Skip to contentSkip to conversation

Packet pacing DEVX API

PATCH Applied Leon Romanovsky 5 participants 10 messages Feb 19, 2020 – Mar 10, 2020

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

cover Packet pacing DEVX API

From: Leon Romanovsky <[email protected]>

Hi,

This series from Yishai extends packet pacing to work over DEVX interface. In first patch, he refactors the mlx5_core internal logic. In second patch, the RDMA APIs are added.

Thanks

Yishai Hadas (2):
net/mlx5: Expose raw packet pacing APIs IB/mlx5: Introduce UAPIs to manage packet pacing

 drivers/infiniband/hw/mlx5/Makefile          |   1 +
 drivers/infiniband/hw/mlx5/main.c            |   1 +
 drivers/infiniband/hw/mlx5/mlx5_ib.h         |   6 +
 drivers/infiniband/hw/mlx5/qos.c             | 136 +++++++++++++++++++
 drivers/net/ethernet/mellanox/mlx5/core/rl.c | 130 +++++++++++++-----
 include/linux/mlx5/driver.h                  |  11 +-
 include/linux/mlx5/mlx5_ifc.h                |  26 ++--
 include/uapi/rdma/mlx5_user_ioctl_cmds.h     |  17 +++
 include/uapi/rdma/mlx5_user_ioctl_verbs.h    |   4 +
 9 files changed, 287 insertions(+), 45 deletions(-)
 create mode 100644 drivers/infiniband/hw/mlx5/qos.c

--
2.24.1

#1 lore raw

PATCH 1/2 net/mlx5: Expose raw packet pacing APIs

From: Yishai Hadas <[email protected]>

Expose raw packet pacing APIs to be used by DEVX based applications. The existing code was refactored to have a single flow with the new raw APIs.

The new raw APIs considered the input of 'pp_rate_limit_context', uid, 'dedicated', upon looking for an existing entry.

This raw mode enables future device specification data in the raw context without changing the existing logic and code.

The ability to ask for a dedicated entry gives control for application to allocate entries according to its needs.

A dedicated entry may not be used by some other process and it also enables the process spreading its resources to some different entries for use different hardware resources as part of enforcing the rate.

The counter per entry mas changed to be u64 to prevent any option to overflow.

---
 drivers/net/ethernet/mellanox/mlx5/core/rl.c | 130 ++++++++++++++-----
 include/linux/mlx5/driver.h                  |  11 +-
 include/linux/mlx5/mlx5_ifc.h                |  26 ++--
 3 files changed, 122 insertions(+), 45 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/rl.c +98 -32
index 01c380425f9d..f3b29d9ade1f 100644
@@ -101,22 +101,39 @@ int mlx5_destroy_scheduling_element_cmd(struct mlx5_core_dev *dev, u8 hierarchy,
 	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 }
 
+static bool mlx5_rl_are_equal_raw(struct mlx5_rl_entry *entry, void *rl_in,
+				  u16 uid)
+{
+	return (!memcmp(entry->rl_raw, rl_in, sizeof(entry->rl_raw)) &&
+		entry->uid == uid);
+}
+
 /* Finds an entry where we can register the given rate
  * If the rate already exists, return the entry where it is registered,
  * otherwise return the first available entry.
  * If the table is full, return NULL
  */
 static struct mlx5_rl_entry *find_rl_entry(struct mlx5_rl_table *table,
-					   struct mlx5_rate_limit *rl)
+					   void *rl_in, u16 uid, bool dedicated)
 {
 	struct mlx5_rl_entry *ret_entry = NULL;
 	bool empty_found = false;
 	int i;
 
 	for (i = 0; i < table->max_size; i++) {
-		if (mlx5_rl_are_equal(&table->rl_entry[i].rl, rl))
-			return &table->rl_entry[i];
-		if (!empty_found && !table->rl_entry[i].rl.rate) {
+		if (dedicated) {
+			if (!table->rl_entry[i].refcount)
+				return &table->rl_entry[i];
+			continue;
+		}
+
+		if (table->rl_entry[i].refcount) {
+			if (table->rl_entry[i].dedicated)
+				continue;
+			if (mlx5_rl_are_equal_raw(&table->rl_entry[i], rl_in,
+						  uid))
+				return &table->rl_entry[i];
+		} else if (!empty_found) {
 			empty_found = true;
 			ret_entry = &table->rl_entry[i];
 		}
@@ -126,18 +143,19 @@ static struct mlx5_rl_entry *find_rl_entry(struct mlx5_rl_table *table,
 }
 
 static int mlx5_set_pp_rate_limit_cmd(struct mlx5_core_dev *dev,
-				      u16 index,
-				      struct mlx5_rate_limit *rl)
+				      struct mlx5_rl_entry *entry, bool set)
 {
-	u32 in[MLX5_ST_SZ_DW(set_pp_rate_limit_in)]   = {0};
-	u32 out[MLX5_ST_SZ_DW(set_pp_rate_limit_out)] = {0};
+	u32 in[MLX5_ST_SZ_DW(set_pp_rate_limit_in)]   = {};
+	u32 out[MLX5_ST_SZ_DW(set_pp_rate_limit_out)] = {};
+	void *pp_context;
 
