
+		subport->busy_grinders--;
 
 		grinder->state = e_GRINDER_PREFETCH_PIPE;
 		return result;
Signature · 1 line
2.21.0
#9 lore raw

PATCH 9/15 sched: update pkt dequeue for subport config flexibility

In reply to Jasvinder Singh

Modify scheduler packet dequeue operation to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 51 ++++++++++++++++++++++++++++--------
 1 file changed, 40 insertions(+), 11 deletions(-)
lib/librte_sched/rte_sched.c +40 -11
index 0451e10ea..e0ef86f40 100644
@@ -245,6 +245,7 @@ struct rte_sched_port {
 	uint32_t busy_grinders;
 	struct rte_mbuf **pkts_out;
 	uint32_t n_pkts_out;
+	uint32_t subport_id;
 
 	/* Queue base calculation */
 	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
@@ -911,6 +912,7 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	/* Grinders */
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
+	port->subport_id = 0;
 
 	return port;
 }
@@ -2616,9 +2618,9 @@ grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 }
 
 static inline uint32_t
-grinder_handle(struct rte_sched_port *port, uint32_t pos)
+grinder_handle(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_subport *subport = port->subport;
 	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
@@ -2717,6 +2719,7 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 	uint64_t cycles = rte_get_tsc_cycles();
 	uint64_t cycles_diff = cycles - port->time_cpu_cycles;
 	uint64_t bytes_diff;
+	uint32_t i;
 
 	/* Compute elapsed time in bytes */
 	bytes_diff = rte_reciprocal_divide(cycles_diff << RTE_SCHED_TIME_SHIFT,
@@ -2729,20 +2732,21 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 		port->time = port->time_cpu_bytes;
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	for (i = 0; i < port->n_subports_per_port; i++)
+		port->subports[i]->pipe_loop = RTE_SCHED_PIPE_INVALID;
 }
 
 static inline int
-rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
+rte_sched_port_exceptions(struct rte_sched_subport *subport, int second_pass)
 {
 	int exceptions;
 
 	/* Check if any exception flag is set */
-	exceptions = (second_pass && port->busy_grinders == 0) ||
-		(port->pipe_exhaustion == 1);
+	exceptions = (second_pass && subport->busy_grinders == 0) ||
+		(subport->pipe_exhaustion == 1);
 
 	/* Clear exception flags */
-	port->pipe_exhaustion = 0;
+	subport->pipe_exhaustion = 0;
 
 	return exceptions;
 }
@@ -2750,7 +2754,9 @@ rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
 int
 rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts)
 {
-	uint32_t i, count;
+	struct rte_sched_subport *subport;
+	uint32_t subport_id = port->subport_id;
+	uint32_t i, n_subports = 0, count;
 
 	port->pkts_out = pkts;
 	port->n_pkts_out = 0;
@@ -2759,9 +2765,32 @@ rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint
 
 	/* Take each queue in the grinder one step further */
 	for (i = 0, count = 0; ; i++)  {
-		count += grinder_handle(port, i & (RTE_SCHED_PORT_N_GRINDERS - 1));
-		if ((count == n_pkts) ||
-		    rte_sched_port_exceptions(port, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+		subport = port->subports[subport_id];
+
+		count += grinder_handle(port, subport,
+				i & (RTE_SCHED_PORT_N_GRINDERS - 1));
+
+		if (count == n_pkts) {
+			subport_id++;
+
+			if (subport_id == port->n_subports_per_port)
+				subport_id = 0;
+
+			port->subport_id = subport_id;
+			break;
+		}
+
+		if (rte_sched_port_exceptions(subport, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+			i = 0;
+			subport_id++;
+			n_subports++;
+		}
+
+		if (subport_id == port->n_subports_per_port)
+			subport_id = 0;
+
+		if (n_subports == port->n_subports_per_port) {
+			port->subport_id = subport_id;
 			break;
 		}
 	}
Signature · 1 line
2.21.0
#10 lore raw

PATCH 10/15 sched: update queue stats read for config flexibility

In reply to Jasvinder Singh

Modify pipe queue stats read function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 28 +++++++++++++++-------------
 1 file changed, 15 insertions(+), 13 deletions(-)
lib/librte_sched/rte_sched.c +15 -13
index e0ef86f40..8a7727286 100644
@@ -284,16 +284,6 @@ enum rte_sched_subport_array {
 	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
 };
 
-#ifdef RTE_SCHED_COLLECT_STATS
-
-static inline uint32_t
-rte_sched_port_queues_per_subport(struct rte_sched_port *port)
-{
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport;
-}
-
-#endif
-
 static inline uint32_t
 rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
 {
@@ -322,7 +312,12 @@ struct rte_sched_subport *subport, uint32_t qindex)
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
+	uint32_t n_queues = 0, i;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		n_queues += rte_sched_subport_pipe_queues(port->subports[i]);
+
+	return n_queues;
 }
 
 static inline uint16_t
@@ -1487,8 +1482,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 	struct rte_sched_queue_stats *stats,
 	uint16_t *qlen)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_queue *q;
 	struct rte_sched_queue_extra *qe;
+	uint32_t subport_id, subport_qmask, subport_qindex;
 
 	/* Check user parameters */
 	if (port == NULL) {
@@ -1514,8 +1511,13 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 			"%s: Incorrect value for parameter qlen\n", __func__);
 		return -EINVAL;
 	}
-	q = port->queue + queue_id;
-	qe = port->queue_extra + queue_id;
+	subport_qmask = port->n_max_pipes_per_subport_log2 + 4;
+	subport_id = (queue_id >> subport_qmask) & (port->n_subports_per_port - 1);
+
+	s = port->subports[subport_id];
+	subport_qindex = ((1 << subport_qmask) - 1) & queue_id;
+	q = s->queue + subport_qindex;
+	qe = s->queue_extra + subport_qindex;
 
 	/* Copy queue stats and clear */
 	memcpy(stats, &qe->stats, sizeof(struct rte_sched_queue_stats));
Signature · 1 line
2.21.0
#11 lore raw

PATCH 11/15 test/sched: modify tests for subport config flexibility

In reply to Jasvinder Singh

Modify tests function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 app/test/test_sched.c | 35 ++++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 17 deletions(-)
app/test/test_sched.c +18 -17
index afe0b0765..6a2896234 100644
@@ -22,18 +22,6 @@
 #define TC              2
 #define QUEUE           0
 
-static struct rte_sched_subport_params subport_param[] = {
-	{
-		.tb_rate = 1250000000,
-		.tb_size = 1000000,
-
-		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000},
-		.tc_period = 10,
-	},
-};
-
 static struct rte_sched_pipe_params pipe_profile[] = {
 	{ /* Profile #0 */
 		.tb_rate = 305175,
@@ -48,16 +36,29 @@ static struct rte_sched_pipe_params pipe_profile[] = {
 	},
 };
 
+static struct rte_sched_subport_params subport_param[] = {
+	{
+		.tb_rate = 1250000000,
+		.tb_size = 1000000,
+
+		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000},
+		.tc_period = 10,
+		.n_pipes_per_subport = 1024,
+		.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
+		.pipe_profiles = pipe_profile,
+		.n_pipe_profiles = 1,
+	},
+};
+
 static struct rte_sched_port_params port_param = {
 	.socket = 0, /* computed */
 	.rate = 0, /* computed */
 	.mtu = 1522,
 	.frame_overhead = RTE_SCHED_FRAME_OVERHEAD_DEFAULT,
 	.n_subports_per_port = 1,
-	.n_pipes_per_subport = 1024,
-	.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
-	.pipe_profiles = pipe_profile,
-	.n_pipe_profiles = 1,
+	.n_max_pipes_per_subport = 1024,
 	.n_max_pipe_profiles = 1,
 };
 
@@ -140,7 +141,7 @@ test_sched(void)
 	err = rte_sched_subport_config(port, SUBPORT, subport_param);
 	TEST_ASSERT_SUCCESS(err, "Error config sched, err=%d\n", err);
 
-	for (pipe = 0; pipe < port_param.n_pipes_per_subport; pipe ++) {
+	for (pipe = 0; pipe < subport_param[0].n_pipes_per_subport; pipe++) {
 		err = rte_sched_pipe_config(port, SUBPORT, pipe, 0);
 		TEST_ASSERT_SUCCESS(err, "Error config sched pipe %u, err=%d\n", pipe, err);
 	}
Signature · 1 line
2.21.0
#12 lore raw

PATCH 12/15 net/softnic: add subport config flexibility to TM function

In reply to Jasvinder Singh

Modify softnic traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 drivers/net/softnic/rte_eth_softnic_tm.c | 51 ++++++++++++------------
 1 file changed, 26 insertions(+), 25 deletions(-)
drivers/net/softnic/rte_eth_softnic_tm.c +26 -25
index 61c3adc82..8fb728d7d 100644
@@ -85,7 +85,8 @@ softnic_tmgr_port_create(struct pmd_internals *p,
 	/* Subport */
 	n_subports = t->port_params.n_subports_per_port;
 	for (subport_id = 0; subport_id < n_subports; subport_id++) {
-		uint32_t n_pipes_per_subport = t->port_params.n_pipes_per_subport;
+		uint32_t n_pipes_per_subport =
+			t->subport_params[subport_id].n_pipes_per_subport;
 		uint32_t pipe_id;
 		int status;
 
@@ -2211,10 +2212,11 @@ tm_tc_wred_profile_get(struct rte_eth_dev *dev, uint32_t tc_id)
 #ifdef RTE_SCHED_RED
 
 static void
-wred_profiles_set(struct rte_eth_dev *dev)
+wred_profiles_set(struct rte_eth_dev *dev, uint32_t subport_id)
 {
 	struct pmd_internals *p = dev->data->dev_private;
-	struct rte_sched_port_params *pp = &p->soft.tm.params.port_params;
+	struct rte_sched_subport_params *pp =
+		&p->soft.tm.params.subport_params[subport_id];
 
 	uint32_t tc_id;
 	enum rte_color color;
@@ -2234,7 +2236,7 @@ wred_profiles_set(struct rte_eth_dev *dev)
 
 #else
 
-#define wred_profiles_set(dev)
+#define wred_profiles_set(dev, subport_id)
 
 #endif
 
@@ -2526,29 +2528,10 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 		.frame_overhead =
 			root->shaper_profile->params.pkt_length_adjust,
 		.n_subports_per_port = root->n_children,
-		.n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
-			h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
-		.qsize = {p->params.tm.qsize[0],
-			p->params.tm.qsize[1],
-			p->params.tm.qsize[2],
-			p->params.tm.qsize[3],
-			p->params.tm.qsize[4],
-			p->params.tm.qsize[5],
-			p->params.tm.qsize[6],
-			p->params.tm.qsize[7],
-			p->params.tm.qsize[8],
-			p->params.tm.qsize[9],
-			p->params.tm.qsize[10],
-			p->params.tm.qsize[11],
-			p->params.tm.qsize[12],
-		},
-		.pipe_profiles = t->pipe_profiles,
-		.n_pipe_profiles = t->n_pipe_profiles,
+		.n_max_pipes_per_subport = TM_MAX_PIPES_PER_SUBPORT,
 		.n_max_pipe_profiles = TM_MAX_PIPE_PROFILE,
 	};
 
-	wred_profiles_set(dev);
-
 	subport_id = 0;
 	TAILQ_FOREACH(n, nl, node) {
 		uint64_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
@@ -2588,8 +2571,26 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 					tc_rate[12],
 				},
 				.tc_period = SUBPORT_TC_PERIOD,
+				.n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+					h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
+				.qsize = {p->params.tm.qsize[0],
+					p->params.tm.qsize[1],
+					p->params.tm.qsize[2],
+					p->params.tm.qsize[3],
+					p->params.tm.qsize[4],
+					p->params.tm.qsize[5],
+					p->params.tm.qsize[6],
+					p->params.tm.qsize[7],
+					p->params.tm.qsize[8],
+					p->params.tm.qsize[9],
+					p->params.tm.qsize[10],
+					p->params.tm.qsize[11],
+					p->params.tm.qsize[12],
+				},
+				.pipe_profiles = t->pipe_profiles,
+				.n_pipe_profiles = t->n_pipe_profiles,
 		};
-
+		wred_profiles_set(dev, subport_id);
 		subport_id++;
 	}
 }
Signature · 1 line
2.21.0
#13 lore raw

PATCH 13/15 ip_pipeline: add subport config flexibility to TM function

In reply to Jasvinder Singh

Modify ip pipeline traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 examples/ip_pipeline/cli.c  | 71 ++++++++++++++++++-------------------
 examples/ip_pipeline/tmgr.c | 23 ++++++------
 examples/ip_pipeline/tmgr.h |  7 ++--
 3 files changed, 49 insertions(+), 52 deletions(-)
examples/ip_pipeline/cli.c +35 -36
index c6cf4204e..1e8dc95b4 100644
@@ -380,7 +380,12 @@ static const char cmd_tmgr_subport_profile_help[] =
 "   <tc0_rate> <tc1_rate> <tc2_rate> <tc3_rate> <tc4_rate>"
 "        <tc5_rate> <tc6_rate> <tc7_rate> <tc8_rate>"
 "        <tc9_rate> <tc10_rate> <tc11_rate> <tc12_rate>\n"
-"   <tc_period>\n";
+"   <tc_period>\n"
+"   pps <n_pipes_per_subport>\n"
+"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
+"       <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
+"       <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
+"       <qsize_tc11> <qsize_tc12>";
 
 static void
 cmd_tmgr_subport_profile(char **tokens,
@@ -391,7 +396,7 @@ cmd_tmgr_subport_profile(char **tokens,
 	struct rte_sched_subport_params p;
 	int status, i;
 
-	if (n_tokens != 19) {
+	if (n_tokens != 35) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -417,6 +422,27 @@ cmd_tmgr_subport_profile(char **tokens,
 		return;
 	}
 
+	if (strcmp(tokens[19], "pps") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
+		return;
+	}
+
+	if (parser_read_uint32(&p.n_pipes_per_subport, tokens[20]) != 0) {
+		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
+		return;
+	}
+
+	if (strcmp(tokens[21], "qsize") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
+		return;
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		if (parser_read_uint16(&p.qsize[i], tokens[22 + i]) != 0) {
+			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
+			return;
+		}
+
 	status = tmgr_subport_profile_add(&p);
 	if (status != 0) {
 		snprintf(out, out_size, MSG_CMD_FAIL, tokens[0]);
@@ -491,11 +517,6 @@ static const char cmd_tmgr_help[] =
 "tmgr <tmgr_name>\n"
 "   rate <rate>\n"
 "   spp <n_subports_per_port>\n"
-"   pps <n_pipes_per_subport>\n"
-"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
-"   <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
-"   <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
-"   <qsize_tc11> <qsize_tc12>\n"
 "   fo <frame_overhead>\n"
 "   mtu <mtu>\n"
 "   cpu <cpu_id>\n";
@@ -509,9 +530,8 @@ cmd_tmgr(char **tokens,
 	struct tmgr_port_params p;
 	char *name;
 	struct tmgr_port *tmgr_port;
-	int i;
 
-	if (n_tokens != 28) {
+	if (n_tokens != 12) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -538,53 +558,32 @@ cmd_tmgr(char **tokens,
 		return;
 	}
 
-	if (strcmp(tokens[6], "pps") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
-		return;
-	}
-
-	if (parser_read_uint32(&p.n_pipes_per_subport, tokens[7]) != 0) {
-		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
-		return;
-	}
-
-	if (strcmp(tokens[8], "qsize") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
-		return;
-	}
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (parser_read_uint16(&p.qsize[i], tokens[9 + i]) != 0) {
-			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
-			return;
-		}
-
-	if (strcmp(tokens[22], "fo") != 0) {
+	if (strcmp(tokens[6], "fo") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "fo");
 		return;
 	}
 
-	if (parser_read_uint32(&p.frame_overhead, tokens[23]) != 0) {
+	if (parser_read_uint32(&p.frame_overhead, tokens[7]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "frame_overhead");
 		return;
 	}
 
-	if (strcmp(tokens[24], "mtu") != 0) {
+	if (strcmp(tokens[8], "mtu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "mtu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.mtu, tokens[25]) != 0) {
+	if (parser_read_uint32(&p.mtu, tokens[9]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "mtu");
 		return;
 	}
 
-	if (strcmp(tokens[26], "cpu") != 0) {
+	if (strcmp(tokens[10], "cpu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "cpu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.cpu_id, tokens[27]) != 0) {
+	if (parser_read_uint32(&p.cpu_id, tokens[11]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "cpu_id");
 		return;
 	}
examples/ip_pipeline/tmgr.c +10 -13
index 40cbf1d0a..cafb9cdcc 100644
@@ -47,7 +47,8 @@ int
 tmgr_subport_profile_add(struct rte_sched_subport_params *p)
 {
 	/* Check input params */
-	if (p == NULL)
+	if (p == NULL ||
+		p->n_pipes_per_subport == 0)
 		return -1;
 
 	/* Save profile */
@@ -90,7 +91,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 		tmgr_port_find(name) ||
 		(params == NULL) ||
 		(params->n_subports_per_port == 0) ||
-		(params->n_pipes_per_subport == 0) ||
 		(params->cpu_id >= RTE_MAX_NUMA_NODES) ||
 		(n_subport_profiles == 0) ||
 		(n_pipe_profiles == 0))
@@ -103,18 +103,16 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	p.mtu = params->mtu;
 	p.frame_overhead = params->frame_overhead;
 	p.n_subports_per_port = params->n_subports_per_port;
-	p.n_pipes_per_subport = params->n_pipes_per_subport;
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		p.qsize[i] = params->qsize[i];
-
-	p.pipe_profiles = pipe_profile;
-	p.n_pipe_profiles = n_pipe_profiles;
+	p.n_max_pipes_per_subport = TMGR_PIPE_SUBPORT_MAX;
+	p.n_max_pipe_profiles = TMGR_PIPE_PROFILE_MAX;
 
 	s = rte_sched_port_config(&p);
 	if (s == NULL)
 		return NULL;
 
+	subport_profile[0].pipe_profiles = pipe_profile;
+	subport_profile[0].n_pipe_profiles = n_pipe_profiles;
+
 	for (i = 0; i < params->n_subports_per_port; i++) {
 		int status;
 
@@ -128,7 +126,7 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 			return NULL;
 		}
 
-		for (j = 0; j < params->n_pipes_per_subport; j++) {
+		for (j = 0; j < subport_profile[0].n_pipes_per_subport; j++) {
 			status = rte_sched_pipe_config(
 				s,
 				i,
@@ -153,7 +151,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	strlcpy(tmgr_port->name, name, sizeof(tmgr_port->name));
 	tmgr_port->s = s;
 	tmgr_port->n_subports_per_port = params->n_subports_per_port;
-	tmgr_port->n_pipes_per_subport = params->n_pipes_per_subport;
 
 	/* Node add to list */
 	TAILQ_INSERT_TAIL(&tmgr_port_list, tmgr_port, node);
@@ -205,8 +202,8 @@ tmgr_pipe_config(const char *port_name,
 	port = tmgr_port_find(port_name);
 	if ((port == NULL) ||
 		(subport_id >= port->n_subports_per_port) ||
-		(pipe_id_first >= port->n_pipes_per_subport) ||
-		(pipe_id_last >= port->n_pipes_per_subport) ||
+		(pipe_id_first >= subport_profile[subport_id].n_pipes_per_subport) ||
+		(pipe_id_last >= subport_profile[subport_id].n_pipes_per_subport) ||
 		(pipe_id_first > pipe_id_last) ||
 		(pipe_profile_id >= n_pipe_profiles))
 		return -1;
examples/ip_pipeline/tmgr.h +4 -3
index 8703a2e00..1fcf66ee1 100644
@@ -12,6 +12,10 @@
 
 #include "common.h"
 
+#ifndef TMGR_PIPE_SUBPORT_MAX
+#define TMGR_PIPE_SUBPORT_MAX                              4096
+#endif
+
 #ifndef TMGR_SUBPORT_PROFILE_MAX
 #define TMGR_SUBPORT_PROFILE_MAX                           256
 #endif
@@ -25,7 +29,6 @@ struct tmgr_port {
 	char name[NAME_SIZE];
 	struct rte_sched_port *s;
 	uint32_t n_subports_per_port;
-	uint32_t n_pipes_per_subport;
 };
 
 TAILQ_HEAD(tmgr_port_list, tmgr_port);
@@ -42,8 +45,6 @@ struct tmgr_port_params {
 	uint32_t frame_overhead;
 	uint32_t mtu;
 	uint32_t cpu_id;
-	uint32_t n_pipes_per_subport;
-	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 };
 
 int
Signature · 1 line
2.21.0
#14 lore raw

PATCH 14/15 examples/qos_sched: add subport configuration flexibility

In reply to Jasvinder Singh

Modify qos sample app to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 examples/qos_sched/app_thread.c   |   4 +-
 examples/qos_sched/cfg_file.c     | 226 ++++++++++++++++--------------
 examples/qos_sched/init.c         |  54 +++----
 examples/qos_sched/main.h         |   1 +
 examples/qos_sched/profile.cfg    |   5 +-
 examples/qos_sched/profile_ov.cfg |   5 +-
 examples/qos_sched/stats.c        |  36 ++---
 7 files changed, 179 insertions(+), 152 deletions(-)
examples/qos_sched/app_thread.c +2 -2
index 18b734a14..e07cc7452 100644
@@ -36,9 +36,9 @@ get_pkt_sched(struct rte_mbuf *m, uint32_t *subport, uint32_t *pipe,
 	uint16_t pipe_queue;
 
 	*subport = (rte_be_to_cpu_16(pdata[SUBPORT_OFFSET]) & 0x0FFF) &
-			(port_params.n_subports_per_port - 1); /* Outer VLAN ID*/
+		(port_params.n_subports_per_port - 1); /* Outer VLAN ID*/
 	*pipe = (rte_be_to_cpu_16(pdata[PIPE_OFFSET]) & 0x0FFF) &
-			(port_params.n_pipes_per_subport - 1); /* Inner VLAN ID */
+		(subport_params[*subport].n_pipes_per_subport - 1); /* Inner VLAN ID */
 	pipe_queue = active_queues[(pdata[QUEUE_OFFSET] >> 8) % n_active_queues];
 	*traffic_class = pipe_queue > RTE_SCHED_TRAFFIC_CLASS_BE ?
 			RTE_SCHED_TRAFFIC_CLASS_BE : pipe_queue; /* Destination IP */
examples/qos_sched/cfg_file.c +121 -105

266 lines not on this page · raw message

examples/qos_sched/init.c +29 -25

79 lines not on this page · raw message

examples/qos_sched/main.h +1 -0
index d8f890b64..baa2b3ead 100644
@@ -152,6 +152,7 @@ uint32_t active_queues[RTE_SCHED_QUEUES_PER_PIPE];
 uint32_t n_active_queues;
 
 extern struct rte_sched_port_params port_params;
+extern struct rte_sched_subport_params subport_params[MAX_SCHED_SUBPORTS];
 
 int app_parse_args(int argc, char **argv);
 int app_init(void);
examples/qos_sched/profile.cfg +3 -2
index 335561370..61b8b7071 100644
@@ -20,11 +20,12 @@
 [port]
 frame overhead = 24
 number of subports per port = 1
-number of pipes per subport = 4096
-queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
 
 ; Subport configuration
 [subport 0]
+number of pipes per subport = 4096
+queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
+
 tb rate = 1250000000           ; Bytes per second
 tb size = 1000000              ; Bytes
 
examples/qos_sched/profile_ov.cfg +3 -2
index 394987399..ab509d28d 100644
@@ -5,11 +5,12 @@
 [port]
 frame overhead = 24
 number of subports per port = 1
-number of pipes per subport = 32
-queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
 
 ; Subport configuration
 [subport 0]
+number of pipes per subport = 32
+queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
+
 tb rate = 8400000           ; Bytes per second
 tb size = 100000            ; Bytes
 
examples/qos_sched/stats.c +20 -16
index e62e4a2f6..0626cf6e6 100644
@@ -24,7 +24,7 @@ qavg_q(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id, uint8_t tc,
 
 	if (i == nb_pfc ||
 		subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport  ||
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport  ||
 		tc >= RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE ||
 		q >= RTE_SCHED_BE_QUEUES_PER_PIPE ||
 		(tc < RTE_SCHED_TRAFFIC_CLASS_BE && q > 0))
@@ -32,7 +32,7 @@ qavg_q(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id, uint8_t tc,
 
 	port = qos_conf[i].sched_port;
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport *
+		queue_id += subport_params[i].n_pipes_per_subport *
 				RTE_SCHED_QUEUES_PER_PIPE;
 	if (tc < RTE_SCHED_TRAFFIC_CLASS_BE)
 		queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE + tc;
@@ -69,14 +69,15 @@ qavg_tcpipe(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id,
 	}
 
 	if (i == nb_pfc || subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport ||
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport ||
 		tc >= RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE)
 		return -1;
 
 	port = qos_conf[i].sched_port;
 
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
+		queue_id +=
+			subport_params[i].n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
 
 	queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE + tc;
 
@@ -123,13 +124,13 @@ qavg_pipe(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id)
 
 	if (i == nb_pfc ||
 		subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport)
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport)
 		return -1;
 
 	port = qos_conf[i].sched_port;
 
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport *
+		queue_id += subport_params[i].n_pipes_per_subport *
 				RTE_SCHED_QUEUES_PER_PIPE;
 
 	queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE;
@@ -177,13 +178,13 @@ qavg_tcsubport(uint16_t port_id, uint32_t subport_id, uint8_t tc)
 
 	for (i = 0; i < subport_id; i++)
 		subport_queue_id +=
-			port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
+			subport_params[i].n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
 
 	average = 0;
 
 	for (count = 0; count < qavg_ntimes; count++) {
 		part_average = 0;
-		for (i = 0; i < port_params.n_pipes_per_subport; i++) {
+		for (i = 0; i < subport_params[subport_id].n_pipes_per_subport; i++) {
 			if (tc < RTE_SCHED_TRAFFIC_CLASS_BE) {
 				queue_id = subport_queue_id +
 					i * RTE_SCHED_QUEUES_PER_PIPE + tc;
@@ -203,10 +204,11 @@ qavg_tcsubport(uint16_t port_id, uint32_t subport_id, uint8_t tc)
 		}
 
 		if (tc < RTE_SCHED_TRAFFIC_CLASS_BE)
-			average += part_average / (port_params.n_pipes_per_subport);
+			average += part_average /
+				(subport_params[subport_id].n_pipes_per_subport);
 		else
-			average +=
-				part_average / (port_params.n_pipes_per_subport) *
+			average += part_average /
+				(subport_params[subport_id].n_pipes_per_subport) *
 				RTE_SCHED_BE_QUEUES_PER_PIPE;
 
 		usleep(qavg_period);
@@ -240,14 +242,14 @@ qavg_subport(uint16_t port_id, uint32_t subport_id)
 	port = qos_conf[i].sched_port;
 
 	for (i = 0; i < subport_id; i++)
-		subport_queue_id += port_params.n_pipes_per_subport *
+		subport_queue_id += subport_params[i].n_pipes_per_subport *
 			RTE_SCHED_QUEUES_PER_PIPE;
 
 	average = 0;
 
 	for (count = 0; count < qavg_ntimes; count++) {
 		part_average = 0;
-		for (i = 0; i < port_params.n_pipes_per_subport; i++) {
+		for (i = 0; i < subport_params[subport_id].n_pipes_per_subport; i++) {
 			queue_id = subport_queue_id + i * RTE_SCHED_QUEUES_PER_PIPE;
 
 			for (j = 0; j < RTE_SCHED_QUEUES_PER_PIPE; j++) {
@@ -258,7 +260,8 @@ qavg_subport(uint16_t port_id, uint32_t subport_id)
 		}
 
 		average += part_average /
-			(port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE);
+			(subport_params[subport_id].n_pipes_per_subport *
+			RTE_SCHED_QUEUES_PER_PIPE);
 		usleep(qavg_period);
 	}
 
@@ -322,12 +325,13 @@ pipe_stat(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id)
 
 	if (i == nb_pfc ||
 		subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport)
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport)
 		return -1;
 
 	port = qos_conf[i].sched_port;
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
+		queue_id += subport_params[i].n_pipes_per_subport *
+			RTE_SCHED_QUEUES_PER_PIPE;
 
 	queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE;
 
Signature · 1 line
2.21.0
#15 lore raw

PATCH 15/15 sched: remove redundant code

In reply to Jasvinder Singh

Remove redundant data structure fields from port level data structures and update release notes.

---
 doc/guides/rel_notes/release_19_11.rst |  6 +++-
 lib/librte_sched/rte_sched.c           | 43 +-------------------------
 lib/librte_sched/rte_sched.h           | 22 -------------
 3 files changed, 6 insertions(+), 65 deletions(-)
doc/guides/rel_notes/release_19_11.rst +5 -1
index 8490d897c..746aeb0ea 100644
@@ -85,6 +85,10 @@ API Changes
    Also, make sure to start the actual text at the margin.
    =========================================================
 
+* sched: The pipe nodes configuration parameters such as number of pipes,
+  pipe queue sizes, pipe profiles, etc., are moved from port level structure
+  to subport level. This allows different subports of the same port to
+  have different configuration for the pipe nodes.
 
 ABI Changes
 -----------
@@ -172,7 +176,7 @@ The libraries prepended with a plus sign were incremented in this version.
      librte_rcu.so.1
      librte_reorder.so.1
      librte_ring.so.2
-     librte_sched.so.3
+   + librte_sched.so.4
      librte_security.so.2
      librte_stack.so.1
      librte_table.so.3
lib/librte_sched/rte_sched.c +1 -42

72 lines not on this page · raw message

lib/librte_sched/rte_sched.h +0 -22
index ea2b07448..db1b0232f 100644
@@ -246,33 +246,11 @@ struct rte_sched_port_params {
 	/** Number of subports */
 	uint32_t n_subports_per_port;
 
-	/** Number of subport_pipes */
-	uint32_t n_pipes_per_subport;
-
 	/** Maximum number of subport_pipes */
 	uint32_t n_max_pipes_per_subport;
 
-	/** Packet queue size for each traffic class.
-	 * All the pipes within the same subport share the similar
-	 * configuration for the queues.
-	 */
-	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-
-	/** Pipe profile table.
-	 * Every pipe is configured using one of the profiles from this table.
-	 */
-	struct rte_sched_pipe_params *pipe_profiles;
-
-	/** Profiles in the pipe profile table */
-	uint32_t n_pipe_profiles;
-
 	/** Max profiles allowed in the pipe profile table */
 	uint32_t n_max_pipe_profiles;
-
-#ifdef RTE_SCHED_RED
-	/** RED parameters */
-	struct rte_red_params red_params[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][RTE_COLORS];
-#endif
 };
 
 /*
Signature · 1 line
2.21.0
#16 lore raw

v2 cover sched: subport level configuration of pipe nodes

In reply to Jasvinder Singh

This patchset refactors the dpdk qos sched library to allow subport level configuration flexibility of the pipe nodes.

Currently, all parameters for the pipe nodes (subscribers) configuration are part of the port level structure which forces all groups of subscribers (pipes) in different subports to have similar configurations in terms of their number, queue sizes, traffic-classes, etc.

The new implementation moves pipe nodes configuration parameters from port level to subport level structure. This allows different subports of the same port to have different configuration for the pipe nodes, for examples- number of pipes, queue sizes, pipe profiles, etc.

In order to keep the implementation complexity under control, all pipes within the same subport share the same configuration for queue sizes.

v2:
- fix qsize parsing in sample application
- fix checkpatch warnings

Jasvinder Singh (15):
sched: add pipe config params to subport struct
sched: modify internal structs for config flexibility
sched: remove pipe params config from port level
shced: add pipe config to subport level
sched: modify pipe functions for config flexibility
sched: modify pkt enqueue for config flexibility
sched: update memory compute to support flexiblity
sched: update grinder functions for config flexibility
sched: update pkt dequeue for flexible config
sched: update queue stats read for config flexibility test/sched: modify tests for subport config flexibility net/softnic: add subport config flexibility to TM
ip_pipeline: add subport config flexibility to TM examples/qos_sched: add subport configuration flexibility
sched: remove redundant code

 app/test/test_sched.c                    |   35 +-
 doc/guides/rel_notes/release_19_11.rst   |    6 +-
 drivers/net/softnic/rte_eth_softnic_tm.c |   51 +-
 examples/ip_pipeline/cli.c               |   71 +-
 examples/ip_pipeline/tmgr.c              |   23 +-
 examples/ip_pipeline/tmgr.h              |    7 +-
 examples/qos_sched/app_thread.c          |    4 +-
 examples/qos_sched/cfg_file.c            |  229 ++--
 examples/qos_sched/init.c                |   54 +-
 examples/qos_sched/main.h                |    1 +
 examples/qos_sched/profile.cfg           |    5 +-
 examples/qos_sched/profile_ov.cfg        |    5 +-
 examples/qos_sched/stats.c               |   36 +-
 lib/librte_sched/Makefile                |    2 +-
 lib/librte_sched/meson.build             |    2 +-
 lib/librte_sched/rte_sched.c             | 1400 +++++++++++++---------
 lib/librte_sched/rte_sched.h             |  114 +-
 17 files changed, 1183 insertions(+), 862 deletions(-)
Signature · 1 line
2.21.0
#17 lore raw

PATCH v2 1/15 sched: add pipe config params to subport struct

Add pipe configuration parameters to subport level structure to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/Makefile    |  2 +-
 lib/librte_sched/meson.build |  2 +-
 lib/librte_sched/rte_sched.h | 81 +++++++++++++++++++++++-------------
 3 files changed, 55 insertions(+), 30 deletions(-)
lib/librte_sched/Makefile +1 -1
index 3d7f410e1..6e4a72d89 100644
@@ -18,7 +18,7 @@ LDLIBS += -lrte_timer
 
 EXPORT_MAP := rte_sched_version.map
 
-LIBABIVER := 3
+LIBABIVER := 4
 
 #
 # all source are stored in SRCS-y
lib/librte_sched/meson.build +1 -1
index 59d43c6d8..9f40a2368 100644
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2017 Intel Corporation
 
-version = 3
+version = 4
 sources = files('rte_sched.c', 'rte_red.c', 'rte_approx.c')
 headers = files('rte_sched.h', 'rte_sched_common.h',
 		'rte_red.h', 'rte_approx.h')
lib/librte_sched/rte_sched.h +53 -28

126 lines not on this page · raw message

Signature · 1 line
2.21.0
#18 lore raw

PATCH v2 2/15 sched: modify internal structs for config flexibility

In reply to Jasvinder Singh

Update internal structures related to port and subport to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 106 ++++++++++++++++++++++++++---------
 1 file changed, 79 insertions(+), 27 deletions(-)
lib/librte_sched/rte_sched.c +79 -27
index d8ab21d64..6a9990e5e 100644
@@ -47,33 +47,6 @@
  */
 #define RTE_SCHED_TIME_SHIFT		      8
 
-struct rte_sched_subport {
-	/* Token bucket (TB) */
-	uint64_t tb_time; /* time of last update */
-	uint32_t tb_period;
-	uint32_t tb_credits_per_period;
-	uint32_t tb_size;
-	uint32_t tb_credits;
-
-	/* Traffic classes (TCs) */
-	uint64_t tc_time; /* time of next update */
-	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_period;
-
-	/* TC oversubscription */
-	uint32_t tc_ov_wm;
-	uint32_t tc_ov_wm_min;
-	uint32_t tc_ov_wm_max;
-	uint8_t tc_ov_period_id;
-	uint8_t tc_ov;
-	uint32_t tc_ov_n;
-	double tc_ov_rate;
-
-	/* Statistics */
-	struct rte_sched_subport_stats stats;
-};
-
 struct rte_sched_pipe_profile {
 	/* Token bucket (TB) */
 	uint32_t tb_period;
@@ -166,10 +139,77 @@ struct rte_sched_grinder {
 	uint8_t wrr_cost[RTE_SCHED_BE_QUEUES_PER_PIPE];
 };
 
+struct rte_sched_subport {
+	/* Token bucket (TB) */
+	uint64_t tb_time; /* time of last update */
+	uint32_t tb_period;
+	uint32_t tb_credits_per_period;
+	uint32_t tb_size;
+	uint32_t tb_credits;
+
+	/* Traffic classes (TCs) */
+	uint64_t tc_time; /* time of next update */
+	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_period;
+
+	/* TC oversubscription */
+	uint32_t tc_ov_wm;
+	uint32_t tc_ov_wm_min;
+	uint32_t tc_ov_wm_max;
+	uint8_t tc_ov_period_id;
+	uint8_t tc_ov;
+	uint32_t tc_ov_n;
+	double tc_ov_rate;
+
+	/* Statistics */
+	struct rte_sched_subport_stats stats;
+
+	/* Subport pipes */
+	uint32_t n_pipes_per_subport;
+	uint32_t n_pipe_profiles;
+
+	/* Pipe best-effort TC rate */
+	uint32_t pipe_tc_be_rate_max;
+
+	/* Pipe queues size */
+	uint16_t qsize[RTE_SCHED_QUEUES_PER_PIPE];
+
+#ifdef RTE_SCHED_RED
+	struct rte_red_config red_config[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][RTE_COLORS];
+#endif
+
+	/* Scheduling loop detection */
+	uint32_t pipe_loop;
+	uint32_t pipe_exhaustion;
+
+	/* Bitmap */
+	struct rte_bitmap *bmp;
+	uint32_t grinder_base_bmp_pos[RTE_SCHED_PORT_N_GRINDERS] __rte_aligned_16;
+
+	/* Grinders */
+	struct rte_sched_grinder grinder[RTE_SCHED_PORT_N_GRINDERS];
+	uint32_t busy_grinders;
+
+	/* Queue base calculation */
+	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
+	uint32_t qsize_sum;
+
+	struct rte_sched_pipe *pipe;
+	struct rte_sched_queue *queue;
+	struct rte_sched_queue_extra *queue_extra;
+	struct rte_sched_pipe_profile *pipe_profiles;
+	uint8_t *bmp_array;
+	struct rte_mbuf **queue_array;
+	uint8_t memory[0] __rte_cache_aligned;
+} __rte_cache_aligned;
+
 struct rte_sched_port {
 	/* User parameters */
 	uint32_t n_subports_per_port;
 	uint32_t n_pipes_per_subport;
+	uint32_t n_max_pipes_per_subport;
+	uint32_t n_max_pipes_per_subport_log2;
 	uint32_t n_pipes_per_subport_log2;
 	uint16_t pipe_queue[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint8_t pipe_tc[RTE_SCHED_QUEUES_PER_PIPE];
@@ -177,6 +217,7 @@ struct rte_sched_port {
 	uint32_t rate;
 	uint32_t mtu;
 	uint32_t frame_overhead;
+	int socket;
 	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t n_pipe_profiles;
 	uint32_t n_max_pipe_profiles;
@@ -210,6 +251,7 @@ struct rte_sched_port {
 	uint32_t qsize_sum;
 
 	/* Large data structures */
+	struct rte_sched_subport *subports[0];
 	struct rte_sched_subport *subport;
 	struct rte_sched_pipe *pipe;
 	struct rte_sched_queue *queue;
@@ -231,6 +273,16 @@ enum rte_sched_port_array {
 	e_RTE_SCHED_PORT_ARRAY_TOTAL,
 };
 
+enum rte_sched_subport_array {
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE = 0,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA,
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES,
+	e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
+};
+
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline uint32_t
Signature · 1 line
2.21.0
#19 lore raw

PATCH v2 3/15 sched: remove pipe params config from port level

In reply to Jasvinder Singh

Remove pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 266 +++++++++++------------------------
 1 file changed, 81 insertions(+), 185 deletions(-)
lib/librte_sched/rte_sched.c +81 -185
index 6a9990e5e..db8b0485a 100644
@@ -293,6 +293,31 @@ rte_sched_port_queues_per_subport(struct rte_sched_port *port)
 
 #endif
 
+static inline uint32_t
+rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
+{
+	return RTE_SCHED_QUEUES_PER_PIPE * subport->n_pipes_per_subport;
+}
+
+static inline struct rte_mbuf **
+rte_sched_subport_pipe_qbase(struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t pindex = qindex >> 4;
+	uint32_t qpos = qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1);
+
+	return (subport->queue_array + pindex *
+		subport->qsize_sum + subport->qsize_add[qpos]);
+}
+
+static inline uint16_t
+rte_sched_subport_pipe_qsize(struct rte_sched_port *port,
+struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
+
+	return subport->qsize[tc];
+}
+
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
@@ -416,8 +441,6 @@ pipe_profile_check(struct rte_sched_pipe_params *params,
 static int
 rte_sched_port_check_params(struct rte_sched_port_params *params)
 {
-	uint32_t i;
-
 	if (params == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter params\n", __func__);
@@ -454,49 +477,21 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 		return -EINVAL;
 	}
 
-	/* n_pipes_per_subport: non-zero, power of 2 */
-	if (params->n_pipes_per_subport == 0 ||
-	    !rte_is_power_of_2(params->n_pipes_per_subport)) {
+	/* n_max_pipes_per_subport: non-zero, power of 2 */
+	if (params->n_max_pipes_per_subport == 0 ||
+	    !rte_is_power_of_2(params->n_max_pipes_per_subport)) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for pipes number\n", __func__);
+			"%s: Incorrect value for maximum pipes number\n", __func__);
 		return -EINVAL;
 	}
 
-	/* qsize: if non-zero, power of 2,
-	 * no bigger than 32K (due to 16-bit read/write pointers)
-	 */
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint16_t qsize = params->qsize[i];
-
-		if ((qsize != 0 && !rte_is_power_of_2(qsize)) ||
-			((i == RTE_SCHED_TRAFFIC_CLASS_BE) && (qsize == 0))) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	/* pipe_profiles and n_pipe_profiles */
-	if (params->pipe_profiles == NULL ||
-	    params->n_pipe_profiles == 0 ||
-		 params->n_pipe_profiles > params->n_max_pipe_profiles) {
+	/* n_max_pipe_profiles: non-zero */
+	if (params->n_max_pipe_profiles == 0) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for number of pipe profiles\n", __func__);
+			"%s: Incorrect value for maximum pipe profiles\n", __func__);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < params->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
-		int status;
-
-		status = pipe_profile_check(p, params->rate, &params->qsize[0]);
-		if (status != 0) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Pipe profile check failed(%d)\n", __func__, status);
-			return -EINVAL;
-		}
-	}
-
 	return 0;
 }
 
@@ -584,32 +579,6 @@ rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
 	return size0 + size1;
 }
 
-static void
-rte_sched_port_config_qsize(struct rte_sched_port *port)
-{
-	uint32_t i;
-
-	port->qsize_add[0] = 0;
-
-	/* Strict prority traffic class */
-	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		port->qsize_add[i] = port->qsize_add[i-1] + port->qsize[i-1];
-
-	/* Best-effort traffic class */
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-	port->qsize_sum = port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-}
-
 static void
 rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
 {
@@ -719,56 +688,42 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
-static void
-rte_sched_port_config_pipe_profile_table(struct rte_sched_port *port,
-	struct rte_sched_port_params *params)
-{
-	uint32_t i;
-
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		struct rte_sched_pipe_profile *dst = port->pipe_profiles + i;
-
-		rte_sched_pipe_profile_convert(port, src, dst, params->rate);
-		rte_sched_port_log_pipe_profile(port, i);
-	}
-
-	port->pipe_tc_be_rate_max = 0;
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-		if (port->pipe_tc_be_rate_max < pipe_tc_be_rate)
-			port->pipe_tc_be_rate_max = pipe_tc_be_rate;
-	}
-}
-
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
 	struct rte_sched_port *port = NULL;
-	uint32_t mem_size, bmp_mem_size, n_queues_per_port, i, j, cycles_per_byte;
+	uint32_t size0, size1;
+	uint32_t cycles_per_byte;
+	uint32_t i, j;
+	int status;
 