+	pp_context = MLX5_ADDR_OF(set_pp_rate_limit_in, in, ctx);
 	MLX5_SET(set_pp_rate_limit_in, in, opcode,
 		 MLX5_CMD_OP_SET_PP_RATE_LIMIT);
-	MLX5_SET(set_pp_rate_limit_in, in, rate_limit_index, index);
-	MLX5_SET(set_pp_rate_limit_in, in, rate_limit, rl->rate);
-	MLX5_SET(set_pp_rate_limit_in, in, burst_upper_bound, rl->max_burst_sz);
-	MLX5_SET(set_pp_rate_limit_in, in, typical_packet_size, rl->typical_pkt_sz);
+	MLX5_SET(set_pp_rate_limit_in, in, uid, entry->uid);
+	MLX5_SET(set_pp_rate_limit_in, in, rate_limit_index, entry->index);
+	if (set)
+		memcpy(pp_context, entry->rl_raw, sizeof(entry->rl_raw));
 	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 }
 
@@ -158,23 +176,25 @@ bool mlx5_rl_are_equal(struct mlx5_rate_limit *rl_0,
 }
 EXPORT_SYMBOL(mlx5_rl_are_equal);
 
-int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
-		     struct mlx5_rate_limit *rl)
+int mlx5_rl_add_rate_raw(struct mlx5_core_dev *dev, void *rl_in, u16 uid,
+			 bool dedicated_entry, u16 *index)
 {
 	struct mlx5_rl_table *table = &dev->priv.rl_table;
 	struct mlx5_rl_entry *entry;
 	int err = 0;
+	u32 rate;
 
+	rate = MLX5_GET(set_pp_rate_limit_context, rl_in, rate_limit);
 	mutex_lock(&table->rl_lock);
 
-	if (!rl->rate || !mlx5_rl_is_in_range(dev, rl->rate)) {
+	if (!rate || !mlx5_rl_is_in_range(dev, rate)) {
 		mlx5_core_err(dev, "Invalid rate: %u, should be %u to %u\n",
-			      rl->rate, table->min_rate, table->max_rate);
+			      rate, table->min_rate, table->max_rate);
 		err = -EINVAL;
 		goto out;
 	}
 
-	entry = find_rl_entry(table, rl);
+	entry = find_rl_entry(table, rl_in, uid, dedicated_entry);
 	if (!entry) {
 		mlx5_core_err(dev, "Max number of %u rates reached\n",
 			      table->max_size);
@@ -185,16 +205,24 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
 		/* rate already configured */
 		entry->refcount++;
 	} else {
+		memcpy(entry->rl_raw, rl_in, sizeof(entry->rl_raw));
+		entry->uid = uid;
 		/* new rate limit */
-		err = mlx5_set_pp_rate_limit_cmd(dev, entry->index, rl);
+		err = mlx5_set_pp_rate_limit_cmd(dev, entry, true);
 		if (err) {
-			mlx5_core_err(dev, "Failed configuring rate limit(err %d): rate %u, max_burst_sz %u, typical_pkt_sz %u\n",
-				      err, rl->rate, rl->max_burst_sz,
-				      rl->typical_pkt_sz);
+			mlx5_core_err(
+				dev,
+				"Failed configuring rate limit(err %d): rate %u, max_burst_sz %u, typical_pkt_sz %u\n",
+				err, rate,
+				MLX5_GET(set_pp_rate_limit_context, rl_in,
+					 burst_upper_bound),
+				MLX5_GET(set_pp_rate_limit_context, rl_in,
+					 typical_packet_size));
 			goto out;
 		}
-		entry->rl = *rl;
+
 		entry->refcount = 1;
+		entry->dedicated = dedicated_entry;
 	}
 	*index = entry->index;
 
@@ -202,20 +230,61 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
 	mutex_unlock(&table->rl_lock);
 	return err;
 }
+EXPORT_SYMBOL(mlx5_rl_add_rate_raw);
+
+void mlx5_rl_remove_rate_raw(struct mlx5_core_dev *dev, u16 index)
+{
+	struct mlx5_rl_table *table = &dev->priv.rl_table;
+	struct mlx5_rl_entry *entry;
+
+	mutex_lock(&table->rl_lock);
+	entry = &table->rl_entry[index - 1];
+	entry->refcount--;
+	if (!entry->refcount)
+		/* need to remove rate */
+		mlx5_set_pp_rate_limit_cmd(dev, entry, false);
+	mutex_unlock(&table->rl_lock);
+}
+EXPORT_SYMBOL(mlx5_rl_remove_rate_raw);
+
+int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
+		     struct mlx5_rate_limit *rl)
+{
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)] = {};
+
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, rate_limit, rl->rate);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, burst_upper_bound,
+		 rl->max_burst_sz);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, typical_packet_size,
+		 rl->typical_pkt_sz);
+
+	return mlx5_rl_add_rate_raw(dev, rl_raw,
+				    MLX5_CAP_QOS(dev, packet_pacing_uid) ?
+					MLX5_SHARED_RESOURCE_UID : 0,
+				    false, index);
+}
 EXPORT_SYMBOL(mlx5_rl_add_rate);
 
 void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, struct mlx5_rate_limit *rl)
 {
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)] = {};
 	struct mlx5_rl_table *table = &dev->priv.rl_table;
 	struct mlx5_rl_entry *entry = NULL;
-	struct mlx5_rate_limit reset_rl = {0};
 
 	/* 0 is a reserved value for unlimited rate */
 	if (rl->rate == 0)
 		return;
 
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, rate_limit, rl->rate);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, burst_upper_bound,
+		 rl->max_burst_sz);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, typical_packet_size,
+		 rl->typical_pkt_sz);
+
 	mutex_lock(&table->rl_lock);
-	entry = find_rl_entry(table, rl);
+	entry = find_rl_entry(table, rl_raw,
+			      MLX5_CAP_QOS(dev, packet_pacing_uid) ?
+				MLX5_SHARED_RESOURCE_UID : 0, false);
 	if (!entry || !entry->refcount) {
 		mlx5_core_warn(dev, "Rate %u, max_burst_sz %u typical_pkt_sz %u are not configured\n",
 			       rl->rate, rl->max_burst_sz, rl->typical_pkt_sz);
@@ -223,11 +292,9 @@ void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, struct mlx5_rate_limit *rl)
 	}
 
 	entry->refcount--;
-	if (!entry->refcount) {
+	if (!entry->refcount)
 		/* need to remove rate */
-		mlx5_set_pp_rate_limit_cmd(dev, entry->index, &reset_rl);
-		entry->rl = reset_rl;
-	}
+		mlx5_set_pp_rate_limit_cmd(dev, entry, false);
 
 out:
 	mutex_unlock(&table->rl_lock);
@@ -273,14 +340,13 @@ int mlx5_init_rl_table(struct mlx5_core_dev *dev)
 void mlx5_cleanup_rl_table(struct mlx5_core_dev *dev)
 {
 	struct mlx5_rl_table *table = &dev->priv.rl_table;
-	struct mlx5_rate_limit rl = {0};
 	int i;
 
 	/* Clear all configured rates */
 	for (i = 0; i < table->max_size; i++)
-		if (table->rl_entry[i].rl.rate)
-			mlx5_set_pp_rate_limit_cmd(dev, table->rl_entry[i].index,
-						   &rl);
+		if (table->rl_entry[i].refcount)
+			mlx5_set_pp_rate_limit_cmd(dev, &table->rl_entry[i],
+						   false);
 
 	kfree(dev->priv.rl_table.rl_entry);
 }
include/linux/mlx5/driver.h +8 -3
index 277a51d3ec40..f2b4225ed650 100644
@@ -518,9 +518,11 @@ struct mlx5_rate_limit {
 };
 
 struct mlx5_rl_entry {
-	struct mlx5_rate_limit	rl;
-	u16                     index;
-	u16                     refcount;
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)];
+	u16 index;
+	u64 refcount;
+	u16 uid;
+	u8 dedicated : 1;
 };
 