-	/* Check user parameters. Determine the amount of memory to allocate */
-	mem_size = rte_sched_port_get_memory_footprint(params);
-	if (mem_size == 0)
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 		return NULL;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = params->n_subports_per_port * sizeof(struct rte_sched_subport *);
 
 	/* Allocate memory to store the data structures */
-	port = rte_zmalloc_socket("qos_params", mem_size, RTE_CACHE_LINE_SIZE,
+	port = rte_zmalloc_socket("qos_params", size0 + size1, RTE_CACHE_LINE_SIZE,
 		params->socket);
-	if (port == NULL)
-		return NULL;
+	if (port == NULL) {
+		RTE_LOG(ERR, SCHED, "%s: Memory allocation fails\n", __func__);
 
-	/* compile time checks */
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS & (RTE_SCHED_PORT_N_GRINDERS - 1));
+		return NULL;
+	}
 
 	/* User parameters */
 	port->n_subports_per_port = params->n_subports_per_port;
-	port->n_pipes_per_subport = params->n_pipes_per_subport;
-	port->n_pipes_per_subport_log2 =
-			__builtin_ctz(params->n_pipes_per_subport);
+	port->n_max_pipes_per_subport = params->n_max_pipes_per_subport;
+	port->n_max_pipes_per_subport_log2 =
+			__builtin_ctz(params->n_max_pipes_per_subport);
+	port->n_max_pipe_profiles = params->n_max_pipe_profiles;
+	port->socket = params->socket;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 		port->pipe_queue[i] = i;
@@ -789,32 +744,6 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	port->rate = params->rate;
 	port->mtu = params->mtu + params->frame_overhead;
 	port->frame_overhead = params->frame_overhead;
-	memcpy(port->qsize, params->qsize, sizeof(params->qsize));
-	port->n_pipe_profiles = params->n_pipe_profiles;
-	port->n_max_pipe_profiles = params->n_max_pipe_profiles;
-
-#ifdef RTE_SCHED_RED
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t j;
-
-		for (j = 0; j < RTE_COLORS; j++) {
-			/* if min/max are both zero, then RED is disabled */
-			if ((params->red_params[i][j].min_th |
-			     params->red_params[i][j].max_th) == 0) {
-				continue;
-			}
-
-			if (rte_red_config_init(&port->red_config[i][j],
-				params->red_params[i][j].wq_log2,
-				params->red_params[i][j].min_th,
-				params->red_params[i][j].max_th,
-				params->red_params[i][j].maxp_inv) != 0) {
-				rte_free(port);
-				return NULL;
-			}
-		}
-	}
-#endif
 
 	/* Timing */
 	port->time_cpu_cycles = rte_get_tsc_cycles();
@@ -825,79 +754,32 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 		/ params->rate;
 	port->inv_cycles_per_byte = rte_reciprocal_value(cycles_per_byte);
 
-	/* Scheduling loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
-	port->pipe_exhaustion = 0;
-
 	/* Grinders */
-	port->busy_grinders = 0;
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
 
-	/* Queue base calculation */
-	rte_sched_port_config_qsize(port);
-
-	/* Large data structures */
-	port->subport = (struct rte_sched_subport *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_SUBPORT));
-	port->pipe = (struct rte_sched_pipe *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE));
-	port->queue = (struct rte_sched_queue *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE));
-	port->queue_extra = (struct rte_sched_queue_extra *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA));
-	port->pipe_profiles = (struct rte_sched_pipe_profile *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES));
-	port->bmp_array =  port->memory
-		+ rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY);
-	port->queue_array = (struct rte_mbuf **)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY));
-
-	/* Pipe profile table */
-	rte_sched_port_config_pipe_profile_table(port, params);
-
-	/* Bitmap */
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
-	bmp_mem_size = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	port->bmp = rte_bitmap_init(n_queues_per_port, port->bmp_array,
-				    bmp_mem_size);
-	if (port->bmp == NULL) {
-		RTE_LOG(ERR, SCHED, "Bitmap init error\n");
-		rte_free(port);
-		return NULL;
-	}
-
-	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
-		port->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
-
-
 	return port;
 }
 
-void
-rte_sched_port_free(struct rte_sched_port *port)
+static inline void
+rte_sched_subport_free(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
+	uint32_t n_subport_pipe_queues;
 	uint32_t qindex;
-	uint32_t n_queues_per_port;
 
-	/* Check user parameters */
-	if (port == NULL)
+	if (subport == NULL)
 		return;
 
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(subport);
 
 	/* Free enqueued mbufs */
-	for (qindex = 0; qindex < n_queues_per_port; qindex++) {
-		struct rte_mbuf **mbufs = rte_sched_port_qbase(port, qindex);
-		uint16_t qsize = rte_sched_port_qsize(port, qindex);
+	for (qindex = 0; qindex < n_subport_pipe_queues; qindex++) {
+		struct rte_mbuf **mbufs =
+			rte_sched_subport_pipe_qbase(subport, qindex);
+		uint16_t qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 		if (qsize != 0) {
-			struct rte_sched_queue *queue = port->queue + qindex;
+			struct rte_sched_queue *queue = subport->queue + qindex;
 			uint16_t qr = queue->qr & (qsize - 1);
 			uint16_t qw = queue->qw & (qsize - 1);
 
@@ -906,7 +788,21 @@ rte_sched_port_free(struct rte_sched_port *port)
 		}
 	}
 
-	rte_bitmap_free(port->bmp);
+	rte_bitmap_free(subport->bmp);
+}
+
+void
+rte_sched_port_free(struct rte_sched_port *port)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (port == NULL)
+		return;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		rte_sched_subport_free(port, port->subports[i]);
+
 	rte_free(port);
 }
 
Signature · 1 line
2.21.0
#20 lore raw

PATCH v2 4/15 shced: add pipe config to subport level

In reply to Jasvinder Singh

Add pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 390 ++++++++++++++++++++++++++++++-----
 1 file changed, 334 insertions(+), 56 deletions(-)
lib/librte_sched/rte_sched.c +334 -56
index db8b0485a..1491cb695 100644
@@ -559,24 +559,92 @@ rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sch
 	return base;
 }
 
-uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+static uint32_t
+rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
+	uint32_t n_max_pipe_profiles,
+	enum rte_sched_subport_array array)
 {
-	uint32_t size0, size1;
-	int status;
+	uint32_t n_pipes_per_subport = params->n_pipes_per_subport;
+	uint32_t n_subport_pipe_queues =
+		RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport;
 
-	status = rte_sched_port_check_params(params);
-	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+	uint32_t size_pipe = n_pipes_per_subport * sizeof(struct rte_sched_pipe);
+	uint32_t size_queue =
+		n_subport_pipe_queues * sizeof(struct rte_sched_queue);
+	uint32_t size_queue_extra
+		= n_subport_pipe_queues * sizeof(struct rte_sched_queue_extra);
+	uint32_t size_pipe_profiles = n_max_pipe_profiles *
+		sizeof(struct rte_sched_pipe_profile);
+	uint32_t size_bmp_array =
+		rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	uint32_t size_per_pipe_queue_array, size_queue_array;
 
-		return 0;
+	uint32_t base, i;
+
+	size_per_pipe_queue_array = 0;
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
+			size_per_pipe_queue_array +=
+				params->qsize[i] * sizeof(struct rte_mbuf *);
+		else
+			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
+				params->qsize[i] * sizeof(struct rte_mbuf *);
 	}
+	size_queue_array = n_pipes_per_subport * size_per_pipe_queue_array;
 
-	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_TOTAL);
+	base = 0;
 
-	return size0 + size1;
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
+
+	return base;
+}
+
+static void
+rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
+{
+	uint32_t i;
+
+	subport->qsize_add[0] = 0;
+
+	/* Strict prority traffic class */
+	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		subport->qsize_add[i] = subport->qsize_add[i-1] + subport->qsize[i-1];
+
+	/* Best-effort traffic class */
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+	subport->qsize_sum = subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
 }
 
 static void
@@ -688,6 +756,126 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static int
+rte_sched_subport_check_params(struct rte_sched_subport_params *params,
+	uint32_t n_max_pipes_per_subport,
+	uint32_t n_max_pipe_profiles,
+	uint32_t rate)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (params == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for parameter params\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_rate == 0 || params->tb_rate > rate) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb rate\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_size == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb size\n", __func__);
+		return -EINVAL;
+	}
+
+	/* qsize: if non-zero, power of 2,
+	 * no bigger than 32K (due to 16-bit read/write pointers)
+	 */
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint16_t qsize = params->qsize[i];
+
+		if (qsize != 0 && !rte_is_power_of_2(qsize)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for qsize\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t tc_rate = params->tc_rate[i];
+		uint16_t qsize = params->qsize[i];
+
+		if ((qsize == 0 && tc_rate != 0) ||
+			(qsize != 0 && tc_rate == 0) ||
+			(tc_rate > params->tb_rate)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for tc rate\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	if (params->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
+		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect qsize or tc rate(best effort)\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tc_period == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tc period\n", __func__);
+		return -EINVAL;
+	}
+
+	/* n_pipes_per_subport: non-zero, power of 2 */
+	if (params->n_pipes_per_subport == 0 ||
+		params->n_pipes_per_subport > n_max_pipes_per_subport ||
+	    !rte_is_power_of_2(params->n_pipes_per_subport)) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipes number\n", __func__);
+		return -EINVAL;
+	}
+
+	/* pipe_profiles and n_pipe_profiles */
+	if (params->pipe_profiles == NULL ||
+	    params->n_pipe_profiles == 0 ||
+		params->n_pipe_profiles > n_max_pipe_profiles) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipe profiles\n", __func__);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < params->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
+		int status;
+
+		status = pipe_profile_check(p, rate, &params->qsize[0]);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Pipe profile check failed(%d)\n", __func__, status);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+uint32_t
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+{
+	uint32_t size0, size1;
+	int status;
+
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"Port scheduler params check failed (%d)\n", status);
+
+		return 0;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = rte_sched_port_get_array_base(params,
+			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	return size0 + size1;
+}
+
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
@@ -809,7 +997,7 @@ rte_sched_port_free(struct rte_sched_port *port)
 static void
 rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 {
-	struct rte_sched_subport *s = port->subport + i;
+	struct rte_sched_subport *s = port->subports[i];
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for subport %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -844,72 +1032,80 @@ rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 		s->tc_ov_wm_max);
 }
 
+static void
+rte_sched_free_memory(struct rte_sched_port *port, uint32_t n_subports)
+{
+	uint32_t i;
+
+	for (i = 0; i < n_subports; i++) {
+		struct rte_sched_subport *subport = port->subports[i];
+
+		rte_sched_subport_free(port, subport);
+	}
+
+	rte_free(port);
+}
+
 int
 rte_sched_subport_config(struct rte_sched_port *port,
 	uint32_t subport_id,
 	struct rte_sched_subport_params *params)
 {
-	struct rte_sched_subport *s;
-	uint32_t i;
+	struct rte_sched_subport *s = NULL;
+	uint32_t n_subports = subport_id;
+	uint32_t n_subport_pipe_queues, i;
+	uint32_t size0, size1, bmp_mem_size;
+	int status;
 
 	/* Check user parameters */
 	if (port == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter port\n", __func__);
-		return -EINVAL;
+		return 0;
 	}
 
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for subport id\n", __func__);
-		return -EINVAL;
-	}
 
-	if (params == NULL) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for parameter params\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (params->tb_rate == 0 || params->tb_rate > port->rate) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb rate\n", __func__);
-		return -EINVAL;
-	}
+	status = rte_sched_subport_check_params(params,
+		port->n_max_pipes_per_subport,
+		port->n_max_pipe_profiles,
+		port->rate);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 
-	if (params->tb_size == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb size\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t tc_rate = params->tc_rate[i];
-		uint16_t qsize = port->qsize[i];
+	/* Determine the amount of memory to allocate */
+	size0 = sizeof(struct rte_sched_subport);
+	size1 = rte_sched_subport_get_array_base(params,
+			port->n_max_pipe_profiles,
+			e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
 
-		if ((qsize == 0 && tc_rate != 0) ||
-			(qsize != 0 && tc_rate == 0) ||
-			(tc_rate > params->tb_rate)) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	if (port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
-		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+	/* Allocate memory to store the data structures */
+	s = rte_zmalloc_socket("subport_params", size0 + size1,
+		RTE_CACHE_LINE_SIZE, port->socket);
+	if (s == NULL) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc rate(best effort)\n", __func__);
-		return -EINVAL;
-	}
+			"%s: Memory allocation fails\n", __func__);
 
-	if (params->tc_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc period\n", __func__);
-		return -EINVAL;
+		rte_sched_free_memory(port, n_subports);
+		return -ENOMEM;
 	}
 
-	s = port->subport + subport_id;
+	n_subports++;
+
+	/* Port */
+	port->subports[subport_id] = s;
 
 	/* Token Bucket (TB) */
 	if (params->tb_rate == port->rate) {
@@ -929,26 +1125,108 @@ rte_sched_subport_config(struct rte_sched_port *port,
 	/* Traffic Classes (TCs) */
 	s->tc_period = rte_sched_time_ms_to_bytes(params->tc_period, port->rate);
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(params->tc_period,
-								 params->tc_rate[i]);
-
+					params->tc_rate[i]);
 	}
 	s->tc_time = port->time + s->tc_period;
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits[i] = s->tc_credits_per_period[i];
 
+	/* compile time checks */
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS &
+		(RTE_SCHED_PORT_N_GRINDERS - 1));
+
+	/* User parameters */
+	s->n_pipes_per_subport = params->n_pipes_per_subport;
+	memcpy(s->qsize, params->qsize, sizeof(params->qsize));
+	s->n_pipe_profiles = params->n_pipe_profiles;
+
+#ifdef RTE_SCHED_RED
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t j;
+
+		for (j = 0; j < RTE_COLORS; j++) {
+			/* if min/max are both zero, then RED is disabled */
+			if ((params->red_params[i][j].min_th |
+			     params->red_params[i][j].max_th) == 0) {
+				continue;
+			}
+
+			if (rte_red_config_init(&s->red_config[i][j],
+				params->red_params[i][j].wq_log2,
+				params->red_params[i][j].min_th,
+				params->red_params[i][j].max_th,
+				params->red_params[i][j].maxp_inv) != 0) {
+				rte_sched_free_memory(port, n_subports);
+
+				RTE_LOG(NOTICE, SCHED,
+				"%s: RED configuration init fails\n", __func__);
+				return -EINVAL;
+			}
+		}
+	}
+#endif
+
+	/* Scheduling loop detection */
+	s->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	s->pipe_exhaustion = 0;
+
+	/* Grinders */
+	s->busy_grinders = 0;
+
+	/* Queue base calculation */
+	rte_sched_subport_config_qsize(s);
+
+	/* Large data structures */
+	s->pipe = (struct rte_sched_pipe *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_PIPE));
+	s->queue = (struct rte_sched_queue *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE));
+	s->queue_extra = (struct rte_sched_queue_extra *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA));
+	s->pipe_profiles = (struct rte_sched_pipe_profile *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES));
+	s->bmp_array =  s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY);
+	s->queue_array = (struct rte_mbuf **)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
+
+	/* Bitmap */
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
+	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	s->bmp = rte_bitmap_init(n_subport_pipe_queues, s->bmp_array,
+				bmp_mem_size);
+	if (s->bmp == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Subport bitmap init error\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
+		s->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
+
+#ifdef RTE_SCHED_SUBPORT_TC_OV
 	/* TC oversubscription */
 	s->tc_ov_wm_min = port->mtu;
 	s->tc_ov_wm_max = rte_sched_time_ms_to_bytes(params->tc_period,
-						     port->pipe_tc_be_rate_max);
+						     s->pipe_tc_be_rate_max);
 	s->tc_ov_wm = s->tc_ov_wm_max;
 	s->tc_ov_period_id = 0;
 	s->tc_ov = 0;
 	s->tc_ov_n = 0;
 	s->tc_ov_rate = 0;
+#endif
 
 	rte_sched_port_log_subport_config(port, subport_id);
 
Signature · 1 line
2.21.0
#21 lore raw

PATCH v2 5/15 sched: modify pipe functions for config flexibility

In reply to Jasvinder Singh

Modify pipe level functions to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 95 +++++++++++++++++++++++++-----------
 lib/librte_sched/rte_sched.h |  5 +-
 2 files changed, 70 insertions(+), 30 deletions(-)
lib/librte_sched/rte_sched.c +66 -29
index 1491cb695..6bafbd8fe 100644
@@ -648,9 +648,9 @@ rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
 }
 
 static void
-rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
+rte_sched_port_log_pipe_profile(struct rte_sched_subport *subport, uint32_t i)
 {
-	struct rte_sched_pipe_profile *p = port->pipe_profiles + i;
+	struct rte_sched_pipe_profile *p = subport->pipe_profiles + i;
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for pipe profile %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -699,7 +699,7 @@ rte_sched_time_ms_to_bytes(uint32_t time_ms, uint32_t rate)
 }
 
 static void
-rte_sched_pipe_profile_convert(struct rte_sched_port *port,
+rte_sched_pipe_profile_convert(struct rte_sched_subport *subport,
 	struct rte_sched_pipe_params *src,
 	struct rte_sched_pipe_profile *dst,
 	uint32_t rate)
@@ -728,7 +728,7 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 						rate);
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (subport->qsize[i])
 			dst->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(src->tc_period,
 					src->tc_rate[i]);
@@ -756,6 +756,30 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static void
+rte_sched_subport_config_pipe_profile_table(struct rte_sched_subport *subport,
+	struct rte_sched_subport_params *params, uint32_t rate)
+{
+	uint32_t i;
+
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		struct rte_sched_pipe_profile *dst = subport->pipe_profiles + i;
+
+		rte_sched_pipe_profile_convert(subport, src, dst, rate);
+		rte_sched_port_log_pipe_profile(subport, i);
+	}
+
+	subport->pipe_tc_be_rate_max = 0;
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+		if (subport->pipe_tc_be_rate_max < pipe_tc_be_rate)
+			subport->pipe_tc_be_rate_max = pipe_tc_be_rate;
+	}
+}
+
 static int
 rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 	uint32_t n_max_pipes_per_subport,
@@ -1200,6 +1224,9 @@ rte_sched_subport_config(struct rte_sched_port *port,
 		(s->memory + rte_sched_subport_get_array_base(params,
 		port->n_max_pipe_profiles, e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
 
+	/* Pipe profile table */
+	rte_sched_subport_config_pipe_profile_table(s, params, port->rate);
+
 	/* Bitmap */
 	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
 	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
@@ -1242,6 +1269,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	struct rte_sched_subport *s;
 	struct rte_sched_pipe *p;
 	struct rte_sched_pipe_profile *params;
+	uint32_t n_subports = subport_id + 1;
 	uint32_t deactivate, profile, i;
 
 	/* Check user parameters */
@@ -1257,34 +1285,32 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter subport id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (pipe_id >= port->n_pipes_per_subport) {
+	s = port->subports[subport_id];
+	if (pipe_id >= s->n_pipes_per_subport) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (!deactivate && profile >= port->n_pipe_profiles) {
+	if (!deactivate && profile >= s->n_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe profile\n", __func__);
-		return -EINVAL;
-	}
 
-	/* Check that subport configuration is valid */
-	s = port->subport + subport_id;
-	if (s->tb_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Subport configuration invalid\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	p = port->pipe + (subport_id * port->n_pipes_per_subport + pipe_id);
-
 	/* Handle the case when pipe already has a valid configuration */
+	p = s->pipe + pipe_id;
 	if (p->tb_time) {
-		params = port->pipe_profiles + p->profile;
+		params = s->pipe_profiles + p->profile;
 
 		double subport_tc_be_rate =
 			(double) s->tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASS_BE]
@@ -1314,7 +1340,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 
 	/* Apply the new pipe configuration */
 	p->profile = profile;
-	params = port->pipe_profiles + p->profile;
+	params = s->pipe_profiles + p->profile;
 
 	/* Token Bucket (TB) */
 	p->tb_time = port->time;
@@ -1324,7 +1350,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	p->tc_time = port->time + params->tc_period;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (s->qsize[i])
 			p->tc_credits[i] = params->tc_credits_per_period[i];
 
 	{
@@ -1355,9 +1381,11 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 
 int
 rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_pipe_profile *pp;
 	uint32_t i;
 	int status;
@@ -1369,40 +1397,49 @@ rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
+	/* Subport id not exceeds the max limit */
+	if (subport_id > port->n_subports_per_port) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for subport id\n", __func__);
+		return -EINVAL;
+	}
+
+	s = port->subports[subport_id];
+
 	/* Pipe profiles not exceeds the max limit */
-	if (port->n_pipe_profiles >= port->n_max_pipe_profiles) {
+	if (s->n_pipe_profiles >= port->n_max_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Number of pipe profiles exceeds the max limit\n", __func__);
 		return -EINVAL;
 	}
 
 	/* Pipe params */
-	status = pipe_profile_check(params, port->rate, &port->qsize[0]);
+	status = pipe_profile_check(params, port->rate, &s->qsize[0]);
 	if (status != 0) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Pipe profile check failed(%d)\n", __func__, status);
 		return -EINVAL;
 	}
 
-	pp = &port->pipe_profiles[port->n_pipe_profiles];
-	rte_sched_pipe_profile_convert(port, params, pp, port->rate);
+	pp = &s->pipe_profiles[s->n_pipe_profiles];
+	rte_sched_pipe_profile_convert(s, params, pp, port->rate);
 
 	/* Pipe profile not exists */
-	for (i = 0; i < port->n_pipe_profiles; i++)
-		if (memcmp(port->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
+	for (i = 0; i < s->n_pipe_profiles; i++)
+		if (memcmp(s->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
 			RTE_LOG(ERR, SCHED,
 				"%s: Pipe profile doesn't exist\n", __func__);
 			return -EINVAL;
 		}
 
 	/* Pipe profile commit */
-	*pipe_profile_id = port->n_pipe_profiles;
-	port->n_pipe_profiles++;
+	*pipe_profile_id = s->n_pipe_profiles;
+	s->n_pipe_profiles++;
 
-	if (port->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
-		port->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+	if (s->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
+		s->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
 
-	rte_sched_port_log_pipe_profile(port, *pipe_profile_id);
+	rte_sched_port_log_pipe_profile(s, *pipe_profile_id);
 
 	return 0;
 }
lib/librte_sched/rte_sched.h +4 -1
index 48be702bc..56e9fbfa7 100644
@@ -308,6 +308,8 @@ rte_sched_port_free(struct rte_sched_port *port);
  *
  * @param port
  *   Handle to port scheduler instance
+ * @param subport_id
+ *   Subport ID
  * @param params
  *   Pipe profile parameters
  * @param pipe_profile_id
@@ -318,6 +320,7 @@ rte_sched_port_free(struct rte_sched_port *port);
 __rte_experimental
 int
 rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id);
 
@@ -348,7 +351,7 @@ rte_sched_subport_config(struct rte_sched_port *port,
  * @param pipe_id
  *   Pipe ID within subport
  * @param pipe_profile
- *   ID of port-level pre-configured pipe profile
+ *   ID of subport-level pre-configured pipe profile
  * @return
  *   0 upon success, error code otherwise
  */
Signature · 1 line
2.21.0
#22 lore raw

PATCH v2 6/15 sched: modify pkt enqueue for config flexibility

In reply to Jasvinder Singh

Modify scheduler packet enqueue operation of the scheduler to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 278 ++++++++++++++++++++++-------------
 1 file changed, 178 insertions(+), 100 deletions(-)
lib/librte_sched/rte_sched.c +178 -100
index 6bafbd8fe..a5e4c45b4 100644
@@ -1452,10 +1452,10 @@ rte_sched_port_qindex(struct rte_sched_port *port,
 	uint32_t queue)
 {
 	return ((subport & (port->n_subports_per_port - 1)) <<
-			(port->n_pipes_per_subport_log2 + 4)) |
-			((pipe & (port->n_pipes_per_subport - 1)) << 4) |
-			((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
-			(RTE_SCHED_QUEUES_PER_PIPE - 1));
+		(port->n_max_pipes_per_subport_log2 + 4)) |
+		((pipe & (port->subports[subport]->n_pipes_per_subport - 1)) << 4) |
+		((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
+		(RTE_SCHED_QUEUES_PER_PIPE - 1));
 }
 
 void
@@ -1479,8 +1479,9 @@ rte_sched_port_pkt_read_tree_path(struct rte_sched_port *port,
 {
 	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
 
-	*subport = queue_id >> (port->n_pipes_per_subport_log2 + 4);
-	*pipe = (queue_id >> 4) & (port->n_pipes_per_subport - 1);
+	*subport = queue_id >> (port->n_max_pipes_per_subport_log2 + 4);
+	*pipe = (queue_id >> 4) &
+		(port->subports[*subport]->n_pipes_per_subport - 1);
 	*traffic_class = rte_sched_port_pipe_tc(port, queue_id);
 	*queue = rte_sched_port_tc_queue(port, queue_id);
 }
@@ -1524,7 +1525,7 @@ rte_sched_subport_read_stats(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
-	s = port->subport + subport_id;
+	s = port->subports[subport_id];
 
 	/* Copy subport stats and clear */
 	memcpy(stats, &s->stats, sizeof(struct rte_sched_subport_stats));
@@ -1597,43 +1598,50 @@ rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline void
-rte_sched_port_update_subport_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_subport_stats(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc[tc_index] += 1;
-	s->stats.n_bytes_tc[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc[tc_index] += 1;
+	subport->stats.n_bytes_tc[tc_index] += pkt_len;
 }
 
 #ifdef RTE_SCHED_RED
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc_dropped[tc_index] += 1;
-	s->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc_dropped[tc_index] += 1;
+	subport->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
 #ifdef RTE_SCHED_RED
-	s->stats.n_pkts_red_dropped[tc_index] += red;
+	subport->stats.n_pkts_red_dropped[tc_index] += red;
 #endif
 }
 
 static inline void
-rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_queue_stats(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts += 1;
@@ -1642,17 +1650,19 @@ rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex,
 
 #ifdef RTE_SCHED_RED
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts_dropped += 1;
@@ -1667,7 +1677,11 @@ rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
 #ifdef RTE_SCHED_RED
 
 static inline int
-rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint32_t qindex, uint16_t qlen)
+rte_sched_port_red_drop(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt,
+	uint32_t qindex,
+	uint16_t qlen)
 {
 	struct rte_sched_queue_extra *qe;
 	struct rte_red_config *red_cfg;
@@ -1677,12 +1691,12 @@ rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint3
 
 	tc_index = rte_sched_port_pipe_tc(port, qindex);
 	color = rte_sched_port_pkt_read_color(pkt);
-	red_cfg = &port->red_config[tc_index][color];
+	red_cfg = &subport->red_config[tc_index][color];
 
 	if ((red_cfg->min_th | red_cfg->max_th) == 0)
 		return 0;
 
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + qindex;
 	red = &qe->red;
 
 	return rte_red_enqueue(red_cfg, red, qlen, port->time);
@@ -1699,7 +1713,14 @@ rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t q
 
 #else
 
-#define rte_sched_port_red_drop(port, pkt, qindex, qlen)             0
+static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unused,
+	struct rte_sched_subport *subport __rte_unused,
+	struct rte_mbuf *pkt __rte_unused,
+	uint32_t qindex __rte_unused,
+	uint16_t qlen __rte_unused)
+{
+	return 0;
+}
 
 #define rte_sched_port_set_queue_empty_timestamp(port, qindex)
 
@@ -1734,63 +1755,79 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 
 #endif /* RTE_SCHED_DEBUG */
 
+static inline struct rte_sched_subport *
+rte_sched_port_subport(struct rte_sched_port *port,
+	struct rte_mbuf *pkt)
+{
+	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_id = queue_id >> (port->n_max_pipes_per_subport_log2 + 4);
+
+	return port->subports[subport_id];
+}
+
 static inline uint32_t
-rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_port *port,
-				       struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt, uint32_t subport_qmask)
 {
 	struct rte_sched_queue *q;
 #ifdef RTE_SCHED_COLLECT_STATS
 	struct rte_sched_queue_extra *qe;
 #endif
 	uint32_t qindex = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_queue_id = subport_qmask & qindex;
 
-	q = port->queue + qindex;
+	q = subport->queue + subport_queue_id;
 	rte_prefetch0(q);
 #ifdef RTE_SCHED_COLLECT_STATS
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + subport_queue_id;
 	rte_prefetch0(qe);
 #endif
 
-	return qindex;
+	return subport_queue_id;
 }
 
 static inline void
 rte_sched_port_enqueue_qwa_prefetch0(struct rte_sched_port *port,
-				     uint32_t qindex, struct rte_mbuf **qbase)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase)
 {
 	struct rte_sched_queue *q;
 	struct rte_mbuf **q_qw;
 	uint16_t qsize;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	q_qw = qbase + (q->qw & (qsize - 1));
 
 	rte_prefetch0(q_qw);
-	rte_bitmap_prefetch0(port->bmp, qindex);
+	rte_bitmap_prefetch0(subport->bmp, qindex);
 }
 
 static inline int
-rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
-			   struct rte_mbuf **qbase, struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qwa(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase,
+	struct rte_mbuf *pkt)
 {
 	struct rte_sched_queue *q;
 	uint16_t qsize;
 	uint16_t qlen;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	qlen = q->qw - q->qr;
 
 	/* Drop the packet (and update drop stats) when queue is full */
-	if (unlikely(rte_sched_port_red_drop(port, pkt, qindex, qlen) ||
+	if (unlikely(rte_sched_port_red_drop(port, subport, pkt, qindex, qlen) ||
 		     (qlen >= qsize))) {
 		rte_pktmbuf_free(pkt);
 #ifdef RTE_SCHED_COLLECT_STATS
-		rte_sched_port_update_subport_stats_on_drop(port, qindex, pkt,
-							    qlen < qsize);
-		rte_sched_port_update_queue_stats_on_drop(port, qindex, pkt,
-							  qlen < qsize);
+		rte_sched_port_update_subport_stats_on_drop(port, subport,
+			qindex, pkt, qlen < qsize);
+		rte_sched_port_update_queue_stats_on_drop(subport, qindex, pkt,
+			qlen < qsize);
 #endif
 		return 0;
 	}
@@ -1799,13 +1836,13 @@ rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
 	qbase[q->qw & (qsize - 1)] = pkt;
 	q->qw++;
 
-	/* Activate queue in the port bitmap */
-	rte_bitmap_set(port->bmp, qindex);
+	/* Activate queue in the subport bitmap */
+	rte_bitmap_set(subport->bmp, qindex);
 
 	/* Statistics */
 #ifdef RTE_SCHED_COLLECT_STATS
-	rte_sched_port_update_subport_stats(port, qindex, pkt);
-	rte_sched_port_update_queue_stats(port, qindex, pkt);
+	rte_sched_port_update_subport_stats(port, subport, qindex, pkt);
+	rte_sched_port_update_queue_stats(subport, qindex, pkt);
 #endif
 
 	return 1;
@@ -1833,17 +1870,22 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		*pkt30, *pkt31, *pkt_last;
 	struct rte_mbuf **q00_base, **q01_base, **q10_base, **q11_base,
 		**q20_base, **q21_base, **q30_base, **q31_base, **q_last_base;
+	struct rte_sched_subport *subport00, *subport01, *subport10, *subport11,
+		*subport20, *subport21, *subport30, *subport31, *subport_last;
 	uint32_t q00, q01, q10, q11, q20, q21, q30, q31, q_last;
 	uint32_t r00, r01, r10, r11, r20, r21, r30, r31, r_last;
+	uint32_t subport_qmask;
 	uint32_t result, i;
 
 	result = 0;
+	subport_qmask = (1 << (port->n_max_pipes_per_subport_log2 + 4)) - 1;
 
 	/*
 	 * Less then 6 input packets available, which is not enough to
 	 * feed the pipeline
 	 */
 	if (unlikely(n_pkts < 6)) {
+		struct rte_sched_subport *subports[5];
 		struct rte_mbuf **q_base[5];
 		uint32_t q[5];
 
@@ -1851,22 +1893,26 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		for (i = 0; i < n_pkts; i++)
 			rte_prefetch0(pkts[i]);
 
+		/* Prefetch the subport structure for each packet */
+		for (i = 0; i < n_pkts; i++)
+			subports[i] = rte_sched_port_subport(port, pkts[i]);
+
 		/* Prefetch the queue structure for each queue */
 		for (i = 0; i < n_pkts; i++)
-			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(port,
-								      pkts[i]);
+			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(subports[i],
+					pkts[i], subport_qmask);
 
 		/* Prefetch the write pointer location of each queue */
 		for (i = 0; i < n_pkts; i++) {
-			q_base[i] = rte_sched_port_qbase(port, q[i]);
-			rte_sched_port_enqueue_qwa_prefetch0(port, q[i],
-							     q_base[i]);
+			q_base[i] = rte_sched_subport_pipe_qbase(subports[i], q[i]);
+			rte_sched_port_enqueue_qwa_prefetch0(port, subports[i],
+				q[i], q_base[i]);
 		}
 
 		/* Write each packet to its queue */
 		for (i = 0; i < n_pkts; i++)
-			result += rte_sched_port_enqueue_qwa(port, q[i],
-							     q_base[i], pkts[i]);
+			result += rte_sched_port_enqueue_qwa(port, subports[i],
+						q[i], q_base[i], pkts[i]);
 
 		return result;
 	}
@@ -1882,21 +1928,29 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	rte_prefetch0(pkt10);
 	rte_prefetch0(pkt11);
 
-	q20 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt20);
-	q21 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt21);
+	subport20 = rte_sched_port_subport(port, pkt20);
+	subport21 = rte_sched_port_subport(port, pkt21);
+	q20 = rte_sched_port_enqueue_qptrs_prefetch0(subport20,
+			pkt20, subport_qmask);
+	q21 = rte_sched_port_enqueue_qptrs_prefetch0(subport21,
+			pkt21, subport_qmask);
 
 	pkt00 = pkts[4];
 	pkt01 = pkts[5];
 	rte_prefetch0(pkt00);
 	rte_prefetch0(pkt01);
 
-	q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-	q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+	subport10 = rte_sched_port_subport(port, pkt10);
+	subport11 = rte_sched_port_subport(port, pkt11);
+	q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+			pkt10, subport_qmask);
+	q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+			pkt11, subport_qmask);
 
-	q20_base = rte_sched_port_qbase(port, q20);
-	q21_base = rte_sched_port_qbase(port, q21);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+	q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+	q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 	/* Run the pipeline */
 	for (i = 6; i < (n_pkts & (~1)); i += 2) {
@@ -1911,6 +1965,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		q31 = q21;
 		q20 = q10;
 		q21 = q11;
+		subport30 = subport20;
+		subport31 = subport21;
+		subport20 = subport10;
+		subport21 = subport11;
 		q30_base = q20_base;
 		q31_base = q21_base;
 
@@ -1920,19 +1978,25 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		rte_prefetch0(pkt00);
 		rte_prefetch0(pkt01);
 
-		/* Stage 1: Prefetch queue structure storing queue pointers */
-		q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-		q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+		/* Stage 1: Prefetch subport and queue structure storing queue pointers */
+		subport10 = rte_sched_port_subport(port, pkt10);
+		subport11 = rte_sched_port_subport(port, pkt11);
+		q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+				pkt10, subport_qmask);
+		q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+				pkt11, subport_qmask);
 
 		/* Stage 2: Prefetch queue write location */
-		q20_base = rte_sched_port_qbase(port, q20);
-		q21_base = rte_sched_port_qbase(port, q21);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+		q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+		q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 		/* Stage 3: Write packet to queue and activate queue */
-		r30 = rte_sched_port_enqueue_qwa(port, q30, q30_base, pkt30);
-		r31 = rte_sched_port_enqueue_qwa(port, q31, q31_base, pkt31);
+		r30 = rte_sched_port_enqueue_qwa(port, subport30,
+				q30, q30_base, pkt30);
+		r31 = rte_sched_port_enqueue_qwa(port, subport31,
+				q31, q31_base, pkt31);
 		result += r30 + r31;
 	}
 
@@ -1944,38 +2008,52 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	pkt_last = pkts[n_pkts - 1];
 	rte_prefetch0(pkt_last);
 
-	q00 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt00);
-	q01 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt01);
-
-	q10_base = rte_sched_port_qbase(port, q10);
-	q11_base = rte_sched_port_qbase(port, q11);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q10, q10_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q11, q11_base);
-
-	r20 = rte_sched_port_enqueue_qwa(port, q20, q20_base, pkt20);
-	r21 = rte_sched_port_enqueue_qwa(port, q21, q21_base, pkt21);
+	subport00 = rte_sched_port_subport(port, pkt00);
+	subport01 = rte_sched_port_subport(port, pkt01);
+	q00 = rte_sched_port_enqueue_qptrs_prefetch0(subport00,
+			pkt00, subport_qmask);
+	q01 = rte_sched_port_enqueue_qptrs_prefetch0(subport01,
+			pkt01, subport_qmask);
+
+	q10_base = rte_sched_subport_pipe_qbase(subport10, q10);
+	q11_base = rte_sched_subport_pipe_qbase(subport11, q11);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport10, q10, q10_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport11, q11, q11_base);
+
+	r20 = rte_sched_port_enqueue_qwa(port, subport20,
+			q20, q20_base, pkt20);
+	r21 = rte_sched_port_enqueue_qwa(port, subport21,
+			q21, q21_base, pkt21);
 	result += r20 + r21;
 
-	q_last = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt_last);
+	subport_last = rte_sched_port_subport(port, pkt_last);
+	q_last = rte_sched_port_enqueue_qptrs_prefetch0(subport_last,
+				pkt_last, subport_qmask);
 
-	q00_base = rte_sched_port_qbase(port, q00);
-	q01_base = rte_sched_port_qbase(port, q01);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q00, q00_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q01, q01_base);
+	q00_base = rte_sched_subport_pipe_qbase(subport00, q00);
+	q01_base = rte_sched_subport_pipe_qbase(subport01, q01);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport00, q00, q00_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport01, q01, q01_base);
 
-	r10 = rte_sched_port_enqueue_qwa(port, q10, q10_base, pkt10);
-	r11 = rte_sched_port_enqueue_qwa(port, q11, q11_base, pkt11);
+	r10 = rte_sched_port_enqueue_qwa(port, subport10, q10,
+			q10_base, pkt10);
+	r11 = rte_sched_port_enqueue_qwa(port, subport11, q11,
+			q11_base, pkt11);
 	result += r10 + r11;
 
-	q_last_base = rte_sched_port_qbase(port, q_last);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q_last, q_last_base);
+	q_last_base = rte_sched_subport_pipe_qbase(subport_last, q_last);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport_last,
+		q_last, q_last_base);
 
-	r00 = rte_sched_port_enqueue_qwa(port, q00, q00_base, pkt00);
-	r01 = rte_sched_port_enqueue_qwa(port, q01, q01_base, pkt01);
+	r00 = rte_sched_port_enqueue_qwa(port, subport00, q00,
+			q00_base, pkt00);
+	r01 = rte_sched_port_enqueue_qwa(port, subport01, q01,
+			q01_base, pkt01);
 	result += r00 + r01;
 
 	if (n_pkts & 1) {
-		r_last = rte_sched_port_enqueue_qwa(port, q_last, q_last_base, pkt_last);
+		r_last = rte_sched_port_enqueue_qwa(port, subport_last,
+					q_last,	q_last_base, pkt_last);
 		result += r_last;
 	}
 
Signature · 1 line
2.21.0
#23 lore raw

PATCH v2 7/15 sched: update memory compute to support flexiblity

In reply to Jasvinder Singh

Update memory footprint compute function for allowing subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 102 +++++++++++------------------------
 lib/librte_sched/rte_sched.h |   8 +--
 2 files changed, 36 insertions(+), 74 deletions(-)
lib/librte_sched/rte_sched.c +31 -71
index a5e4c45b4..f8d13c66d 100644
@@ -495,70 +495,6 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 	return 0;
 }
 
-static uint32_t
-rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sched_port_array array)
-{
-	uint32_t n_subports_per_port = params->n_subports_per_port;
-	uint32_t n_pipes_per_subport = params->n_pipes_per_subport;
-	uint32_t n_pipes_per_port = n_pipes_per_subport * n_subports_per_port;
-	uint32_t n_queues_per_port = RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport * n_subports_per_port;
-
-	uint32_t size_subport = n_subports_per_port * sizeof(struct rte_sched_subport);
-	uint32_t size_pipe = n_pipes_per_port * sizeof(struct rte_sched_pipe);
-	uint32_t size_queue = n_queues_per_port * sizeof(struct rte_sched_queue);
-	uint32_t size_queue_extra
-		= n_queues_per_port * sizeof(struct rte_sched_queue_extra);
-	uint32_t size_pipe_profiles
-		= params->n_max_pipe_profiles * sizeof(struct rte_sched_pipe_profile);
-	uint32_t size_bmp_array = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	uint32_t size_per_pipe_queue_array, size_queue_array;
-
-	uint32_t base, i;
-
-	size_per_pipe_queue_array = 0;
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
-			size_per_pipe_queue_array +=
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-		else
-			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-	}
-	size_queue_array = n_pipes_per_port * size_per_pipe_queue_array;
-
-	base = 0;
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_SUBPORT)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_subport);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
-
-	return base;
-}
-
 static uint32_t
 rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
 	uint32_t n_max_pipe_profiles,
@@ -880,22 +816,46 @@ rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 }
 
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params)
 {
-	uint32_t size0, size1;
+	uint32_t size0 = 0, size1 = 0, i;
 	int status;
 
-	status = rte_sched_port_check_params(params);
+	status = rte_sched_port_check_params(port_params);
 	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler port params check failed (%d)\n",
+			__func__, status);
 
 		return 0;
 	}
 
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		status = rte_sched_subport_check_params(sp,
+				port_params->n_max_pipes_per_subport,
+				port_params->n_max_pipe_profiles,
+				port_params->rate);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Port scheduler subport params check failed (%d)\n",
+				__func__, status);
+
+			return 0;
+		}
+	}
+
 	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params,
-			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		size1 += rte_sched_subport_get_array_base(sp,
+					port_params->n_max_pipe_profiles,
+					e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
+	}
 
 	return size0 + size1;
 }
lib/librte_sched/rte_sched.h +5 -3
index 56e9fbfa7..ea2b07448 100644
@@ -364,14 +364,16 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 /**
  * Hierarchical scheduler memory footprint size per port
  *
- * @param params
+ * @param port_params
  *   Port scheduler configuration parameter structure
+ * @param subport_params
+ *   Array of subport parameter structures
  * @return
  *   Memory footprint size in bytes upon success, 0 otherwise
  */
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params);
-
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params);
 /*
  * Statistics
  *
Signature · 1 line
2.21.0
#24 lore raw