 struct mlx5_rl_table {
@@ -1007,6 +1009,9 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
 		     struct mlx5_rate_limit *rl);
 void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, struct mlx5_rate_limit *rl);
 bool mlx5_rl_is_in_range(struct mlx5_core_dev *dev, u32 rate);
+int mlx5_rl_add_rate_raw(struct mlx5_core_dev *dev, void *rl_in, u16 uid,
+			 bool dedicated_entry, u16 *index);
+void mlx5_rl_remove_rate_raw(struct mlx5_core_dev *dev, u16 index);
 bool mlx5_rl_are_equal(struct mlx5_rate_limit *rl_0,
 		       struct mlx5_rate_limit *rl_1);
 int mlx5_alloc_bfreg(struct mlx5_core_dev *mdev, struct mlx5_sq_bfreg *bfreg,
include/linux/mlx5/mlx5_ifc.h +16 -10
index ff8c9d527bb4..7d89ab64b372 100644
@@ -810,7 +810,9 @@ struct mlx5_ifc_qos_cap_bits {
 	u8         reserved_at_4[0x1];
 	u8         packet_pacing_burst_bound[0x1];
 	u8         packet_pacing_typical_size[0x1];
-	u8         reserved_at_7[0x19];
+	u8         reserved_at_7[0x4];
+	u8         packet_pacing_uid[0x1];
+	u8         reserved_at_c[0x14];
 
 	u8         reserved_at_20[0x20];
 
@@ -8262,9 +8264,20 @@ struct mlx5_ifc_set_pp_rate_limit_out_bits {
 	u8         reserved_at_40[0x40];
 };
 
+struct mlx5_ifc_set_pp_rate_limit_context_bits {
+	u8         rate_limit[0x20];
+
+	u8	   burst_upper_bound[0x20];
+
+	u8         reserved_at_40[0x10];
+	u8	   typical_packet_size[0x10];
+
+	u8         reserved_at_60[0x120];
+};
+
 struct mlx5_ifc_set_pp_rate_limit_in_bits {
 	u8         opcode[0x10];
-	u8         reserved_at_10[0x10];
+	u8         uid[0x10];
 
 	u8         reserved_at_20[0x10];
 	u8         op_mod[0x10];
@@ -8274,14 +8287,7 @@ struct mlx5_ifc_set_pp_rate_limit_in_bits {
 
 	u8         reserved_at_60[0x20];
 
-	u8         rate_limit[0x20];
-
-	u8	   burst_upper_bound[0x20];
-
-	u8         reserved_at_c0[0x10];
-	u8	   typical_packet_size[0x10];
-
-	u8         reserved_at_e0[0x120];
+	struct mlx5_ifc_set_pp_rate_limit_context_bits ctx;
 };
 
 struct mlx5_ifc_access_register_out_bits {
Signature · 1 line
2.24.1
#2 lore raw

Re: [PATCH mlx5-next 1/2] net/mlx5: Expose raw packet pacing APIs

Quoted 14 lines · Leon Romanovsky

From: Yishai Hadas <[email protected]>

Expose raw packet pacing APIs to be used by DEVX based applications. The existing code was refactored to have a single flow with the new raw
APIs.

The new raw APIs considered the input of 'pp_rate_limit_context', uid,
'dedicated', upon looking for an existing entry.

This raw mode enables future device specification data in the raw context without changing the existing logic and code.

Quoted 8 lines

The ability to ask for a dedicated entry gives control for application
to allocate entries according to its needs.

A dedicated entry may not be used by some other process and it also enables the process spreading its resources to some different entries for use different hardware resources as part of enforcing the rate.

It sounds like the dedicated means "no sharing" which means you don't need to use the mlx5_core API and you can go directly to FW.. The problem is that the entry indices are managed by driver, and i guess this is the reason why you had to expand the mlx5_core API..

I would like to suggest some alternatives to simplify the approach and allow using RAW PRM for DEVX properly.

1. preserve RL entries for DEVX and let DEVX access FW directly with PRM commands.
2. keep mlx5_core API simple and instead of adding this raw/non raw api and complicating the RL API with this dedicated bit:

just add mlx5_rl_{alloc/free}_index(), this will dedicate for you the RL index form the end of the RL indices database and you are free to access the FW with this index the way you like via direct PRM commands.

Quoted 1 line

The counter per entry mas changed to be u64 to prevent any option to

typo ^^^ was

Quoted 369 lines

overflow.

---
 drivers/net/ethernet/mellanox/mlx5/core/rl.c | 130 ++++++++++++++---
--
 include/linux/mlx5/driver.h                  |  11 +-
 include/linux/mlx5/mlx5_ifc.h                |  26 ++--
 3 files changed, 122 insertions(+), 45 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/rl.c +18 -4
index 01c380425f9d..f3b29d9ade1f 100644
@@ -101,22 +101,39 @@ int mlx5_destroy_scheduling_element_cmd(struct
 mlx5_core_dev *dev, u8 hierarchy,
 	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 }
 
+static bool mlx5_rl_are_equal_raw(struct mlx5_rl_entry *entry, void
 *rl_in,
+				  u16 uid)
+{
+	return (!memcmp(entry->rl_raw, rl_in, sizeof(entry->rl_raw)) &&
+		entry->uid == uid);
+}
+
 /* Finds an entry where we can register the given rate
  * If the rate already exists, return the entry where it is
 registered,
  * otherwise return the first available entry.
  * If the table is full, return NULL
  */
 static struct mlx5_rl_entry *find_rl_entry(struct mlx5_rl_table
 *table,
-					   struct mlx5_rate_limit *rl)
+					   void *rl_in, u16 uid, bool
 dedicated)
 {
 	struct mlx5_rl_entry *ret_entry = NULL;
 	bool empty_found = false;
 	int i;
 
 	for (i = 0; i < table->max_size; i++) {
-		if (mlx5_rl_are_equal(&table->rl_entry[i].rl, rl))
-			return &table->rl_entry[i];
-		if (!empty_found && !table->rl_entry[i].rl.rate) {
+		if (dedicated) {
+			if (!table->rl_entry[i].refcount)
+				return &table->rl_entry[i];
+			continue;
+		}
+
+		if (table->rl_entry[i].refcount) {
+			if (table->rl_entry[i].dedicated)
+				continue;
+			if (mlx5_rl_are_equal_raw(&table->rl_entry[i],

rl_in,

drivers/net/ethernet/mellanox/mlx5/core/rl.c +8 -8
+						  uid))
+				return &table->rl_entry[i];
+		} else if (!empty_found) {
 			empty_found = true;
 			ret_entry = &table->rl_entry[i];
 		}
@@ -126,18 +143,19 @@ static struct mlx5_rl_entry
 *find_rl_entry(struct mlx5_rl_table *table,
 }
 
 static int mlx5_set_pp_rate_limit_cmd(struct mlx5_core_dev *dev,
-				      u16 index,
-				      struct mlx5_rate_limit *rl)
+				      struct mlx5_rl_entry *entry, bool
 set)
 {
-	u32 in[MLX5_ST_SZ_DW(set_pp_rate_limit_in)]   = {0};
-	u32 out[MLX5_ST_SZ_DW(set_pp_rate_limit_out)] = {0};
+	u32 in[MLX5_ST_SZ_DW(set_pp_rate_limit_in)]   = {};
+	u32 out[MLX5_ST_SZ_DW(set_pp_rate_limit_out)] = {};
+	void *pp_context;
 
+	pp_context = MLX5_ADDR_OF(set_pp_rate_limit_in, in, ctx);
 	MLX5_SET(set_pp_rate_limit_in, in, opcode,
 		 MLX5_CMD_OP_SET_PP_RATE_LIMIT);
-	MLX5_SET(set_pp_rate_limit_in, in, rate_limit_index, index);
-	MLX5_SET(set_pp_rate_limit_in, in, rate_limit, rl->rate);
-	MLX5_SET(set_pp_rate_limit_in, in, burst_upper_bound, rl-
 >max_burst_sz);
-	MLX5_SET(set_pp_rate_limit_in, in, typical_packet_size, rl-
Quoted 1 line

typical_pkt_sz);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -0
+	MLX5_SET(set_pp_rate_limit_in, in, uid, entry->uid);
+	MLX5_SET(set_pp_rate_limit_in, in, rate_limit_index, entry-
Quoted 1 line

index);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -0
+	if (set)
+		memcpy(pp_context, entry->rl_raw, sizeof(entry-
Quoted 1 line

rl_raw));

drivers/net/ethernet/mellanox/mlx5/core/rl.c +50 -10
 	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 }
@@ -158,23 +176,25 @@ bool mlx5_rl_are_equal(struct mlx5_rate_limit
 *rl_0,
 }
 EXPORT_SYMBOL(mlx5_rl_are_equal);
 
-int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
-		     struct mlx5_rate_limit *rl)
+int mlx5_rl_add_rate_raw(struct mlx5_core_dev *dev, void *rl_in, u16
 uid,
+			 bool dedicated_entry, u16 *index)
 {
 	struct mlx5_rl_table *table = &dev->priv.rl_table;
 	struct mlx5_rl_entry *entry;
 	int err = 0;
+	u32 rate;
 
+	rate = MLX5_GET(set_pp_rate_limit_context, rl_in, rate_limit);
 	mutex_lock(&table->rl_lock);
 
-	if (!rl->rate || !mlx5_rl_is_in_range(dev, rl->rate)) {
+	if (!rate || !mlx5_rl_is_in_range(dev, rate)) {
 		mlx5_core_err(dev, "Invalid rate: %u, should be %u to
 %u\n",
-			      rl->rate, table->min_rate, table-
 >max_rate);
+			      rate, table->min_rate, table->max_rate);
 		err = -EINVAL;
 		goto out;
 	}
 