PATCH v2 8/15 sched: update grinder functions for config flexibility

In reply to Jasvinder Singh

Modify packet grinder functions of the schedule to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 223 +++++++++++++++++------------------
 1 file changed, 106 insertions(+), 117 deletions(-)
lib/librte_sched/rte_sched.c +106 -117
index f8d13c66d..0451e10ea 100644
@@ -324,24 +324,6 @@ rte_sched_port_queues_per_port(struct rte_sched_port *port)
 	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
 }
 
-static inline struct rte_mbuf **
-rte_sched_port_qbase(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t pindex = qindex >> 4;
-	uint32_t qpos = qindex & 0xF;
-
-	return (port->queue_array + pindex *
-		port->qsize_sum + port->qsize_add[qpos]);
-}
-
-static inline uint16_t
-rte_sched_port_qsize(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
-
-	return port->qsize[tc];
-}
-
 static inline uint16_t
 rte_sched_port_pipe_queue(struct rte_sched_port *port, uint32_t traffic_class)
 {
@@ -1546,9 +1528,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 #ifdef RTE_SCHED_DEBUG
 
 static inline int
-rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_queue_is_empty(struct rte_sched_subport *subport,
+	uint32_t qindex)
 {
-	struct rte_sched_queue *queue = port->queue + qindex;
+	struct rte_sched_queue *queue = subport->queue + qindex;
 
 	return queue->qr == queue->qw;
 }
@@ -1663,9 +1646,10 @@ rte_sched_port_red_drop(struct rte_sched_port *port,
 }
 
 static inline void
-rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t qindex)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	struct rte_red *red = &qe->red;
 
 	rte_red_mark_queue_empty(red, port->time);
@@ -1682,14 +1666,14 @@ static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unus
 	return 0;
 }
 
-#define rte_sched_port_set_queue_empty_timestamp(port, qindex)
+#define rte_sched_port_set_queue_empty_timestamp(port, subport, qindex)
 
 #endif /* RTE_SCHED_RED */
 
 #ifdef RTE_SCHED_DEBUG
 
 static inline void
-debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
+debug_check_queue_slab(struct rte_sched_subport *subport, uint32_t bmp_pos,
 		       uint64_t bmp_slab)
 {
 	uint64_t mask;
@@ -1701,7 +1685,7 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 	panic = 0;
 	for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 		if (mask & bmp_slab) {
-			if (rte_sched_port_queue_is_empty(port, bmp_pos + i)) {
+			if (rte_sched_port_queue_is_empty(subport, bmp_pos + i)) {
 				printf("Queue %u (slab offset %u) is empty\n", bmp_pos + i, i);
 				panic = 1;
 			}
@@ -2023,10 +2007,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2064,10 +2048,9 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline uint32_t
-grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_tc_ov_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
 	uint32_t tc_ov_consumption[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t tc_consumption = 0, tc_ov_consumption_max;
 	uint32_t tc_ov_wm = subport->tc_ov_wm;
@@ -2107,10 +2090,10 @@ grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2130,7 +2113,7 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 
 	/* Subport TCs */
 	if (unlikely(port->time >= subport->tc_time)) {
-		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, pos);
+		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, subport);
 
 		for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 			subport->tc_credits[i] = subport->tc_credits_per_period[i];
@@ -2160,10 +2143,10 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2195,10 +2178,10 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2243,15 +2226,16 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline int
-grinder_schedule(struct rte_sched_port *port, uint32_t pos)
+grinder_schedule(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_queue *queue = grinder->queue[grinder->qpos];
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t pkt_len = pkt->pkt_len + port->frame_overhead;
 	uint32_t be_tc_active;
 
-	if (!grinder_credits_check(port, pos))
+	if (!grinder_credits_check(port, subport, pos))
 		return 0;
 
 	/* Advance port time */
@@ -2268,15 +2252,15 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 	if (queue->qr == queue->qw) {
 		uint32_t qindex = grinder->qindex[grinder->qpos];
 
-		rte_bitmap_clear(port->bmp, qindex);
+		rte_bitmap_clear(subport->bmp, qindex);
 		grinder->qmask &= ~(1 << grinder->qpos);
 		if (be_tc_active)
 			grinder->wrr_mask[grinder->qpos] = 0;
-		rte_sched_port_set_queue_empty_timestamp(port, qindex);
+		rte_sched_port_set_queue_empty_timestamp(port, subport, qindex);
 	}
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	grinder->productive = 1;
 
 	return 1;
@@ -2285,13 +2269,13 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 #ifdef SCHED_VECTOR_SSE4
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	__m128i index = _mm_set1_epi32(base_pipe);
-	__m128i pipes = _mm_load_si128((__m128i *)port->grinder_base_bmp_pos);
+	__m128i pipes = _mm_load_si128((__m128i *)subport->grinder_base_bmp_pos);
 	__m128i res = _mm_cmpeq_epi32(pipes, index);
 
-	pipes = _mm_load_si128((__m128i *)(port->grinder_base_bmp_pos + 4));
+	pipes = _mm_load_si128((__m128i *)(subport->grinder_base_bmp_pos + 4));
 	pipes = _mm_cmpeq_epi32(pipes, index);
 	res = _mm_or_si128(res, pipes);
 
@@ -2304,10 +2288,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #elif defined(SCHED_VECTOR_NEON)
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32x4_t index, pipes;
-	uint32_t *pos = (uint32_t *)port->grinder_base_bmp_pos;
+	uint32_t *pos = (uint32_t *)subport->grinder_base_bmp_pos;
 
 	index = vmovq_n_u32(base_pipe);
 	pipes = vld1q_u32(pos);
@@ -2324,12 +2308,12 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #else
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32_t i;
 
 	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++) {
-		if (port->grinder_base_bmp_pos[i] == base_pipe)
+		if (subport->grinder_base_bmp_pos[i] == base_pipe)
 			return 1;
 	}
 
@@ -2339,9 +2323,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #endif /* RTE_SCHED_OPTIMIZATIONS */
 
 static inline void
-grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
+grinder_pcache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t w[4];
 
 	grinder->pcache_w = 0;
@@ -2370,9 +2355,10 @@ grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_
 }
 
 static inline void
-grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qindex, uint16_t qmask)
+grinder_tccache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t qindex, uint16_t qmask)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint8_t b, i;
 
 	grinder->tccache_w = 0;
@@ -2393,9 +2379,10 @@ grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qin
 }
 
 static inline int
-grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
+grinder_next_tc(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_mbuf **qbase;
 	uint32_t qindex;
 	uint16_t qsize;
@@ -2404,15 +2391,15 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 0;
 
 	qindex = grinder->tccache_qindex[grinder->tccache_r];
-	qbase = rte_sched_port_qbase(port, qindex);
-	qsize = rte_sched_port_qsize(port, qindex);
+	qbase = rte_sched_subport_pipe_qbase(subport, qindex);
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 
 	grinder->tc_index = rte_sched_port_pipe_tc(port, qindex);
 	grinder->qmask = grinder->tccache_qmask[grinder->tccache_r];
 	grinder->qsize = qsize;
 
 	if (grinder->tc_index < RTE_SCHED_TRAFFIC_CLASS_BE) {
-		grinder->queue[0] = port->queue + qindex;
+		grinder->queue[0] = subport->queue + qindex;
 		grinder->qbase[0] = qbase;
 		grinder->qindex[0] = qindex;
 		grinder->tccache_r++;
@@ -2420,10 +2407,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 1;
 	}
 
-	grinder->queue[0] = port->queue + qindex;
-	grinder->queue[1] = port->queue + qindex + 1;
-	grinder->queue[2] = port->queue + qindex + 2;
-	grinder->queue[3] = port->queue + qindex + 3;
+	grinder->queue[0] = subport->queue + qindex;
+	grinder->queue[1] = subport->queue + qindex + 1;
+	grinder->queue[2] = subport->queue + qindex + 2;
+	grinder->queue[3] = subport->queue + qindex + 3;
 
 	grinder->qbase[0] = qbase;
 	grinder->qbase[1] = qbase + qsize;
@@ -2440,9 +2427,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline int
-grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_next_pipe(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t pipe_qindex;
 	uint16_t pipe_qmask;
 
@@ -2455,22 +2443,22 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 		uint32_t bmp_pos = 0;
 
 		/* Get another non-empty pipe group */
-		if (unlikely(rte_bitmap_scan(port->bmp, &bmp_pos, &bmp_slab) <= 0))
+		if (unlikely(rte_bitmap_scan(subport->bmp, &bmp_pos, &bmp_slab) <= 0))
 			return 0;
 
 #ifdef RTE_SCHED_DEBUG
-		debug_check_queue_slab(port, bmp_pos, bmp_slab);
+		debug_check_queue_slab(subport, bmp_pos, bmp_slab);
 #endif
 
 		/* Return if pipe group already in one of the other grinders */
-		port->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
-		if (unlikely(grinder_pipe_exists(port, bmp_pos)))
+		subport->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
+		if (unlikely(grinder_pipe_exists(subport, bmp_pos)))
 			return 0;
 
-		port->grinder_base_bmp_pos[pos] = bmp_pos;
+		subport->grinder_base_bmp_pos[pos] = bmp_pos;
 
 		/* Install new pipe group into grinder's pipe cache */
-		grinder_pcache_populate(port, pos, bmp_pos, bmp_slab);
+		grinder_pcache_populate(subport, pos, bmp_pos, bmp_slab);
 
 		pipe_qmask = grinder->pcache_qmask[0];
 		pipe_qindex = grinder->pcache_qindex[0];
@@ -2479,18 +2467,18 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 	/* Install new pipe in the grinder */
 	grinder->pindex = pipe_qindex >> 4;
-	grinder->subport = port->subport + (grinder->pindex / port->n_pipes_per_subport);
-	grinder->pipe = port->pipe + grinder->pindex;
+	grinder->subport = subport;
+	grinder->pipe = subport->pipe + grinder->pindex;
 	grinder->pipe_params = NULL; /* to be set after the pipe structure is prefetched */
 	grinder->productive = 0;
 
-	grinder_tccache_populate(port, pos, pipe_qindex, pipe_qmask);
-	grinder_next_tc(port, pos);
+	grinder_tccache_populate(subport, pos, pipe_qindex, pipe_qmask);
+	grinder_next_tc(port, subport, pos);
 
 	/* Check for pipe exhaustion */
-	if (grinder->pindex == port->pipe_loop) {
-		port->pipe_exhaustion = 1;
-		port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	if (grinder->pindex == subport->pipe_loop) {
+		subport->pipe_exhaustion = 1;
+		subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	}
 
 	return 1;
@@ -2498,9 +2486,9 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline void
-grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_load(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *pipe_params = grinder->pipe_params;
 	uint32_t qmask = grinder->qmask;
@@ -2526,9 +2514,9 @@ grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_store(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 
 	pipe->wrr_tokens[0] =
@@ -2546,9 +2534,9 @@ grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t wrr_tokens_min;
 
 	grinder->wrr_tokens[0] |= ~grinder->wrr_mask[0];
@@ -2566,21 +2554,21 @@ grinder_wrr(struct rte_sched_port *port, uint32_t pos)
 }
 
 
-#define grinder_evict(port, pos)
+#define grinder_evict(subport, pos)
 
 static inline void
-grinder_prefetch_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_pipe(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	rte_prefetch0(grinder->pipe);
 	rte_prefetch0(grinder->queue[0]);
 }
 
 static inline void
-grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_tc_queue_arrays(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t qsize, qr[RTE_SCHED_MAX_QUEUES_PER_TC];
 
 	qsize = grinder->qsize;
@@ -2601,17 +2589,17 @@ grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
 	rte_prefetch0(grinder->qbase[0] + qr[0]);
 	rte_prefetch0(grinder->qbase[1] + qr[1]);
 
-	grinder_wrr_load(port, pos);
-	grinder_wrr(port, pos);
+	grinder_wrr_load(subport, pos);
+	grinder_wrr(subport, pos);
 
 	rte_prefetch0(grinder->qbase[2] + qr[2]);
 	rte_prefetch0(grinder->qbase[3] + qr[3]);
 }
 
 static inline void
-grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t qpos = grinder->qpos;
 	struct rte_mbuf **qbase = grinder->qbase[qpos];
 	uint16_t qsize = grinder->qsize;
@@ -2630,14 +2618,15 @@ grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
 static inline uint32_t
 grinder_handle(struct rte_sched_port *port, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_subport *subport = port->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
 	case e_GRINDER_PREFETCH_PIPE:
 	{
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
-			port->busy_grinders++;
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
+			subport->busy_grinders++;
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return 0;
@@ -2650,9 +2639,9 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		struct rte_sched_pipe *pipe = grinder->pipe;
 
-		grinder->pipe_params = port->pipe_profiles + pipe->profile;
-		grinder_prefetch_tc_queue_arrays(port, pos);
-		grinder_credits_update(port, pos);
+		grinder->pipe_params = subport->pipe_profiles + pipe->profile;
+		grinder_prefetch_tc_queue_arrays(subport, pos);
+		grinder_credits_update(port, subport, pos);
 
 		grinder->state = e_GRINDER_PREFETCH_MBUF;
 		return 0;
@@ -2660,7 +2649,7 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 
 	case e_GRINDER_PREFETCH_MBUF:
 	{
-		grinder_prefetch_mbuf(port, pos);
+		grinder_prefetch_mbuf(subport, pos);
 
 		grinder->state = e_GRINDER_READ_MBUF;
 		return 0;
@@ -2670,47 +2659,47 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		uint32_t wrr_active, result = 0;
 
-		result = grinder_schedule(port, pos);
+		result = grinder_schedule(port, subport, pos);
 
 		wrr_active = (grinder->tc_index == RTE_SCHED_TRAFFIC_CLASS_BE);
 
 		/* Look for next packet within the same TC */
 		if (result && grinder->qmask) {
 			if (wrr_active)
-				grinder_wrr(port, pos);
+				grinder_wrr(subport, pos);
 
-			grinder_prefetch_mbuf(port, pos);
+			grinder_prefetch_mbuf(subport, pos);
 
 			return 1;
 		}
 
 		if (wrr_active)
-			grinder_wrr_store(port, pos);
+			grinder_wrr_store(subport, pos);
 
 		/* Look for another active TC within same pipe */
-		if (grinder_next_tc(port, pos)) {
-			grinder_prefetch_tc_queue_arrays(port, pos);
+		if (grinder_next_tc(port, subport, pos)) {
+			grinder_prefetch_tc_queue_arrays(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_MBUF;
 			return result;
 		}
 
 		if (grinder->productive == 0 &&
-		    port->pipe_loop == RTE_SCHED_PIPE_INVALID)
-			port->pipe_loop = grinder->pindex;
+		    subport->pipe_loop == RTE_SCHED_PIPE_INVALID)
+			subport->pipe_loop = grinder->pindex;
 
-		grinder_evict(port, pos);
+		grinder_evict(subport, pos);
 
 		/* Look for another active pipe */
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return result;
 		}
 
 		/* No active pipe found */
-		port->busy_grinders--;
+		subport->busy_grinders--;
 
 		grinder->state = e_GRINDER_PREFETCH_PIPE;
 		return result;
Signature · 1 line
2.21.0
#25 lore raw

PATCH v2 9/15 sched: update pkt dequeue for flexible config

In reply to Jasvinder Singh

Modify scheduler packet dequeue operation to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 51 ++++++++++++++++++++++++++++--------
 1 file changed, 40 insertions(+), 11 deletions(-)
lib/librte_sched/rte_sched.c +40 -11
index 0451e10ea..e0ef86f40 100644
@@ -245,6 +245,7 @@ struct rte_sched_port {
 	uint32_t busy_grinders;
 	struct rte_mbuf **pkts_out;
 	uint32_t n_pkts_out;
+	uint32_t subport_id;
 
 	/* Queue base calculation */
 	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
@@ -911,6 +912,7 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	/* Grinders */
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
+	port->subport_id = 0;
 
 	return port;
 }
@@ -2616,9 +2618,9 @@ grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 }
 
 static inline uint32_t
-grinder_handle(struct rte_sched_port *port, uint32_t pos)
+grinder_handle(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_subport *subport = port->subport;
 	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
@@ -2717,6 +2719,7 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 	uint64_t cycles = rte_get_tsc_cycles();
 	uint64_t cycles_diff = cycles - port->time_cpu_cycles;
 	uint64_t bytes_diff;
+	uint32_t i;
 
 	/* Compute elapsed time in bytes */
 	bytes_diff = rte_reciprocal_divide(cycles_diff << RTE_SCHED_TIME_SHIFT,
@@ -2729,20 +2732,21 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 		port->time = port->time_cpu_bytes;
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	for (i = 0; i < port->n_subports_per_port; i++)
+		port->subports[i]->pipe_loop = RTE_SCHED_PIPE_INVALID;
 }
 
 static inline int
-rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
+rte_sched_port_exceptions(struct rte_sched_subport *subport, int second_pass)
 {
 	int exceptions;
 
 	/* Check if any exception flag is set */
-	exceptions = (second_pass && port->busy_grinders == 0) ||
-		(port->pipe_exhaustion == 1);
+	exceptions = (second_pass && subport->busy_grinders == 0) ||
+		(subport->pipe_exhaustion == 1);
 
 	/* Clear exception flags */
-	port->pipe_exhaustion = 0;
+	subport->pipe_exhaustion = 0;
 
 	return exceptions;
 }
@@ -2750,7 +2754,9 @@ rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
 int
 rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts)
 {
-	uint32_t i, count;
+	struct rte_sched_subport *subport;
+	uint32_t subport_id = port->subport_id;
+	uint32_t i, n_subports = 0, count;
 
 	port->pkts_out = pkts;
 	port->n_pkts_out = 0;
@@ -2759,9 +2765,32 @@ rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint
 
 	/* Take each queue in the grinder one step further */
 	for (i = 0, count = 0; ; i++)  {
-		count += grinder_handle(port, i & (RTE_SCHED_PORT_N_GRINDERS - 1));
-		if ((count == n_pkts) ||
-		    rte_sched_port_exceptions(port, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+		subport = port->subports[subport_id];
+
+		count += grinder_handle(port, subport,
+				i & (RTE_SCHED_PORT_N_GRINDERS - 1));
+
+		if (count == n_pkts) {
+			subport_id++;
+
+			if (subport_id == port->n_subports_per_port)
+				subport_id = 0;
+
+			port->subport_id = subport_id;
+			break;
+		}
+
+		if (rte_sched_port_exceptions(subport, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+			i = 0;
+			subport_id++;
+			n_subports++;
+		}
+
+		if (subport_id == port->n_subports_per_port)
+			subport_id = 0;
+
+		if (n_subports == port->n_subports_per_port) {
+			port->subport_id = subport_id;
 			break;
 		}
 	}
Signature · 1 line
2.21.0
#26 lore raw

PATCH v2 10/15 sched: update queue stats read for config flexibility

In reply to Jasvinder Singh

Modify pipe queue stats read function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 28 +++++++++++++++-------------
 1 file changed, 15 insertions(+), 13 deletions(-)
lib/librte_sched/rte_sched.c +15 -13
index e0ef86f40..8a7727286 100644
@@ -284,16 +284,6 @@ enum rte_sched_subport_array {
 	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
 };
 
-#ifdef RTE_SCHED_COLLECT_STATS
-
-static inline uint32_t
-rte_sched_port_queues_per_subport(struct rte_sched_port *port)
-{
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport;
-}
-
-#endif
-
 static inline uint32_t
 rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
 {
@@ -322,7 +312,12 @@ struct rte_sched_subport *subport, uint32_t qindex)
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
+	uint32_t n_queues = 0, i;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		n_queues += rte_sched_subport_pipe_queues(port->subports[i]);
+
+	return n_queues;
 }
 
 static inline uint16_t
@@ -1487,8 +1482,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 	struct rte_sched_queue_stats *stats,
 	uint16_t *qlen)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_queue *q;
 	struct rte_sched_queue_extra *qe;
+	uint32_t subport_id, subport_qmask, subport_qindex;
 
 	/* Check user parameters */
 	if (port == NULL) {
@@ -1514,8 +1511,13 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 			"%s: Incorrect value for parameter qlen\n", __func__);
 		return -EINVAL;
 	}
-	q = port->queue + queue_id;
-	qe = port->queue_extra + queue_id;
+	subport_qmask = port->n_max_pipes_per_subport_log2 + 4;
+	subport_id = (queue_id >> subport_qmask) & (port->n_subports_per_port - 1);
+
+	s = port->subports[subport_id];
+	subport_qindex = ((1 << subport_qmask) - 1) & queue_id;
+	q = s->queue + subport_qindex;
+	qe = s->queue_extra + subport_qindex;
 
 	/* Copy queue stats and clear */
 	memcpy(stats, &qe->stats, sizeof(struct rte_sched_queue_stats));
Signature · 1 line
2.21.0
#27 lore raw

PATCH v2 11/15 test/sched: modify tests for subport config flexibility

In reply to Jasvinder Singh

Modify tests function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 app/test/test_sched.c | 35 ++++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 17 deletions(-)
app/test/test_sched.c +18 -17
index afe0b0765..6a2896234 100644
@@ -22,18 +22,6 @@
 #define TC              2
 #define QUEUE           0
 
-static struct rte_sched_subport_params subport_param[] = {
-	{
-		.tb_rate = 1250000000,
-		.tb_size = 1000000,
-
-		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000},
-		.tc_period = 10,
-	},
-};
-
 static struct rte_sched_pipe_params pipe_profile[] = {
 	{ /* Profile #0 */
 		.tb_rate = 305175,
@@ -48,16 +36,29 @@ static struct rte_sched_pipe_params pipe_profile[] = {
 	},
 };
 
+static struct rte_sched_subport_params subport_param[] = {
+	{
+		.tb_rate = 1250000000,
+		.tb_size = 1000000,
+
+		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000},
+		.tc_period = 10,
+		.n_pipes_per_subport = 1024,
+		.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
+		.pipe_profiles = pipe_profile,
+		.n_pipe_profiles = 1,
+	},
+};
+
 static struct rte_sched_port_params port_param = {
 	.socket = 0, /* computed */
 	.rate = 0, /* computed */
 	.mtu = 1522,
 	.frame_overhead = RTE_SCHED_FRAME_OVERHEAD_DEFAULT,
 	.n_subports_per_port = 1,
-	.n_pipes_per_subport = 1024,
-	.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
-	.pipe_profiles = pipe_profile,
-	.n_pipe_profiles = 1,
+	.n_max_pipes_per_subport = 1024,
 	.n_max_pipe_profiles = 1,
 };
 
@@ -140,7 +141,7 @@ test_sched(void)
 	err = rte_sched_subport_config(port, SUBPORT, subport_param);
 	TEST_ASSERT_SUCCESS(err, "Error config sched, err=%d\n", err);
 
-	for (pipe = 0; pipe < port_param.n_pipes_per_subport; pipe ++) {
+	for (pipe = 0; pipe < subport_param[0].n_pipes_per_subport; pipe++) {
 		err = rte_sched_pipe_config(port, SUBPORT, pipe, 0);
 		TEST_ASSERT_SUCCESS(err, "Error config sched pipe %u, err=%d\n", pipe, err);
 	}
Signature · 1 line
2.21.0
#28 lore raw

PATCH v2 12/15 net/softnic: add subport config flexibility to TM

In reply to Jasvinder Singh

Modify softnic traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 drivers/net/softnic/rte_eth_softnic_tm.c | 51 ++++++++++++------------
 1 file changed, 26 insertions(+), 25 deletions(-)
drivers/net/softnic/rte_eth_softnic_tm.c +26 -25
index 61c3adc82..8fb728d7d 100644
@@ -85,7 +85,8 @@ softnic_tmgr_port_create(struct pmd_internals *p,
 	/* Subport */
 	n_subports = t->port_params.n_subports_per_port;
 	for (subport_id = 0; subport_id < n_subports; subport_id++) {
-		uint32_t n_pipes_per_subport = t->port_params.n_pipes_per_subport;
+		uint32_t n_pipes_per_subport =
+			t->subport_params[subport_id].n_pipes_per_subport;
 		uint32_t pipe_id;
 		int status;
 
@@ -2211,10 +2212,11 @@ tm_tc_wred_profile_get(struct rte_eth_dev *dev, uint32_t tc_id)
 #ifdef RTE_SCHED_RED
 
 static void
-wred_profiles_set(struct rte_eth_dev *dev)
+wred_profiles_set(struct rte_eth_dev *dev, uint32_t subport_id)
 {
 	struct pmd_internals *p = dev->data->dev_private;
-	struct rte_sched_port_params *pp = &p->soft.tm.params.port_params;
+	struct rte_sched_subport_params *pp =
+		&p->soft.tm.params.subport_params[subport_id];
 
 	uint32_t tc_id;
 	enum rte_color color;
@@ -2234,7 +2236,7 @@ wred_profiles_set(struct rte_eth_dev *dev)
 
 #else
 
-#define wred_profiles_set(dev)
+#define wred_profiles_set(dev, subport_id)
 
 #endif
 
@@ -2526,29 +2528,10 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 		.frame_overhead =
 			root->shaper_profile->params.pkt_length_adjust,
 		.n_subports_per_port = root->n_children,
-		.n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
-			h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
-		.qsize = {p->params.tm.qsize[0],
-			p->params.tm.qsize[1],
-			p->params.tm.qsize[2],
-			p->params.tm.qsize[3],
-			p->params.tm.qsize[4],
-			p->params.tm.qsize[5],
-			p->params.tm.qsize[6],
-			p->params.tm.qsize[7],
-			p->params.tm.qsize[8],
-			p->params.tm.qsize[9],
-			p->params.tm.qsize[10],
-			p->params.tm.qsize[11],
-			p->params.tm.qsize[12],
-		},
-		.pipe_profiles = t->pipe_profiles,
-		.n_pipe_profiles = t->n_pipe_profiles,
+		.n_max_pipes_per_subport = TM_MAX_PIPES_PER_SUBPORT,
 		.n_max_pipe_profiles = TM_MAX_PIPE_PROFILE,
 	};
 
-	wred_profiles_set(dev);
-
 	subport_id = 0;
 	TAILQ_FOREACH(n, nl, node) {
 		uint64_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
@@ -2588,8 +2571,26 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 					tc_rate[12],
 				},
 				.tc_period = SUBPORT_TC_PERIOD,
+				.n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+					h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
+				.qsize = {p->params.tm.qsize[0],
+					p->params.tm.qsize[1],
+					p->params.tm.qsize[2],
+					p->params.tm.qsize[3],
+					p->params.tm.qsize[4],
+					p->params.tm.qsize[5],
+					p->params.tm.qsize[6],
+					p->params.tm.qsize[7],
+					p->params.tm.qsize[8],
+					p->params.tm.qsize[9],
+					p->params.tm.qsize[10],
+					p->params.tm.qsize[11],
+					p->params.tm.qsize[12],
+				},
+				.pipe_profiles = t->pipe_profiles,
+				.n_pipe_profiles = t->n_pipe_profiles,
 		};
-
+		wred_profiles_set(dev, subport_id);
 		subport_id++;
 	}
 }
Signature · 1 line
2.21.0
#29 lore raw

PATCH v2 13/15 ip_pipeline: add subport config flexibility to TM

In reply to Jasvinder Singh

Modify ip pipeline traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 examples/ip_pipeline/cli.c  | 71 ++++++++++++++++++-------------------
 examples/ip_pipeline/tmgr.c | 23 ++++++------
 examples/ip_pipeline/tmgr.h |  7 ++--
 3 files changed, 49 insertions(+), 52 deletions(-)
examples/ip_pipeline/cli.c +35 -36
index c6cf4204e..1e8dc95b4 100644
@@ -380,7 +380,12 @@ static const char cmd_tmgr_subport_profile_help[] =
 "   <tc0_rate> <tc1_rate> <tc2_rate> <tc3_rate> <tc4_rate>"
 "        <tc5_rate> <tc6_rate> <tc7_rate> <tc8_rate>"
 "        <tc9_rate> <tc10_rate> <tc11_rate> <tc12_rate>\n"
-"   <tc_period>\n";
+"   <tc_period>\n"
+"   pps <n_pipes_per_subport>\n"
+"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
+"       <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
+"       <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
+"       <qsize_tc11> <qsize_tc12>";
 
 static void
 cmd_tmgr_subport_profile(char **tokens,
@@ -391,7 +396,7 @@ cmd_tmgr_subport_profile(char **tokens,
 	struct rte_sched_subport_params p;
 	int status, i;
 
-	if (n_tokens != 19) {
+	if (n_tokens != 35) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -417,6 +422,27 @@ cmd_tmgr_subport_profile(char **tokens,
 		return;
 	}
 
+	if (strcmp(tokens[19], "pps") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
+		return;
+	}
+
+	if (parser_read_uint32(&p.n_pipes_per_subport, tokens[20]) != 0) {
+		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
+		return;
+	}
+
+	if (strcmp(tokens[21], "qsize") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
+		return;
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		if (parser_read_uint16(&p.qsize[i], tokens[22 + i]) != 0) {
+			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
+			return;
+		}
+
 	status = tmgr_subport_profile_add(&p);
 	if (status != 0) {
 		snprintf(out, out_size, MSG_CMD_FAIL, tokens[0]);
@@ -491,11 +517,6 @@ static const char cmd_tmgr_help[] =
 "tmgr <tmgr_name>\n"
 "   rate <rate>\n"
 "   spp <n_subports_per_port>\n"
-"   pps <n_pipes_per_subport>\n"
-"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
-"   <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
-"   <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
-"   <qsize_tc11> <qsize_tc12>\n"
 "   fo <frame_overhead>\n"
 "   mtu <mtu>\n"
 "   cpu <cpu_id>\n";
@@ -509,9 +530,8 @@ cmd_tmgr(char **tokens,
 	struct tmgr_port_params p;
 	char *name;
 	struct tmgr_port *tmgr_port;
-	int i;
 
-	if (n_tokens != 28) {
+	if (n_tokens != 12) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -538,53 +558,32 @@ cmd_tmgr(char **tokens,
 		return;
 	}
 
-	if (strcmp(tokens[6], "pps") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
-		return;
-	}
-
-	if (parser_read_uint32(&p.n_pipes_per_subport, tokens[7]) != 0) {
-		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
-		return;
-	}
-
-	if (strcmp(tokens[8], "qsize") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
-		return;
-	}
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (parser_read_uint16(&p.qsize[i], tokens[9 + i]) != 0) {
-			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
-			return;
-		}
-
-	if (strcmp(tokens[22], "fo") != 0) {
+	if (strcmp(tokens[6], "fo") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "fo");
 		return;
 	}
 
-	if (parser_read_uint32(&p.frame_overhead, tokens[23]) != 0) {
+	if (parser_read_uint32(&p.frame_overhead, tokens[7]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "frame_overhead");
 		return;
 	}
 
-	if (strcmp(tokens[24], "mtu") != 0) {
+	if (strcmp(tokens[8], "mtu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "mtu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.mtu, tokens[25]) != 0) {
+	if (parser_read_uint32(&p.mtu, tokens[9]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "mtu");
 		return;
 	}
 
-	if (strcmp(tokens[26], "cpu") != 0) {
+	if (strcmp(tokens[10], "cpu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "cpu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.cpu_id, tokens[27]) != 0) {
+	if (parser_read_uint32(&p.cpu_id, tokens[11]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "cpu_id");
 		return;
 	}
examples/ip_pipeline/tmgr.c +10 -13
index 40cbf1d0a..cafb9cdcc 100644
@@ -47,7 +47,8 @@ int
 tmgr_subport_profile_add(struct rte_sched_subport_params *p)
 {
 	/* Check input params */
-	if (p == NULL)
+	if (p == NULL ||
+		p->n_pipes_per_subport == 0)
 		return -1;
 
 	/* Save profile */
@@ -90,7 +91,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 		tmgr_port_find(name) ||
 		(params == NULL) ||
 		(params->n_subports_per_port == 0) ||
-		(params->n_pipes_per_subport == 0) ||
 		(params->cpu_id >= RTE_MAX_NUMA_NODES) ||
 		(n_subport_profiles == 0) ||
 		(n_pipe_profiles == 0))
@@ -103,18 +103,16 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	p.mtu = params->mtu;
 	p.frame_overhead = params->frame_overhead;
 	p.n_subports_per_port = params->n_subports_per_port;
-	p.n_pipes_per_subport = params->n_pipes_per_subport;
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		p.qsize[i] = params->qsize[i];
-
-	p.pipe_profiles = pipe_profile;
-	p.n_pipe_profiles = n_pipe_profiles;
+	p.n_max_pipes_per_subport = TMGR_PIPE_SUBPORT_MAX;
+	p.n_max_pipe_profiles = TMGR_PIPE_PROFILE_MAX;
 
 	s = rte_sched_port_config(&p);
 	if (s == NULL)
 		return NULL;
 
+	subport_profile[0].pipe_profiles = pipe_profile;
+	subport_profile[0].n_pipe_profiles = n_pipe_profiles;
+
 	for (i = 0; i < params->n_subports_per_port; i++) {
 		int status;
 
@@ -128,7 +126,7 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 			return NULL;
 		}
 
-		for (j = 0; j < params->n_pipes_per_subport; j++) {
+		for (j = 0; j < subport_profile[0].n_pipes_per_subport; j++) {
 			status = rte_sched_pipe_config(
 				s,
 				i,
@@ -153,7 +151,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	strlcpy(tmgr_port->name, name, sizeof(tmgr_port->name));
 	tmgr_port->s = s;
 	tmgr_port->n_subports_per_port = params->n_subports_per_port;
-	tmgr_port->n_pipes_per_subport = params->n_pipes_per_subport;
 
 	/* Node add to list */
 	TAILQ_INSERT_TAIL(&tmgr_port_list, tmgr_port, node);
@@ -205,8 +202,8 @@ tmgr_pipe_config(const char *port_name,
 	port = tmgr_port_find(port_name);
 	if ((port == NULL) ||
 		(subport_id >= port->n_subports_per_port) ||
-		(pipe_id_first >= port->n_pipes_per_subport) ||
-		(pipe_id_last >= port->n_pipes_per_subport) ||
+		(pipe_id_first >= subport_profile[subport_id].n_pipes_per_subport) ||
+		(pipe_id_last >= subport_profile[subport_id].n_pipes_per_subport) ||
 		(pipe_id_first > pipe_id_last) ||
 		(pipe_profile_id >= n_pipe_profiles))
 		return -1;
examples/ip_pipeline/tmgr.h +4 -3
index 8703a2e00..1fcf66ee1 100644
@@ -12,6 +12,10 @@
 
 #include "common.h"
 
+#ifndef TMGR_PIPE_SUBPORT_MAX
+#define TMGR_PIPE_SUBPORT_MAX                              4096
+#endif
+
 #ifndef TMGR_SUBPORT_PROFILE_MAX
 #define TMGR_SUBPORT_PROFILE_MAX                           256
 #endif
@@ -25,7 +29,6 @@ struct tmgr_port {
 	char name[NAME_SIZE];
 	struct rte_sched_port *s;
 	uint32_t n_subports_per_port;
-	uint32_t n_pipes_per_subport;
 };
 
 TAILQ_HEAD(tmgr_port_list, tmgr_port);
@@ -42,8 +45,6 @@ struct tmgr_port_params {
 	uint32_t frame_overhead;
 	uint32_t mtu;
 	uint32_t cpu_id;
-	uint32_t n_pipes_per_subport;
-	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 };
 
 int
Signature · 1 line
2.21.0
#30 lore raw

PATCH v2 14/15 examples/qos_sched: add subport configuration flexibility

In reply to Jasvinder Singh

Modify qos sample app to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 examples/qos_sched/app_thread.c   |   4 +-
 examples/qos_sched/cfg_file.c     | 229 ++++++++++++++++--------------
 examples/qos_sched/init.c         |  54 +++----
 examples/qos_sched/main.h         |   1 +
 examples/qos_sched/profile.cfg    |   5 +-
 examples/qos_sched/profile_ov.cfg |   5 +-
 examples/qos_sched/stats.c        |  36 ++---
 7 files changed, 182 insertions(+), 152 deletions(-)
examples/qos_sched/app_thread.c +2 -2
index 18b734a14..e07cc7452 100644
@@ -36,9 +36,9 @@ get_pkt_sched(struct rte_mbuf *m, uint32_t *subport, uint32_t *pipe,
 	uint16_t pipe_queue;
 
 	*subport = (rte_be_to_cpu_16(pdata[SUBPORT_OFFSET]) & 0x0FFF) &
-			(port_params.n_subports_per_port - 1); /* Outer VLAN ID*/
+		(port_params.n_subports_per_port - 1); /* Outer VLAN ID*/
 	*pipe = (rte_be_to_cpu_16(pdata[PIPE_OFFSET]) & 0x0FFF) &
-			(port_params.n_pipes_per_subport - 1); /* Inner VLAN ID */
+		(subport_params[*subport].n_pipes_per_subport - 1); /* Inner VLAN ID */
 	pipe_queue = active_queues[(pdata[QUEUE_OFFSET] >> 8) % n_active_queues];
 	*traffic_class = pipe_queue > RTE_SCHED_TRAFFIC_CLASS_BE ?
 			RTE_SCHED_TRAFFIC_CLASS_BE : pipe_queue; /* Destination IP */
examples/qos_sched/cfg_file.c +124 -105

269 lines not on this page · raw message

examples/qos_sched/init.c +29 -25

79 lines not on this page · raw message

examples/qos_sched/main.h +1 -0
index d8f890b64..baa2b3ead 100644
@@ -152,6 +152,7 @@ uint32_t active_queues[RTE_SCHED_QUEUES_PER_PIPE];
 uint32_t n_active_queues;
 
 extern struct rte_sched_port_params port_params;
+extern struct rte_sched_subport_params subport_params[MAX_SCHED_SUBPORTS];
 
 int app_parse_args(int argc, char **argv);
 int app_init(void);
examples/qos_sched/profile.cfg +3 -2
index 335561370..61b8b7071 100644
@@ -20,11 +20,12 @@
 [port]
 frame overhead = 24
 number of subports per port = 1
-number of pipes per subport = 4096
-queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
 
 ; Subport configuration
 [subport 0]
+number of pipes per subport = 4096
+queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
+
 tb rate = 1250000000           ; Bytes per second
 tb size = 1000000              ; Bytes
 
examples/qos_sched/profile_ov.cfg +3 -2
index 394987399..ab509d28d 100644
@@ -5,11 +5,12 @@
 [port]
 frame overhead = 24
 number of subports per port = 1
-number of pipes per subport = 32
-queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
 
 ; Subport configuration
 [subport 0]
+number of pipes per subport = 32
+queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
+
 tb rate = 8400000           ; Bytes per second
 tb size = 100000            ; Bytes
 
examples/qos_sched/stats.c +20 -16
index e62e4a2f6..0626cf6e6 100644
@@ -24,7 +24,7 @@ qavg_q(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id, uint8_t tc,
 
 	if (i == nb_pfc ||
 		subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport  ||
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport  ||
 		tc >= RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE ||
 		q >= RTE_SCHED_BE_QUEUES_PER_PIPE ||
 		(tc < RTE_SCHED_TRAFFIC_CLASS_BE && q > 0))
@@ -32,7 +32,7 @@ qavg_q(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id, uint8_t tc,
 
 	port = qos_conf[i].sched_port;
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport *
+		queue_id += subport_params[i].n_pipes_per_subport *
 				RTE_SCHED_QUEUES_PER_PIPE;
 	if (tc < RTE_SCHED_TRAFFIC_CLASS_BE)
 		queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE + tc;
@@ -69,14 +69,15 @@ qavg_tcpipe(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id,
 	}
 
 	if (i == nb_pfc || subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport ||
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport ||
 		tc >= RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE)
 		return -1;
 
 	port = qos_conf[i].sched_port;
 
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
+		queue_id +=
+			subport_params[i].n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
 
 	queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE + tc;
 
@@ -123,13 +124,13 @@ qavg_pipe(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id)
 
 	if (i == nb_pfc ||
 		subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport)
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport)
 		return -1;
 
 	port = qos_conf[i].sched_port;
 
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport *
+		queue_id += subport_params[i].n_pipes_per_subport *
 				RTE_SCHED_QUEUES_PER_PIPE;
 
 	queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE;
@@ -177,13 +178,13 @@ qavg_tcsubport(uint16_t port_id, uint32_t subport_id, uint8_t tc)
 
 	for (i = 0; i < subport_id; i++)
 		subport_queue_id +=
-			port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
+			subport_params[i].n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
 
 	average = 0;
 
 	for (count = 0; count < qavg_ntimes; count++) {
 		part_average = 0;
-		for (i = 0; i < port_params.n_pipes_per_subport; i++) {
+		for (i = 0; i < subport_params[subport_id].n_pipes_per_subport; i++) {
 			if (tc < RTE_SCHED_TRAFFIC_CLASS_BE) {
 				queue_id = subport_queue_id +
 					i * RTE_SCHED_QUEUES_PER_PIPE + tc;
@@ -203,10 +204,11 @@ qavg_tcsubport(uint16_t port_id, uint32_t subport_id, uint8_t tc)
 		}
 
 		if (tc < RTE_SCHED_TRAFFIC_CLASS_BE)
-			average += part_average / (port_params.n_pipes_per_subport);
+			average += part_average /
+				(subport_params[subport_id].n_pipes_per_subport);
 		else
-			average +=
-				part_average / (port_params.n_pipes_per_subport) *
+			average += part_average /
+				(subport_params[subport_id].n_pipes_per_subport) *
 				RTE_SCHED_BE_QUEUES_PER_PIPE;
 
 		usleep(qavg_period);
@@ -240,14 +242,14 @@ qavg_subport(uint16_t port_id, uint32_t subport_id)
 	port = qos_conf[i].sched_port;
 
 	for (i = 0; i < subport_id; i++)
-		subport_queue_id += port_params.n_pipes_per_subport *
+		subport_queue_id += subport_params[i].n_pipes_per_subport *
 			RTE_SCHED_QUEUES_PER_PIPE;
 
 	average = 0;
 
 	for (count = 0; count < qavg_ntimes; count++) {
 		part_average = 0;
-		for (i = 0; i < port_params.n_pipes_per_subport; i++) {
+		for (i = 0; i < subport_params[subport_id].n_pipes_per_subport; i++) {
 			queue_id = subport_queue_id + i * RTE_SCHED_QUEUES_PER_PIPE;
 
 			for (j = 0; j < RTE_SCHED_QUEUES_PER_PIPE; j++) {
@@ -258,7 +260,8 @@ qavg_subport(uint16_t port_id, uint32_t subport_id)
 		}
 
 		average += part_average /
-			(port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE);
+			(subport_params[subport_id].n_pipes_per_subport *
+			RTE_SCHED_QUEUES_PER_PIPE);
 		usleep(qavg_period);
 	}
 
@@ -322,12 +325,13 @@ pipe_stat(uint16_t port_id, uint32_t subport_id, uint32_t pipe_id)
 
 	if (i == nb_pfc ||
 		subport_id >= port_params.n_subports_per_port ||
-		pipe_id >= port_params.n_pipes_per_subport)
+		pipe_id >= subport_params[subport_id].n_pipes_per_subport)
 		return -1;
 
 	port = qos_conf[i].sched_port;
 	for (i = 0; i < subport_id; i++)
-		queue_id += port_params.n_pipes_per_subport * RTE_SCHED_QUEUES_PER_PIPE;
+		queue_id += subport_params[i].n_pipes_per_subport *
+			RTE_SCHED_QUEUES_PER_PIPE;
 
 	queue_id += pipe_id * RTE_SCHED_QUEUES_PER_PIPE;
 
Signature · 1 line
2.21.0
#31 lore raw

PATCH v2 15/15 sched: remove redundant code

In reply to Jasvinder Singh

Remove redundant data structure fields from port level data structures and update release notes.