-	entry = find_rl_entry(table, rl);
+	entry = find_rl_entry(table, rl_in, uid, dedicated_entry);
 	if (!entry) {
 		mlx5_core_err(dev, "Max number of %u rates reached\n",
 			      table->max_size);
@@ -185,16 +205,24 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev,
 u16 *index,
 		/* rate already configured */
 		entry->refcount++;
 	} else {
+		memcpy(entry->rl_raw, rl_in, sizeof(entry->rl_raw));
+		entry->uid = uid;
 		/* new rate limit */
-		err = mlx5_set_pp_rate_limit_cmd(dev, entry->index,
 rl);
+		err = mlx5_set_pp_rate_limit_cmd(dev, entry, true);
 		if (err) {
-			mlx5_core_err(dev, "Failed configuring rate
 limit(err %d): rate %u, max_burst_sz %u, typical_pkt_sz %u\n",
-				      err, rl->rate, rl->max_burst_sz,
-				      rl->typical_pkt_sz);
+			mlx5_core_err(
+				dev,
+				"Failed configuring rate limit(err %d):
 rate %u, max_burst_sz %u, typical_pkt_sz %u\n",
+				err, rate,
+				MLX5_GET(set_pp_rate_limit_context,
 rl_in,
+					 burst_upper_bound),
+				MLX5_GET(set_pp_rate_limit_context,
 rl_in,
+					 typical_packet_size));
 			goto out;
 		}
-		entry->rl = *rl;
+
 		entry->refcount = 1;
+		entry->dedicated = dedicated_entry;
 	}
 	*index = entry->index;
@@ -202,20 +230,61 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev,
 u16 *index,
 	mutex_unlock(&table->rl_lock);
 	return err;
 }
+EXPORT_SYMBOL(mlx5_rl_add_rate_raw);
+
+void mlx5_rl_remove_rate_raw(struct mlx5_core_dev *dev, u16 index)
+{
+	struct mlx5_rl_table *table = &dev->priv.rl_table;
+	struct mlx5_rl_entry *entry;
+
+	mutex_lock(&table->rl_lock);
+	entry = &table->rl_entry[index - 1];
+	entry->refcount--;
+	if (!entry->refcount)
+		/* need to remove rate */
+		mlx5_set_pp_rate_limit_cmd(dev, entry, false);
+	mutex_unlock(&table->rl_lock);
+}
+EXPORT_SYMBOL(mlx5_rl_remove_rate_raw);
+
+int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
+		     struct mlx5_rate_limit *rl)
+{
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)] = {};
+
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, rate_limit, rl-
 >rate);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, burst_upper_bound,
+		 rl->max_burst_sz);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw,
 typical_packet_size,
+		 rl->typical_pkt_sz);
+
+	return mlx5_rl_add_rate_raw(dev, rl_raw,
+				    MLX5_CAP_QOS(dev,

packet_pacing_uid) ?

drivers/net/ethernet/mellanox/mlx5/core/rl.c +3 -0
+					MLX5_SHARED_RESOURCE_UID : 0,
+				    false, index);
+}
 EXPORT_SYMBOL(mlx5_rl_add_rate);
 
 void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, struct

mlx5_rate_limit *rl)
{

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -1
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)] = {};
 	struct mlx5_rl_table *table = &dev->priv.rl_table;
 	struct mlx5_rl_entry *entry = NULL;
-	struct mlx5_rate_limit reset_rl = {0};
 
 	/* 0 is a reserved value for unlimited rate */
 	if (rl->rate == 0)
 		return;
 
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, rate_limit, rl-
Quoted 1 line

rate);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +3 -0
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, burst_upper_bound,
+		 rl->max_burst_sz);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw,

typical_packet_size,

drivers/net/ethernet/mellanox/mlx5/core/rl.c +5 -1
+		 rl->typical_pkt_sz);
+
 	mutex_lock(&table->rl_lock);
-	entry = find_rl_entry(table, rl);
+	entry = find_rl_entry(table, rl_raw,
+			      MLX5_CAP_QOS(dev, packet_pacing_uid) ?
+				MLX5_SHARED_RESOURCE_UID : 0, false);
 	if (!entry || !entry->refcount) {
 		mlx5_core_warn(dev, "Rate %u, max_burst_sz %u

typical_pkt_sz %u are not configured\n",
rl->rate, rl->max_burst_sz, rl-

Quoted 1 line

typical_pkt_sz);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -4
@@ -223,11 +292,9 @@ void mlx5_rl_remove_rate(struct mlx5_core_dev
 *dev, struct mlx5_rate_limit *rl)
 	}
 
 	entry->refcount--;
-	if (!entry->refcount) {
+	if (!entry->refcount)
 		/* need to remove rate */
-		mlx5_set_pp_rate_limit_cmd(dev, entry->index,
 &reset_rl);
-		entry->rl = reset_rl;
-	}
+		mlx5_set_pp_rate_limit_cmd(dev, entry, false);
 

out:

drivers/net/ethernet/mellanox/mlx5/core/rl.c +3 -4
 	mutex_unlock(&table->rl_lock);
@@ -273,14 +340,13 @@ int mlx5_init_rl_table(struct mlx5_core_dev
 *dev)
 void mlx5_cleanup_rl_table(struct mlx5_core_dev *dev)
 {
 	struct mlx5_rl_table *table = &dev->priv.rl_table;
-	struct mlx5_rate_limit rl = {0};
 	int i;
 
 	/* Clear all configured rates */
 	for (i = 0; i < table->max_size; i++)
-		if (table->rl_entry[i].rl.rate)
-			mlx5_set_pp_rate_limit_cmd(dev, table-
 >rl_entry[i].index,
-						   &rl);
+		if (table->rl_entry[i].refcount)
+			mlx5_set_pp_rate_limit_cmd(dev, &table-
 >rl_entry[i],
+						   false);
drivers/net/ethernet/mellanox/mlx5/core/rl.c +0 -0
 	kfree(dev->priv.rl_table.rl_entry);
 }
include/linux/mlx5/driver.h +8 -3
index 277a51d3ec40..f2b4225ed650 100644
@@ -518,9 +518,11 @@ struct mlx5_rate_limit {
 };
 
 struct mlx5_rl_entry {
-	struct mlx5_rate_limit	rl;
-	u16                     index;
-	u16                     refcount;
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)];
+	u16 index;
+	u64 refcount;
+	u16 uid;
+	u8 dedicated : 1;
 };
 
 struct mlx5_rl_table {
@@ -1007,6 +1009,9 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev,
 u16 *index,
 		     struct mlx5_rate_limit *rl);
 void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, struct
 mlx5_rate_limit *rl);
 bool mlx5_rl_is_in_range(struct mlx5_core_dev *dev, u32 rate);
+int mlx5_rl_add_rate_raw(struct mlx5_core_dev *dev, void *rl_in, u16
 uid,
+			 bool dedicated_entry, u16 *index);
+void mlx5_rl_remove_rate_raw(struct mlx5_core_dev *dev, u16 index);

bool mlx5_rl_are_equal(struct mlx5_rate_limit *rl_0,
struct mlx5_rate_limit *rl_1);
int mlx5_alloc_bfreg(struct mlx5_core_dev *mdev, struct
mlx5_sq_bfreg *bfreg,

include/linux/mlx5/mlx5_ifc.h +16 -10
index ff8c9d527bb4..7d89ab64b372 100644
@@ -810,7 +810,9 @@ struct mlx5_ifc_qos_cap_bits {
 	u8         reserved_at_4[0x1];
 	u8         packet_pacing_burst_bound[0x1];
 	u8         packet_pacing_typical_size[0x1];
-	u8         reserved_at_7[0x19];
+	u8         reserved_at_7[0x4];
+	u8         packet_pacing_uid[0x1];
+	u8         reserved_at_c[0x14];
 
 	u8         reserved_at_20[0x20];
 
@@ -8262,9 +8264,20 @@ struct mlx5_ifc_set_pp_rate_limit_out_bits {
 	u8         reserved_at_40[0x40];
 };
 
+struct mlx5_ifc_set_pp_rate_limit_context_bits {
+	u8         rate_limit[0x20];
+
+	u8	   burst_upper_bound[0x20];
+
+	u8         reserved_at_40[0x10];
+	u8	   typical_packet_size[0x10];
+
+	u8         reserved_at_60[0x120];
+};
+
 struct mlx5_ifc_set_pp_rate_limit_in_bits {
 	u8         opcode[0x10];
-	u8         reserved_at_10[0x10];
+	u8         uid[0x10];
 
 	u8         reserved_at_20[0x10];
 	u8         op_mod[0x10];
@@ -8274,14 +8287,7 @@ struct mlx5_ifc_set_pp_rate_limit_in_bits {
 
 	u8         reserved_at_60[0x20];
 
-	u8         rate_limit[0x20];
-
-	u8	   burst_upper_bound[0x20];
-
-	u8         reserved_at_c0[0x10];
-	u8	   typical_packet_size[0x10];
-
-	u8         reserved_at_e0[0x120];
+	struct mlx5_ifc_set_pp_rate_limit_context_bits ctx;
 };
 
 struct mlx5_ifc_access_register_out_bits {
#3 lore raw

Re: [PATCH mlx5-next 1/2] net/mlx5: Expose raw packet pacing APIs

Quoted 24 lines · Saeed Mahameed
Quoted 14 lines · Leon Romanovsky

From: Yishai Hadas <[email protected]>

Expose raw packet pacing APIs to be used by DEVX based applications. The existing code was refactored to have a single flow with the new raw
APIs.