---
 doc/guides/rel_notes/release_19_11.rst |  6 +++-
 lib/librte_sched/rte_sched.c           | 43 +-------------------------
 lib/librte_sched/rte_sched.h           | 22 -------------
 3 files changed, 6 insertions(+), 65 deletions(-)
doc/guides/rel_notes/release_19_11.rst +5 -1
index 8490d897c..746aeb0ea 100644
@@ -85,6 +85,10 @@ API Changes
    Also, make sure to start the actual text at the margin.
    =========================================================
 
+* sched: The pipe nodes configuration parameters such as number of pipes,
+  pipe queue sizes, pipe profiles, etc., are moved from port level structure
+  to subport level. This allows different subports of the same port to
+  have different configuration for the pipe nodes.
 
 ABI Changes
 -----------
@@ -172,7 +176,7 @@ The libraries prepended with a plus sign were incremented in this version.
      librte_rcu.so.1
      librte_reorder.so.1
      librte_ring.so.2
-     librte_sched.so.3
+   + librte_sched.so.4
      librte_security.so.2
      librte_stack.so.1
      librte_table.so.3
lib/librte_sched/rte_sched.c +1 -42

72 lines not on this page · raw message

lib/librte_sched/rte_sched.h +0 -22
index ea2b07448..db1b0232f 100644
@@ -246,33 +246,11 @@ struct rte_sched_port_params {
 	/** Number of subports */
 	uint32_t n_subports_per_port;
 
-	/** Number of subport_pipes */
-	uint32_t n_pipes_per_subport;
-
 	/** Maximum number of subport_pipes */
 	uint32_t n_max_pipes_per_subport;
 
-	/** Packet queue size for each traffic class.
-	 * All the pipes within the same subport share the similar
-	 * configuration for the queues.
-	 */
-	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-
-	/** Pipe profile table.
-	 * Every pipe is configured using one of the profiles from this table.
-	 */
-	struct rte_sched_pipe_params *pipe_profiles;
-
-	/** Profiles in the pipe profile table */
-	uint32_t n_pipe_profiles;
-
 	/** Max profiles allowed in the pipe profile table */
 	uint32_t n_max_pipe_profiles;
-
-#ifdef RTE_SCHED_RED
-	/** RED parameters */
-	struct rte_red_params red_params[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][RTE_COLORS];
-#endif
 };
 
 /*
Signature · 1 line
2.21.0
Acked-by on the v2 series
#32 lore raw
In reply to Jasvinder Singh

Hi Jasvinder,

Quoted 66 lines

-----Original Message-----
From: Singh, Jasvinder
Sent: Monday, September 9, 2019 11:05 AM
To: [email protected]
Cc: Dumitrescu, Cristian <[email protected]>
Subject: [PATCH v2 00/15] sched: subport level configuration of pipe nodes

This patchset refactors the dpdk qos sched library to allow subport level configuration flexibility of the pipe nodes.

Currently, all parameters for the pipe nodes (subscribers) configuration are part of the port level structure which forces all groups of subscribers (pipes) in different subports to have similar configurations in terms of their number, queue sizes, traffic-classes, etc.

The new implementation moves pipe nodes configuration parameters from port level to subport level structure. This allows different subports of the same port to have different configuration for the pipe nodes, for examples- number of pipes, queue sizes, pipe profiles, etc.

In order to keep the implementation complexity under control, all pipes within the same subport share the same configuration for queue sizes.

v2:
- fix qsize parsing in sample application
- fix checkpatch warnings

Jasvinder Singh (15):
sched: add pipe config params to subport struct
sched: modify internal structs for config flexibility
sched: remove pipe params config from port level
shced: add pipe config to subport level
sched: modify pipe functions for config flexibility
sched: modify pkt enqueue for config flexibility
sched: update memory compute to support flexiblity
sched: update grinder functions for config flexibility
sched: update pkt dequeue for flexible config
sched: update queue stats read for config flexibility test/sched: modify tests for subport config flexibility net/softnic: add subport config flexibility to TM
ip_pipeline: add subport config flexibility to TM examples/qos_sched: add subport configuration flexibility
sched: remove redundant code

 app/test/test_sched.c                    |   35 +-
 doc/guides/rel_notes/release_19_11.rst   |    6 +-
 drivers/net/softnic/rte_eth_softnic_tm.c |   51 +-
 examples/ip_pipeline/cli.c               |   71 +-
 examples/ip_pipeline/tmgr.c              |   23 +-
 examples/ip_pipeline/tmgr.h              |    7 +-
 examples/qos_sched/app_thread.c          |    4 +-
 examples/qos_sched/cfg_file.c            |  229 ++--
 examples/qos_sched/init.c                |   54 +-
 examples/qos_sched/main.h                |    1 +
 examples/qos_sched/profile.cfg           |    5 +-
 examples/qos_sched/profile_ov.cfg        |    5 +-
 examples/qos_sched/stats.c               |   36 +-
 lib/librte_sched/Makefile                |    2 +-
 lib/librte_sched/meson.build             |    2 +-
 lib/librte_sched/rte_sched.c             | 1400 +++++++++++++---------
 lib/librte_sched/rte_sched.h             |  114 +-
 17 files changed, 1183 insertions(+), 862 deletions(-)
Quoted 2 lines

--
2.21.0

A few comments/suggestion, mainly on some API name changes:

1. We need a good comment explaining the difference between rte_sched_port_params::n_max_pipes_per_subport and rte_sched_subport_params::n_pipes_per_subport. The former reserves a fixed number of bits in struct rte_mbuf::sched.queue_id for the pipe_id for all the subports of the same port, while the latter provides a mechanism to enable/allocate fewer pipes for each subport, as needed, with the benefit of avoiding memory allocation for the queues of the pipes that are not really needed. Another way to look at it is to say all subports have the same number of pipes (n_max_pipes_per_subport), but some of these pipes might not be implemented by all the subports. Maybe we should name the port parameter as n_pipes_per_subport and the subport parameter as n_pipes_per_subport_enabled?
PS: I did check the critical functions rte_sched_port_qindex() and rte_sched_port_pkt_read_tree_path(), and they look good to me.

2. The rte_sched_PORT_pipe_profile_add() should probably be now renamed as rte_sched_SUBPORT_pipe_profile_add(), right?

3. The port parameter n_max_pipe_profiles does not make sense to me, as we are now introducing the n_pipe_profiles subport parameter. Is this a code leftover or is this used to simplify the memory allocation? Assuming the latter, my vote is to remove it.

Regards,
Cristian

#33 lore raw
Quoted 83 lines

-----Original Message-----
From: Dumitrescu, Cristian
Sent: Monday, September 23, 2019 2:06 PM
To: Singh, Jasvinder <[email protected]>; [email protected]
Subject: RE: [PATCH v2 00/15] sched: subport level configuration of pipe nodes

Hi Jasvinder,

Quoted 66 lines

-----Original Message-----
From: Singh, Jasvinder
Sent: Monday, September 9, 2019 11:05 AM
To: [email protected]
Cc: Dumitrescu, Cristian <[email protected]>
Subject: [PATCH v2 00/15] sched: subport level configuration of pipe nodes

This patchset refactors the dpdk qos sched library to allow subport level configuration flexibility of the pipe nodes.

Currently, all parameters for the pipe nodes (subscribers) configuration are part of the port level structure which forces all groups of subscribers (pipes) in different subports to have similar configurations in terms of their number, queue sizes, traffic-classes, etc.

The new implementation moves pipe nodes configuration parameters from port level to subport level structure. This allows different subports of the same port to have different configuration for the pipe nodes, for examples- number of pipes, queue sizes, pipe profiles, etc.

In order to keep the implementation complexity under control, all pipes within the same subport share the same configuration for queue sizes.

v2:
- fix qsize parsing in sample application
- fix checkpatch warnings

Jasvinder Singh (15):
sched: add pipe config params to subport struct
sched: modify internal structs for config flexibility
sched: remove pipe params config from port level
shced: add pipe config to subport level
sched: modify pipe functions for config flexibility
sched: modify pkt enqueue for config flexibility
sched: update memory compute to support flexiblity
sched: update grinder functions for config flexibility
sched: update pkt dequeue for flexible config
sched: update queue stats read for config flexibility test/sched: modify tests for subport config flexibility net/softnic: add subport config flexibility to TM
ip_pipeline: add subport config flexibility to TM examples/qos_sched: add subport configuration flexibility
sched: remove redundant code

 app/test/test_sched.c                    |   35 +-
 doc/guides/rel_notes/release_19_11.rst   |    6 +-
 drivers/net/softnic/rte_eth_softnic_tm.c |   51 +-
 examples/ip_pipeline/cli.c               |   71 +-
 examples/ip_pipeline/tmgr.c              |   23 +-
 examples/ip_pipeline/tmgr.h              |    7 +-
 examples/qos_sched/app_thread.c          |    4 +-
 examples/qos_sched/cfg_file.c            |  229 ++--
 examples/qos_sched/init.c                |   54 +-
 examples/qos_sched/main.h                |    1 +
 examples/qos_sched/profile.cfg           |    5 +-
 examples/qos_sched/profile_ov.cfg        |    5 +-
 examples/qos_sched/stats.c               |   36 +-
 lib/librte_sched/Makefile                |    2 +-
 lib/librte_sched/meson.build             |    2 +-
 lib/librte_sched/rte_sched.c             | 1400 +++++++++++++---------
 lib/librte_sched/rte_sched.h             |  114 +-
 17 files changed, 1183 insertions(+), 862 deletions(-)
Quoted 2 lines

--
2.21.0

A few comments/suggestion, mainly on some API name changes:

Quoted 10 lines

1. We need a good comment explaining the difference between rte_sched_port_params::n_max_pipes_per_subport and rte_sched_subport_params::n_pipes_per_subport. The former reserves a fixed number of bits in struct rte_mbuf::sched.queue_id for the pipe_id for all the subports of the same port, while the latter provides a mechanism to enable/allocate fewer pipes for each subport, as needed, with the benefit of avoiding memory allocation for the queues of the pipes that are not really needed. Another way to look at it is to say all subports have the same number of pipes (n_max_pipes_per_subport), but some of these pipes might not be implemented by all the subports.

I will improve doxygen comments as suggested above, thanks.

Maybe we should name the port parameter

Quoted 2 lines

as n_pipes_per_subport and the subport parameter as n_pipes_per_subport_enabled?

Will rename these parameters as suggested.

Quoted 5 lines

PS: I did check the critical functions rte_sched_port_qindex() and rte_sched_port_pkt_read_tree_path(), and they look good to me.

2. The rte_sched_PORT_pipe_profile_add() should probably be now renamed as rte_sched_SUBPORT_pipe_profile_add(), right?

Thanks for spotting this :)

Quoted 4 lines

3. The port parameter n_max_pipe_profiles does not make sense to me, as we are now introducing the n_pipe_profiles subport parameter. Is this a code leftover or is this used to simplify the memory allocation? Assuming the latter, my vote is to remove it.

I will remove this field from the port structure.

Thank you once again for review and comments, will fix above in v3 version. Jasvinder

#34 lore raw

v3 cover sched: subport level configuration of pipe nodes

In reply to Jasvinder Singh

This patchset refactors the dpdk qos sched library to allow subport level configuration flexibility of the pipe nodes.

Currently, all parameters for the pipe nodes (subscribers) configuration are part of the port level structure which forces all groups of subscribers (pipes) in different subports to have similar configurations in terms of their number, queue sizes, traffic-classes, etc.

The new implementation moves pipe nodes configuration parameters from port level to subport level structure. This allows different subports of the same port to have different configuration for the pipe nodes, for examples- number of pipes, queue sizes, pipe profiles, etc.

In order to keep the implementation complexity under control, all pipes within the same subport share the same configuration for queue sizes.

v3:
- improve doxygen comments
- remove n_max_pipe_profiles parameter from struct rte_sched_port_params
- rename n_max_pipes_per_subport in struct rte_sched_port_params to n_pipes_per_subport
- rename n_pipes_per_subport in struct rte_sched_subport_params to n_pipes_per_subport_enabled
- rename rte_sched_port_pipe_profile_add() to rte_sched_subport_pipe_profile_add()

v2:
- fix qsize parsing in sample application
- fix checkpatch warnings

Jasvinder Singh (15):
sched: add pipe config params to subport struct
sched: modify internal structs for config flexibility
sched: remove pipe params config from port level
shced: add pipe config to subport level
sched: modify pipe functions for config flexibility
sched: modify pkt enqueue for config flexibility
sched: update memory compute to support flexiblity
sched: update grinder functions for config flexibility
sched: update pkt dequeue for flexible config
sched: update queue stats read for config flexibility test/sched: modify tests for subport config flexibility net/softnic: add subport config flexibility to TM
ip_pipeline: add subport config flexibility to TM examples/qos_sched: add subport configuration flexibility
sched: remove redundant code

 app/test/test_sched.c                    |   35 +-
 doc/guides/rel_notes/release_19_11.rst   |    6 +-
 drivers/net/softnic/rte_eth_softnic_tm.c |   54 +-
 examples/ip_pipeline/cli.c               |   71 +-
 examples/ip_pipeline/tmgr.c              |   25 +-
 examples/ip_pipeline/tmgr.h              |    7 +-
 examples/qos_sched/app_thread.c          |   20 +-
 examples/qos_sched/cfg_file.c            |  229 ++--
 examples/qos_sched/init.c                |   54 +-
 examples/qos_sched/main.h                |    1 +
 examples/qos_sched/profile.cfg           |    5 +-
 examples/qos_sched/profile_ov.cfg        |    5 +-
 examples/qos_sched/stats.c               |   44 +-
 lib/librte_sched/Makefile                |    2 +-
 lib/librte_sched/meson.build             |    2 +-
 lib/librte_sched/rte_sched.c             | 1394 +++++++++++++---------
 lib/librte_sched/rte_sched.h             |  129 +-
 lib/librte_sched/rte_sched_version.map   |    2 +-
 18 files changed, 1212 insertions(+), 873 deletions(-)
Signature · 1 line
2.21.0
#35 lore raw

PATCH v3 1/15 sched: add pipe config params to subport struct

Add pipe configuration parameters to subport level structure to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/Makefile    |  2 +-
 lib/librte_sched/meson.build |  2 +-
 lib/librte_sched/rte_sched.h | 92 ++++++++++++++++++++++++------------
 3 files changed, 65 insertions(+), 31 deletions(-)
lib/librte_sched/Makefile +1 -1
index 3d7f410e1..6e4a72d89 100644
@@ -18,7 +18,7 @@ LDLIBS += -lrte_timer
 
 EXPORT_MAP := rte_sched_version.map
 
-LIBABIVER := 3
+LIBABIVER := 4
 
 #
 # all source are stored in SRCS-y
lib/librte_sched/meson.build +1 -1
index 59d43c6d8..9f40a2368 100644
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2017 Intel Corporation
 
-version = 3
+version = 4
 sources = files('rte_sched.c', 'rte_red.c', 'rte_approx.c')
 headers = files('rte_sched.h', 'rte_sched_common.h',
 		'rte_red.h', 'rte_approx.h')
lib/librte_sched/rte_sched.h +63 -29

137 lines not on this page · raw message

Signature · 1 line
2.21.0
#36 lore raw

PATCH v3 2/15 sched: modify internal structs for config flexibility

In reply to Jasvinder Singh

Update internal structures related to port and subport to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 106 ++++++++++++++++++++++++++---------
 1 file changed, 78 insertions(+), 28 deletions(-)
lib/librte_sched/rte_sched.c +78 -28
index d8ab21d64..672412b77 100644
@@ -47,33 +47,6 @@
  */
 #define RTE_SCHED_TIME_SHIFT		      8
 
-struct rte_sched_subport {
-	/* Token bucket (TB) */
-	uint64_t tb_time; /* time of last update */
-	uint32_t tb_period;
-	uint32_t tb_credits_per_period;
-	uint32_t tb_size;
-	uint32_t tb_credits;
-
-	/* Traffic classes (TCs) */
-	uint64_t tc_time; /* time of next update */
-	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_period;
-
-	/* TC oversubscription */
-	uint32_t tc_ov_wm;
-	uint32_t tc_ov_wm_min;
-	uint32_t tc_ov_wm_max;
-	uint8_t tc_ov_period_id;
-	uint8_t tc_ov;
-	uint32_t tc_ov_n;
-	double tc_ov_rate;
-
-	/* Statistics */
-	struct rte_sched_subport_stats stats;
-};
-
 struct rte_sched_pipe_profile {
 	/* Token bucket (TB) */
 	uint32_t tb_period;
@@ -107,7 +80,6 @@ struct rte_sched_pipe {
 	/* TC oversubscription */
 	uint32_t tc_ov_credits;
 	uint8_t tc_ov_period_id;
-	uint8_t reserved[3];
 } __rte_cache_aligned;
 
 struct rte_sched_queue {
@@ -166,6 +138,72 @@ struct rte_sched_grinder {
 	uint8_t wrr_cost[RTE_SCHED_BE_QUEUES_PER_PIPE];
 };
 
+struct rte_sched_subport {
+	/* Token bucket (TB) */
+	uint64_t tb_time; /* time of last update */
+	uint32_t tb_period;
+	uint32_t tb_credits_per_period;
+	uint32_t tb_size;
+	uint32_t tb_credits;
+
+	/* Traffic classes (TCs) */
+	uint64_t tc_time; /* time of next update */
+	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_period;
+
+	/* TC oversubscription */
+	uint32_t tc_ov_wm;
+	uint32_t tc_ov_wm_min;
+	uint32_t tc_ov_wm_max;
+	uint8_t tc_ov_period_id;
+	uint8_t tc_ov;
+	uint32_t tc_ov_n;
+	double tc_ov_rate;
+
+	/* Statistics */
+	struct rte_sched_subport_stats stats;
+
+	/* Subport pipes */
+	uint32_t n_pipes_per_subport_enabled;
+	uint32_t n_pipe_profiles;
+	uint32_t n_max_pipe_profiles;
+
+	/* Pipe best-effort TC rate */
+	uint32_t pipe_tc_be_rate_max;
+
+	/* Pipe queues size */
+	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+
+#ifdef RTE_SCHED_RED
+	struct rte_red_config red_config[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][RTE_COLORS];
+#endif
+
+	/* Scheduling loop detection */
+	uint32_t pipe_loop;
+	uint32_t pipe_exhaustion;
+
+	/* Bitmap */
+	struct rte_bitmap *bmp;
+	uint32_t grinder_base_bmp_pos[RTE_SCHED_PORT_N_GRINDERS] __rte_aligned_16;
+
+	/* Grinders */
+	struct rte_sched_grinder grinder[RTE_SCHED_PORT_N_GRINDERS];
+	uint32_t busy_grinders;
+
+	/* Queue base calculation */
+	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
+	uint32_t qsize_sum;
+
+	struct rte_sched_pipe *pipe;
+	struct rte_sched_queue *queue;
+	struct rte_sched_queue_extra *queue_extra;
+	struct rte_sched_pipe_profile *pipe_profiles;
+	uint8_t *bmp_array;
+	struct rte_mbuf **queue_array;
+	uint8_t memory[0] __rte_cache_aligned;
+} __rte_cache_aligned;
+
 struct rte_sched_port {
 	/* User parameters */
 	uint32_t n_subports_per_port;
@@ -177,6 +215,7 @@ struct rte_sched_port {
 	uint32_t rate;
 	uint32_t mtu;
 	uint32_t frame_overhead;
+	int socket;
 	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t n_pipe_profiles;
 	uint32_t n_max_pipe_profiles;
@@ -210,6 +249,7 @@ struct rte_sched_port {
 	uint32_t qsize_sum;
 
 	/* Large data structures */
+	struct rte_sched_subport *subports[0];
 	struct rte_sched_subport *subport;
 	struct rte_sched_pipe *pipe;
 	struct rte_sched_queue *queue;
@@ -231,6 +271,16 @@ enum rte_sched_port_array {
 	e_RTE_SCHED_PORT_ARRAY_TOTAL,
 };
 
+enum rte_sched_subport_array {
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE = 0,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA,
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES,
+	e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
+};
+
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline uint32_t
Signature · 1 line
2.21.0
#37 lore raw

PATCH v3 3/15 sched: remove pipe params config from port level

In reply to Jasvinder Singh

Remove pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 254 ++++++++++-------------------------
 1 file changed, 71 insertions(+), 183 deletions(-)
lib/librte_sched/rte_sched.c +71 -183
index 672412b77..952e449f7 100644
@@ -291,6 +291,31 @@ rte_sched_port_queues_per_subport(struct rte_sched_port *port)
 
 #endif
 
+static inline uint32_t
+rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
+{
+	return RTE_SCHED_QUEUES_PER_PIPE * subport->n_pipes_per_subport_enabled;
+}
+
+static inline struct rte_mbuf **
+rte_sched_subport_pipe_qbase(struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t pindex = qindex >> 4;
+	uint32_t qpos = qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1);
+
+	return (subport->queue_array + pindex *
+		subport->qsize_sum + subport->qsize_add[qpos]);
+}
+
+static inline uint16_t
+rte_sched_subport_pipe_qsize(struct rte_sched_port *port,
+struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
+
+	return subport->qsize[tc];
+}
+
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
@@ -414,8 +439,6 @@ pipe_profile_check(struct rte_sched_pipe_params *params,
 static int
 rte_sched_port_check_params(struct rte_sched_port_params *params)
 {
-	uint32_t i;
-
 	if (params == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter params\n", __func__);
@@ -456,45 +479,10 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 	if (params->n_pipes_per_subport == 0 ||
 	    !rte_is_power_of_2(params->n_pipes_per_subport)) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for pipes number\n", __func__);
-		return -EINVAL;
-	}
-
-	/* qsize: if non-zero, power of 2,
-	 * no bigger than 32K (due to 16-bit read/write pointers)
-	 */
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint16_t qsize = params->qsize[i];
-
-		if ((qsize != 0 && !rte_is_power_of_2(qsize)) ||
-			((i == RTE_SCHED_TRAFFIC_CLASS_BE) && (qsize == 0))) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	/* pipe_profiles and n_pipe_profiles */
-	if (params->pipe_profiles == NULL ||
-	    params->n_pipe_profiles == 0 ||
-		 params->n_pipe_profiles > params->n_max_pipe_profiles) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for number of pipe profiles\n", __func__);
+			"%s: Incorrect value for maximum pipes number\n", __func__);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < params->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
-		int status;
-
-		status = pipe_profile_check(p, params->rate, &params->qsize[0]);
-		if (status != 0) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Pipe profile check failed(%d)\n", __func__, status);
-			return -EINVAL;
-		}
-	}
-
 	return 0;
 }
 
@@ -582,32 +570,6 @@ rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
 	return size0 + size1;
 }
 
-static void
-rte_sched_port_config_qsize(struct rte_sched_port *port)
-{
-	uint32_t i;
-
-	port->qsize_add[0] = 0;
-
-	/* Strict prority traffic class */
-	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		port->qsize_add[i] = port->qsize_add[i-1] + port->qsize[i-1];
-
-	/* Best-effort traffic class */
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-	port->qsize_sum = port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-}
-
 static void
 rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
 {
@@ -717,56 +679,41 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
-static void
-rte_sched_port_config_pipe_profile_table(struct rte_sched_port *port,
-	struct rte_sched_port_params *params)
-{
-	uint32_t i;
-
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		struct rte_sched_pipe_profile *dst = port->pipe_profiles + i;
-
-		rte_sched_pipe_profile_convert(port, src, dst, params->rate);
-		rte_sched_port_log_pipe_profile(port, i);
-	}
-
-	port->pipe_tc_be_rate_max = 0;
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-		if (port->pipe_tc_be_rate_max < pipe_tc_be_rate)
-			port->pipe_tc_be_rate_max = pipe_tc_be_rate;
-	}
-}
-
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
 	struct rte_sched_port *port = NULL;
-	uint32_t mem_size, bmp_mem_size, n_queues_per_port, i, j, cycles_per_byte;
+	uint32_t size0, size1;
+	uint32_t cycles_per_byte;
+	uint32_t i, j;
+	int status;
 
-	/* Check user parameters. Determine the amount of memory to allocate */
-	mem_size = rte_sched_port_get_memory_footprint(params);
-	if (mem_size == 0)
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 		return NULL;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = params->n_subports_per_port * sizeof(struct rte_sched_subport *);
 
 	/* Allocate memory to store the data structures */
-	port = rte_zmalloc_socket("qos_params", mem_size, RTE_CACHE_LINE_SIZE,
+	port = rte_zmalloc_socket("qos_params", size0 + size1, RTE_CACHE_LINE_SIZE,
 		params->socket);
-	if (port == NULL)
-		return NULL;
+	if (port == NULL) {
+		RTE_LOG(ERR, SCHED, "%s: Memory allocation fails\n", __func__);
 
-	/* compile time checks */
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS & (RTE_SCHED_PORT_N_GRINDERS - 1));
+		return NULL;
+	}
 
 	/* User parameters */
 	port->n_subports_per_port = params->n_subports_per_port;
 	port->n_pipes_per_subport = params->n_pipes_per_subport;
 	port->n_pipes_per_subport_log2 =
 			__builtin_ctz(params->n_pipes_per_subport);
+	port->socket = params->socket;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 		port->pipe_queue[i] = i;
@@ -787,32 +734,6 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	port->rate = params->rate;
 	port->mtu = params->mtu + params->frame_overhead;
 	port->frame_overhead = params->frame_overhead;
-	memcpy(port->qsize, params->qsize, sizeof(params->qsize));
-	port->n_pipe_profiles = params->n_pipe_profiles;
-	port->n_max_pipe_profiles = params->n_max_pipe_profiles;
-
-#ifdef RTE_SCHED_RED
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t j;
-
-		for (j = 0; j < RTE_COLORS; j++) {
-			/* if min/max are both zero, then RED is disabled */
-			if ((params->red_params[i][j].min_th |
-			     params->red_params[i][j].max_th) == 0) {
-				continue;
-			}
-
-			if (rte_red_config_init(&port->red_config[i][j],
-				params->red_params[i][j].wq_log2,
-				params->red_params[i][j].min_th,
-				params->red_params[i][j].max_th,
-				params->red_params[i][j].maxp_inv) != 0) {
-				rte_free(port);
-				return NULL;
-			}
-		}
-	}
-#endif
 
 	/* Timing */
 	port->time_cpu_cycles = rte_get_tsc_cycles();
@@ -823,79 +744,32 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 		/ params->rate;
 	port->inv_cycles_per_byte = rte_reciprocal_value(cycles_per_byte);
 
-	/* Scheduling loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
-	port->pipe_exhaustion = 0;
-
 	/* Grinders */
-	port->busy_grinders = 0;
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
 
-	/* Queue base calculation */
-	rte_sched_port_config_qsize(port);
-
-	/* Large data structures */
-	port->subport = (struct rte_sched_subport *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_SUBPORT));
-	port->pipe = (struct rte_sched_pipe *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE));
-	port->queue = (struct rte_sched_queue *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE));
-	port->queue_extra = (struct rte_sched_queue_extra *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA));
-	port->pipe_profiles = (struct rte_sched_pipe_profile *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES));
-	port->bmp_array =  port->memory
-		+ rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY);
-	port->queue_array = (struct rte_mbuf **)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY));
-
-	/* Pipe profile table */
-	rte_sched_port_config_pipe_profile_table(port, params);
-
-	/* Bitmap */
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
-	bmp_mem_size = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	port->bmp = rte_bitmap_init(n_queues_per_port, port->bmp_array,
-				    bmp_mem_size);
-	if (port->bmp == NULL) {
-		RTE_LOG(ERR, SCHED, "Bitmap init error\n");
-		rte_free(port);
-		return NULL;
-	}
-
-	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
-		port->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
-
-
 	return port;
 }
 
-void
-rte_sched_port_free(struct rte_sched_port *port)
+static inline void
+rte_sched_subport_free(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
+	uint32_t n_subport_pipe_queues;
 	uint32_t qindex;
-	uint32_t n_queues_per_port;
 
-	/* Check user parameters */
-	if (port == NULL)
+	if (subport == NULL)
 		return;
 
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(subport);
 
 	/* Free enqueued mbufs */
-	for (qindex = 0; qindex < n_queues_per_port; qindex++) {
-		struct rte_mbuf **mbufs = rte_sched_port_qbase(port, qindex);
-		uint16_t qsize = rte_sched_port_qsize(port, qindex);
+	for (qindex = 0; qindex < n_subport_pipe_queues; qindex++) {
+		struct rte_mbuf **mbufs =
+			rte_sched_subport_pipe_qbase(subport, qindex);
+		uint16_t qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 		if (qsize != 0) {
-			struct rte_sched_queue *queue = port->queue + qindex;
+			struct rte_sched_queue *queue = subport->queue + qindex;
 			uint16_t qr = queue->qr & (qsize - 1);
 			uint16_t qw = queue->qw & (qsize - 1);
 
@@ -904,7 +778,21 @@ rte_sched_port_free(struct rte_sched_port *port)
 		}
 	}
 
-	rte_bitmap_free(port->bmp);
+	rte_bitmap_free(subport->bmp);
+}
+
+void
+rte_sched_port_free(struct rte_sched_port *port)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (port == NULL)
+		return;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		rte_sched_subport_free(port, port->subports[i]);
+
 	rte_free(port);
 }
 
Signature · 1 line
2.21.0
#38 lore raw

PATCH v3 4/15 sched: add pipe config to subport level

In reply to Jasvinder Singh

Add pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 388 ++++++++++++++++++++++++++++++-----
 1 file changed, 332 insertions(+), 56 deletions(-)
lib/librte_sched/rte_sched.c +332 -56
index 952e449f7..60dfc6232 100644
@@ -550,24 +550,91 @@ rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sch
 	return base;
 }
 
-uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+static uint32_t
+rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
+	enum rte_sched_subport_array array)
 {
-	uint32_t size0, size1;
-	int status;
+	uint32_t n_pipes_per_subport = params->n_pipes_per_subport_enabled;
+	uint32_t n_subport_pipe_queues =
+		RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport;
 
-	status = rte_sched_port_check_params(params);
-	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+	uint32_t size_pipe = n_pipes_per_subport * sizeof(struct rte_sched_pipe);
+	uint32_t size_queue =
+		n_subport_pipe_queues * sizeof(struct rte_sched_queue);
+	uint32_t size_queue_extra
+		= n_subport_pipe_queues * sizeof(struct rte_sched_queue_extra);
+	uint32_t size_pipe_profiles = params->n_max_pipe_profiles *
+		sizeof(struct rte_sched_pipe_profile);
+	uint32_t size_bmp_array =
+		rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	uint32_t size_per_pipe_queue_array, size_queue_array;
 
-		return 0;
+	uint32_t base, i;
+
+	size_per_pipe_queue_array = 0;
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
+			size_per_pipe_queue_array +=
+				params->qsize[i] * sizeof(struct rte_mbuf *);
+		else
+			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
+				params->qsize[i] * sizeof(struct rte_mbuf *);
 	}
+	size_queue_array = n_pipes_per_subport * size_per_pipe_queue_array;
 
-	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_TOTAL);
+	base = 0;
 
-	return size0 + size1;
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
+
+	return base;
+}
+
+static void
+rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
+{
+	uint32_t i;
+
+	subport->qsize_add[0] = 0;
+
+	/* Strict prority traffic class */
+	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		subport->qsize_add[i] = subport->qsize_add[i-1] + subport->qsize[i-1];
+
+	/* Best-effort traffic class */
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+	subport->qsize_sum = subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
 }
 
 static void
@@ -679,6 +746,126 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static int
+rte_sched_subport_check_params(struct rte_sched_subport_params *params,
+	uint32_t n_max_pipes_per_subport,
+	uint32_t rate)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (params == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for parameter params\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_rate == 0 || params->tb_rate > rate) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb rate\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_size == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb size\n", __func__);
+		return -EINVAL;
+	}
+
+	/* qsize: if non-zero, power of 2,
+	 * no bigger than 32K (due to 16-bit read/write pointers)
+	 */
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint16_t qsize = params->qsize[i];
+
+		if (qsize != 0 && !rte_is_power_of_2(qsize)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for qsize\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t tc_rate = params->tc_rate[i];
+		uint16_t qsize = params->qsize[i];
+
+		if ((qsize == 0 && tc_rate != 0) ||
+			(qsize != 0 && tc_rate == 0) ||
+			(tc_rate > params->tb_rate)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for tc rate\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	if (params->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
+		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect qsize or tc rate(best effort)\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tc_period == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tc period\n", __func__);
+		return -EINVAL;
+	}
+
+	/* n_pipes_per_subport: non-zero, power of 2 */
+	if (params->n_pipes_per_subport_enabled == 0 ||
+		params->n_pipes_per_subport_enabled > n_max_pipes_per_subport ||
+	    !rte_is_power_of_2(params->n_pipes_per_subport_enabled)) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipes number\n", __func__);
+		return -EINVAL;
+	}
+
+	/* pipe_profiles and n_pipe_profiles */
+	if (params->pipe_profiles == NULL ||
+	    params->n_pipe_profiles == 0 ||
+		params->n_max_pipe_profiles == 0 ||
+		params->n_pipe_profiles > params->n_max_pipe_profiles) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipe profiles\n", __func__);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < params->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
+		int status;
+
+		status = pipe_profile_check(p, rate, &params->qsize[0]);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Pipe profile check failed(%d)\n", __func__, status);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+uint32_t
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+{
+	uint32_t size0, size1;
+	int status;
+
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"Port scheduler params check failed (%d)\n", status);
+
+		return 0;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = rte_sched_port_get_array_base(params,
+			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	return size0 + size1;
+}
+
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
@@ -799,7 +986,7 @@ rte_sched_port_free(struct rte_sched_port *port)
 static void
 rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 {
-	struct rte_sched_subport *s = port->subport + i;
+	struct rte_sched_subport *s = port->subports[i];
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for subport %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -834,72 +1021,78 @@ rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 		s->tc_ov_wm_max);
 }
 
+static void
+rte_sched_free_memory(struct rte_sched_port *port, uint32_t n_subports)
+{
+	uint32_t i;
+
+	for (i = 0; i < n_subports; i++) {
+		struct rte_sched_subport *subport = port->subports[i];
+
+		rte_sched_subport_free(port, subport);
+	}
+
+	rte_free(port);
+}
+
 int
 rte_sched_subport_config(struct rte_sched_port *port,
 	uint32_t subport_id,
 	struct rte_sched_subport_params *params)
 {
-	struct rte_sched_subport *s;
-	uint32_t i;
+	struct rte_sched_subport *s = NULL;
+	uint32_t n_subports = subport_id;
+	uint32_t n_subport_pipe_queues, i;
+	uint32_t size0, size1, bmp_mem_size;
+	int status;
 
 	/* Check user parameters */
 	if (port == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter port\n", __func__);
-		return -EINVAL;
+		return 0;
 	}
 
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for subport id\n", __func__);
-		return -EINVAL;
-	}
 
-	if (params == NULL) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for parameter params\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (params->tb_rate == 0 || params->tb_rate > port->rate) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb rate\n", __func__);
-		return -EINVAL;
-	}
+	status = rte_sched_subport_check_params(params,
+		port->n_pipes_per_subport,
+		port->rate);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 
-	if (params->tb_size == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb size\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t tc_rate = params->tc_rate[i];
-		uint16_t qsize = port->qsize[i];
+	/* Determine the amount of memory to allocate */
+	size0 = sizeof(struct rte_sched_subport);
+	size1 = rte_sched_subport_get_array_base(params,
+				e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
 
-		if ((qsize == 0 && tc_rate != 0) ||
-			(qsize != 0 && tc_rate == 0) ||
-			(tc_rate > params->tb_rate)) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	if (port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
-		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+	/* Allocate memory to store the data structures */
+	s = rte_zmalloc_socket("subport_params", size0 + size1,
+		RTE_CACHE_LINE_SIZE, port->socket);
+	if (s == NULL) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc rate(best effort)\n", __func__);
-		return -EINVAL;
-	}
+			"%s: Memory allocation fails\n", __func__);
 
-	if (params->tc_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc period\n", __func__);
-		return -EINVAL;
+		rte_sched_free_memory(port, n_subports);
+		return -ENOMEM;
 	}
 
-	s = port->subport + subport_id;
+	n_subports++;
+
+	/* Port */
+	port->subports[subport_id] = s;
 
 	/* Token Bucket (TB) */
 	if (params->tb_rate == port->rate) {
@@ -919,26 +1112,109 @@ rte_sched_subport_config(struct rte_sched_port *port,
 	/* Traffic Classes (TCs) */
 	s->tc_period = rte_sched_time_ms_to_bytes(params->tc_period, port->rate);
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(params->tc_period,
-								 params->tc_rate[i]);
-
+					params->tc_rate[i]);
 	}
 	s->tc_time = port->time + s->tc_period;
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits[i] = s->tc_credits_per_period[i];
 
+	/* compile time checks */
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS &
+		(RTE_SCHED_PORT_N_GRINDERS - 1));
+
+	/* User parameters */
+	s->n_pipes_per_subport_enabled = params->n_pipes_per_subport_enabled;
+	memcpy(s->qsize, params->qsize, sizeof(params->qsize));
+	s->n_pipe_profiles = params->n_pipe_profiles;
+	s->n_max_pipe_profiles = params->n_max_pipe_profiles;
+
+#ifdef RTE_SCHED_RED
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t j;
+
+		for (j = 0; j < RTE_COLORS; j++) {
+			/* if min/max are both zero, then RED is disabled */
+			if ((params->red_params[i][j].min_th |
+			     params->red_params[i][j].max_th) == 0) {
+				continue;
+			}
+
+			if (rte_red_config_init(&s->red_config[i][j],
+				params->red_params[i][j].wq_log2,
+				params->red_params[i][j].min_th,
+				params->red_params[i][j].max_th,
+				params->red_params[i][j].maxp_inv) != 0) {
+				rte_sched_free_memory(port, n_subports);
+
+				RTE_LOG(NOTICE, SCHED,
+				"%s: RED configuration init fails\n", __func__);
+				return -EINVAL;
+			}
+		}
+	}
+#endif
+
+	/* Scheduling loop detection */
+	s->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	s->pipe_exhaustion = 0;
+
+	/* Grinders */
+	s->busy_grinders = 0;
+
+	/* Queue base calculation */
+	rte_sched_subport_config_qsize(s);
+
+	/* Large data structures */
+	s->pipe = (struct rte_sched_pipe *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_PIPE));
+	s->queue = (struct rte_sched_queue *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE));
+	s->queue_extra = (struct rte_sched_queue_extra *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA));
+	s->pipe_profiles = (struct rte_sched_pipe_profile *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES));
+	s->bmp_array =  s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY);
+	s->queue_array = (struct rte_mbuf **)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
+
+	/* Bitmap */
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
+	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	s->bmp = rte_bitmap_init(n_subport_pipe_queues, s->bmp_array,
+				bmp_mem_size);
+	if (s->bmp == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Subport bitmap init error\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
+		s->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
+
+#ifdef RTE_SCHED_SUBPORT_TC_OV
 	/* TC oversubscription */
 	s->tc_ov_wm_min = port->mtu;
 	s->tc_ov_wm_max = rte_sched_time_ms_to_bytes(params->tc_period,
-						     port->pipe_tc_be_rate_max);
+						     s->pipe_tc_be_rate_max);
 	s->tc_ov_wm = s->tc_ov_wm_max;
 	s->tc_ov_period_id = 0;
 	s->tc_ov = 0;
 	s->tc_ov_n = 0;
 	s->tc_ov_rate = 0;
+#endif
 
 	rte_sched_port_log_subport_config(port, subport_id);
 
Signature · 1 line
2.21.0
#39 lore raw

PATCH v3 5/15 sched: modify pipe functions for config flexibility

In reply to Jasvinder Singh

Modify pipe level functions to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c           | 103 +++++++++++++++++--------
 lib/librte_sched/rte_sched.h           |   7 +-
 lib/librte_sched/rte_sched_version.map |   2 +-
 3 files changed, 76 insertions(+), 36 deletions(-)
lib/librte_sched/rte_sched.c +70 -33
index 60dfc6232..6b6219e45 100644
@@ -638,9 +638,9 @@ rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
 }
 
 static void
-rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
+rte_sched_port_log_pipe_profile(struct rte_sched_subport *subport, uint32_t i)
 {
-	struct rte_sched_pipe_profile *p = port->pipe_profiles + i;
+	struct rte_sched_pipe_profile *p = subport->pipe_profiles + i;
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for pipe profile %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -689,7 +689,7 @@ rte_sched_time_ms_to_bytes(uint32_t time_ms, uint32_t rate)
 }
 
 static void
-rte_sched_pipe_profile_convert(struct rte_sched_port *port,
+rte_sched_pipe_profile_convert(struct rte_sched_subport *subport,
 	struct rte_sched_pipe_params *src,
 	struct rte_sched_pipe_profile *dst,
 	uint32_t rate)
@@ -718,7 +718,7 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 						rate);
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (subport->qsize[i])
 			dst->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(src->tc_period,
 					src->tc_rate[i]);
@@ -746,6 +746,30 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static void
+rte_sched_subport_config_pipe_profile_table(struct rte_sched_subport *subport,
+	struct rte_sched_subport_params *params, uint32_t rate)
+{
+	uint32_t i;
+
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		struct rte_sched_pipe_profile *dst = subport->pipe_profiles + i;
+
+		rte_sched_pipe_profile_convert(subport, src, dst, rate);
+		rte_sched_port_log_pipe_profile(subport, i);
+	}
+
+	subport->pipe_tc_be_rate_max = 0;
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+		if (subport->pipe_tc_be_rate_max < pipe_tc_be_rate)
+			subport->pipe_tc_be_rate_max = pipe_tc_be_rate;
+	}
+}
+
 static int
 rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 	uint32_t n_max_pipes_per_subport,
@@ -1188,6 +1212,9 @@ rte_sched_subport_config(struct rte_sched_port *port,
 		(s->memory + rte_sched_subport_get_array_base(params,
 		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
 
+	/* Pipe profile table */
+	rte_sched_subport_config_pipe_profile_table(s, params, port->rate);
+
 	/* Bitmap */
 	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
 	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
@@ -1230,6 +1257,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	struct rte_sched_subport *s;
 	struct rte_sched_pipe *p;
 	struct rte_sched_pipe_profile *params;
+	uint32_t n_subports = subport_id + 1;
 	uint32_t deactivate, profile, i;
 
 	/* Check user parameters */
@@ -1245,34 +1273,32 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter subport id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (pipe_id >= port->n_pipes_per_subport) {
+	s = port->subports[subport_id];
+	if (pipe_id >= s->n_pipes_per_subport_enabled) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (!deactivate && profile >= port->n_pipe_profiles) {
+	if (!deactivate && profile >= s->n_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe profile\n", __func__);
-		return -EINVAL;
-	}
 
-	/* Check that subport configuration is valid */
-	s = port->subport + subport_id;
-	if (s->tb_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Subport configuration invalid\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	p = port->pipe + (subport_id * port->n_pipes_per_subport + pipe_id);
-
 	/* Handle the case when pipe already has a valid configuration */
+	p = s->pipe + pipe_id;
 	if (p->tb_time) {
-		params = port->pipe_profiles + p->profile;
+		params = s->pipe_profiles + p->profile;
 
 		double subport_tc_be_rate =
 			(double) s->tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASS_BE]
@@ -1302,7 +1328,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 
 	/* Apply the new pipe configuration */
 	p->profile = profile;
-	params = port->pipe_profiles + p->profile;
+	params = s->pipe_profiles + p->profile;
 
 	/* Token Bucket (TB) */
 	p->tb_time = port->time;
@@ -1312,7 +1338,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	p->tc_time = port->time + params->tc_period;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (s->qsize[i])
 			p->tc_credits[i] = params->tc_credits_per_period[i];
 
 	{
@@ -1342,10 +1368,12 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 }
 
 int
-rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+rte_sched_subport_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_pipe_profile *pp;
 	uint32_t i;
 	int status;
@@ -1357,40 +1385,49 @@ rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
-	/* Pipe profiles not exceeds the max limit */
-	if (port->n_pipe_profiles >= port->n_max_pipe_profiles) {
+	/* Subport id not exceeds the max limit */
+	if (subport_id > port->n_subports_per_port) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for subport id\n", __func__);
+		return -EINVAL;
+	}
+
+	s = port->subports[subport_id];
+
+	/* Pipe profiles exceeds the max limit */
+	if (s->n_pipe_profiles >= s->n_max_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Number of pipe profiles exceeds the max limit\n", __func__);
 		return -EINVAL;
 	}
 
 	/* Pipe params */
-	status = pipe_profile_check(params, port->rate, &port->qsize[0]);
+	status = pipe_profile_check(params, port->rate, &s->qsize[0]);
 	if (status != 0) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Pipe profile check failed(%d)\n", __func__, status);
 		return -EINVAL;
 	}
 
-	pp = &port->pipe_profiles[port->n_pipe_profiles];
-	rte_sched_pipe_profile_convert(port, params, pp, port->rate);
+	pp = &s->pipe_profiles[s->n_pipe_profiles];
+	rte_sched_pipe_profile_convert(s, params, pp, port->rate);
 
-	/* Pipe profile not exists */
-	for (i = 0; i < port->n_pipe_profiles; i++)
-		if (memcmp(port->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
+	/* Pipe profile should not exists */
+	for (i = 0; i < s->n_pipe_profiles; i++)
+		if (memcmp(s->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
 			RTE_LOG(ERR, SCHED,
-				"%s: Pipe profile doesn't exist\n", __func__);
+				"%s: Pipe profile exists\n", __func__);
 			return -EINVAL;
 		}
 
 	/* Pipe profile commit */
-	*pipe_profile_id = port->n_pipe_profiles;
-	port->n_pipe_profiles++;
+	*pipe_profile_id = s->n_pipe_profiles;
+	s->n_pipe_profiles++;
 
-	if (port->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
-		port->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+	if (s->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
+		s->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
 
-	rte_sched_port_log_pipe_profile(port, *pipe_profile_id);
+	rte_sched_port_log_pipe_profile(s, *pipe_profile_id);
 
 	return 0;
 }
lib/librte_sched/rte_sched.h +5 -2
index fbb0e23fa..5001f09ed 100644
@@ -317,6 +317,8 @@ rte_sched_port_free(struct rte_sched_port *port);
  *
  * @param port
  *   Handle to port scheduler instance
+ * @param subport_id
+ *   Subport ID
  * @param params
  *   Pipe profile parameters
  * @param pipe_profile_id
@@ -326,7 +328,8 @@ rte_sched_port_free(struct rte_sched_port *port);
  */
 __rte_experimental
 int
-rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+rte_sched_subport_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id);
 
@@ -357,7 +360,7 @@ rte_sched_subport_config(struct rte_sched_port *port,
  * @param pipe_id
  *   Pipe ID within subport
  * @param pipe_profile
- *   ID of port-level pre-configured pipe profile
+ *   ID of subport-level pre-configured pipe profile
  * @return
  *   0 upon success, error code otherwise
  */
lib/librte_sched/rte_sched_version.map +1 -1
index 729588794..f33761e63 100644
@@ -33,5 +33,5 @@ DPDK_2.1 {
 EXPERIMENTAL {
 	global:
 
-	rte_sched_port_pipe_profile_add;
+	rte_sched_subport_pipe_profile_add;
 };
Signature · 1 line
2.21.0
#40 lore raw

PATCH v3 6/15 sched: modify pkt enqueue for config flexibility

In reply to Jasvinder Singh

Modify scheduler packet enqueue operation of the scheduler to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 277 ++++++++++++++++++++++-------------
 1 file changed, 178 insertions(+), 99 deletions(-)
lib/librte_sched/rte_sched.c +178 -99
index 6b6219e45..a8174afad 100644
@@ -1440,10 +1440,11 @@ rte_sched_port_qindex(struct rte_sched_port *port,
 	uint32_t queue)
 {
 	return ((subport & (port->n_subports_per_port - 1)) <<
-			(port->n_pipes_per_subport_log2 + 4)) |
-			((pipe & (port->n_pipes_per_subport - 1)) << 4) |
-			((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
-			(RTE_SCHED_QUEUES_PER_PIPE - 1));
+		(port->n_pipes_per_subport_log2 + 4)) |
+		((pipe &
+		(port->subports[subport]->n_pipes_per_subport_enabled - 1)) << 4) |
+		((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
+		(RTE_SCHED_QUEUES_PER_PIPE - 1));
 }
 
 void
@@ -1468,7 +1469,8 @@ rte_sched_port_pkt_read_tree_path(struct rte_sched_port *port,
 	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
 
 	*subport = queue_id >> (port->n_pipes_per_subport_log2 + 4);
-	*pipe = (queue_id >> 4) & (port->n_pipes_per_subport - 1);
+	*pipe = (queue_id >> 4) &
+		(port->subports[*subport]->n_pipes_per_subport_enabled - 1);
 	*traffic_class = rte_sched_port_pipe_tc(port, queue_id);
 	*queue = rte_sched_port_tc_queue(port, queue_id);
 }
@@ -1512,7 +1514,7 @@ rte_sched_subport_read_stats(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
-	s = port->subport + subport_id;
+	s = port->subports[subport_id];
 
 	/* Copy subport stats and clear */
 	memcpy(stats, &s->stats, sizeof(struct rte_sched_subport_stats));
@@ -1585,43 +1587,50 @@ rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline void
-rte_sched_port_update_subport_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_subport_stats(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc[tc_index] += 1;
-	s->stats.n_bytes_tc[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc[tc_index] += 1;
+	subport->stats.n_bytes_tc[tc_index] += pkt_len;
 }
 
 #ifdef RTE_SCHED_RED
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc_dropped[tc_index] += 1;
-	s->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc_dropped[tc_index] += 1;
+	subport->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
 #ifdef RTE_SCHED_RED
-	s->stats.n_pkts_red_dropped[tc_index] += red;
+	subport->stats.n_pkts_red_dropped[tc_index] += red;
 #endif
 }
 
 static inline void
-rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_queue_stats(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts += 1;
@@ -1630,17 +1639,19 @@ rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex,
 
 #ifdef RTE_SCHED_RED
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts_dropped += 1;
@@ -1655,7 +1666,11 @@ rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
 #ifdef RTE_SCHED_RED
 
 static inline int
-rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint32_t qindex, uint16_t qlen)
+rte_sched_port_red_drop(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt,
+	uint32_t qindex,
+	uint16_t qlen)
 {
 	struct rte_sched_queue_extra *qe;
 	struct rte_red_config *red_cfg;
@@ -1665,12 +1680,12 @@ rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint3
 
 	tc_index = rte_sched_port_pipe_tc(port, qindex);
 	color = rte_sched_port_pkt_read_color(pkt);
-	red_cfg = &port->red_config[tc_index][color];
+	red_cfg = &subport->red_config[tc_index][color];
 
 	if ((red_cfg->min_th | red_cfg->max_th) == 0)
 		return 0;
 
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + qindex;
 	red = &qe->red;
 
 	return rte_red_enqueue(red_cfg, red, qlen, port->time);
@@ -1687,7 +1702,14 @@ rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t q
 
 #else
 
-#define rte_sched_port_red_drop(port, pkt, qindex, qlen)             0
+static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unused,
+	struct rte_sched_subport *subport __rte_unused,
+	struct rte_mbuf *pkt __rte_unused,
+	uint32_t qindex __rte_unused,
+	uint16_t qlen __rte_unused)
+{
+	return 0;
+}
 
 #define rte_sched_port_set_queue_empty_timestamp(port, qindex)
 
@@ -1722,63 +1744,79 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 
 #endif /* RTE_SCHED_DEBUG */
 
+static inline struct rte_sched_subport *
+rte_sched_port_subport(struct rte_sched_port *port,
+	struct rte_mbuf *pkt)
+{
+	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_id = queue_id >> (port->n_pipes_per_subport_log2 + 4);
+
+	return port->subports[subport_id];
+}
+
 static inline uint32_t
-rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_port *port,
-				       struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt, uint32_t subport_qmask)
 {
 	struct rte_sched_queue *q;
 #ifdef RTE_SCHED_COLLECT_STATS
 	struct rte_sched_queue_extra *qe;
 #endif
 	uint32_t qindex = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_queue_id = subport_qmask & qindex;
 
-	q = port->queue + qindex;
+	q = subport->queue + subport_queue_id;
 	rte_prefetch0(q);
 #ifdef RTE_SCHED_COLLECT_STATS
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + subport_queue_id;
 	rte_prefetch0(qe);
 #endif
 
-	return qindex;
+	return subport_queue_id;
 }
 
 static inline void
 rte_sched_port_enqueue_qwa_prefetch0(struct rte_sched_port *port,
-				     uint32_t qindex, struct rte_mbuf **qbase)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase)
 {
 	struct rte_sched_queue *q;
 	struct rte_mbuf **q_qw;
 	uint16_t qsize;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	q_qw = qbase + (q->qw & (qsize - 1));
 
 	rte_prefetch0(q_qw);
-	rte_bitmap_prefetch0(port->bmp, qindex);
+	rte_bitmap_prefetch0(subport->bmp, qindex);
 }
 
 static inline int
-rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
-			   struct rte_mbuf **qbase, struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qwa(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase,
+	struct rte_mbuf *pkt)
 {
 	struct rte_sched_queue *q;
 	uint16_t qsize;
 	uint16_t qlen;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	qlen = q->qw - q->qr;
 
 	/* Drop the packet (and update drop stats) when queue is full */
-	if (unlikely(rte_sched_port_red_drop(port, pkt, qindex, qlen) ||
+	if (unlikely(rte_sched_port_red_drop(port, subport, pkt, qindex, qlen) ||
 		     (qlen >= qsize))) {
 		rte_pktmbuf_free(pkt);
 #ifdef RTE_SCHED_COLLECT_STATS
-		rte_sched_port_update_subport_stats_on_drop(port, qindex, pkt,
-							    qlen < qsize);
-		rte_sched_port_update_queue_stats_on_drop(port, qindex, pkt,
-							  qlen < qsize);
+		rte_sched_port_update_subport_stats_on_drop(port, subport,
+			qindex, pkt, qlen < qsize);
+		rte_sched_port_update_queue_stats_on_drop(subport, qindex, pkt,
+			qlen < qsize);
 #endif
 		return 0;
 	}
@@ -1787,13 +1825,13 @@ rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
 	qbase[q->qw & (qsize - 1)] = pkt;
 	q->qw++;
 
-	/* Activate queue in the port bitmap */
-	rte_bitmap_set(port->bmp, qindex);
+	/* Activate queue in the subport bitmap */
+	rte_bitmap_set(subport->bmp, qindex);
 
 	/* Statistics */
 #ifdef RTE_SCHED_COLLECT_STATS
-	rte_sched_port_update_subport_stats(port, qindex, pkt);
-	rte_sched_port_update_queue_stats(port, qindex, pkt);
+	rte_sched_port_update_subport_stats(port, subport, qindex, pkt);
+	rte_sched_port_update_queue_stats(subport, qindex, pkt);
 #endif
 
 	return 1;
@@ -1821,17 +1859,22 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		*pkt30, *pkt31, *pkt_last;
 	struct rte_mbuf **q00_base, **q01_base, **q10_base, **q11_base,
 		**q20_base, **q21_base, **q30_base, **q31_base, **q_last_base;
+	struct rte_sched_subport *subport00, *subport01, *subport10, *subport11,
+		*subport20, *subport21, *subport30, *subport31, *subport_last;
 	uint32_t q00, q01, q10, q11, q20, q21, q30, q31, q_last;
 	uint32_t r00, r01, r10, r11, r20, r21, r30, r31, r_last;
+	uint32_t subport_qmask;
 	uint32_t result, i;
 
 	result = 0;
+	subport_qmask = (1 << (port->n_pipes_per_subport_log2 + 4)) - 1;
 
 	/*
 	 * Less then 6 input packets available, which is not enough to
 	 * feed the pipeline
 	 */
 	if (unlikely(n_pkts < 6)) {
+		struct rte_sched_subport *subports[5];
 		struct rte_mbuf **q_base[5];
 		uint32_t q[5];
 
@@ -1839,22 +1882,26 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		for (i = 0; i < n_pkts; i++)
 			rte_prefetch0(pkts[i]);
 
+		/* Prefetch the subport structure for each packet */
+		for (i = 0; i < n_pkts; i++)
+			subports[i] = rte_sched_port_subport(port, pkts[i]);
+
 		/* Prefetch the queue structure for each queue */
 		for (i = 0; i < n_pkts; i++)
-			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(port,
-								      pkts[i]);
+			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(subports[i],
+					pkts[i], subport_qmask);
 
 		/* Prefetch the write pointer location of each queue */
 		for (i = 0; i < n_pkts; i++) {
-			q_base[i] = rte_sched_port_qbase(port, q[i]);
-			rte_sched_port_enqueue_qwa_prefetch0(port, q[i],
-							     q_base[i]);
+			q_base[i] = rte_sched_subport_pipe_qbase(subports[i], q[i]);
+			rte_sched_port_enqueue_qwa_prefetch0(port, subports[i],
+				q[i], q_base[i]);
 		}
 
 		/* Write each packet to its queue */
 		for (i = 0; i < n_pkts; i++)
-			result += rte_sched_port_enqueue_qwa(port, q[i],
-							     q_base[i], pkts[i]);
+			result += rte_sched_port_enqueue_qwa(port, subports[i],
+						q[i], q_base[i], pkts[i]);
 
 		return result;
 	}
@@ -1870,21 +1917,29 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	rte_prefetch0(pkt10);
 	rte_prefetch0(pkt11);
 
-	q20 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt20);
-	q21 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt21);
+	subport20 = rte_sched_port_subport(port, pkt20);
+	subport21 = rte_sched_port_subport(port, pkt21);
+	q20 = rte_sched_port_enqueue_qptrs_prefetch0(subport20,
+			pkt20, subport_qmask);
+	q21 = rte_sched_port_enqueue_qptrs_prefetch0(subport21,
+			pkt21, subport_qmask);
 
 	pkt00 = pkts[4];
 	pkt01 = pkts[5];
 	rte_prefetch0(pkt00);
 	rte_prefetch0(pkt01);
 
-	q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-	q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+	subport10 = rte_sched_port_subport(port, pkt10);
+	subport11 = rte_sched_port_subport(port, pkt11);
+	q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+			pkt10, subport_qmask);
+	q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+			pkt11, subport_qmask);
 
-	q20_base = rte_sched_port_qbase(port, q20);
-	q21_base = rte_sched_port_qbase(port, q21);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+	q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+	q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 	/* Run the pipeline */
 	for (i = 6; i < (n_pkts & (~1)); i += 2) {
@@ -1899,6 +1954,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		q31 = q21;
 		q20 = q10;
 		q21 = q11;
+		subport30 = subport20;
+		subport31 = subport21;
+		subport20 = subport10;
+		subport21 = subport11;
 		q30_base = q20_base;
 		q31_base = q21_base;
 
@@ -1908,19 +1967,25 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		rte_prefetch0(pkt00);
 		rte_prefetch0(pkt01);
 
-		/* Stage 1: Prefetch queue structure storing queue pointers */
-		q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-		q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+		/* Stage 1: Prefetch subport and queue structure storing queue pointers */
+		subport10 = rte_sched_port_subport(port, pkt10);
+		subport11 = rte_sched_port_subport(port, pkt11);
+		q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+				pkt10, subport_qmask);
+		q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+				pkt11, subport_qmask);
 
 		/* Stage 2: Prefetch queue write location */
-		q20_base = rte_sched_port_qbase(port, q20);
-		q21_base = rte_sched_port_qbase(port, q21);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+		q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+		q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 		/* Stage 3: Write packet to queue and activate queue */
-		r30 = rte_sched_port_enqueue_qwa(port, q30, q30_base, pkt30);
-		r31 = rte_sched_port_enqueue_qwa(port, q31, q31_base, pkt31);
+		r30 = rte_sched_port_enqueue_qwa(port, subport30,
+				q30, q30_base, pkt30);
+		r31 = rte_sched_port_enqueue_qwa(port, subport31,
+				q31, q31_base, pkt31);
 		result += r30 + r31;
 	}
 
@@ -1932,38 +1997,52 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	pkt_last = pkts[n_pkts - 1];
 	rte_prefetch0(pkt_last);
 
-	q00 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt00);
-	q01 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt01);
-
-	q10_base = rte_sched_port_qbase(port, q10);
-	q11_base = rte_sched_port_qbase(port, q11);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q10, q10_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q11, q11_base);
-
-	r20 = rte_sched_port_enqueue_qwa(port, q20, q20_base, pkt20);
-	r21 = rte_sched_port_enqueue_qwa(port, q21, q21_base, pkt21);
+	subport00 = rte_sched_port_subport(port, pkt00);
+	subport01 = rte_sched_port_subport(port, pkt01);
+	q00 = rte_sched_port_enqueue_qptrs_prefetch0(subport00,
+			pkt00, subport_qmask);
+	q01 = rte_sched_port_enqueue_qptrs_prefetch0(subport01,
+			pkt01, subport_qmask);
+
+	q10_base = rte_sched_subport_pipe_qbase(subport10, q10);
+	q11_base = rte_sched_subport_pipe_qbase(subport11, q11);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport10, q10, q10_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport11, q11, q11_base);
+
+	r20 = rte_sched_port_enqueue_qwa(port, subport20,
+			q20, q20_base, pkt20);
+	r21 = rte_sched_port_enqueue_qwa(port, subport21,
+			q21, q21_base, pkt21);
 	result += r20 + r21;
 
-	q_last = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt_last);
+	subport_last = rte_sched_port_subport(port, pkt_last);
+	q_last = rte_sched_port_enqueue_qptrs_prefetch0(subport_last,
+				pkt_last, subport_qmask);
 
-	q00_base = rte_sched_port_qbase(port, q00);
-	q01_base = rte_sched_port_qbase(port, q01);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q00, q00_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q01, q01_base);
+	q00_base = rte_sched_subport_pipe_qbase(subport00, q00);
+	q01_base = rte_sched_subport_pipe_qbase(subport01, q01);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport00, q00, q00_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport01, q01, q01_base);
 
-	r10 = rte_sched_port_enqueue_qwa(port, q10, q10_base, pkt10);
-	r11 = rte_sched_port_enqueue_qwa(port, q11, q11_base, pkt11);
+	r10 = rte_sched_port_enqueue_qwa(port, subport10, q10,
+			q10_base, pkt10);
+	r11 = rte_sched_port_enqueue_qwa(port, subport11, q11,
+			q11_base, pkt11);
 	result += r10 + r11;
 
-	q_last_base = rte_sched_port_qbase(port, q_last);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q_last, q_last_base);
+	q_last_base = rte_sched_subport_pipe_qbase(subport_last, q_last);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport_last,
+		q_last, q_last_base);
 
-	r00 = rte_sched_port_enqueue_qwa(port, q00, q00_base, pkt00);
-	r01 = rte_sched_port_enqueue_qwa(port, q01, q01_base, pkt01);
+	r00 = rte_sched_port_enqueue_qwa(port, subport00, q00,
+			q00_base, pkt00);
+	r01 = rte_sched_port_enqueue_qwa(port, subport01, q01,
+			q01_base, pkt01);
 	result += r00 + r01;
 
 	if (n_pkts & 1) {
-		r_last = rte_sched_port_enqueue_qwa(port, q_last, q_last_base, pkt_last);
+		r_last = rte_sched_port_enqueue_qwa(port, subport_last,
+					q_last,	q_last_base, pkt_last);
 		result += r_last;
 	}
 
Signature · 1 line
2.21.0
#41 lore raw

PATCH v3 7/15 sched: update memory compute to support flexiblity

In reply to Jasvinder Singh

Update memory footprint compute function for allowing subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 100 ++++++++++-------------------------
 lib/librte_sched/rte_sched.h |   8 +--
 2 files changed, 34 insertions(+), 74 deletions(-)
lib/librte_sched/rte_sched.c +29 -71
index a8174afad..58f34359c 100644
@@ -486,70 +486,6 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 	return 0;
 }
 
-static uint32_t
-rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sched_port_array array)
-{
-	uint32_t n_subports_per_port = params->n_subports_per_port;
-	uint32_t n_pipes_per_subport = params->n_pipes_per_subport;
-	uint32_t n_pipes_per_port = n_pipes_per_subport * n_subports_per_port;
-	uint32_t n_queues_per_port = RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport * n_subports_per_port;
-
-	uint32_t size_subport = n_subports_per_port * sizeof(struct rte_sched_subport);
-	uint32_t size_pipe = n_pipes_per_port * sizeof(struct rte_sched_pipe);
-	uint32_t size_queue = n_queues_per_port * sizeof(struct rte_sched_queue);
-	uint32_t size_queue_extra
-		= n_queues_per_port * sizeof(struct rte_sched_queue_extra);
-	uint32_t size_pipe_profiles
-		= params->n_max_pipe_profiles * sizeof(struct rte_sched_pipe_profile);
-	uint32_t size_bmp_array = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	uint32_t size_per_pipe_queue_array, size_queue_array;
-
-	uint32_t base, i;
-
-	size_per_pipe_queue_array = 0;
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
-			size_per_pipe_queue_array +=
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-		else
-			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-	}
-	size_queue_array = n_pipes_per_port * size_per_pipe_queue_array;
-
-	base = 0;
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_SUBPORT)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_subport);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
-
-	return base;
-}
-
 static uint32_t
 rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
 	enum rte_sched_subport_array array)
@@ -870,22 +806,44 @@ rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 }
 
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params)
 {
-	uint32_t size0, size1;
+	uint32_t size0 = 0, size1 = 0, i;
 	int status;
 
-	status = rte_sched_port_check_params(params);
+	status = rte_sched_port_check_params(port_params);
 	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler port params check failed (%d)\n",
+			__func__, status);
 
 		return 0;
 	}
 
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		status = rte_sched_subport_check_params(sp,
+				port_params->n_pipes_per_subport,
+				port_params->rate);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Port scheduler subport params check failed (%d)\n",
+				__func__, status);
+
+			return 0;
+		}
+	}
+
 	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params,
-			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		size1 += rte_sched_subport_get_array_base(sp,
+					e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
+	}
 
 	return size0 + size1;
 }
lib/librte_sched/rte_sched.h +5 -3
index 5001f09ed..40f02f124 100644
@@ -373,14 +373,16 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 /**
  * Hierarchical scheduler memory footprint size per port
  *
- * @param params
+ * @param port_params
  *   Port scheduler configuration parameter structure
+ * @param subport_params
+ *   Array of subport parameter structures
  * @return
  *   Memory footprint size in bytes upon success, 0 otherwise
  */
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params);
-
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params);
 /*
  * Statistics
  *
Signature · 1 line
2.21.0
#42 lore raw

PATCH v3 8/15 sched: update grinder functions for config flexibility

In reply to Jasvinder Singh

Modify packet grinder functions of the schedule to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 223 +++++++++++++++++------------------
 1 file changed, 106 insertions(+), 117 deletions(-)
lib/librte_sched/rte_sched.c +106 -117
index 58f34359c..d7abe9661 100644
@@ -322,24 +322,6 @@ rte_sched_port_queues_per_port(struct rte_sched_port *port)
 	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
 }
 
-static inline struct rte_mbuf **
-rte_sched_port_qbase(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t pindex = qindex >> 4;
-	uint32_t qpos = qindex & 0xF;
-
-	return (port->queue_array + pindex *
-		port->qsize_sum + port->qsize_add[qpos]);
-}
-
-static inline uint16_t
-rte_sched_port_qsize(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
-
-	return port->qsize[tc];
-}
-
 static inline uint16_t
 rte_sched_port_pipe_queue(struct rte_sched_port *port, uint32_t traffic_class)
 {
@@ -1533,9 +1515,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 #ifdef RTE_SCHED_DEBUG
 
 static inline int
-rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_queue_is_empty(struct rte_sched_subport *subport,
+	uint32_t qindex)
 {
-	struct rte_sched_queue *queue = port->queue + qindex;
+	struct rte_sched_queue *queue = subport->queue + qindex;
 
 	return queue->qr == queue->qw;
 }
@@ -1650,9 +1633,10 @@ rte_sched_port_red_drop(struct rte_sched_port *port,
 }
 
 static inline void
-rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t qindex)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	struct rte_red *red = &qe->red;
 
 	rte_red_mark_queue_empty(red, port->time);
@@ -1669,14 +1653,14 @@ static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unus
 	return 0;
 }
 
-#define rte_sched_port_set_queue_empty_timestamp(port, qindex)
+#define rte_sched_port_set_queue_empty_timestamp(port, subport, qindex)
 
 #endif /* RTE_SCHED_RED */
 
 #ifdef RTE_SCHED_DEBUG
 
 static inline void
-debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
+debug_check_queue_slab(struct rte_sched_subport *subport, uint32_t bmp_pos,
 		       uint64_t bmp_slab)
 {
 	uint64_t mask;
@@ -1688,7 +1672,7 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 	panic = 0;
 	for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 		if (mask & bmp_slab) {
-			if (rte_sched_port_queue_is_empty(port, bmp_pos + i)) {
+			if (rte_sched_port_queue_is_empty(subport, bmp_pos + i)) {
 				printf("Queue %u (slab offset %u) is empty\n", bmp_pos + i, i);
 				panic = 1;
 			}
@@ -2010,10 +1994,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2051,10 +2035,9 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline uint32_t
-grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_tc_ov_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
 	uint32_t tc_ov_consumption[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t tc_consumption = 0, tc_ov_consumption_max;
 	uint32_t tc_ov_wm = subport->tc_ov_wm;
@@ -2094,10 +2077,10 @@ grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2117,7 +2100,7 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 
 	/* Subport TCs */
 	if (unlikely(port->time >= subport->tc_time)) {
-		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, pos);
+		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, subport);
 
 		for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 			subport->tc_credits[i] = subport->tc_credits_per_period[i];
@@ -2147,10 +2130,10 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2182,10 +2165,10 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2230,15 +2213,16 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline int
-grinder_schedule(struct rte_sched_port *port, uint32_t pos)
+grinder_schedule(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_queue *queue = grinder->queue[grinder->qpos];
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t pkt_len = pkt->pkt_len + port->frame_overhead;
 	uint32_t be_tc_active;
 
-	if (!grinder_credits_check(port, pos))
+	if (!grinder_credits_check(port, subport, pos))
 		return 0;
 
 	/* Advance port time */
@@ -2255,15 +2239,15 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 	if (queue->qr == queue->qw) {
 		uint32_t qindex = grinder->qindex[grinder->qpos];
 
-		rte_bitmap_clear(port->bmp, qindex);
+		rte_bitmap_clear(subport->bmp, qindex);
 		grinder->qmask &= ~(1 << grinder->qpos);
 		if (be_tc_active)
 			grinder->wrr_mask[grinder->qpos] = 0;
-		rte_sched_port_set_queue_empty_timestamp(port, qindex);
+		rte_sched_port_set_queue_empty_timestamp(port, subport, qindex);
 	}
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	grinder->productive = 1;
 
 	return 1;
@@ -2272,13 +2256,13 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 #ifdef SCHED_VECTOR_SSE4
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	__m128i index = _mm_set1_epi32(base_pipe);
-	__m128i pipes = _mm_load_si128((__m128i *)port->grinder_base_bmp_pos);
+	__m128i pipes = _mm_load_si128((__m128i *)subport->grinder_base_bmp_pos);
 	__m128i res = _mm_cmpeq_epi32(pipes, index);
 
-	pipes = _mm_load_si128((__m128i *)(port->grinder_base_bmp_pos + 4));
+	pipes = _mm_load_si128((__m128i *)(subport->grinder_base_bmp_pos + 4));
 	pipes = _mm_cmpeq_epi32(pipes, index);
 	res = _mm_or_si128(res, pipes);
 
@@ -2291,10 +2275,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #elif defined(SCHED_VECTOR_NEON)
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32x4_t index, pipes;
-	uint32_t *pos = (uint32_t *)port->grinder_base_bmp_pos;
+	uint32_t *pos = (uint32_t *)subport->grinder_base_bmp_pos;
 
 	index = vmovq_n_u32(base_pipe);
 	pipes = vld1q_u32(pos);
@@ -2311,12 +2295,12 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #else
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32_t i;
 
 	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++) {
-		if (port->grinder_base_bmp_pos[i] == base_pipe)
+		if (subport->grinder_base_bmp_pos[i] == base_pipe)
 			return 1;
 	}
 
@@ -2326,9 +2310,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #endif /* RTE_SCHED_OPTIMIZATIONS */
 
 static inline void
-grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
+grinder_pcache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t w[4];
 
 	grinder->pcache_w = 0;
@@ -2357,9 +2342,10 @@ grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_
 }
 
 static inline void
-grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qindex, uint16_t qmask)
+grinder_tccache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t qindex, uint16_t qmask)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint8_t b, i;
 
 	grinder->tccache_w = 0;
@@ -2380,9 +2366,10 @@ grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qin
 }
 
 static inline int
-grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
+grinder_next_tc(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_mbuf **qbase;
 	uint32_t qindex;
 	uint16_t qsize;
@@ -2391,15 +2378,15 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 0;
 
 	qindex = grinder->tccache_qindex[grinder->tccache_r];
-	qbase = rte_sched_port_qbase(port, qindex);
-	qsize = rte_sched_port_qsize(port, qindex);
+	qbase = rte_sched_subport_pipe_qbase(subport, qindex);
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 
 	grinder->tc_index = rte_sched_port_pipe_tc(port, qindex);
 	grinder->qmask = grinder->tccache_qmask[grinder->tccache_r];
 	grinder->qsize = qsize;
 
 	if (grinder->tc_index < RTE_SCHED_TRAFFIC_CLASS_BE) {
-		grinder->queue[0] = port->queue + qindex;
+		grinder->queue[0] = subport->queue + qindex;
 		grinder->qbase[0] = qbase;
 		grinder->qindex[0] = qindex;
 		grinder->tccache_r++;
@@ -2407,10 +2394,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 1;
 	}
 
-	grinder->queue[0] = port->queue + qindex;
-	grinder->queue[1] = port->queue + qindex + 1;
-	grinder->queue[2] = port->queue + qindex + 2;
-	grinder->queue[3] = port->queue + qindex + 3;
+	grinder->queue[0] = subport->queue + qindex;
+	grinder->queue[1] = subport->queue + qindex + 1;
+	grinder->queue[2] = subport->queue + qindex + 2;
+	grinder->queue[3] = subport->queue + qindex + 3;
 
 	grinder->qbase[0] = qbase;
 	grinder->qbase[1] = qbase + qsize;
@@ -2427,9 +2414,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline int
-grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_next_pipe(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t pipe_qindex;
 	uint16_t pipe_qmask;
 
@@ -2442,22 +2430,22 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 		uint32_t bmp_pos = 0;
 
 		/* Get another non-empty pipe group */
-		if (unlikely(rte_bitmap_scan(port->bmp, &bmp_pos, &bmp_slab) <= 0))
+		if (unlikely(rte_bitmap_scan(subport->bmp, &bmp_pos, &bmp_slab) <= 0))
 			return 0;
 
 #ifdef RTE_SCHED_DEBUG
-		debug_check_queue_slab(port, bmp_pos, bmp_slab);
+		debug_check_queue_slab(subport, bmp_pos, bmp_slab);
 #endif
 
 		/* Return if pipe group already in one of the other grinders */
-		port->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
-		if (unlikely(grinder_pipe_exists(port, bmp_pos)))
+		subport->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
+		if (unlikely(grinder_pipe_exists(subport, bmp_pos)))
 			return 0;
 
-		port->grinder_base_bmp_pos[pos] = bmp_pos;
+		subport->grinder_base_bmp_pos[pos] = bmp_pos;
 
 		/* Install new pipe group into grinder's pipe cache */
-		grinder_pcache_populate(port, pos, bmp_pos, bmp_slab);
+		grinder_pcache_populate(subport, pos, bmp_pos, bmp_slab);
 
 		pipe_qmask = grinder->pcache_qmask[0];
 		pipe_qindex = grinder->pcache_qindex[0];
@@ -2466,18 +2454,18 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 	/* Install new pipe in the grinder */
 	grinder->pindex = pipe_qindex >> 4;
-	grinder->subport = port->subport + (grinder->pindex / port->n_pipes_per_subport);
-	grinder->pipe = port->pipe + grinder->pindex;
+	grinder->subport = subport;
+	grinder->pipe = subport->pipe + grinder->pindex;
 	grinder->pipe_params = NULL; /* to be set after the pipe structure is prefetched */
 	grinder->productive = 0;
 
-	grinder_tccache_populate(port, pos, pipe_qindex, pipe_qmask);
-	grinder_next_tc(port, pos);
+	grinder_tccache_populate(subport, pos, pipe_qindex, pipe_qmask);
+	grinder_next_tc(port, subport, pos);
 
 	/* Check for pipe exhaustion */
-	if (grinder->pindex == port->pipe_loop) {
-		port->pipe_exhaustion = 1;
-		port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	if (grinder->pindex == subport->pipe_loop) {
+		subport->pipe_exhaustion = 1;
+		subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	}
 
 	return 1;
@@ -2485,9 +2473,9 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline void
-grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_load(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *pipe_params = grinder->pipe_params;
 	uint32_t qmask = grinder->qmask;
@@ -2513,9 +2501,9 @@ grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_store(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 
 	pipe->wrr_tokens[0] =
@@ -2533,9 +2521,9 @@ grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t wrr_tokens_min;
 
 	grinder->wrr_tokens[0] |= ~grinder->wrr_mask[0];
@@ -2553,21 +2541,21 @@ grinder_wrr(struct rte_sched_port *port, uint32_t pos)
 }
 
 
-#define grinder_evict(port, pos)
+#define grinder_evict(subport, pos)
 
 static inline void
-grinder_prefetch_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_pipe(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	rte_prefetch0(grinder->pipe);
 	rte_prefetch0(grinder->queue[0]);
 }
 
 static inline void
-grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_tc_queue_arrays(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t qsize, qr[RTE_SCHED_MAX_QUEUES_PER_TC];
 
 	qsize = grinder->qsize;
@@ -2588,17 +2576,17 @@ grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
 	rte_prefetch0(grinder->qbase[0] + qr[0]);
 	rte_prefetch0(grinder->qbase[1] + qr[1]);
 
-	grinder_wrr_load(port, pos);
-	grinder_wrr(port, pos);
+	grinder_wrr_load(subport, pos);
+	grinder_wrr(subport, pos);
 
 	rte_prefetch0(grinder->qbase[2] + qr[2]);
 	rte_prefetch0(grinder->qbase[3] + qr[3]);
 }
 
 static inline void
-grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t qpos = grinder->qpos;
 	struct rte_mbuf **qbase = grinder->qbase[qpos];
 	uint16_t qsize = grinder->qsize;
@@ -2617,14 +2605,15 @@ grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
 static inline uint32_t
 grinder_handle(struct rte_sched_port *port, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_subport *subport = port->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
 	case e_GRINDER_PREFETCH_PIPE:
 	{
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
-			port->busy_grinders++;
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
+			subport->busy_grinders++;
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return 0;
@@ -2637,9 +2626,9 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		struct rte_sched_pipe *pipe = grinder->pipe;
 
-		grinder->pipe_params = port->pipe_profiles + pipe->profile;
-		grinder_prefetch_tc_queue_arrays(port, pos);
-		grinder_credits_update(port, pos);
+		grinder->pipe_params = subport->pipe_profiles + pipe->profile;
+		grinder_prefetch_tc_queue_arrays(subport, pos);
+		grinder_credits_update(port, subport, pos);
 
 		grinder->state = e_GRINDER_PREFETCH_MBUF;
 		return 0;
@@ -2647,7 +2636,7 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 
 	case e_GRINDER_PREFETCH_MBUF:
 	{
-		grinder_prefetch_mbuf(port, pos);
+		grinder_prefetch_mbuf(subport, pos);
 
 		grinder->state = e_GRINDER_READ_MBUF;
 		return 0;
@@ -2657,47 +2646,47 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		uint32_t wrr_active, result = 0;
 
-		result = grinder_schedule(port, pos);
+		result = grinder_schedule(port, subport, pos);
 
 		wrr_active = (grinder->tc_index == RTE_SCHED_TRAFFIC_CLASS_BE);
 
 		/* Look for next packet within the same TC */
 		if (result && grinder->qmask) {
 			if (wrr_active)
-				grinder_wrr(port, pos);
+				grinder_wrr(subport, pos);
 
-			grinder_prefetch_mbuf(port, pos);
+			grinder_prefetch_mbuf(subport, pos);
 
 			return 1;
 		}
 
 		if (wrr_active)
-			grinder_wrr_store(port, pos);
+			grinder_wrr_store(subport, pos);
 
 		/* Look for another active TC within same pipe */
-		if (grinder_next_tc(port, pos)) {
-			grinder_prefetch_tc_queue_arrays(port, pos);
+		if (grinder_next_tc(port, subport, pos)) {
+			grinder_prefetch_tc_queue_arrays(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_MBUF;
 			return result;
 		}
 
 		if (grinder->productive == 0 &&
-		    port->pipe_loop == RTE_SCHED_PIPE_INVALID)
-			port->pipe_loop = grinder->pindex;
+		    subport->pipe_loop == RTE_SCHED_PIPE_INVALID)
+			subport->pipe_loop = grinder->pindex;
 
-		grinder_evict(port, pos);
+		grinder_evict(subport, pos);
 
 		/* Look for another active pipe */
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return result;
 		}
 
 		/* No active pipe found */
-		port->busy_grinders--;
+		subport->busy_grinders--;
 
 		grinder->state = e_GRINDER_PREFETCH_PIPE;
 		return result;
Signature · 1 line
2.21.0
#43 lore raw

PATCH v3 9/15 sched: update pkt dequeue for flexible config

In reply to Jasvinder Singh

Modify scheduler packet dequeue operation to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 51 ++++++++++++++++++++++++++++--------
 1 file changed, 40 insertions(+), 11 deletions(-)
lib/librte_sched/rte_sched.c +40 -11
index d7abe9661..e14506b2d 100644
@@ -243,6 +243,7 @@ struct rte_sched_port {
 	uint32_t busy_grinders;
 	struct rte_mbuf **pkts_out;
 	uint32_t n_pkts_out;
+	uint32_t subport_id;
 
 	/* Queue base calculation */
 	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
@@ -898,6 +899,7 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	/* Grinders */
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
+	port->subport_id = 0;
 
 	return port;
 }
@@ -2603,9 +2605,9 @@ grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 }
 
 static inline uint32_t
-grinder_handle(struct rte_sched_port *port, uint32_t pos)
+grinder_handle(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_subport *subport = port->subport;
 	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
@@ -2704,6 +2706,7 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 	uint64_t cycles = rte_get_tsc_cycles();
 	uint64_t cycles_diff = cycles - port->time_cpu_cycles;
 	uint64_t bytes_diff;
+	uint32_t i;
 
 	/* Compute elapsed time in bytes */
 	bytes_diff = rte_reciprocal_divide(cycles_diff << RTE_SCHED_TIME_SHIFT,
@@ -2716,20 +2719,21 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 		port->time = port->time_cpu_bytes;
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	for (i = 0; i < port->n_subports_per_port; i++)
+		port->subports[i]->pipe_loop = RTE_SCHED_PIPE_INVALID;
 }
 
 static inline int
-rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
+rte_sched_port_exceptions(struct rte_sched_subport *subport, int second_pass)
 {
 	int exceptions;
 
 	/* Check if any exception flag is set */
-	exceptions = (second_pass && port->busy_grinders == 0) ||
-		(port->pipe_exhaustion == 1);
+	exceptions = (second_pass && subport->busy_grinders == 0) ||
+		(subport->pipe_exhaustion == 1);
 
 	/* Clear exception flags */
-	port->pipe_exhaustion = 0;
+	subport->pipe_exhaustion = 0;
 
 	return exceptions;
 }
@@ -2737,7 +2741,9 @@ rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
 int
 rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts)
 {
-	uint32_t i, count;
+	struct rte_sched_subport *subport;
+	uint32_t subport_id = port->subport_id;
+	uint32_t i, n_subports = 0, count;
 
 	port->pkts_out = pkts;
 	port->n_pkts_out = 0;
@@ -2746,9 +2752,32 @@ rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint
 
 	/* Take each queue in the grinder one step further */
 	for (i = 0, count = 0; ; i++)  {
-		count += grinder_handle(port, i & (RTE_SCHED_PORT_N_GRINDERS - 1));
-		if ((count == n_pkts) ||
-		    rte_sched_port_exceptions(port, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+		subport = port->subports[subport_id];
+
+		count += grinder_handle(port, subport,
+				i & (RTE_SCHED_PORT_N_GRINDERS - 1));
+
+		if (count == n_pkts) {
+			subport_id++;
+
+			if (subport_id == port->n_subports_per_port)
+				subport_id = 0;
+
+			port->subport_id = subport_id;
+			break;
+		}
+
+		if (rte_sched_port_exceptions(subport, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+			i = 0;
+			subport_id++;
+			n_subports++;
+		}
+
+		if (subport_id == port->n_subports_per_port)
+			subport_id = 0;
+
+		if (n_subports == port->n_subports_per_port) {
+			port->subport_id = subport_id;
 			break;
 		}
 	}
Signature · 1 line
2.21.0
#44 lore raw

PATCH v3 10/15 sched: update queue stats read for config flexibility

In reply to Jasvinder Singh

Modify pipe queue stats read function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 28 +++++++++++++++-------------
 1 file changed, 15 insertions(+), 13 deletions(-)
lib/librte_sched/rte_sched.c +15 -13
index e14506b2d..1faa580d0 100644
@@ -282,16 +282,6 @@ enum rte_sched_subport_array {
 	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
 };
 
-#ifdef RTE_SCHED_COLLECT_STATS
-
-static inline uint32_t
-rte_sched_port_queues_per_subport(struct rte_sched_port *port)
-{
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport;
-}
-
-#endif
-
 static inline uint32_t
 rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
 {
@@ -320,7 +310,12 @@ struct rte_sched_subport *subport, uint32_t qindex)
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
+	uint32_t n_queues = 0, i;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		n_queues += rte_sched_subport_pipe_queues(port->subports[i]);
+
+	return n_queues;
 }
 
 static inline uint16_t
@@ -1474,8 +1469,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 	struct rte_sched_queue_stats *stats,
 	uint16_t *qlen)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_queue *q;
 	struct rte_sched_queue_extra *qe;
+	uint32_t subport_id, subport_qmask, subport_qindex;
 
 	/* Check user parameters */
 	if (port == NULL) {
@@ -1501,8 +1498,13 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 			"%s: Incorrect value for parameter qlen\n", __func__);
 		return -EINVAL;
 	}
-	q = port->queue + queue_id;
-	qe = port->queue_extra + queue_id;
+	subport_qmask = port->n_pipes_per_subport_log2 + 4;
+	subport_id = (queue_id >> subport_qmask) & (port->n_subports_per_port - 1);
+
+	s = port->subports[subport_id];
+	subport_qindex = ((1 << subport_qmask) - 1) & queue_id;
+	q = s->queue + subport_qindex;
+	qe = s->queue_extra + subport_qindex;
 
 	/* Copy queue stats and clear */
 	memcpy(stats, &qe->stats, sizeof(struct rte_sched_queue_stats));
Signature · 1 line
2.21.0
#45 lore raw

PATCH v3 11/15 test/sched: modify tests for subport config flexibility

In reply to Jasvinder Singh

Modify tests function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 app/test/test_sched.c | 35 ++++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 17 deletions(-)
app/test/test_sched.c +18 -17
index afe0b0765..fc31080ef 100644
@@ -22,18 +22,6 @@
 #define TC              2
 #define QUEUE           0
 
-static struct rte_sched_subport_params subport_param[] = {
-	{
-		.tb_rate = 1250000000,
-		.tb_size = 1000000,
-
-		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000},
-		.tc_period = 10,
-	},
-};
-
 static struct rte_sched_pipe_params pipe_profile[] = {
 	{ /* Profile #0 */
 		.tb_rate = 305175,
@@ -48,6 +36,23 @@ static struct rte_sched_pipe_params pipe_profile[] = {
 	},
 };
 
+static struct rte_sched_subport_params subport_param[] = {
+	{
+		.tb_rate = 1250000000,
+		.tb_size = 1000000,
+
+		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000},
+		.tc_period = 10,
+		.n_pipes_per_subport_enabled = 1024,
+		.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
+		.pipe_profiles = pipe_profile,
+		.n_pipe_profiles = 1,
+		.n_max_pipe_profiles = 1,
+	},
+};
+
 static struct rte_sched_port_params port_param = {
 	.socket = 0, /* computed */
 	.rate = 0, /* computed */
@@ -55,10 +60,6 @@ static struct rte_sched_port_params port_param = {
 	.frame_overhead = RTE_SCHED_FRAME_OVERHEAD_DEFAULT,
 	.n_subports_per_port = 1,
 	.n_pipes_per_subport = 1024,
-	.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
-	.pipe_profiles = pipe_profile,
-	.n_pipe_profiles = 1,
-	.n_max_pipe_profiles = 1,
 };
 
 #define NB_MBUF          32
@@ -140,7 +141,7 @@ test_sched(void)
 	err = rte_sched_subport_config(port, SUBPORT, subport_param);
 	TEST_ASSERT_SUCCESS(err, "Error config sched, err=%d\n", err);
 
-	for (pipe = 0; pipe < port_param.n_pipes_per_subport; pipe ++) {
+	for (pipe = 0; pipe < subport_param[0].n_pipes_per_subport_enabled; pipe++) {
 		err = rte_sched_pipe_config(port, SUBPORT, pipe, 0);
 		TEST_ASSERT_SUCCESS(err, "Error config sched pipe %u, err=%d\n", pipe, err);
 	}
Signature · 1 line
2.21.0
#46 lore raw

PATCH v3 12/15 net/softnic: add subport config flexibility to TM

In reply to Jasvinder Singh

Modify softnic traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 drivers/net/softnic/rte_eth_softnic_tm.c | 54 ++++++++++++------------
 1 file changed, 28 insertions(+), 26 deletions(-)
drivers/net/softnic/rte_eth_softnic_tm.c +28 -26
index 61c3adc82..80a470c9e 100644
@@ -85,7 +85,8 @@ softnic_tmgr_port_create(struct pmd_internals *p,
 	/* Subport */
 	n_subports = t->port_params.n_subports_per_port;
 	for (subport_id = 0; subport_id < n_subports; subport_id++) {
-		uint32_t n_pipes_per_subport = t->port_params.n_pipes_per_subport;
+		uint32_t n_pipes_per_subport =
+			t->subport_params[subport_id].n_pipes_per_subport_enabled;
 		uint32_t pipe_id;
 		int status;
 
@@ -2211,10 +2212,11 @@ tm_tc_wred_profile_get(struct rte_eth_dev *dev, uint32_t tc_id)
 #ifdef RTE_SCHED_RED
 
 static void
-wred_profiles_set(struct rte_eth_dev *dev)
+wred_profiles_set(struct rte_eth_dev *dev, uint32_t subport_id)
 {
 	struct pmd_internals *p = dev->data->dev_private;
-	struct rte_sched_port_params *pp = &p->soft.tm.params.port_params;
+	struct rte_sched_subport_params *pp =
+		&p->soft.tm.params.subport_params[subport_id];
 
 	uint32_t tc_id;
 	enum rte_color color;
@@ -2234,7 +2236,7 @@ wred_profiles_set(struct rte_eth_dev *dev)
 
 #else
 
-#define wred_profiles_set(dev)
+#define wred_profiles_set(dev, subport_id)
 
 #endif
 
@@ -2526,29 +2528,9 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 		.frame_overhead =
 			root->shaper_profile->params.pkt_length_adjust,
 		.n_subports_per_port = root->n_children,
-		.n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
-			h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
-		.qsize = {p->params.tm.qsize[0],
-			p->params.tm.qsize[1],
-			p->params.tm.qsize[2],
-			p->params.tm.qsize[3],
-			p->params.tm.qsize[4],
-			p->params.tm.qsize[5],
-			p->params.tm.qsize[6],
-			p->params.tm.qsize[7],
-			p->params.tm.qsize[8],
-			p->params.tm.qsize[9],
-			p->params.tm.qsize[10],
-			p->params.tm.qsize[11],
-			p->params.tm.qsize[12],
-		},
-		.pipe_profiles = t->pipe_profiles,
-		.n_pipe_profiles = t->n_pipe_profiles,
-		.n_max_pipe_profiles = TM_MAX_PIPE_PROFILE,
+		.n_pipes_per_subport = TM_MAX_PIPES_PER_SUBPORT,
 	};
 
-	wred_profiles_set(dev);
-
 	subport_id = 0;
 	TAILQ_FOREACH(n, nl, node) {
 		uint64_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
@@ -2588,8 +2570,28 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 					tc_rate[12],
 				},
 				.tc_period = SUBPORT_TC_PERIOD,
+				.n_pipes_per_subport_enabled =
+					h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+					h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
+				.qsize = {p->params.tm.qsize[0],
+					p->params.tm.qsize[1],
+					p->params.tm.qsize[2],
+					p->params.tm.qsize[3],
+					p->params.tm.qsize[4],
+					p->params.tm.qsize[5],
+					p->params.tm.qsize[6],
+					p->params.tm.qsize[7],
+					p->params.tm.qsize[8],
+					p->params.tm.qsize[9],
+					p->params.tm.qsize[10],
+					p->params.tm.qsize[11],
+					p->params.tm.qsize[12],
+				},
+				.pipe_profiles = t->pipe_profiles,
+				.n_pipe_profiles = t->n_pipe_profiles,
+				.n_max_pipe_profiles = TM_MAX_PIPE_PROFILE,
 		};
-
+		wred_profiles_set(dev, subport_id);
 		subport_id++;
 	}
 }
Signature · 1 line
2.21.0
#47 lore raw

PATCH v3 13/15 ip_pipeline: add subport config flexibility to TM

In reply to Jasvinder Singh

Modify ip pipeline traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 examples/ip_pipeline/cli.c  | 71 ++++++++++++++++++-------------------
 examples/ip_pipeline/tmgr.c | 25 +++++++------
 examples/ip_pipeline/tmgr.h |  7 ++--
 3 files changed, 51 insertions(+), 52 deletions(-)
examples/ip_pipeline/cli.c +35 -36
index c6cf4204e..e37a84444 100644
@@ -380,7 +380,12 @@ static const char cmd_tmgr_subport_profile_help[] =
 "   <tc0_rate> <tc1_rate> <tc2_rate> <tc3_rate> <tc4_rate>"
 "        <tc5_rate> <tc6_rate> <tc7_rate> <tc8_rate>"
 "        <tc9_rate> <tc10_rate> <tc11_rate> <tc12_rate>\n"
-"   <tc_period>\n";
+"   <tc_period>\n"
+"   pps <n_pipes_per_subport>\n"
+"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
+"       <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
+"       <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
+"       <qsize_tc11> <qsize_tc12>";
 
 static void
 cmd_tmgr_subport_profile(char **tokens,
@@ -391,7 +396,7 @@ cmd_tmgr_subport_profile(char **tokens,
 	struct rte_sched_subport_params p;
 	int status, i;
 
-	if (n_tokens != 19) {
+	if (n_tokens != 35) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -417,6 +422,27 @@ cmd_tmgr_subport_profile(char **tokens,
 		return;
 	}
 
+	if (strcmp(tokens[19], "pps") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
+		return;
+	}
+
+	if (parser_read_uint32(&p.n_pipes_per_subport_enabled, tokens[20]) != 0) {
+		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
+		return;
+	}
+
+	if (strcmp(tokens[21], "qsize") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
+		return;
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		if (parser_read_uint16(&p.qsize[i], tokens[22 + i]) != 0) {
+			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
+			return;
+		}
+
 	status = tmgr_subport_profile_add(&p);
 	if (status != 0) {
 		snprintf(out, out_size, MSG_CMD_FAIL, tokens[0]);
@@ -491,11 +517,6 @@ static const char cmd_tmgr_help[] =
 "tmgr <tmgr_name>\n"
 "   rate <rate>\n"
 "   spp <n_subports_per_port>\n"
-"   pps <n_pipes_per_subport>\n"
-"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
-"   <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
-"   <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
-"   <qsize_tc11> <qsize_tc12>\n"
 "   fo <frame_overhead>\n"
 "   mtu <mtu>\n"
 "   cpu <cpu_id>\n";
@@ -509,9 +530,8 @@ cmd_tmgr(char **tokens,
 	struct tmgr_port_params p;
 	char *name;
 	struct tmgr_port *tmgr_port;
-	int i;
 
-	if (n_tokens != 28) {
+	if (n_tokens != 12) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -538,53 +558,32 @@ cmd_tmgr(char **tokens,
 		return;
 	}
 
-	if (strcmp(tokens[6], "pps") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
-		return;
-	}
-
-	if (parser_read_uint32(&p.n_pipes_per_subport, tokens[7]) != 0) {
-		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
-		return;
-	}
-
-	if (strcmp(tokens[8], "qsize") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
-		return;
-	}
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (parser_read_uint16(&p.qsize[i], tokens[9 + i]) != 0) {
-			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
-			return;
-		}
-
-	if (strcmp(tokens[22], "fo") != 0) {
+	if (strcmp(tokens[6], "fo") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "fo");
 		return;
 	}
 
-	if (parser_read_uint32(&p.frame_overhead, tokens[23]) != 0) {
+	if (parser_read_uint32(&p.frame_overhead, tokens[7]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "frame_overhead");
 		return;
 	}
 
-	if (strcmp(tokens[24], "mtu") != 0) {
+	if (strcmp(tokens[8], "mtu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "mtu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.mtu, tokens[25]) != 0) {
+	if (parser_read_uint32(&p.mtu, tokens[9]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "mtu");
 		return;
 	}
 
-	if (strcmp(tokens[26], "cpu") != 0) {
+	if (strcmp(tokens[10], "cpu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "cpu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.cpu_id, tokens[27]) != 0) {
+	if (parser_read_uint32(&p.cpu_id, tokens[11]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "cpu_id");
 		return;
 	}
examples/ip_pipeline/tmgr.c +12 -13
index 40cbf1d0a..91ccbf60f 100644
@@ -47,7 +47,8 @@ int
 tmgr_subport_profile_add(struct rte_sched_subport_params *p)
 {
 	/* Check input params */
-	if (p == NULL)
+	if (p == NULL ||
+		p->n_pipes_per_subport_enabled == 0)
 		return -1;
 
 	/* Save profile */
@@ -90,7 +91,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 		tmgr_port_find(name) ||
 		(params == NULL) ||
 		(params->n_subports_per_port == 0) ||
-		(params->n_pipes_per_subport == 0) ||
 		(params->cpu_id >= RTE_MAX_NUMA_NODES) ||
 		(n_subport_profiles == 0) ||
 		(n_pipe_profiles == 0))
@@ -103,18 +103,16 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	p.mtu = params->mtu;
 	p.frame_overhead = params->frame_overhead;
 	p.n_subports_per_port = params->n_subports_per_port;
-	p.n_pipes_per_subport = params->n_pipes_per_subport;
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		p.qsize[i] = params->qsize[i];
-
-	p.pipe_profiles = pipe_profile;
-	p.n_pipe_profiles = n_pipe_profiles;
+	p.n_pipes_per_subport = TMGR_PIPE_SUBPORT_MAX;
 
 	s = rte_sched_port_config(&p);
 	if (s == NULL)
 		return NULL;
 
+	subport_profile[0].pipe_profiles = pipe_profile;
+	subport_profile[0].n_pipe_profiles = n_pipe_profiles;
+	subport_profile[0].n_max_pipe_profiles = TMGR_PIPE_PROFILE_MAX;
+
 	for (i = 0; i < params->n_subports_per_port; i++) {
 		int status;
 
@@ -128,7 +126,7 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 			return NULL;
 		}
 
-		for (j = 0; j < params->n_pipes_per_subport; j++) {
+		for (j = 0; j < subport_profile[0].n_pipes_per_subport_enabled; j++) {
 			status = rte_sched_pipe_config(
 				s,
 				i,
@@ -153,7 +151,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	strlcpy(tmgr_port->name, name, sizeof(tmgr_port->name));
 	tmgr_port->s = s;
 	tmgr_port->n_subports_per_port = params->n_subports_per_port;
-	tmgr_port->n_pipes_per_subport = params->n_pipes_per_subport;
 
 	/* Node add to list */
 	TAILQ_INSERT_TAIL(&tmgr_port_list, tmgr_port, node);
@@ -205,8 +202,10 @@ tmgr_pipe_config(const char *port_name,
 	port = tmgr_port_find(port_name);
 	if ((port == NULL) ||
 		(subport_id >= port->n_subports_per_port) ||
-		(pipe_id_first >= port->n_pipes_per_subport) ||
-		(pipe_id_last >= port->n_pipes_per_subport) ||
+		(pipe_id_first >=
+			subport_profile[subport_id].n_pipes_per_subport_enabled) ||
+		(pipe_id_last >=
+			subport_profile[subport_id].n_pipes_per_subport_enabled) ||
 		(pipe_id_first > pipe_id_last) ||
 		(pipe_profile_id >= n_pipe_profiles))
 		return -1;
examples/ip_pipeline/tmgr.h +4 -3
index 8703a2e00..1fcf66ee1 100644
@@ -12,6 +12,10 @@
 
 #include "common.h"
 
+#ifndef TMGR_PIPE_SUBPORT_MAX
+#define TMGR_PIPE_SUBPORT_MAX                              4096
+#endif
+
 #ifndef TMGR_SUBPORT_PROFILE_MAX
 #define TMGR_SUBPORT_PROFILE_MAX                           256
 #endif
@@ -25,7 +29,6 @@ struct tmgr_port {
 	char name[NAME_SIZE];
 	struct rte_sched_port *s;
 	uint32_t n_subports_per_port;
-	uint32_t n_pipes_per_subport;
 };
 
 TAILQ_HEAD(tmgr_port_list, tmgr_port);
@@ -42,8 +45,6 @@ struct tmgr_port_params {
 	uint32_t frame_overhead;
 	uint32_t mtu;
 	uint32_t cpu_id;
-	uint32_t n_pipes_per_subport;
-	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 };
 
 int
Signature · 1 line
2.21.0
#48 lore raw

PATCH v3 14/15 examples/qos_sched: add subport configuration flexibility

In reply to Jasvinder Singh

Modify qos sample app to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 examples/qos_sched/app_thread.c   |  20 ++-
 examples/qos_sched/cfg_file.c     | 229 ++++++++++++++++--------------
 examples/qos_sched/init.c         |  54 +++----
 examples/qos_sched/main.h         |   1 +
 examples/qos_sched/profile.cfg    |   5 +-
 examples/qos_sched/profile_ov.cfg |   5 +-
 examples/qos_sched/stats.c        |  44 +++---
 7 files changed, 203 insertions(+), 155 deletions(-)
examples/qos_sched/app_thread.c +15 -5
index 18b734a14..dbc878b55 100644
@@ -35,15 +35,25 @@ get_pkt_sched(struct rte_mbuf *m, uint32_t *subport, uint32_t *pipe,
 	uint16_t *pdata = rte_pktmbuf_mtod(m, uint16_t *);
 	uint16_t pipe_queue;
 
+	/* Outer VLAN ID*/
 	*subport = (rte_be_to_cpu_16(pdata[SUBPORT_OFFSET]) & 0x0FFF) &
-			(port_params.n_subports_per_port - 1); /* Outer VLAN ID*/
+		(port_params.n_subports_per_port - 1);
+
+	/* Inner VLAN ID */
 	*pipe = (rte_be_to_cpu_16(pdata[PIPE_OFFSET]) & 0x0FFF) &
-			(port_params.n_pipes_per_subport - 1); /* Inner VLAN ID */
+		(subport_params[*subport].n_pipes_per_subport_enabled - 1);
+
 	pipe_queue = active_queues[(pdata[QUEUE_OFFSET] >> 8) % n_active_queues];
+
+	/* Traffic class (Destination IP) */
 	*traffic_class = pipe_queue > RTE_SCHED_TRAFFIC_CLASS_BE ?
-			RTE_SCHED_TRAFFIC_CLASS_BE : pipe_queue; /* Destination IP */
-	*queue = pipe_queue - *traffic_class; /* Destination IP */
-	*color = pdata[COLOR_OFFSET] & 0x03; /* Destination IP */
+			RTE_SCHED_TRAFFIC_CLASS_BE : pipe_queue;
+
+	/* Traffic class queue (Destination IP) */
+	*queue = pipe_queue - *traffic_class;
+
+	/* Color (Destination IP) */
+	*color = pdata[COLOR_OFFSET] & 0x03;
 
 	return 0;
 }
examples/qos_sched/cfg_file.c +124 -105

269 lines not on this page · raw message

examples/qos_sched/init.c +29 -25

79 lines not on this page · raw message

examples/qos_sched/main.h +1 -0
index d8f890b64..baa2b3ead 100644
@@ -152,6 +152,7 @@ uint32_t active_queues[RTE_SCHED_QUEUES_PER_PIPE];
 uint32_t n_active_queues;
 
 extern struct rte_sched_port_params port_params;
+extern struct rte_sched_subport_params subport_params[MAX_SCHED_SUBPORTS];
 
 int app_parse_args(int argc, char **argv);
 int app_init(void);
examples/qos_sched/profile.cfg +3 -2
index 335561370..61b8b7071 100644
@@ -20,11 +20,12 @@
 [port]
 frame overhead = 24
 number of subports per port = 1
-number of pipes per subport = 4096
-queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
 
 ; Subport configuration
 [subport 0]
+number of pipes per subport = 4096
+queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
+
 tb rate = 1250000000           ; Bytes per second
 tb size = 1000000              ; Bytes
 
examples/qos_sched/profile_ov.cfg +3 -2
index 394987399..ab509d28d 100644
@@ -5,11 +5,12 @@
 [port]
 frame overhead = 24
 number of subports per port = 1
-number of pipes per subport = 32
-queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
 
 ; Subport configuration
 [subport 0]
+number of pipes per subport = 32
+queue sizes = 64 64 64 64 64 64 64 64 64 64 64 64 64
+
 tb rate = 8400000           ; Bytes per second
 tb size = 100000            ; Bytes
 
examples/qos_sched/stats.c +28 -16

125 lines not on this page · raw message

Signature · 1 line
2.21.0
#49 lore raw

PATCH v3 15/15 sched: remove redundant code

In reply to Jasvinder Singh

Remove redundant data structure fields from port level data structures and update release notes.

---
 doc/guides/rel_notes/release_19_11.rst |  6 +++-
 lib/librte_sched/rte_sched.c           | 42 +-------------------------
 lib/librte_sched/rte_sched.h           | 22 --------------
 3 files changed, 6 insertions(+), 64 deletions(-)
doc/guides/rel_notes/release_19_11.rst +5 -1
index 27cfbd9e3..dd122f00a 100644
@@ -94,6 +94,10 @@ API Changes
    Also, make sure to start the actual text at the margin.
    =========================================================
 
+* sched: The pipe nodes configuration parameters such as number of pipes,
+  pipe queue sizes, pipe profiles, etc., are moved from port level structure
+  to subport level. This allows different subports of the same port to
+  have different configuration for the pipe nodes.
 
 ABI Changes
 -----------
@@ -181,7 +185,7 @@ The libraries prepended with a plus sign were incremented in this version.
      librte_rcu.so.1
      librte_reorder.so.1
      librte_ring.so.2
-     librte_sched.so.3
+   + librte_sched.so.4
      librte_security.so.2
      librte_stack.so.1
      librte_table.so.3
lib/librte_sched/rte_sched.c +1 -41

62 lines not on this page · raw message

lib/librte_sched/rte_sched.h +0 -22
index 40f02f124..c82c23c14 100644
@@ -260,28 +260,6 @@ struct rte_sched_port_params {
 	 * the subports of the same port.
 	 */
 	uint32_t n_pipes_per_subport;
-
-	/** Packet queue size for each traffic class.
-	 * All the pipes within the same subport share the similar
-	 * configuration for the queues.
-	 */
-	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-
-	/** Pipe profile table.
-	 * Every pipe is configured using one of the profiles from this table.
-	 */
-	struct rte_sched_pipe_params *pipe_profiles;
-
-	/** Profiles in the pipe profile table */
-	uint32_t n_pipe_profiles;
-
-	/** Max profiles allowed in the pipe profile table */
-	uint32_t n_max_pipe_profiles;
-
-#ifdef RTE_SCHED_RED
-	/** RED parameters */
-	struct rte_red_params red_params[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][RTE_COLORS];
-#endif
 };
 
 /*
Signature · 1 line
2.21.0
#50 lore raw

v4 cover sched: subport level configuration of pipe nodes

In reply to Jasvinder Singh

This patchset refactors the dpdk qos sched library to allow subport level configuration flexibility of the pipe nodes.

Currently, all parameters for the pipe nodes (subscribers) configuration are part of the port level structure which forces all groups of subscribers (pipes) in different subports to have similar configurations in terms of their number, queue sizes, traffic-classes, etc.

The new implementation moves pipe nodes configuration parameters from port level to subport level structure. This allows different subports of the same port to have different configuration for the pipe nodes, for examples- number of pipes, queue sizes, pipe profiles, etc.

In order to keep the implementation complexity under control, all pipes within the same subport share the same configuration for queue sizes.

v4:
- remove deprecation note
- rebase on current dpdk head
- add 64-bit values support

v3:
- improve doxygen comments

v2:
- fix qsize parsing in sample application
- fix checkpatch warnings

Jasvinder Singh (17):
sched: add pipe config params to subport struct
sched: modify internal structs for config flexibility
sched: remove pipe params config from port level
sched: add pipe config to subport level
sched: modify pipe functions for config flexibility
sched: modify pkt enqueue for config flexibility
sched: update memory compute to support flexiblity
sched: update grinder functions for config flexibility
sched: update pkt dequeue for flexible config
sched: update queue stats read for config flexibility test/sched: modify tests for subport config flexibility net/softnic: add subport config flexibility to TM
ip_pipeline: add subport config flexibility to TM examples/qos_sched: add subport configuration flexibility
sched: remove redundant code
sched: add support for 64 bit values
sched: modify internal structs and functions for 64 bit values

 app/test/test_sched.c                    |   35 +-
 doc/guides/rel_notes/deprecation.rst     |    6 -
 doc/guides/rel_notes/release_19_11.rst   |    7 +-
 drivers/net/softnic/rte_eth_softnic_tm.c |   54 +-
 examples/ip_pipeline/cli.c               |   89 +-
 examples/ip_pipeline/tmgr.c              |   25 +-
 examples/ip_pipeline/tmgr.h              |    9 +-
 examples/qos_sched/app_thread.c          |   20 +-
 examples/qos_sched/cfg_file.c            |  293 ++--
 examples/qos_sched/init.c                |   54 +-
 examples/qos_sched/main.h                |    1 +
 examples/qos_sched/profile.cfg           |    5 +-
 examples/qos_sched/profile_ov.cfg        |    5 +-
 examples/qos_sched/stats.c               |   50 +-
 lib/librte_sched/Makefile                |    2 +-
 lib/librte_sched/meson.build             |    2 +-
 lib/librte_sched/rte_approx.c            |   57 +-
 lib/librte_sched/rte_approx.h            |    3 +-
 lib/librte_sched/rte_sched.c             | 1575 +++++++++++++---------
 lib/librte_sched/rte_sched.h             |  157 ++-
 lib/librte_sched/rte_sched_common.h      |   12 +-
 lib/librte_sched/rte_sched_version.map   |    2 +-
 22 files changed, 1413 insertions(+), 1050 deletions(-)
Signature · 1 line
2.21.0
#51 lore raw

PATCH v4 1/17 sched: add pipe config params to subport struct

Add pipe configuration parameters to subport level structure to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/Makefile    |  2 +-
 lib/librte_sched/meson.build |  2 +-
 lib/librte_sched/rte_sched.h | 92 ++++++++++++++++++++++++------------
 3 files changed, 65 insertions(+), 31 deletions(-)
lib/librte_sched/Makefile +1 -1
index 3d7f410e1..6e4a72d89 100644
@@ -18,7 +18,7 @@ LDLIBS += -lrte_timer
 
 EXPORT_MAP := rte_sched_version.map
 
-LIBABIVER := 3
+LIBABIVER := 4
 
 #
 # all source are stored in SRCS-y
lib/librte_sched/meson.build +1 -1
index 59d43c6d8..9f40a2368 100644
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2017 Intel Corporation
 
-version = 3
+version = 4
 sources = files('rte_sched.c', 'rte_red.c', 'rte_approx.c')
 headers = files('rte_sched.h', 'rte_sched_common.h',
 		'rte_red.h', 'rte_approx.h')
lib/librte_sched/rte_sched.h +63 -29

137 lines not on this page · raw message

Signature · 1 line
2.21.0
#52 lore raw

PATCH v4 2/17 sched: modify internal structs for config flexibility

In reply to Jasvinder Singh

Update internal structures related to port and subport to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 106 ++++++++++++++++++++++++++---------
 1 file changed, 78 insertions(+), 28 deletions(-)
lib/librte_sched/rte_sched.c +78 -28
index d8ab21d64..672412b77 100644
@@ -47,33 +47,6 @@
  */
 #define RTE_SCHED_TIME_SHIFT		      8
 
-struct rte_sched_subport {
-	/* Token bucket (TB) */
-	uint64_t tb_time; /* time of last update */
-	uint32_t tb_period;
-	uint32_t tb_credits_per_period;
-	uint32_t tb_size;
-	uint32_t tb_credits;
-
-	/* Traffic classes (TCs) */
-	uint64_t tc_time; /* time of next update */
-	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
-	uint32_t tc_period;
-
-	/* TC oversubscription */
-	uint32_t tc_ov_wm;
-	uint32_t tc_ov_wm_min;
-	uint32_t tc_ov_wm_max;
-	uint8_t tc_ov_period_id;
-	uint8_t tc_ov;
-	uint32_t tc_ov_n;
-	double tc_ov_rate;
-
-	/* Statistics */
-	struct rte_sched_subport_stats stats;
-};
-
 struct rte_sched_pipe_profile {
 	/* Token bucket (TB) */
 	uint32_t tb_period;
@@ -107,7 +80,6 @@ struct rte_sched_pipe {
 	/* TC oversubscription */
 	uint32_t tc_ov_credits;
 	uint8_t tc_ov_period_id;
-	uint8_t reserved[3];
 } __rte_cache_aligned;
 
 struct rte_sched_queue {
@@ -166,6 +138,72 @@ struct rte_sched_grinder {
 	uint8_t wrr_cost[RTE_SCHED_BE_QUEUES_PER_PIPE];
 };
 
+struct rte_sched_subport {
+	/* Token bucket (TB) */
+	uint64_t tb_time; /* time of last update */
+	uint32_t tb_period;
+	uint32_t tb_credits_per_period;
+	uint32_t tb_size;
+	uint32_t tb_credits;
+
+	/* Traffic classes (TCs) */
+	uint64_t tc_time; /* time of next update */
+	uint32_t tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_credits[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+	uint32_t tc_period;
+
+	/* TC oversubscription */
+	uint32_t tc_ov_wm;
+	uint32_t tc_ov_wm_min;
+	uint32_t tc_ov_wm_max;
+	uint8_t tc_ov_period_id;
+	uint8_t tc_ov;
+	uint32_t tc_ov_n;
+	double tc_ov_rate;
+
+	/* Statistics */
+	struct rte_sched_subport_stats stats;
+
+	/* Subport pipes */
+	uint32_t n_pipes_per_subport_enabled;
+	uint32_t n_pipe_profiles;
+	uint32_t n_max_pipe_profiles;
+
+	/* Pipe best-effort TC rate */
+	uint32_t pipe_tc_be_rate_max;
+
+	/* Pipe queues size */
+	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
+
+#ifdef RTE_SCHED_RED
+	struct rte_red_config red_config[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE][RTE_COLORS];
+#endif
+
+	/* Scheduling loop detection */
+	uint32_t pipe_loop;
+	uint32_t pipe_exhaustion;
+
+	/* Bitmap */
+	struct rte_bitmap *bmp;
+	uint32_t grinder_base_bmp_pos[RTE_SCHED_PORT_N_GRINDERS] __rte_aligned_16;
+
+	/* Grinders */
+	struct rte_sched_grinder grinder[RTE_SCHED_PORT_N_GRINDERS];
+	uint32_t busy_grinders;
+
+	/* Queue base calculation */
+	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
+	uint32_t qsize_sum;
+
+	struct rte_sched_pipe *pipe;
+	struct rte_sched_queue *queue;
+	struct rte_sched_queue_extra *queue_extra;
+	struct rte_sched_pipe_profile *pipe_profiles;
+	uint8_t *bmp_array;
+	struct rte_mbuf **queue_array;
+	uint8_t memory[0] __rte_cache_aligned;
+} __rte_cache_aligned;
+
 struct rte_sched_port {
 	/* User parameters */
 	uint32_t n_subports_per_port;
@@ -177,6 +215,7 @@ struct rte_sched_port {
 	uint32_t rate;
 	uint32_t mtu;
 	uint32_t frame_overhead;
+	int socket;
 	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t n_pipe_profiles;
 	uint32_t n_max_pipe_profiles;
@@ -210,6 +249,7 @@ struct rte_sched_port {
 	uint32_t qsize_sum;
 
 	/* Large data structures */
+	struct rte_sched_subport *subports[0];
 	struct rte_sched_subport *subport;
 	struct rte_sched_pipe *pipe;
 	struct rte_sched_queue *queue;
@@ -231,6 +271,16 @@ enum rte_sched_port_array {
 	e_RTE_SCHED_PORT_ARRAY_TOTAL,
 };
 
+enum rte_sched_subport_array {
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE = 0,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA,
+	e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES,
+	e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY,
+	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
+};
+
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline uint32_t
Signature · 1 line
2.21.0
#53 lore raw

PATCH v4 3/17 sched: remove pipe params config from port level

In reply to Jasvinder Singh

Remove pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 254 ++++++++++-------------------------
 1 file changed, 71 insertions(+), 183 deletions(-)
lib/librte_sched/rte_sched.c +71 -183
index 672412b77..952e449f7 100644
@@ -291,6 +291,31 @@ rte_sched_port_queues_per_subport(struct rte_sched_port *port)
 
 #endif
 
+static inline uint32_t
+rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
+{
+	return RTE_SCHED_QUEUES_PER_PIPE * subport->n_pipes_per_subport_enabled;
+}
+
+static inline struct rte_mbuf **
+rte_sched_subport_pipe_qbase(struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t pindex = qindex >> 4;
+	uint32_t qpos = qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1);
+
+	return (subport->queue_array + pindex *
+		subport->qsize_sum + subport->qsize_add[qpos]);
+}
+
+static inline uint16_t
+rte_sched_subport_pipe_qsize(struct rte_sched_port *port,
+struct rte_sched_subport *subport, uint32_t qindex)
+{
+	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
+
+	return subport->qsize[tc];
+}
+
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
@@ -414,8 +439,6 @@ pipe_profile_check(struct rte_sched_pipe_params *params,
 static int
 rte_sched_port_check_params(struct rte_sched_port_params *params)
 {
-	uint32_t i;
-
 	if (params == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter params\n", __func__);
@@ -456,45 +479,10 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 	if (params->n_pipes_per_subport == 0 ||
 	    !rte_is_power_of_2(params->n_pipes_per_subport)) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for pipes number\n", __func__);
-		return -EINVAL;
-	}
-
-	/* qsize: if non-zero, power of 2,
-	 * no bigger than 32K (due to 16-bit read/write pointers)
-	 */
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint16_t qsize = params->qsize[i];
-
-		if ((qsize != 0 && !rte_is_power_of_2(qsize)) ||
-			((i == RTE_SCHED_TRAFFIC_CLASS_BE) && (qsize == 0))) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	/* pipe_profiles and n_pipe_profiles */
-	if (params->pipe_profiles == NULL ||
-	    params->n_pipe_profiles == 0 ||
-		 params->n_pipe_profiles > params->n_max_pipe_profiles) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for number of pipe profiles\n", __func__);
+			"%s: Incorrect value for maximum pipes number\n", __func__);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < params->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
-		int status;
-
-		status = pipe_profile_check(p, params->rate, &params->qsize[0]);
-		if (status != 0) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Pipe profile check failed(%d)\n", __func__, status);
-			return -EINVAL;
-		}
-	}
-
 	return 0;
 }
 
@@ -582,32 +570,6 @@ rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
 	return size0 + size1;
 }
 
-static void
-rte_sched_port_config_qsize(struct rte_sched_port *port)
-{
-	uint32_t i;
-
-	port->qsize_add[0] = 0;
-
-	/* Strict prority traffic class */
-	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		port->qsize_add[i] = port->qsize_add[i-1] + port->qsize[i-1];
-
-	/* Best-effort traffic class */
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-	port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
-		port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-	port->qsize_sum = port->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
-		port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
-}
-
 static void
 rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
 {
@@ -717,56 +679,41 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
-static void
-rte_sched_port_config_pipe_profile_table(struct rte_sched_port *port,
-	struct rte_sched_port_params *params)
-{
-	uint32_t i;
-
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		struct rte_sched_pipe_profile *dst = port->pipe_profiles + i;
-
-		rte_sched_pipe_profile_convert(port, src, dst, params->rate);
-		rte_sched_port_log_pipe_profile(port, i);
-	}
-
-	port->pipe_tc_be_rate_max = 0;
-	for (i = 0; i < port->n_pipe_profiles; i++) {
-		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
-		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
-
-		if (port->pipe_tc_be_rate_max < pipe_tc_be_rate)
-			port->pipe_tc_be_rate_max = pipe_tc_be_rate;
-	}
-}
-
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
 	struct rte_sched_port *port = NULL;
-	uint32_t mem_size, bmp_mem_size, n_queues_per_port, i, j, cycles_per_byte;
+	uint32_t size0, size1;
+	uint32_t cycles_per_byte;
+	uint32_t i, j;
+	int status;
 
-	/* Check user parameters. Determine the amount of memory to allocate */
-	mem_size = rte_sched_port_get_memory_footprint(params);
-	if (mem_size == 0)
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 		return NULL;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = params->n_subports_per_port * sizeof(struct rte_sched_subport *);
 
 	/* Allocate memory to store the data structures */
-	port = rte_zmalloc_socket("qos_params", mem_size, RTE_CACHE_LINE_SIZE,
+	port = rte_zmalloc_socket("qos_params", size0 + size1, RTE_CACHE_LINE_SIZE,
 		params->socket);
-	if (port == NULL)
-		return NULL;
+	if (port == NULL) {
+		RTE_LOG(ERR, SCHED, "%s: Memory allocation fails\n", __func__);
 
-	/* compile time checks */
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
-	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS & (RTE_SCHED_PORT_N_GRINDERS - 1));
+		return NULL;
+	}
 
 	/* User parameters */
 	port->n_subports_per_port = params->n_subports_per_port;
 	port->n_pipes_per_subport = params->n_pipes_per_subport;
 	port->n_pipes_per_subport_log2 =
 			__builtin_ctz(params->n_pipes_per_subport);
+	port->socket = params->socket;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 		port->pipe_queue[i] = i;
@@ -787,32 +734,6 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	port->rate = params->rate;
 	port->mtu = params->mtu + params->frame_overhead;
 	port->frame_overhead = params->frame_overhead;
-	memcpy(port->qsize, params->qsize, sizeof(params->qsize));
-	port->n_pipe_profiles = params->n_pipe_profiles;
-	port->n_max_pipe_profiles = params->n_max_pipe_profiles;
-
-#ifdef RTE_SCHED_RED
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t j;
-
-		for (j = 0; j < RTE_COLORS; j++) {
-			/* if min/max are both zero, then RED is disabled */
-			if ((params->red_params[i][j].min_th |
-			     params->red_params[i][j].max_th) == 0) {
-				continue;
-			}
-
-			if (rte_red_config_init(&port->red_config[i][j],
-				params->red_params[i][j].wq_log2,
-				params->red_params[i][j].min_th,
-				params->red_params[i][j].max_th,
-				params->red_params[i][j].maxp_inv) != 0) {
-				rte_free(port);
-				return NULL;
-			}
-		}
-	}
-#endif
 
 	/* Timing */
 	port->time_cpu_cycles = rte_get_tsc_cycles();
@@ -823,79 +744,32 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 		/ params->rate;
 	port->inv_cycles_per_byte = rte_reciprocal_value(cycles_per_byte);
 
-	/* Scheduling loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
-	port->pipe_exhaustion = 0;
-
 	/* Grinders */
-	port->busy_grinders = 0;
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
 
-	/* Queue base calculation */
-	rte_sched_port_config_qsize(port);
-
-	/* Large data structures */
-	port->subport = (struct rte_sched_subport *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_SUBPORT));
-	port->pipe = (struct rte_sched_pipe *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE));
-	port->queue = (struct rte_sched_queue *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE));
-	port->queue_extra = (struct rte_sched_queue_extra *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA));
-	port->pipe_profiles = (struct rte_sched_pipe_profile *)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES));
-	port->bmp_array =  port->memory
-		+ rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY);
-	port->queue_array = (struct rte_mbuf **)
-		(port->memory + rte_sched_port_get_array_base(params,
-							      e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY));
-
-	/* Pipe profile table */
-	rte_sched_port_config_pipe_profile_table(port, params);
-
-	/* Bitmap */
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
-	bmp_mem_size = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	port->bmp = rte_bitmap_init(n_queues_per_port, port->bmp_array,
-				    bmp_mem_size);
-	if (port->bmp == NULL) {
-		RTE_LOG(ERR, SCHED, "Bitmap init error\n");
-		rte_free(port);
-		return NULL;
-	}
-
-	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
-		port->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
-
-
 	return port;
 }
 
-void
-rte_sched_port_free(struct rte_sched_port *port)
+static inline void
+rte_sched_subport_free(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
+	uint32_t n_subport_pipe_queues;
 	uint32_t qindex;
-	uint32_t n_queues_per_port;
 
-	/* Check user parameters */
-	if (port == NULL)
+	if (subport == NULL)
 		return;
 
-	n_queues_per_port = rte_sched_port_queues_per_port(port);
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(subport);
 
 	/* Free enqueued mbufs */
-	for (qindex = 0; qindex < n_queues_per_port; qindex++) {
-		struct rte_mbuf **mbufs = rte_sched_port_qbase(port, qindex);
-		uint16_t qsize = rte_sched_port_qsize(port, qindex);
+	for (qindex = 0; qindex < n_subport_pipe_queues; qindex++) {
+		struct rte_mbuf **mbufs =
+			rte_sched_subport_pipe_qbase(subport, qindex);
+		uint16_t qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 		if (qsize != 0) {
-			struct rte_sched_queue *queue = port->queue + qindex;
+			struct rte_sched_queue *queue = subport->queue + qindex;
 			uint16_t qr = queue->qr & (qsize - 1);
 			uint16_t qw = queue->qw & (qsize - 1);
 
@@ -904,7 +778,21 @@ rte_sched_port_free(struct rte_sched_port *port)
 		}
 	}
 
-	rte_bitmap_free(port->bmp);
+	rte_bitmap_free(subport->bmp);
+}
+
+void
+rte_sched_port_free(struct rte_sched_port *port)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (port == NULL)
+		return;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		rte_sched_subport_free(port, port->subports[i]);
+
 	rte_free(port);
 }
 
Signature · 1 line
2.21.0
#54 lore raw

PATCH v4 4/17 sched: add pipe config to subport level

In reply to Jasvinder Singh

Add pipes configuration from the port level to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 388 ++++++++++++++++++++++++++++++-----
 1 file changed, 332 insertions(+), 56 deletions(-)
lib/librte_sched/rte_sched.c +332 -56
index 952e449f7..60dfc6232 100644
@@ -550,24 +550,91 @@ rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sch
 	return base;
 }
 
-uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+static uint32_t
+rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
+	enum rte_sched_subport_array array)
 {
-	uint32_t size0, size1;
-	int status;
+	uint32_t n_pipes_per_subport = params->n_pipes_per_subport_enabled;
+	uint32_t n_subport_pipe_queues =
+		RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport;
 
-	status = rte_sched_port_check_params(params);
-	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+	uint32_t size_pipe = n_pipes_per_subport * sizeof(struct rte_sched_pipe);
+	uint32_t size_queue =
+		n_subport_pipe_queues * sizeof(struct rte_sched_queue);
+	uint32_t size_queue_extra
+		= n_subport_pipe_queues * sizeof(struct rte_sched_queue_extra);
+	uint32_t size_pipe_profiles = params->n_max_pipe_profiles *
+		sizeof(struct rte_sched_pipe_profile);
+	uint32_t size_bmp_array =
+		rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	uint32_t size_per_pipe_queue_array, size_queue_array;
 
-		return 0;
+	uint32_t base, i;
+
+	size_per_pipe_queue_array = 0;
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
+			size_per_pipe_queue_array +=
+				params->qsize[i] * sizeof(struct rte_mbuf *);
+		else
+			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
+				params->qsize[i] * sizeof(struct rte_mbuf *);
 	}
+	size_queue_array = n_pipes_per_subport * size_per_pipe_queue_array;
 
-	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params, e_RTE_SCHED_PORT_ARRAY_TOTAL);
+	base = 0;
 
-	return size0 + size1;
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
+
+	if (array == e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY)
+		return base;
+	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
+
+	return base;
+}
+
+static void
+rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
+{
+	uint32_t i;
+
+	subport->qsize_add[0] = 0;
+
+	/* Strict prority traffic class */
+	for (i = 1; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		subport->qsize_add[i] = subport->qsize_add[i-1] + subport->qsize[i-1];
+
+	/* Best-effort traffic class */
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 1] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+	subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] =
+		subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 2] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+	subport->qsize_sum = subport->qsize_add[RTE_SCHED_TRAFFIC_CLASS_BE + 3] +
+		subport->qsize[RTE_SCHED_TRAFFIC_CLASS_BE];
 }
 
 static void
@@ -679,6 +746,126 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static int
+rte_sched_subport_check_params(struct rte_sched_subport_params *params,
+	uint32_t n_max_pipes_per_subport,
+	uint32_t rate)
+{
+	uint32_t i;
+
+	/* Check user parameters */
+	if (params == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for parameter params\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_rate == 0 || params->tb_rate > rate) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb rate\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tb_size == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tb size\n", __func__);
+		return -EINVAL;
+	}
+
+	/* qsize: if non-zero, power of 2,
+	 * no bigger than 32K (due to 16-bit read/write pointers)
+	 */
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint16_t qsize = params->qsize[i];
+
+		if (qsize != 0 && !rte_is_power_of_2(qsize)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for qsize\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t tc_rate = params->tc_rate[i];
+		uint16_t qsize = params->qsize[i];
+
+		if ((qsize == 0 && tc_rate != 0) ||
+			(qsize != 0 && tc_rate == 0) ||
+			(tc_rate > params->tb_rate)) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Incorrect value for tc rate\n", __func__);
+			return -EINVAL;
+		}
+	}
+
+	if (params->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
+		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect qsize or tc rate(best effort)\n", __func__);
+		return -EINVAL;
+	}
+
+	if (params->tc_period == 0) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for tc period\n", __func__);
+		return -EINVAL;
+	}
+
+	/* n_pipes_per_subport: non-zero, power of 2 */
+	if (params->n_pipes_per_subport_enabled == 0 ||
+		params->n_pipes_per_subport_enabled > n_max_pipes_per_subport ||
+	    !rte_is_power_of_2(params->n_pipes_per_subport_enabled)) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipes number\n", __func__);
+		return -EINVAL;
+	}
+
+	/* pipe_profiles and n_pipe_profiles */
+	if (params->pipe_profiles == NULL ||
+	    params->n_pipe_profiles == 0 ||
+		params->n_max_pipe_profiles == 0 ||
+		params->n_pipe_profiles > params->n_max_pipe_profiles) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for pipe profiles\n", __func__);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < params->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *p = params->pipe_profiles + i;
+		int status;
+
+		status = pipe_profile_check(p, rate, &params->qsize[0]);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Pipe profile check failed(%d)\n", __func__, status);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+uint32_t
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+{
+	uint32_t size0, size1;
+	int status;
+
+	status = rte_sched_port_check_params(params);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"Port scheduler params check failed (%d)\n", status);
+
+		return 0;
+	}
+
+	size0 = sizeof(struct rte_sched_port);
+	size1 = rte_sched_port_get_array_base(params,
+			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	return size0 + size1;
+}
+
 struct rte_sched_port *
 rte_sched_port_config(struct rte_sched_port_params *params)
 {
@@ -799,7 +986,7 @@ rte_sched_port_free(struct rte_sched_port *port)
 static void
 rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 {
-	struct rte_sched_subport *s = port->subport + i;
+	struct rte_sched_subport *s = port->subports[i];
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for subport %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -834,72 +1021,78 @@ rte_sched_port_log_subport_config(struct rte_sched_port *port, uint32_t i)
 		s->tc_ov_wm_max);
 }
 
+static void
+rte_sched_free_memory(struct rte_sched_port *port, uint32_t n_subports)
+{
+	uint32_t i;
+
+	for (i = 0; i < n_subports; i++) {
+		struct rte_sched_subport *subport = port->subports[i];
+
+		rte_sched_subport_free(port, subport);
+	}
+
+	rte_free(port);
+}
+
 int
 rte_sched_subport_config(struct rte_sched_port *port,
 	uint32_t subport_id,
 	struct rte_sched_subport_params *params)
 {
-	struct rte_sched_subport *s;
-	uint32_t i;
+	struct rte_sched_subport *s = NULL;
+	uint32_t n_subports = subport_id;
+	uint32_t n_subport_pipe_queues, i;
+	uint32_t size0, size1, bmp_mem_size;
+	int status;
 
 	/* Check user parameters */
 	if (port == NULL) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter port\n", __func__);
-		return -EINVAL;
+		return 0;
 	}
 
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for subport id\n", __func__);
-		return -EINVAL;
-	}
 
-	if (params == NULL) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for parameter params\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (params->tb_rate == 0 || params->tb_rate > port->rate) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb rate\n", __func__);
-		return -EINVAL;
-	}
+	status = rte_sched_subport_check_params(params,
+		port->n_pipes_per_subport,
+		port->rate);
+	if (status != 0) {
+		RTE_LOG(NOTICE, SCHED,
+			"%s: Port scheduler params check failed (%d)\n",
+			__func__, status);
 
-	if (params->tb_size == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tb size\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		uint32_t tc_rate = params->tc_rate[i];
-		uint16_t qsize = port->qsize[i];
+	/* Determine the amount of memory to allocate */
+	size0 = sizeof(struct rte_sched_subport);
+	size1 = rte_sched_subport_get_array_base(params,
+				e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
 
-		if ((qsize == 0 && tc_rate != 0) ||
-			(qsize != 0 && tc_rate == 0) ||
-			(tc_rate > params->tb_rate)) {
-			RTE_LOG(ERR, SCHED,
-				"%s: Incorrect value for tc rate\n", __func__);
-			return -EINVAL;
-		}
-	}
-
-	if (port->qsize[RTE_SCHED_TRAFFIC_CLASS_BE] == 0 ||
-		params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE] == 0) {
+	/* Allocate memory to store the data structures */
+	s = rte_zmalloc_socket("subport_params", size0 + size1,
+		RTE_CACHE_LINE_SIZE, port->socket);
+	if (s == NULL) {
 		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc rate(best effort)\n", __func__);
-		return -EINVAL;
-	}
+			"%s: Memory allocation fails\n", __func__);
 
-	if (params->tc_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Incorrect value for tc period\n", __func__);
-		return -EINVAL;
+		rte_sched_free_memory(port, n_subports);
+		return -ENOMEM;
 	}
 
-	s = port->subport + subport_id;
+	n_subports++;
+
+	/* Port */
+	port->subports[subport_id] = s;
 
 	/* Token Bucket (TB) */
 	if (params->tb_rate == port->rate) {
@@ -919,26 +1112,109 @@ rte_sched_subport_config(struct rte_sched_port *port,
 	/* Traffic Classes (TCs) */
 	s->tc_period = rte_sched_time_ms_to_bytes(params->tc_period, port->rate);
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(params->tc_period,
-								 params->tc_rate[i]);
-
+					params->tc_rate[i]);
 	}
 	s->tc_time = port->time + s->tc_period;
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (params->qsize[i])
 			s->tc_credits[i] = s->tc_credits_per_period[i];
 
+	/* compile time checks */
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS == 0);
+	RTE_BUILD_BUG_ON(RTE_SCHED_PORT_N_GRINDERS &
+		(RTE_SCHED_PORT_N_GRINDERS - 1));
+
+	/* User parameters */
+	s->n_pipes_per_subport_enabled = params->n_pipes_per_subport_enabled;
+	memcpy(s->qsize, params->qsize, sizeof(params->qsize));
+	s->n_pipe_profiles = params->n_pipe_profiles;
+	s->n_max_pipe_profiles = params->n_max_pipe_profiles;
+
+#ifdef RTE_SCHED_RED
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
+		uint32_t j;
+
+		for (j = 0; j < RTE_COLORS; j++) {
+			/* if min/max are both zero, then RED is disabled */
+			if ((params->red_params[i][j].min_th |
+			     params->red_params[i][j].max_th) == 0) {
+				continue;
+			}
+
+			if (rte_red_config_init(&s->red_config[i][j],
+				params->red_params[i][j].wq_log2,
+				params->red_params[i][j].min_th,
+				params->red_params[i][j].max_th,
+				params->red_params[i][j].maxp_inv) != 0) {
+				rte_sched_free_memory(port, n_subports);
+
+				RTE_LOG(NOTICE, SCHED,
+				"%s: RED configuration init fails\n", __func__);
+				return -EINVAL;
+			}
+		}
+	}
+#endif
+
+	/* Scheduling loop detection */
+	s->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	s->pipe_exhaustion = 0;
+
+	/* Grinders */
+	s->busy_grinders = 0;
+
+	/* Queue base calculation */
+	rte_sched_subport_config_qsize(s);
+
+	/* Large data structures */
+	s->pipe = (struct rte_sched_pipe *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_PIPE));
+	s->queue = (struct rte_sched_queue *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE));
+	s->queue_extra = (struct rte_sched_queue_extra *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_EXTRA));
+	s->pipe_profiles = (struct rte_sched_pipe_profile *)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_PIPE_PROFILES));
+	s->bmp_array =  s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_BMP_ARRAY);
+	s->queue_array = (struct rte_mbuf **)
+		(s->memory + rte_sched_subport_get_array_base(params,
+		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
+
+	/* Bitmap */
+	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
+	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
+	s->bmp = rte_bitmap_init(n_subport_pipe_queues, s->bmp_array,
+				bmp_mem_size);
+	if (s->bmp == NULL) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Subport bitmap init error\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++)
+		s->grinder_base_bmp_pos[i] = RTE_SCHED_PIPE_INVALID;
+
+#ifdef RTE_SCHED_SUBPORT_TC_OV
 	/* TC oversubscription */
 	s->tc_ov_wm_min = port->mtu;
 	s->tc_ov_wm_max = rte_sched_time_ms_to_bytes(params->tc_period,
-						     port->pipe_tc_be_rate_max);
+						     s->pipe_tc_be_rate_max);
 	s->tc_ov_wm = s->tc_ov_wm_max;
 	s->tc_ov_period_id = 0;
 	s->tc_ov = 0;
 	s->tc_ov_n = 0;
 	s->tc_ov_rate = 0;
+#endif
 
 	rte_sched_port_log_subport_config(port, subport_id);
 
Signature · 1 line
2.21.0
#55 lore raw

PATCH v4 5/17 sched: modify pipe functions for config flexibility

In reply to Jasvinder Singh

Modify pipe level functions to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c           | 103 +++++++++++++++++--------
 lib/librte_sched/rte_sched.h           |   7 +-
 lib/librte_sched/rte_sched_version.map |   2 +-
 3 files changed, 76 insertions(+), 36 deletions(-)
lib/librte_sched/rte_sched.c +70 -33
index 60dfc6232..6b6219e45 100644
@@ -638,9 +638,9 @@ rte_sched_subport_config_qsize(struct rte_sched_subport *subport)
 }
 
 static void
-rte_sched_port_log_pipe_profile(struct rte_sched_port *port, uint32_t i)
+rte_sched_port_log_pipe_profile(struct rte_sched_subport *subport, uint32_t i)
 {
-	struct rte_sched_pipe_profile *p = port->pipe_profiles + i;
+	struct rte_sched_pipe_profile *p = subport->pipe_profiles + i;
 
 	RTE_LOG(DEBUG, SCHED, "Low level config for pipe profile %u:\n"
 		"	Token bucket: period = %u, credits per period = %u, size = %u\n"
@@ -689,7 +689,7 @@ rte_sched_time_ms_to_bytes(uint32_t time_ms, uint32_t rate)
 }
 
 static void
-rte_sched_pipe_profile_convert(struct rte_sched_port *port,
+rte_sched_pipe_profile_convert(struct rte_sched_subport *subport,
 	struct rte_sched_pipe_params *src,
 	struct rte_sched_pipe_profile *dst,
 	uint32_t rate)
@@ -718,7 +718,7 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 						rate);
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (subport->qsize[i])
 			dst->tc_credits_per_period[i]
 				= rte_sched_time_ms_to_bytes(src->tc_period,
 					src->tc_rate[i]);
@@ -746,6 +746,30 @@ rte_sched_pipe_profile_convert(struct rte_sched_port *port,
 	dst->wrr_cost[3] = (uint8_t) wrr_cost[3];
 }
 
+static void
+rte_sched_subport_config_pipe_profile_table(struct rte_sched_subport *subport,
+	struct rte_sched_subport_params *params, uint32_t rate)
+{
+	uint32_t i;
+
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		struct rte_sched_pipe_profile *dst = subport->pipe_profiles + i;
+
+		rte_sched_pipe_profile_convert(subport, src, dst, rate);
+		rte_sched_port_log_pipe_profile(subport, i);
+	}
+
+	subport->pipe_tc_be_rate_max = 0;
+	for (i = 0; i < subport->n_pipe_profiles; i++) {
+		struct rte_sched_pipe_params *src = params->pipe_profiles + i;
+		uint32_t pipe_tc_be_rate = src->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+
+		if (subport->pipe_tc_be_rate_max < pipe_tc_be_rate)
+			subport->pipe_tc_be_rate_max = pipe_tc_be_rate;
+	}
+}
+
 static int
 rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 	uint32_t n_max_pipes_per_subport,
@@ -1188,6 +1212,9 @@ rte_sched_subport_config(struct rte_sched_port *port,
 		(s->memory + rte_sched_subport_get_array_base(params,
 		e_RTE_SCHED_SUBPORT_ARRAY_QUEUE_ARRAY));
 
+	/* Pipe profile table */
+	rte_sched_subport_config_pipe_profile_table(s, params, port->rate);
+
 	/* Bitmap */
 	n_subport_pipe_queues = rte_sched_subport_pipe_queues(s);
 	bmp_mem_size = rte_bitmap_get_memory_footprint(n_subport_pipe_queues);
@@ -1230,6 +1257,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	struct rte_sched_subport *s;
 	struct rte_sched_pipe *p;
 	struct rte_sched_pipe_profile *params;
+	uint32_t n_subports = subport_id + 1;
 	uint32_t deactivate, profile, i;
 
 	/* Check user parameters */
@@ -1245,34 +1273,32 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	if (subport_id >= port->n_subports_per_port) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter subport id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (pipe_id >= port->n_pipes_per_subport) {
+	s = port->subports[subport_id];
+	if (pipe_id >= s->n_pipes_per_subport_enabled) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe id\n", __func__);
+
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	if (!deactivate && profile >= port->n_pipe_profiles) {
+	if (!deactivate && profile >= s->n_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Incorrect value for parameter pipe profile\n", __func__);
-		return -EINVAL;
-	}
 
-	/* Check that subport configuration is valid */
-	s = port->subport + subport_id;
-	if (s->tb_period == 0) {
-		RTE_LOG(ERR, SCHED,
-			"%s: Subport configuration invalid\n", __func__);
+		rte_sched_free_memory(port, n_subports);
 		return -EINVAL;
 	}
 
-	p = port->pipe + (subport_id * port->n_pipes_per_subport + pipe_id);
-
 	/* Handle the case when pipe already has a valid configuration */
+	p = s->pipe + pipe_id;
 	if (p->tb_time) {
-		params = port->pipe_profiles + p->profile;
+		params = s->pipe_profiles + p->profile;
 
 		double subport_tc_be_rate =
 			(double) s->tc_credits_per_period[RTE_SCHED_TRAFFIC_CLASS_BE]
@@ -1302,7 +1328,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 
 	/* Apply the new pipe configuration */
 	p->profile = profile;
-	params = port->pipe_profiles + p->profile;
+	params = s->pipe_profiles + p->profile;
 
 	/* Token Bucket (TB) */
 	p->tb_time = port->time;
@@ -1312,7 +1338,7 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 	p->tc_time = port->time + params->tc_period;
 
 	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (port->qsize[i])
+		if (s->qsize[i])
 			p->tc_credits[i] = params->tc_credits_per_period[i];
 
 	{
@@ -1342,10 +1368,12 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 }
 
 int
-rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+rte_sched_subport_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_pipe_profile *pp;
 	uint32_t i;
 	int status;
@@ -1357,40 +1385,49 @@ rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
-	/* Pipe profiles not exceeds the max limit */
-	if (port->n_pipe_profiles >= port->n_max_pipe_profiles) {
+	/* Subport id not exceeds the max limit */
+	if (subport_id > port->n_subports_per_port) {
+		RTE_LOG(ERR, SCHED,
+			"%s: Incorrect value for subport id\n", __func__);
+		return -EINVAL;
+	}
+
+	s = port->subports[subport_id];
+
+	/* Pipe profiles exceeds the max limit */
+	if (s->n_pipe_profiles >= s->n_max_pipe_profiles) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Number of pipe profiles exceeds the max limit\n", __func__);
 		return -EINVAL;
 	}
 
 	/* Pipe params */
-	status = pipe_profile_check(params, port->rate, &port->qsize[0]);
+	status = pipe_profile_check(params, port->rate, &s->qsize[0]);
 	if (status != 0) {
 		RTE_LOG(ERR, SCHED,
 			"%s: Pipe profile check failed(%d)\n", __func__, status);
 		return -EINVAL;
 	}
 
-	pp = &port->pipe_profiles[port->n_pipe_profiles];
-	rte_sched_pipe_profile_convert(port, params, pp, port->rate);
+	pp = &s->pipe_profiles[s->n_pipe_profiles];
+	rte_sched_pipe_profile_convert(s, params, pp, port->rate);
 
-	/* Pipe profile not exists */
-	for (i = 0; i < port->n_pipe_profiles; i++)
-		if (memcmp(port->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
+	/* Pipe profile should not exists */
+	for (i = 0; i < s->n_pipe_profiles; i++)
+		if (memcmp(s->pipe_profiles + i, pp, sizeof(*pp)) == 0) {
 			RTE_LOG(ERR, SCHED,
-				"%s: Pipe profile doesn't exist\n", __func__);
+				"%s: Pipe profile exists\n", __func__);
 			return -EINVAL;
 		}
 
 	/* Pipe profile commit */
-	*pipe_profile_id = port->n_pipe_profiles;
-	port->n_pipe_profiles++;
+	*pipe_profile_id = s->n_pipe_profiles;
+	s->n_pipe_profiles++;
 
-	if (port->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
-		port->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
+	if (s->pipe_tc_be_rate_max < params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE])
+		s->pipe_tc_be_rate_max = params->tc_rate[RTE_SCHED_TRAFFIC_CLASS_BE];
 
-	rte_sched_port_log_pipe_profile(port, *pipe_profile_id);
+	rte_sched_port_log_pipe_profile(s, *pipe_profile_id);
 
 	return 0;
 }
lib/librte_sched/rte_sched.h +5 -2
index fbb0e23fa..5001f09ed 100644
@@ -317,6 +317,8 @@ rte_sched_port_free(struct rte_sched_port *port);
  *
  * @param port
  *   Handle to port scheduler instance
+ * @param subport_id
+ *   Subport ID
  * @param params
  *   Pipe profile parameters
  * @param pipe_profile_id
@@ -326,7 +328,8 @@ rte_sched_port_free(struct rte_sched_port *port);
  */
 __rte_experimental
 int
-rte_sched_port_pipe_profile_add(struct rte_sched_port *port,
+rte_sched_subport_pipe_profile_add(struct rte_sched_port *port,
+	uint32_t subport_id,
 	struct rte_sched_pipe_params *params,
 	uint32_t *pipe_profile_id);
 
@@ -357,7 +360,7 @@ rte_sched_subport_config(struct rte_sched_port *port,
  * @param pipe_id
  *   Pipe ID within subport
  * @param pipe_profile
- *   ID of port-level pre-configured pipe profile
+ *   ID of subport-level pre-configured pipe profile
  * @return
  *   0 upon success, error code otherwise
  */
lib/librte_sched/rte_sched_version.map +1 -1
index 729588794..f33761e63 100644
@@ -33,5 +33,5 @@ DPDK_2.1 {
 EXPERIMENTAL {
 	global:
 
-	rte_sched_port_pipe_profile_add;
+	rte_sched_subport_pipe_profile_add;
 };
Signature · 1 line
2.21.0
#56 lore raw

PATCH v4 6/17 sched: modify pkt enqueue for config flexibility

In reply to Jasvinder Singh

Modify scheduler packet enqueue operation of the scheduler to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 277 ++++++++++++++++++++++-------------
 1 file changed, 178 insertions(+), 99 deletions(-)
lib/librte_sched/rte_sched.c +178 -99
index 6b6219e45..a8174afad 100644
@@ -1440,10 +1440,11 @@ rte_sched_port_qindex(struct rte_sched_port *port,
 	uint32_t queue)
 {
 	return ((subport & (port->n_subports_per_port - 1)) <<
-			(port->n_pipes_per_subport_log2 + 4)) |
-			((pipe & (port->n_pipes_per_subport - 1)) << 4) |
-			((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
-			(RTE_SCHED_QUEUES_PER_PIPE - 1));
+		(port->n_pipes_per_subport_log2 + 4)) |
+		((pipe &
+		(port->subports[subport]->n_pipes_per_subport_enabled - 1)) << 4) |
+		((rte_sched_port_pipe_queue(port, traffic_class) + queue) &
+		(RTE_SCHED_QUEUES_PER_PIPE - 1));
 }
 
 void
@@ -1468,7 +1469,8 @@ rte_sched_port_pkt_read_tree_path(struct rte_sched_port *port,
 	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
 
 	*subport = queue_id >> (port->n_pipes_per_subport_log2 + 4);
-	*pipe = (queue_id >> 4) & (port->n_pipes_per_subport - 1);
+	*pipe = (queue_id >> 4) &
+		(port->subports[*subport]->n_pipes_per_subport_enabled - 1);
 	*traffic_class = rte_sched_port_pipe_tc(port, queue_id);
 	*queue = rte_sched_port_tc_queue(port, queue_id);
 }
@@ -1512,7 +1514,7 @@ rte_sched_subport_read_stats(struct rte_sched_port *port,
 		return -EINVAL;
 	}
 
-	s = port->subport + subport_id;
+	s = port->subports[subport_id];
 
 	/* Copy subport stats and clear */
 	memcpy(stats, &s->stats, sizeof(struct rte_sched_subport_stats));
@@ -1585,43 +1587,50 @@ rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
 #ifdef RTE_SCHED_COLLECT_STATS
 
 static inline void
-rte_sched_port_update_subport_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_subport_stats(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc[tc_index] += 1;
-	s->stats.n_bytes_tc[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc[tc_index] += 1;
+	subport->stats.n_bytes_tc[tc_index] += pkt_len;
 }
 
 #ifdef RTE_SCHED_RED
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
 rte_sched_port_update_subport_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_subport *s = port->subport + (qindex / rte_sched_port_queues_per_subport(port));
 	uint32_t tc_index = rte_sched_port_pipe_tc(port, qindex);
 	uint32_t pkt_len = pkt->pkt_len;
 
-	s->stats.n_pkts_tc_dropped[tc_index] += 1;
-	s->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
+	subport->stats.n_pkts_tc_dropped[tc_index] += 1;
+	subport->stats.n_bytes_tc_dropped[tc_index] += pkt_len;
 #ifdef RTE_SCHED_RED
-	s->stats.n_pkts_red_dropped[tc_index] += red;
+	subport->stats.n_pkts_red_dropped[tc_index] += red;
 #endif
 }
 
 static inline void
-rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex, struct rte_mbuf *pkt)
+rte_sched_port_update_queue_stats(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts += 1;
@@ -1630,17 +1639,19 @@ rte_sched_port_update_queue_stats(struct rte_sched_port *port, uint32_t qindex,
 
 #ifdef RTE_SCHED_RED
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	uint32_t red)
 #else
 static inline void
-rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
-						uint32_t qindex,
-						struct rte_mbuf *pkt, __rte_unused uint32_t red)
+rte_sched_port_update_queue_stats_on_drop(struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf *pkt,
+	__rte_unused uint32_t red)
 #endif
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	uint32_t pkt_len = pkt->pkt_len;
 
 	qe->stats.n_pkts_dropped += 1;
@@ -1655,7 +1666,11 @@ rte_sched_port_update_queue_stats_on_drop(struct rte_sched_port *port,
 #ifdef RTE_SCHED_RED
 
 static inline int
-rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint32_t qindex, uint16_t qlen)
+rte_sched_port_red_drop(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt,
+	uint32_t qindex,
+	uint16_t qlen)
 {
 	struct rte_sched_queue_extra *qe;
 	struct rte_red_config *red_cfg;
@@ -1665,12 +1680,12 @@ rte_sched_port_red_drop(struct rte_sched_port *port, struct rte_mbuf *pkt, uint3
 
 	tc_index = rte_sched_port_pipe_tc(port, qindex);
 	color = rte_sched_port_pkt_read_color(pkt);
-	red_cfg = &port->red_config[tc_index][color];
+	red_cfg = &subport->red_config[tc_index][color];
 
 	if ((red_cfg->min_th | red_cfg->max_th) == 0)
 		return 0;
 
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + qindex;
 	red = &qe->red;
 
 	return rte_red_enqueue(red_cfg, red, qlen, port->time);
@@ -1687,7 +1702,14 @@ rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t q
 
 #else
 
-#define rte_sched_port_red_drop(port, pkt, qindex, qlen)             0
+static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unused,
+	struct rte_sched_subport *subport __rte_unused,
+	struct rte_mbuf *pkt __rte_unused,
+	uint32_t qindex __rte_unused,
+	uint16_t qlen __rte_unused)
+{
+	return 0;
+}
 
 #define rte_sched_port_set_queue_empty_timestamp(port, qindex)
 
@@ -1722,63 +1744,79 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 
 #endif /* RTE_SCHED_DEBUG */
 
+static inline struct rte_sched_subport *
+rte_sched_port_subport(struct rte_sched_port *port,
+	struct rte_mbuf *pkt)
+{
+	uint32_t queue_id = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_id = queue_id >> (port->n_pipes_per_subport_log2 + 4);
+
+	return port->subports[subport_id];
+}
+
 static inline uint32_t
-rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_port *port,
-				       struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qptrs_prefetch0(struct rte_sched_subport *subport,
+	struct rte_mbuf *pkt, uint32_t subport_qmask)
 {
 	struct rte_sched_queue *q;
 #ifdef RTE_SCHED_COLLECT_STATS
 	struct rte_sched_queue_extra *qe;
 #endif
 	uint32_t qindex = rte_mbuf_sched_queue_get(pkt);
+	uint32_t subport_queue_id = subport_qmask & qindex;
 
-	q = port->queue + qindex;
+	q = subport->queue + subport_queue_id;
 	rte_prefetch0(q);
 #ifdef RTE_SCHED_COLLECT_STATS
-	qe = port->queue_extra + qindex;
+	qe = subport->queue_extra + subport_queue_id;
 	rte_prefetch0(qe);
 #endif
 
-	return qindex;
+	return subport_queue_id;
 }
 
 static inline void
 rte_sched_port_enqueue_qwa_prefetch0(struct rte_sched_port *port,
-				     uint32_t qindex, struct rte_mbuf **qbase)
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase)
 {
 	struct rte_sched_queue *q;
 	struct rte_mbuf **q_qw;
 	uint16_t qsize;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	q_qw = qbase + (q->qw & (qsize - 1));
 
 	rte_prefetch0(q_qw);
-	rte_bitmap_prefetch0(port->bmp, qindex);
+	rte_bitmap_prefetch0(subport->bmp, qindex);
 }
 
 static inline int
-rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
-			   struct rte_mbuf **qbase, struct rte_mbuf *pkt)
+rte_sched_port_enqueue_qwa(struct rte_sched_port *port,
+	struct rte_sched_subport *subport,
+	uint32_t qindex,
+	struct rte_mbuf **qbase,
+	struct rte_mbuf *pkt)
 {
 	struct rte_sched_queue *q;
 	uint16_t qsize;
 	uint16_t qlen;
 
-	q = port->queue + qindex;
-	qsize = rte_sched_port_qsize(port, qindex);
+	q = subport->queue + qindex;
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 	qlen = q->qw - q->qr;
 
 	/* Drop the packet (and update drop stats) when queue is full */
-	if (unlikely(rte_sched_port_red_drop(port, pkt, qindex, qlen) ||
+	if (unlikely(rte_sched_port_red_drop(port, subport, pkt, qindex, qlen) ||
 		     (qlen >= qsize))) {
 		rte_pktmbuf_free(pkt);
 #ifdef RTE_SCHED_COLLECT_STATS
-		rte_sched_port_update_subport_stats_on_drop(port, qindex, pkt,
-							    qlen < qsize);
-		rte_sched_port_update_queue_stats_on_drop(port, qindex, pkt,
-							  qlen < qsize);
+		rte_sched_port_update_subport_stats_on_drop(port, subport,
+			qindex, pkt, qlen < qsize);
+		rte_sched_port_update_queue_stats_on_drop(subport, qindex, pkt,
+			qlen < qsize);
 #endif
 		return 0;
 	}
@@ -1787,13 +1825,13 @@ rte_sched_port_enqueue_qwa(struct rte_sched_port *port, uint32_t qindex,
 	qbase[q->qw & (qsize - 1)] = pkt;
 	q->qw++;
 
-	/* Activate queue in the port bitmap */
-	rte_bitmap_set(port->bmp, qindex);
+	/* Activate queue in the subport bitmap */
+	rte_bitmap_set(subport->bmp, qindex);
 
 	/* Statistics */
 #ifdef RTE_SCHED_COLLECT_STATS
-	rte_sched_port_update_subport_stats(port, qindex, pkt);
-	rte_sched_port_update_queue_stats(port, qindex, pkt);
+	rte_sched_port_update_subport_stats(port, subport, qindex, pkt);
+	rte_sched_port_update_queue_stats(subport, qindex, pkt);
 #endif
 
 	return 1;
@@ -1821,17 +1859,22 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		*pkt30, *pkt31, *pkt_last;
 	struct rte_mbuf **q00_base, **q01_base, **q10_base, **q11_base,
 		**q20_base, **q21_base, **q30_base, **q31_base, **q_last_base;
+	struct rte_sched_subport *subport00, *subport01, *subport10, *subport11,
+		*subport20, *subport21, *subport30, *subport31, *subport_last;
 	uint32_t q00, q01, q10, q11, q20, q21, q30, q31, q_last;
 	uint32_t r00, r01, r10, r11, r20, r21, r30, r31, r_last;
+	uint32_t subport_qmask;
 	uint32_t result, i;
 
 	result = 0;
+	subport_qmask = (1 << (port->n_pipes_per_subport_log2 + 4)) - 1;
 
 	/*
 	 * Less then 6 input packets available, which is not enough to
 	 * feed the pipeline
 	 */
 	if (unlikely(n_pkts < 6)) {
+		struct rte_sched_subport *subports[5];
 		struct rte_mbuf **q_base[5];
 		uint32_t q[5];
 
@@ -1839,22 +1882,26 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		for (i = 0; i < n_pkts; i++)
 			rte_prefetch0(pkts[i]);
 
+		/* Prefetch the subport structure for each packet */
+		for (i = 0; i < n_pkts; i++)
+			subports[i] = rte_sched_port_subport(port, pkts[i]);
+
 		/* Prefetch the queue structure for each queue */
 		for (i = 0; i < n_pkts; i++)
-			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(port,
-								      pkts[i]);
+			q[i] = rte_sched_port_enqueue_qptrs_prefetch0(subports[i],
+					pkts[i], subport_qmask);
 
 		/* Prefetch the write pointer location of each queue */
 		for (i = 0; i < n_pkts; i++) {
-			q_base[i] = rte_sched_port_qbase(port, q[i]);
-			rte_sched_port_enqueue_qwa_prefetch0(port, q[i],
-							     q_base[i]);
+			q_base[i] = rte_sched_subport_pipe_qbase(subports[i], q[i]);
+			rte_sched_port_enqueue_qwa_prefetch0(port, subports[i],
+				q[i], q_base[i]);
 		}
 
 		/* Write each packet to its queue */
 		for (i = 0; i < n_pkts; i++)
-			result += rte_sched_port_enqueue_qwa(port, q[i],
-							     q_base[i], pkts[i]);
+			result += rte_sched_port_enqueue_qwa(port, subports[i],
+						q[i], q_base[i], pkts[i]);
 
 		return result;
 	}
@@ -1870,21 +1917,29 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	rte_prefetch0(pkt10);
 	rte_prefetch0(pkt11);
 
-	q20 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt20);
-	q21 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt21);
+	subport20 = rte_sched_port_subport(port, pkt20);
+	subport21 = rte_sched_port_subport(port, pkt21);
+	q20 = rte_sched_port_enqueue_qptrs_prefetch0(subport20,
+			pkt20, subport_qmask);
+	q21 = rte_sched_port_enqueue_qptrs_prefetch0(subport21,
+			pkt21, subport_qmask);
 
 	pkt00 = pkts[4];
 	pkt01 = pkts[5];
 	rte_prefetch0(pkt00);
 	rte_prefetch0(pkt01);
 
-	q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-	q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+	subport10 = rte_sched_port_subport(port, pkt10);
+	subport11 = rte_sched_port_subport(port, pkt11);
+	q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+			pkt10, subport_qmask);
+	q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+			pkt11, subport_qmask);
 
-	q20_base = rte_sched_port_qbase(port, q20);
-	q21_base = rte_sched_port_qbase(port, q21);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+	q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+	q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 	/* Run the pipeline */
 	for (i = 6; i < (n_pkts & (~1)); i += 2) {
@@ -1899,6 +1954,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		q31 = q21;
 		q20 = q10;
 		q21 = q11;
+		subport30 = subport20;
+		subport31 = subport21;
+		subport20 = subport10;
+		subport21 = subport11;
 		q30_base = q20_base;
 		q31_base = q21_base;
 
@@ -1908,19 +1967,25 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 		rte_prefetch0(pkt00);
 		rte_prefetch0(pkt01);
 
-		/* Stage 1: Prefetch queue structure storing queue pointers */
-		q10 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt10);
-		q11 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt11);
+		/* Stage 1: Prefetch subport and queue structure storing queue pointers */
+		subport10 = rte_sched_port_subport(port, pkt10);
+		subport11 = rte_sched_port_subport(port, pkt11);
+		q10 = rte_sched_port_enqueue_qptrs_prefetch0(subport10,
+				pkt10, subport_qmask);
+		q11 = rte_sched_port_enqueue_qptrs_prefetch0(subport11,
+				pkt11, subport_qmask);
 
 		/* Stage 2: Prefetch queue write location */
-		q20_base = rte_sched_port_qbase(port, q20);
-		q21_base = rte_sched_port_qbase(port, q21);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q20, q20_base);
-		rte_sched_port_enqueue_qwa_prefetch0(port, q21, q21_base);
+		q20_base = rte_sched_subport_pipe_qbase(subport20, q20);
+		q21_base = rte_sched_subport_pipe_qbase(subport21, q21);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport20, q20, q20_base);
+		rte_sched_port_enqueue_qwa_prefetch0(port, subport21, q21, q21_base);
 
 		/* Stage 3: Write packet to queue and activate queue */
-		r30 = rte_sched_port_enqueue_qwa(port, q30, q30_base, pkt30);
-		r31 = rte_sched_port_enqueue_qwa(port, q31, q31_base, pkt31);
+		r30 = rte_sched_port_enqueue_qwa(port, subport30,
+				q30, q30_base, pkt30);
+		r31 = rte_sched_port_enqueue_qwa(port, subport31,
+				q31, q31_base, pkt31);
 		result += r30 + r31;
 	}
 
@@ -1932,38 +1997,52 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 	pkt_last = pkts[n_pkts - 1];
 	rte_prefetch0(pkt_last);
 
-	q00 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt00);
-	q01 = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt01);
-
-	q10_base = rte_sched_port_qbase(port, q10);
-	q11_base = rte_sched_port_qbase(port, q11);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q10, q10_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q11, q11_base);
-
-	r20 = rte_sched_port_enqueue_qwa(port, q20, q20_base, pkt20);
-	r21 = rte_sched_port_enqueue_qwa(port, q21, q21_base, pkt21);
+	subport00 = rte_sched_port_subport(port, pkt00);
+	subport01 = rte_sched_port_subport(port, pkt01);
+	q00 = rte_sched_port_enqueue_qptrs_prefetch0(subport00,
+			pkt00, subport_qmask);
+	q01 = rte_sched_port_enqueue_qptrs_prefetch0(subport01,
+			pkt01, subport_qmask);
+
+	q10_base = rte_sched_subport_pipe_qbase(subport10, q10);
+	q11_base = rte_sched_subport_pipe_qbase(subport11, q11);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport10, q10, q10_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport11, q11, q11_base);
+
+	r20 = rte_sched_port_enqueue_qwa(port, subport20,
+			q20, q20_base, pkt20);
+	r21 = rte_sched_port_enqueue_qwa(port, subport21,
+			q21, q21_base, pkt21);
 	result += r20 + r21;
 
-	q_last = rte_sched_port_enqueue_qptrs_prefetch0(port, pkt_last);
+	subport_last = rte_sched_port_subport(port, pkt_last);
+	q_last = rte_sched_port_enqueue_qptrs_prefetch0(subport_last,
+				pkt_last, subport_qmask);
 
-	q00_base = rte_sched_port_qbase(port, q00);
-	q01_base = rte_sched_port_qbase(port, q01);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q00, q00_base);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q01, q01_base);
+	q00_base = rte_sched_subport_pipe_qbase(subport00, q00);
+	q01_base = rte_sched_subport_pipe_qbase(subport01, q01);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport00, q00, q00_base);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport01, q01, q01_base);
 
-	r10 = rte_sched_port_enqueue_qwa(port, q10, q10_base, pkt10);
-	r11 = rte_sched_port_enqueue_qwa(port, q11, q11_base, pkt11);
+	r10 = rte_sched_port_enqueue_qwa(port, subport10, q10,
+			q10_base, pkt10);
+	r11 = rte_sched_port_enqueue_qwa(port, subport11, q11,
+			q11_base, pkt11);
 	result += r10 + r11;
 
-	q_last_base = rte_sched_port_qbase(port, q_last);
-	rte_sched_port_enqueue_qwa_prefetch0(port, q_last, q_last_base);
+	q_last_base = rte_sched_subport_pipe_qbase(subport_last, q_last);
+	rte_sched_port_enqueue_qwa_prefetch0(port, subport_last,
+		q_last, q_last_base);
 
-	r00 = rte_sched_port_enqueue_qwa(port, q00, q00_base, pkt00);
-	r01 = rte_sched_port_enqueue_qwa(port, q01, q01_base, pkt01);
+	r00 = rte_sched_port_enqueue_qwa(port, subport00, q00,
+			q00_base, pkt00);
+	r01 = rte_sched_port_enqueue_qwa(port, subport01, q01,
+			q01_base, pkt01);
 	result += r00 + r01;
 
 	if (n_pkts & 1) {
-		r_last = rte_sched_port_enqueue_qwa(port, q_last, q_last_base, pkt_last);
+		r_last = rte_sched_port_enqueue_qwa(port, subport_last,
+					q_last,	q_last_base, pkt_last);
 		result += r_last;
 	}
 
Signature · 1 line
2.21.0
#57 lore raw

PATCH v4 7/17 sched: update memory compute to support flexiblity

In reply to Jasvinder Singh

Update memory footprint compute function for allowing subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 100 ++++++++++-------------------------
 lib/librte_sched/rte_sched.h |   8 +--
 2 files changed, 34 insertions(+), 74 deletions(-)
lib/librte_sched/rte_sched.c +29 -71
index a8174afad..58f34359c 100644
@@ -486,70 +486,6 @@ rte_sched_port_check_params(struct rte_sched_port_params *params)
 	return 0;
 }
 
-static uint32_t
-rte_sched_port_get_array_base(struct rte_sched_port_params *params, enum rte_sched_port_array array)
-{
-	uint32_t n_subports_per_port = params->n_subports_per_port;
-	uint32_t n_pipes_per_subport = params->n_pipes_per_subport;
-	uint32_t n_pipes_per_port = n_pipes_per_subport * n_subports_per_port;
-	uint32_t n_queues_per_port = RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport * n_subports_per_port;
-
-	uint32_t size_subport = n_subports_per_port * sizeof(struct rte_sched_subport);
-	uint32_t size_pipe = n_pipes_per_port * sizeof(struct rte_sched_pipe);
-	uint32_t size_queue = n_queues_per_port * sizeof(struct rte_sched_queue);
-	uint32_t size_queue_extra
-		= n_queues_per_port * sizeof(struct rte_sched_queue_extra);
-	uint32_t size_pipe_profiles
-		= params->n_max_pipe_profiles * sizeof(struct rte_sched_pipe_profile);
-	uint32_t size_bmp_array = rte_bitmap_get_memory_footprint(n_queues_per_port);
-	uint32_t size_per_pipe_queue_array, size_queue_array;
-
-	uint32_t base, i;
-
-	size_per_pipe_queue_array = 0;
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++) {
-		if (i < RTE_SCHED_TRAFFIC_CLASS_BE)
-			size_per_pipe_queue_array +=
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-		else
-			size_per_pipe_queue_array += RTE_SCHED_MAX_QUEUES_PER_TC *
-				params->qsize[i] * sizeof(struct rte_mbuf *);
-	}
-	size_queue_array = n_pipes_per_port * size_per_pipe_queue_array;
-
-	base = 0;
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_SUBPORT)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_subport);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_EXTRA)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_extra);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_PIPE_PROFILES)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_pipe_profiles);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_BMP_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_bmp_array);
-
-	if (array == e_RTE_SCHED_PORT_ARRAY_QUEUE_ARRAY)
-		return base;
-	base += RTE_CACHE_LINE_ROUNDUP(size_queue_array);
-
-	return base;
-}
-
 static uint32_t
 rte_sched_subport_get_array_base(struct rte_sched_subport_params *params,
 	enum rte_sched_subport_array array)
@@ -870,22 +806,44 @@ rte_sched_subport_check_params(struct rte_sched_subport_params *params,
 }
 
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params)
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params)
 {
-	uint32_t size0, size1;
+	uint32_t size0 = 0, size1 = 0, i;
 	int status;
 
-	status = rte_sched_port_check_params(params);
+	status = rte_sched_port_check_params(port_params);
 	if (status != 0) {
-		RTE_LOG(NOTICE, SCHED,
-			"Port scheduler params check failed (%d)\n", status);
+		RTE_LOG(ERR, SCHED,
+			"%s: Port scheduler port params check failed (%d)\n",
+			__func__, status);
 
 		return 0;
 	}
 
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		status = rte_sched_subport_check_params(sp,
+				port_params->n_pipes_per_subport,
+				port_params->rate);
+		if (status != 0) {
+			RTE_LOG(ERR, SCHED,
+				"%s: Port scheduler subport params check failed (%d)\n",
+				__func__, status);
+
+			return 0;
+		}
+	}
+
 	size0 = sizeof(struct rte_sched_port);
-	size1 = rte_sched_port_get_array_base(params,
-			e_RTE_SCHED_PORT_ARRAY_TOTAL);
+
+	for (i = 0; i < port_params->n_subports_per_port; i++) {
+		struct rte_sched_subport_params *sp = subport_params[i];
+
+		size1 += rte_sched_subport_get_array_base(sp,
+					e_RTE_SCHED_SUBPORT_ARRAY_TOTAL);
+	}
 
 	return size0 + size1;
 }
lib/librte_sched/rte_sched.h +5 -3
index 5001f09ed..40f02f124 100644
@@ -373,14 +373,16 @@ rte_sched_pipe_config(struct rte_sched_port *port,
 /**
  * Hierarchical scheduler memory footprint size per port
  *
- * @param params
+ * @param port_params
  *   Port scheduler configuration parameter structure
+ * @param subport_params
+ *   Array of subport parameter structures
  * @return
  *   Memory footprint size in bytes upon success, 0 otherwise
  */
 uint32_t
-rte_sched_port_get_memory_footprint(struct rte_sched_port_params *params);
-
+rte_sched_port_get_memory_footprint(struct rte_sched_port_params *port_params,
+	struct rte_sched_subport_params **subport_params);
 /*
  * Statistics
  *
Signature · 1 line
2.21.0
#58 lore raw

PATCH v4 8/17 sched: update grinder functions for config flexibility

In reply to Jasvinder Singh

Modify packet grinder functions of the schedule to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 223 +++++++++++++++++------------------
 1 file changed, 106 insertions(+), 117 deletions(-)
lib/librte_sched/rte_sched.c +106 -117
index 58f34359c..d7abe9661 100644
@@ -322,24 +322,6 @@ rte_sched_port_queues_per_port(struct rte_sched_port *port)
 	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
 }
 
-static inline struct rte_mbuf **
-rte_sched_port_qbase(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t pindex = qindex >> 4;
-	uint32_t qpos = qindex & 0xF;
-
-	return (port->queue_array + pindex *
-		port->qsize_sum + port->qsize_add[qpos]);
-}
-
-static inline uint16_t
-rte_sched_port_qsize(struct rte_sched_port *port, uint32_t qindex)
-{
-	uint32_t tc = port->pipe_tc[qindex & (RTE_SCHED_QUEUES_PER_PIPE - 1)];
-
-	return port->qsize[tc];
-}
-
 static inline uint16_t
 rte_sched_port_pipe_queue(struct rte_sched_port *port, uint32_t traffic_class)
 {
@@ -1533,9 +1515,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 #ifdef RTE_SCHED_DEBUG
 
 static inline int
-rte_sched_port_queue_is_empty(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_queue_is_empty(struct rte_sched_subport *subport,
+	uint32_t qindex)
 {
-	struct rte_sched_queue *queue = port->queue + qindex;
+	struct rte_sched_queue *queue = subport->queue + qindex;
 
 	return queue->qr == queue->qw;
 }
@@ -1650,9 +1633,10 @@ rte_sched_port_red_drop(struct rte_sched_port *port,
 }
 
 static inline void
-rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port, uint32_t qindex)
+rte_sched_port_set_queue_empty_timestamp(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t qindex)
 {
-	struct rte_sched_queue_extra *qe = port->queue_extra + qindex;
+	struct rte_sched_queue_extra *qe = subport->queue_extra + qindex;
 	struct rte_red *red = &qe->red;
 
 	rte_red_mark_queue_empty(red, port->time);
@@ -1669,14 +1653,14 @@ static inline int rte_sched_port_red_drop(struct rte_sched_port *port __rte_unus
 	return 0;
 }
 
-#define rte_sched_port_set_queue_empty_timestamp(port, qindex)
+#define rte_sched_port_set_queue_empty_timestamp(port, subport, qindex)
 
 #endif /* RTE_SCHED_RED */
 
 #ifdef RTE_SCHED_DEBUG
 
 static inline void
-debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
+debug_check_queue_slab(struct rte_sched_subport *subport, uint32_t bmp_pos,
 		       uint64_t bmp_slab)
 {
 	uint64_t mask;
@@ -1688,7 +1672,7 @@ debug_check_queue_slab(struct rte_sched_port *port, uint32_t bmp_pos,
 	panic = 0;
 	for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 		if (mask & bmp_slab) {
-			if (rte_sched_port_queue_is_empty(port, bmp_pos + i)) {
+			if (rte_sched_port_queue_is_empty(subport, bmp_pos + i)) {
 				printf("Queue %u (slab offset %u) is empty\n", bmp_pos + i, i);
 				panic = 1;
 			}
@@ -2010,10 +1994,10 @@ rte_sched_port_enqueue(struct rte_sched_port *port, struct rte_mbuf **pkts,
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2051,10 +2035,9 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline uint32_t
-grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_tc_ov_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
 	uint32_t tc_ov_consumption[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 	uint32_t tc_consumption = 0, tc_ov_consumption_max;
 	uint32_t tc_ov_wm = subport->tc_ov_wm;
@@ -2094,10 +2077,10 @@ grinder_tc_ov_credits_update(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_update(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *params = grinder->pipe_params;
 	uint64_t n_periods;
@@ -2117,7 +2100,7 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 
 	/* Subport TCs */
 	if (unlikely(port->time >= subport->tc_time)) {
-		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, pos);
+		subport->tc_ov_wm = grinder_tc_ov_credits_update(port, subport);
 
 		for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
 			subport->tc_credits[i] = subport->tc_credits_per_period[i];
@@ -2147,10 +2130,10 @@ grinder_credits_update(struct rte_sched_port *port, uint32_t pos)
 #ifndef RTE_SCHED_SUBPORT_TC_OV
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2182,10 +2165,10 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 #else
 
 static inline int
-grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
+grinder_credits_check(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
-	struct rte_sched_subport *subport = grinder->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t tc_index = grinder->tc_index;
@@ -2230,15 +2213,16 @@ grinder_credits_check(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline int
-grinder_schedule(struct rte_sched_port *port, uint32_t pos)
+grinder_schedule(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_queue *queue = grinder->queue[grinder->qpos];
 	struct rte_mbuf *pkt = grinder->pkt;
 	uint32_t pkt_len = pkt->pkt_len + port->frame_overhead;
 	uint32_t be_tc_active;
 
-	if (!grinder_credits_check(port, pos))
+	if (!grinder_credits_check(port, subport, pos))
 		return 0;
 
 	/* Advance port time */
@@ -2255,15 +2239,15 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 	if (queue->qr == queue->qw) {
 		uint32_t qindex = grinder->qindex[grinder->qpos];
 
-		rte_bitmap_clear(port->bmp, qindex);
+		rte_bitmap_clear(subport->bmp, qindex);
 		grinder->qmask &= ~(1 << grinder->qpos);
 		if (be_tc_active)
 			grinder->wrr_mask[grinder->qpos] = 0;
-		rte_sched_port_set_queue_empty_timestamp(port, qindex);
+		rte_sched_port_set_queue_empty_timestamp(port, subport, qindex);
 	}
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	grinder->productive = 1;
 
 	return 1;
@@ -2272,13 +2256,13 @@ grinder_schedule(struct rte_sched_port *port, uint32_t pos)
 #ifdef SCHED_VECTOR_SSE4
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	__m128i index = _mm_set1_epi32(base_pipe);
-	__m128i pipes = _mm_load_si128((__m128i *)port->grinder_base_bmp_pos);
+	__m128i pipes = _mm_load_si128((__m128i *)subport->grinder_base_bmp_pos);
 	__m128i res = _mm_cmpeq_epi32(pipes, index);
 
-	pipes = _mm_load_si128((__m128i *)(port->grinder_base_bmp_pos + 4));
+	pipes = _mm_load_si128((__m128i *)(subport->grinder_base_bmp_pos + 4));
 	pipes = _mm_cmpeq_epi32(pipes, index);
 	res = _mm_or_si128(res, pipes);
 
@@ -2291,10 +2275,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #elif defined(SCHED_VECTOR_NEON)
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32x4_t index, pipes;
-	uint32_t *pos = (uint32_t *)port->grinder_base_bmp_pos;
+	uint32_t *pos = (uint32_t *)subport->grinder_base_bmp_pos;
 
 	index = vmovq_n_u32(base_pipe);
 	pipes = vld1q_u32(pos);
@@ -2311,12 +2295,12 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #else
 
 static inline int
-grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
+grinder_pipe_exists(struct rte_sched_subport *subport, uint32_t base_pipe)
 {
 	uint32_t i;
 
 	for (i = 0; i < RTE_SCHED_PORT_N_GRINDERS; i++) {
-		if (port->grinder_base_bmp_pos[i] == base_pipe)
+		if (subport->grinder_base_bmp_pos[i] == base_pipe)
 			return 1;
 	}
 
@@ -2326,9 +2310,10 @@ grinder_pipe_exists(struct rte_sched_port *port, uint32_t base_pipe)
 #endif /* RTE_SCHED_OPTIMIZATIONS */
 
 static inline void
-grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
+grinder_pcache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t bmp_pos, uint64_t bmp_slab)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t w[4];
 
 	grinder->pcache_w = 0;
@@ -2357,9 +2342,10 @@ grinder_pcache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t bmp_
 }
 
 static inline void
-grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qindex, uint16_t qmask)
+grinder_tccache_populate(struct rte_sched_subport *subport,
+	uint32_t pos, uint32_t qindex, uint16_t qmask)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint8_t b, i;
 
 	grinder->tccache_w = 0;
@@ -2380,9 +2366,10 @@ grinder_tccache_populate(struct rte_sched_port *port, uint32_t pos, uint32_t qin
 }
 
 static inline int
-grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
+grinder_next_tc(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_mbuf **qbase;
 	uint32_t qindex;
 	uint16_t qsize;
@@ -2391,15 +2378,15 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 0;
 
 	qindex = grinder->tccache_qindex[grinder->tccache_r];
-	qbase = rte_sched_port_qbase(port, qindex);
-	qsize = rte_sched_port_qsize(port, qindex);
+	qbase = rte_sched_subport_pipe_qbase(subport, qindex);
+	qsize = rte_sched_subport_pipe_qsize(port, subport, qindex);
 
 	grinder->tc_index = rte_sched_port_pipe_tc(port, qindex);
 	grinder->qmask = grinder->tccache_qmask[grinder->tccache_r];
 	grinder->qsize = qsize;
 
 	if (grinder->tc_index < RTE_SCHED_TRAFFIC_CLASS_BE) {
-		grinder->queue[0] = port->queue + qindex;
+		grinder->queue[0] = subport->queue + qindex;
 		grinder->qbase[0] = qbase;
 		grinder->qindex[0] = qindex;
 		grinder->tccache_r++;
@@ -2407,10 +2394,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 		return 1;
 	}
 
-	grinder->queue[0] = port->queue + qindex;
-	grinder->queue[1] = port->queue + qindex + 1;
-	grinder->queue[2] = port->queue + qindex + 2;
-	grinder->queue[3] = port->queue + qindex + 3;
+	grinder->queue[0] = subport->queue + qindex;
+	grinder->queue[1] = subport->queue + qindex + 1;
+	grinder->queue[2] = subport->queue + qindex + 2;
+	grinder->queue[3] = subport->queue + qindex + 3;
 
 	grinder->qbase[0] = qbase;
 	grinder->qbase[1] = qbase + qsize;
@@ -2427,9 +2414,10 @@ grinder_next_tc(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline int
-grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_next_pipe(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t pipe_qindex;
 	uint16_t pipe_qmask;
 
@@ -2442,22 +2430,22 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 		uint32_t bmp_pos = 0;
 
 		/* Get another non-empty pipe group */
-		if (unlikely(rte_bitmap_scan(port->bmp, &bmp_pos, &bmp_slab) <= 0))
+		if (unlikely(rte_bitmap_scan(subport->bmp, &bmp_pos, &bmp_slab) <= 0))
 			return 0;
 
 #ifdef RTE_SCHED_DEBUG
-		debug_check_queue_slab(port, bmp_pos, bmp_slab);
+		debug_check_queue_slab(subport, bmp_pos, bmp_slab);
 #endif
 
 		/* Return if pipe group already in one of the other grinders */
-		port->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
-		if (unlikely(grinder_pipe_exists(port, bmp_pos)))
+		subport->grinder_base_bmp_pos[pos] = RTE_SCHED_BMP_POS_INVALID;
+		if (unlikely(grinder_pipe_exists(subport, bmp_pos)))
 			return 0;
 
-		port->grinder_base_bmp_pos[pos] = bmp_pos;
+		subport->grinder_base_bmp_pos[pos] = bmp_pos;
 
 		/* Install new pipe group into grinder's pipe cache */
-		grinder_pcache_populate(port, pos, bmp_pos, bmp_slab);
+		grinder_pcache_populate(subport, pos, bmp_pos, bmp_slab);
 
 		pipe_qmask = grinder->pcache_qmask[0];
 		pipe_qindex = grinder->pcache_qindex[0];
@@ -2466,18 +2454,18 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 	/* Install new pipe in the grinder */
 	grinder->pindex = pipe_qindex >> 4;
-	grinder->subport = port->subport + (grinder->pindex / port->n_pipes_per_subport);
-	grinder->pipe = port->pipe + grinder->pindex;
+	grinder->subport = subport;
+	grinder->pipe = subport->pipe + grinder->pindex;
 	grinder->pipe_params = NULL; /* to be set after the pipe structure is prefetched */
 	grinder->productive = 0;
 
-	grinder_tccache_populate(port, pos, pipe_qindex, pipe_qmask);
-	grinder_next_tc(port, pos);
+	grinder_tccache_populate(subport, pos, pipe_qindex, pipe_qmask);
+	grinder_next_tc(port, subport, pos);
 
 	/* Check for pipe exhaustion */
-	if (grinder->pindex == port->pipe_loop) {
-		port->pipe_exhaustion = 1;
-		port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	if (grinder->pindex == subport->pipe_loop) {
+		subport->pipe_exhaustion = 1;
+		subport->pipe_loop = RTE_SCHED_PIPE_INVALID;
 	}
 
 	return 1;
@@ -2485,9 +2473,9 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 
 static inline void
-grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_load(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 	struct rte_sched_pipe_profile *pipe_params = grinder->pipe_params;
 	uint32_t qmask = grinder->qmask;
@@ -2513,9 +2501,9 @@ grinder_wrr_load(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr_store(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	struct rte_sched_pipe *pipe = grinder->pipe;
 
 	pipe->wrr_tokens[0] =
@@ -2533,9 +2521,9 @@ grinder_wrr_store(struct rte_sched_port *port, uint32_t pos)
 }
 
 static inline void
-grinder_wrr(struct rte_sched_port *port, uint32_t pos)
+grinder_wrr(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t wrr_tokens_min;
 
 	grinder->wrr_tokens[0] |= ~grinder->wrr_mask[0];
@@ -2553,21 +2541,21 @@ grinder_wrr(struct rte_sched_port *port, uint32_t pos)
 }
 
 
-#define grinder_evict(port, pos)
+#define grinder_evict(subport, pos)
 
 static inline void
-grinder_prefetch_pipe(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_pipe(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	rte_prefetch0(grinder->pipe);
 	rte_prefetch0(grinder->queue[0]);
 }
 
 static inline void
-grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_tc_queue_arrays(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint16_t qsize, qr[RTE_SCHED_MAX_QUEUES_PER_TC];
 
 	qsize = grinder->qsize;
@@ -2588,17 +2576,17 @@ grinder_prefetch_tc_queue_arrays(struct rte_sched_port *port, uint32_t pos)
 	rte_prefetch0(grinder->qbase[0] + qr[0]);
 	rte_prefetch0(grinder->qbase[1] + qr[1]);
 
-	grinder_wrr_load(port, pos);
-	grinder_wrr(port, pos);
+	grinder_wrr_load(subport, pos);
+	grinder_wrr(subport, pos);
 
 	rte_prefetch0(grinder->qbase[2] + qr[2]);
 	rte_prefetch0(grinder->qbase[3] + qr[3]);
 }
 
 static inline void
-grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
+grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 	uint32_t qpos = grinder->qpos;
 	struct rte_mbuf **qbase = grinder->qbase[qpos];
 	uint16_t qsize = grinder->qsize;
@@ -2617,14 +2605,15 @@ grinder_prefetch_mbuf(struct rte_sched_port *port, uint32_t pos)
 static inline uint32_t
 grinder_handle(struct rte_sched_port *port, uint32_t pos)
 {
-	struct rte_sched_grinder *grinder = port->grinder + pos;
+	struct rte_sched_subport *subport = port->subport;
+	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
 	case e_GRINDER_PREFETCH_PIPE:
 	{
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
-			port->busy_grinders++;
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
+			subport->busy_grinders++;
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return 0;
@@ -2637,9 +2626,9 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		struct rte_sched_pipe *pipe = grinder->pipe;
 
-		grinder->pipe_params = port->pipe_profiles + pipe->profile;
-		grinder_prefetch_tc_queue_arrays(port, pos);
-		grinder_credits_update(port, pos);
+		grinder->pipe_params = subport->pipe_profiles + pipe->profile;
+		grinder_prefetch_tc_queue_arrays(subport, pos);
+		grinder_credits_update(port, subport, pos);
 
 		grinder->state = e_GRINDER_PREFETCH_MBUF;
 		return 0;
@@ -2647,7 +2636,7 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 
 	case e_GRINDER_PREFETCH_MBUF:
 	{
-		grinder_prefetch_mbuf(port, pos);
+		grinder_prefetch_mbuf(subport, pos);
 
 		grinder->state = e_GRINDER_READ_MBUF;
 		return 0;
@@ -2657,47 +2646,47 @@ grinder_handle(struct rte_sched_port *port, uint32_t pos)
 	{
 		uint32_t wrr_active, result = 0;
 
-		result = grinder_schedule(port, pos);
+		result = grinder_schedule(port, subport, pos);
 
 		wrr_active = (grinder->tc_index == RTE_SCHED_TRAFFIC_CLASS_BE);
 
 		/* Look for next packet within the same TC */
 		if (result && grinder->qmask) {
 			if (wrr_active)
-				grinder_wrr(port, pos);
+				grinder_wrr(subport, pos);
 
-			grinder_prefetch_mbuf(port, pos);
+			grinder_prefetch_mbuf(subport, pos);
 
 			return 1;
 		}
 
 		if (wrr_active)
-			grinder_wrr_store(port, pos);
+			grinder_wrr_store(subport, pos);
 
 		/* Look for another active TC within same pipe */
-		if (grinder_next_tc(port, pos)) {
-			grinder_prefetch_tc_queue_arrays(port, pos);
+		if (grinder_next_tc(port, subport, pos)) {
+			grinder_prefetch_tc_queue_arrays(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_MBUF;
 			return result;
 		}
 
 		if (grinder->productive == 0 &&
-		    port->pipe_loop == RTE_SCHED_PIPE_INVALID)
-			port->pipe_loop = grinder->pindex;
+		    subport->pipe_loop == RTE_SCHED_PIPE_INVALID)
+			subport->pipe_loop = grinder->pindex;
 
-		grinder_evict(port, pos);
+		grinder_evict(subport, pos);
 
 		/* Look for another active pipe */
-		if (grinder_next_pipe(port, pos)) {
-			grinder_prefetch_pipe(port, pos);
+		if (grinder_next_pipe(port, subport, pos)) {
+			grinder_prefetch_pipe(subport, pos);
 
 			grinder->state = e_GRINDER_PREFETCH_TC_QUEUE_ARRAYS;
 			return result;
 		}
 
 		/* No active pipe found */
-		port->busy_grinders--;
+		subport->busy_grinders--;
 
 		grinder->state = e_GRINDER_PREFETCH_PIPE;
 		return result;
Signature · 1 line
2.21.0
#59 lore raw

PATCH v4 9/17 sched: update pkt dequeue for flexible config

In reply to Jasvinder Singh

Modify scheduler packet dequeue operation to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 51 ++++++++++++++++++++++++++++--------
 1 file changed, 40 insertions(+), 11 deletions(-)
lib/librte_sched/rte_sched.c +40 -11
index d7abe9661..e14506b2d 100644
@@ -243,6 +243,7 @@ struct rte_sched_port {
 	uint32_t busy_grinders;
 	struct rte_mbuf **pkts_out;
 	uint32_t n_pkts_out;
+	uint32_t subport_id;
 
 	/* Queue base calculation */
 	uint32_t qsize_add[RTE_SCHED_QUEUES_PER_PIPE];
@@ -898,6 +899,7 @@ rte_sched_port_config(struct rte_sched_port_params *params)
 	/* Grinders */
 	port->pkts_out = NULL;
 	port->n_pkts_out = 0;
+	port->subport_id = 0;
 
 	return port;
 }
@@ -2603,9 +2605,9 @@ grinder_prefetch_mbuf(struct rte_sched_subport *subport, uint32_t pos)
 }
 
 static inline uint32_t
-grinder_handle(struct rte_sched_port *port, uint32_t pos)
+grinder_handle(struct rte_sched_port *port,
+	struct rte_sched_subport *subport, uint32_t pos)
 {
-	struct rte_sched_subport *subport = port->subport;
 	struct rte_sched_grinder *grinder = subport->grinder + pos;
 
 	switch (grinder->state) {
@@ -2704,6 +2706,7 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 	uint64_t cycles = rte_get_tsc_cycles();
 	uint64_t cycles_diff = cycles - port->time_cpu_cycles;
 	uint64_t bytes_diff;
+	uint32_t i;
 
 	/* Compute elapsed time in bytes */
 	bytes_diff = rte_reciprocal_divide(cycles_diff << RTE_SCHED_TIME_SHIFT,
@@ -2716,20 +2719,21 @@ rte_sched_port_time_resync(struct rte_sched_port *port)
 		port->time = port->time_cpu_bytes;
 
 	/* Reset pipe loop detection */
-	port->pipe_loop = RTE_SCHED_PIPE_INVALID;
+	for (i = 0; i < port->n_subports_per_port; i++)
+		port->subports[i]->pipe_loop = RTE_SCHED_PIPE_INVALID;
 }
 
 static inline int
-rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
+rte_sched_port_exceptions(struct rte_sched_subport *subport, int second_pass)
 {
 	int exceptions;
 
 	/* Check if any exception flag is set */
-	exceptions = (second_pass && port->busy_grinders == 0) ||
-		(port->pipe_exhaustion == 1);
+	exceptions = (second_pass && subport->busy_grinders == 0) ||
+		(subport->pipe_exhaustion == 1);
 
 	/* Clear exception flags */
-	port->pipe_exhaustion = 0;
+	subport->pipe_exhaustion = 0;
 
 	return exceptions;
 }
@@ -2737,7 +2741,9 @@ rte_sched_port_exceptions(struct rte_sched_port *port, int second_pass)
 int
 rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint32_t n_pkts)
 {
-	uint32_t i, count;
+	struct rte_sched_subport *subport;
+	uint32_t subport_id = port->subport_id;
+	uint32_t i, n_subports = 0, count;
 
 	port->pkts_out = pkts;
 	port->n_pkts_out = 0;
@@ -2746,9 +2752,32 @@ rte_sched_port_dequeue(struct rte_sched_port *port, struct rte_mbuf **pkts, uint
 
 	/* Take each queue in the grinder one step further */
 	for (i = 0, count = 0; ; i++)  {
-		count += grinder_handle(port, i & (RTE_SCHED_PORT_N_GRINDERS - 1));
-		if ((count == n_pkts) ||
-		    rte_sched_port_exceptions(port, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+		subport = port->subports[subport_id];
+
+		count += grinder_handle(port, subport,
+				i & (RTE_SCHED_PORT_N_GRINDERS - 1));
+
+		if (count == n_pkts) {
+			subport_id++;
+
+			if (subport_id == port->n_subports_per_port)
+				subport_id = 0;
+
+			port->subport_id = subport_id;
+			break;
+		}
+
+		if (rte_sched_port_exceptions(subport, i >= RTE_SCHED_PORT_N_GRINDERS)) {
+			i = 0;
+			subport_id++;
+			n_subports++;
+		}
+
+		if (subport_id == port->n_subports_per_port)
+			subport_id = 0;
+
+		if (n_subports == port->n_subports_per_port) {
+			port->subport_id = subport_id;
 			break;
 		}
 	}
Signature · 1 line
2.21.0
#60 lore raw

PATCH v4 10/17 sched: update queue stats read for config flexibility

In reply to Jasvinder Singh

Modify pipe queue stats read function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 lib/librte_sched/rte_sched.c | 28 +++++++++++++++-------------
 1 file changed, 15 insertions(+), 13 deletions(-)
lib/librte_sched/rte_sched.c +15 -13
index e14506b2d..1faa580d0 100644
@@ -282,16 +282,6 @@ enum rte_sched_subport_array {
 	e_RTE_SCHED_SUBPORT_ARRAY_TOTAL,
 };
 
-#ifdef RTE_SCHED_COLLECT_STATS
-
-static inline uint32_t
-rte_sched_port_queues_per_subport(struct rte_sched_port *port)
-{
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport;
-}
-
-#endif
-
 static inline uint32_t
 rte_sched_subport_pipe_queues(struct rte_sched_subport *subport)
 {
@@ -320,7 +310,12 @@ struct rte_sched_subport *subport, uint32_t qindex)
 static inline uint32_t
 rte_sched_port_queues_per_port(struct rte_sched_port *port)
 {
-	return RTE_SCHED_QUEUES_PER_PIPE * port->n_pipes_per_subport * port->n_subports_per_port;
+	uint32_t n_queues = 0, i;
+
+	for (i = 0; i < port->n_subports_per_port; i++)
+		n_queues += rte_sched_subport_pipe_queues(port->subports[i]);
+
+	return n_queues;
 }
 
 static inline uint16_t
@@ -1474,8 +1469,10 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 	struct rte_sched_queue_stats *stats,
 	uint16_t *qlen)
 {
+	struct rte_sched_subport *s;
 	struct rte_sched_queue *q;
 	struct rte_sched_queue_extra *qe;
+	uint32_t subport_id, subport_qmask, subport_qindex;
 
 	/* Check user parameters */
 	if (port == NULL) {
@@ -1501,8 +1498,13 @@ rte_sched_queue_read_stats(struct rte_sched_port *port,
 			"%s: Incorrect value for parameter qlen\n", __func__);
 		return -EINVAL;
 	}
-	q = port->queue + queue_id;
-	qe = port->queue_extra + queue_id;
+	subport_qmask = port->n_pipes_per_subport_log2 + 4;
+	subport_id = (queue_id >> subport_qmask) & (port->n_subports_per_port - 1);
+
+	s = port->subports[subport_id];
+	subport_qindex = ((1 << subport_qmask) - 1) & queue_id;
+	q = s->queue + subport_qindex;
+	qe = s->queue_extra + subport_qindex;
 
 	/* Copy queue stats and clear */
 	memcpy(stats, &qe->stats, sizeof(struct rte_sched_queue_stats));
Signature · 1 line
2.21.0
#61 lore raw

PATCH v4 11/17 test/sched: modify tests for subport config flexibility

In reply to Jasvinder Singh

Modify tests function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 app/test/test_sched.c | 35 ++++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 17 deletions(-)
app/test/test_sched.c +18 -17
index afe0b0765..fc31080ef 100644
@@ -22,18 +22,6 @@
 #define TC              2
 #define QUEUE           0
 
-static struct rte_sched_subport_params subport_param[] = {
-	{
-		.tb_rate = 1250000000,
-		.tb_size = 1000000,
-
-		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
-			1250000000, 1250000000, 1250000000, 1250000000},
-		.tc_period = 10,
-	},
-};
-
 static struct rte_sched_pipe_params pipe_profile[] = {
 	{ /* Profile #0 */
 		.tb_rate = 305175,
@@ -48,6 +36,23 @@ static struct rte_sched_pipe_params pipe_profile[] = {
 	},
 };
 
+static struct rte_sched_subport_params subport_param[] = {
+	{
+		.tb_rate = 1250000000,
+		.tb_size = 1000000,
+
+		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000, 1250000000,
+			1250000000, 1250000000, 1250000000, 1250000000},
+		.tc_period = 10,
+		.n_pipes_per_subport_enabled = 1024,
+		.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
+		.pipe_profiles = pipe_profile,
+		.n_pipe_profiles = 1,
+		.n_max_pipe_profiles = 1,
+	},
+};
+
 static struct rte_sched_port_params port_param = {
 	.socket = 0, /* computed */
 	.rate = 0, /* computed */
@@ -55,10 +60,6 @@ static struct rte_sched_port_params port_param = {
 	.frame_overhead = RTE_SCHED_FRAME_OVERHEAD_DEFAULT,
 	.n_subports_per_port = 1,
 	.n_pipes_per_subport = 1024,
-	.qsize = {32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32},
-	.pipe_profiles = pipe_profile,
-	.n_pipe_profiles = 1,
-	.n_max_pipe_profiles = 1,
 };
 
 #define NB_MBUF          32
@@ -140,7 +141,7 @@ test_sched(void)
 	err = rte_sched_subport_config(port, SUBPORT, subport_param);
 	TEST_ASSERT_SUCCESS(err, "Error config sched, err=%d\n", err);
 
-	for (pipe = 0; pipe < port_param.n_pipes_per_subport; pipe ++) {
+	for (pipe = 0; pipe < subport_param[0].n_pipes_per_subport_enabled; pipe++) {
 		err = rte_sched_pipe_config(port, SUBPORT, pipe, 0);
 		TEST_ASSERT_SUCCESS(err, "Error config sched pipe %u, err=%d\n", pipe, err);
 	}
Signature · 1 line
2.21.0
#62 lore raw

PATCH v4 12/17 net/softnic: add subport config flexibility to TM

In reply to Jasvinder Singh

Modify softnic traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 drivers/net/softnic/rte_eth_softnic_tm.c | 54 ++++++++++++------------
 1 file changed, 28 insertions(+), 26 deletions(-)
drivers/net/softnic/rte_eth_softnic_tm.c +28 -26
index 61c3adc82..80a470c9e 100644
@@ -85,7 +85,8 @@ softnic_tmgr_port_create(struct pmd_internals *p,
 	/* Subport */
 	n_subports = t->port_params.n_subports_per_port;
 	for (subport_id = 0; subport_id < n_subports; subport_id++) {
-		uint32_t n_pipes_per_subport = t->port_params.n_pipes_per_subport;
+		uint32_t n_pipes_per_subport =
+			t->subport_params[subport_id].n_pipes_per_subport_enabled;
 		uint32_t pipe_id;
 		int status;
 
@@ -2211,10 +2212,11 @@ tm_tc_wred_profile_get(struct rte_eth_dev *dev, uint32_t tc_id)
 #ifdef RTE_SCHED_RED
 
 static void
-wred_profiles_set(struct rte_eth_dev *dev)
+wred_profiles_set(struct rte_eth_dev *dev, uint32_t subport_id)
 {
 	struct pmd_internals *p = dev->data->dev_private;
-	struct rte_sched_port_params *pp = &p->soft.tm.params.port_params;
+	struct rte_sched_subport_params *pp =
+		&p->soft.tm.params.subport_params[subport_id];
 
 	uint32_t tc_id;
 	enum rte_color color;
@@ -2234,7 +2236,7 @@ wred_profiles_set(struct rte_eth_dev *dev)
 
 #else
 
-#define wred_profiles_set(dev)
+#define wred_profiles_set(dev, subport_id)
 
 #endif
 
@@ -2526,29 +2528,9 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 		.frame_overhead =
 			root->shaper_profile->params.pkt_length_adjust,
 		.n_subports_per_port = root->n_children,
-		.n_pipes_per_subport = h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
-			h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
-		.qsize = {p->params.tm.qsize[0],
-			p->params.tm.qsize[1],
-			p->params.tm.qsize[2],
-			p->params.tm.qsize[3],
-			p->params.tm.qsize[4],
-			p->params.tm.qsize[5],
-			p->params.tm.qsize[6],
-			p->params.tm.qsize[7],
-			p->params.tm.qsize[8],
-			p->params.tm.qsize[9],
-			p->params.tm.qsize[10],
-			p->params.tm.qsize[11],
-			p->params.tm.qsize[12],
-		},
-		.pipe_profiles = t->pipe_profiles,
-		.n_pipe_profiles = t->n_pipe_profiles,
-		.n_max_pipe_profiles = TM_MAX_PIPE_PROFILE,
+		.n_pipes_per_subport = TM_MAX_PIPES_PER_SUBPORT,
 	};
 
-	wred_profiles_set(dev);
-
 	subport_id = 0;
 	TAILQ_FOREACH(n, nl, node) {
 		uint64_t tc_rate[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
@@ -2588,8 +2570,28 @@ hierarchy_blueprints_create(struct rte_eth_dev *dev)
 					tc_rate[12],
 				},
 				.tc_period = SUBPORT_TC_PERIOD,
+				.n_pipes_per_subport_enabled =
+					h->n_tm_nodes[TM_NODE_LEVEL_PIPE] /
+					h->n_tm_nodes[TM_NODE_LEVEL_SUBPORT],
+				.qsize = {p->params.tm.qsize[0],
+					p->params.tm.qsize[1],
+					p->params.tm.qsize[2],
+					p->params.tm.qsize[3],
+					p->params.tm.qsize[4],
+					p->params.tm.qsize[5],
+					p->params.tm.qsize[6],
+					p->params.tm.qsize[7],
+					p->params.tm.qsize[8],
+					p->params.tm.qsize[9],
+					p->params.tm.qsize[10],
+					p->params.tm.qsize[11],
+					p->params.tm.qsize[12],
+				},
+				.pipe_profiles = t->pipe_profiles,
+				.n_pipe_profiles = t->n_pipe_profiles,
+				.n_max_pipe_profiles = TM_MAX_PIPE_PROFILE,
 		};
-
+		wred_profiles_set(dev, subport_id);
 		subport_id++;
 	}
 }
Signature · 1 line
2.21.0
#63 lore raw

PATCH v4 13/17 ip_pipeline: add subport config flexibility to TM

In reply to Jasvinder Singh

Modify ip pipeline traffic management function to allow different subports of the same port to have different configuration in terms of number of pipes, pipe queue sizes, etc.

---
 examples/ip_pipeline/cli.c  | 71 ++++++++++++++++++-------------------
 examples/ip_pipeline/tmgr.c | 25 +++++++------
 examples/ip_pipeline/tmgr.h |  7 ++--
 3 files changed, 51 insertions(+), 52 deletions(-)
examples/ip_pipeline/cli.c +35 -36
index 02dc11495..c72030682 100644
@@ -393,7 +393,12 @@ static const char cmd_tmgr_subport_profile_help[] =
 "   <tc0_rate> <tc1_rate> <tc2_rate> <tc3_rate> <tc4_rate>"
 "        <tc5_rate> <tc6_rate> <tc7_rate> <tc8_rate>"
 "        <tc9_rate> <tc10_rate> <tc11_rate> <tc12_rate>\n"
-"   <tc_period>\n";
+"   <tc_period>\n"
+"   pps <n_pipes_per_subport>\n"
+"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
+"       <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
+"       <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
+"       <qsize_tc11> <qsize_tc12>";
 
 static void
 cmd_tmgr_subport_profile(char **tokens,
@@ -404,7 +409,7 @@ cmd_tmgr_subport_profile(char **tokens,
 	struct rte_sched_subport_params p;
 	int status, i;
 
-	if (n_tokens != 19) {
+	if (n_tokens != 35) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -430,6 +435,27 @@ cmd_tmgr_subport_profile(char **tokens,
 		return;
 	}
 
+	if (strcmp(tokens[19], "pps") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
+		return;
+	}
+
+	if (parser_read_uint32(&p.n_pipes_per_subport_enabled, tokens[20]) != 0) {
+		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
+		return;
+	}
+
+	if (strcmp(tokens[21], "qsize") != 0) {
+		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
+		return;
+	}
+
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		if (parser_read_uint16(&p.qsize[i], tokens[22 + i]) != 0) {
+			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
+			return;
+		}
+
 	status = tmgr_subport_profile_add(&p);
 	if (status != 0) {
 		snprintf(out, out_size, MSG_CMD_FAIL, tokens[0]);
@@ -504,11 +530,6 @@ static const char cmd_tmgr_help[] =
 "tmgr <tmgr_name>\n"
 "   rate <rate>\n"
 "   spp <n_subports_per_port>\n"
-"   pps <n_pipes_per_subport>\n"
-"   qsize <qsize_tc0> <qsize_tc1> <qsize_tc2>"
-"   <qsize_tc3> <qsize_tc4> <qsize_tc5> <qsize_tc6>"
-"   <qsize_tc7> <qsize_tc8> <qsize_tc9> <qsize_tc10>"
-"   <qsize_tc11> <qsize_tc12>\n"
 "   fo <frame_overhead>\n"
 "   mtu <mtu>\n"
 "   cpu <cpu_id>\n";
@@ -522,9 +543,8 @@ cmd_tmgr(char **tokens,
 	struct tmgr_port_params p;
 	char *name;
 	struct tmgr_port *tmgr_port;
-	int i;
 
-	if (n_tokens != 28) {
+	if (n_tokens != 12) {
 		snprintf(out, out_size, MSG_ARG_MISMATCH, tokens[0]);
 		return;
 	}
@@ -551,53 +571,32 @@ cmd_tmgr(char **tokens,
 		return;
 	}
 
-	if (strcmp(tokens[6], "pps") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "pps");
-		return;
-	}
-
-	if (parser_read_uint32(&p.n_pipes_per_subport, tokens[7]) != 0) {
-		snprintf(out, out_size, MSG_ARG_INVALID, "n_pipes_per_subport");
-		return;
-	}
-
-	if (strcmp(tokens[8], "qsize") != 0) {
-		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "qsize");
-		return;
-	}
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		if (parser_read_uint16(&p.qsize[i], tokens[9 + i]) != 0) {
-			snprintf(out, out_size, MSG_ARG_INVALID, "qsize");
-			return;
-		}
-
-	if (strcmp(tokens[22], "fo") != 0) {
+	if (strcmp(tokens[6], "fo") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "fo");
 		return;
 	}
 
-	if (parser_read_uint32(&p.frame_overhead, tokens[23]) != 0) {
+	if (parser_read_uint32(&p.frame_overhead, tokens[7]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "frame_overhead");
 		return;
 	}
 
-	if (strcmp(tokens[24], "mtu") != 0) {
+	if (strcmp(tokens[8], "mtu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "mtu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.mtu, tokens[25]) != 0) {
+	if (parser_read_uint32(&p.mtu, tokens[9]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "mtu");
 		return;
 	}
 
-	if (strcmp(tokens[26], "cpu") != 0) {
+	if (strcmp(tokens[10], "cpu") != 0) {
 		snprintf(out, out_size, MSG_ARG_NOT_FOUND, "cpu");
 		return;
 	}
 
-	if (parser_read_uint32(&p.cpu_id, tokens[27]) != 0) {
+	if (parser_read_uint32(&p.cpu_id, tokens[11]) != 0) {
 		snprintf(out, out_size, MSG_ARG_INVALID, "cpu_id");
 		return;
 	}
examples/ip_pipeline/tmgr.c +12 -13
index 40cbf1d0a..91ccbf60f 100644
@@ -47,7 +47,8 @@ int
 tmgr_subport_profile_add(struct rte_sched_subport_params *p)
 {
 	/* Check input params */
-	if (p == NULL)
+	if (p == NULL ||
+		p->n_pipes_per_subport_enabled == 0)
 		return -1;
 
 	/* Save profile */
@@ -90,7 +91,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 		tmgr_port_find(name) ||
 		(params == NULL) ||
 		(params->n_subports_per_port == 0) ||
-		(params->n_pipes_per_subport == 0) ||
 		(params->cpu_id >= RTE_MAX_NUMA_NODES) ||
 		(n_subport_profiles == 0) ||
 		(n_pipe_profiles == 0))
@@ -103,18 +103,16 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	p.mtu = params->mtu;
 	p.frame_overhead = params->frame_overhead;
 	p.n_subports_per_port = params->n_subports_per_port;
-	p.n_pipes_per_subport = params->n_pipes_per_subport;
-
-	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
-		p.qsize[i] = params->qsize[i];
-
-	p.pipe_profiles = pipe_profile;
-	p.n_pipe_profiles = n_pipe_profiles;
+	p.n_pipes_per_subport = TMGR_PIPE_SUBPORT_MAX;
 
 	s = rte_sched_port_config(&p);
 	if (s == NULL)
 		return NULL;
 
+	subport_profile[0].pipe_profiles = pipe_profile;
+	subport_profile[0].n_pipe_profiles = n_pipe_profiles;
+	subport_profile[0].n_max_pipe_profiles = TMGR_PIPE_PROFILE_MAX;
+
 	for (i = 0; i < params->n_subports_per_port; i++) {
 		int status;
 
@@ -128,7 +126,7 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 			return NULL;
 		}
 
-		for (j = 0; j < params->n_pipes_per_subport; j++) {
+		for (j = 0; j < subport_profile[0].n_pipes_per_subport_enabled; j++) {
 			status = rte_sched_pipe_config(
 				s,
 				i,
@@ -153,7 +151,6 @@ tmgr_port_create(const char *name, struct tmgr_port_params *params)
 	strlcpy(tmgr_port->name, name, sizeof(tmgr_port->name));
 	tmgr_port->s = s;
 	tmgr_port->n_subports_per_port = params->n_subports_per_port;
-	tmgr_port->n_pipes_per_subport = params->n_pipes_per_subport;
 
 	/* Node add to list */
 	TAILQ_INSERT_TAIL(&tmgr_port_list, tmgr_port, node);
@@ -205,8 +202,10 @@ tmgr_pipe_config(const char *port_name,
 	port = tmgr_port_find(port_name);
 	if ((port == NULL) ||
 		(subport_id >= port->n_subports_per_port) ||
-		(pipe_id_first >= port->n_pipes_per_subport) ||
-		(pipe_id_last >= port->n_pipes_per_subport) ||
+		(pipe_id_first >=
+			subport_profile[subport_id].n_pipes_per_subport_enabled) ||
+		(pipe_id_last >=
+			subport_profile[subport_id].n_pipes_per_subport_enabled) ||
 		(pipe_id_first > pipe_id_last) ||
 		(pipe_profile_id >= n_pipe_profiles))
 		return -1;
examples/ip_pipeline/tmgr.h +4 -3
index 8703a2e00..1fcf66ee1 100644
@@ -12,6 +12,10 @@
 
 #include "common.h"
 
+#ifndef TMGR_PIPE_SUBPORT_MAX
+#define TMGR_PIPE_SUBPORT_MAX                              4096
+#endif
+
 #ifndef TMGR_SUBPORT_PROFILE_MAX
 #define TMGR_SUBPORT_PROFILE_MAX                           256
 #endif
@@ -25,7 +29,6 @@ struct tmgr_port {
 	char name[NAME_SIZE];
 	struct rte_sched_port *s;
 	uint32_t n_subports_per_port;
-	uint32_t n_pipes_per_subport;
 };
 
 TAILQ_HEAD(tmgr_port_list, tmgr_port);
@@ -42,8 +45,6 @@ struct tmgr_port_params {
 	uint32_t frame_overhead;
 	uint32_t mtu;
 	uint32_t cpu_id;
-	uint32_t n_pipes_per_subport;
-	uint16_t qsize[RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE];
 };
 
 int
Signature · 1 line
2.21.0