The new raw APIs considered the input of 'pp_rate_limit_context', uid,
'dedicated', upon looking for an existing entry.

This raw mode enables future device specification data in the raw context without changing the existing logic and code.

Quoted 8 lines

The ability to ask for a dedicated entry gives control for application
to allocate entries according to its needs.

A dedicated entry may not be used by some other process and it also enables the process spreading its resources to some different entries for use different hardware resources as part of enforcing the rate.

Quoted 5 lines

It sounds like the dedicated means "no sharing" which means you don't need to use the mlx5_core API and you can go directly to FW.. The problem is that the entry indices are managed by driver, and i guess this is the reason why you had to expand the mlx5_core API..

The main reason for introducing the new mlx5_core APIs was the need to support the "shared mode" in a "raw data" format to prevent future touching the kernel once PRM will support extra fields. As the RL indices are managed by the driver (mlx5_core) including the sharing, we couldn’t go directly to FW, the legacy API was refactored inside the core to have one flow with the new raw APIs. So we may need those APIs regardless the dedicated mode.

Quoted 12 lines

I would like to suggest some alternatives to simplify the approach and allow using RAW PRM for DEVX properly.

1. preserve RL entries for DEVX and let DEVX access FW directly with PRM commands.
2. keep mlx5_core API simple and instead of adding this raw/non raw api and complicating the RL API with this dedicated bit:

just add mlx5_rl_{alloc/free}_index(), this will dedicate for you the RL index form the end of the RL indices database and you are free to access the FW with this index the way you like via direct PRM commands.

As mentioned above, we may still need the new mlx5_core raw APIs for the shared mode which is the main usage of the API, we found it reasonable to have the dedicate flag in the new raw alloc API instead of exposing more two new APIs only for that.

Please note that even if we'll go with those 2 extra APIs for the dedicated mode, we may still need to maintain in the core this information to prevent returning this entry for other cases.

Also the idea to preserve some entries at the end might be wasteful as there is no guarantee that DEVX will really be used, and even so it may not ask for entries in a dedicated mode.

Presering them for this optional use case might prevent using them for all other cases.

Quoted 2 lines
Quoted 1 line

The counter per entry mas changed to be u64 to prevent any option to

typo ^^^ was

Sure, thanks.

Quoted 370 lines
Quoted 370 lines

overflow.

---
  drivers/net/ethernet/mellanox/mlx5/core/rl.c | 130 ++++++++++++++---
--
  include/linux/mlx5/driver.h                  |  11 +-
  include/linux/mlx5/mlx5_ifc.h                |  26 ++--
  3 files changed, 122 insertions(+), 45 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/rl.c +18 -4
index 01c380425f9d..f3b29d9ade1f 100644
@@ -101,22 +101,39 @@ int mlx5_destroy_scheduling_element_cmd(struct
 mlx5_core_dev *dev, u8 hierarchy,
  	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
  }
 
+static bool mlx5_rl_are_equal_raw(struct mlx5_rl_entry *entry, void
 *rl_in,
+				  u16 uid)
+{
+	return (!memcmp(entry->rl_raw, rl_in, sizeof(entry->rl_raw)) &&
+		entry->uid == uid);
+}
+
  /* Finds an entry where we can register the given rate
   * If the rate already exists, return the entry where it is
 registered,
   * otherwise return the first available entry.
   * If the table is full, return NULL
   */
  static struct mlx5_rl_entry *find_rl_entry(struct mlx5_rl_table
 *table,
-					   struct mlx5_rate_limit *rl)
+					   void *rl_in, u16 uid, bool
 dedicated)
  {
  	struct mlx5_rl_entry *ret_entry = NULL;
  	bool empty_found = false;
  	int i;
 
  	for (i = 0; i < table->max_size; i++) {
-		if (mlx5_rl_are_equal(&table->rl_entry[i].rl, rl))
-			return &table->rl_entry[i];
-		if (!empty_found && !table->rl_entry[i].rl.rate) {
+		if (dedicated) {
+			if (!table->rl_entry[i].refcount)
+				return &table->rl_entry[i];
+			continue;
+		}
+
+		if (table->rl_entry[i].refcount) {
+			if (table->rl_entry[i].dedicated)
+				continue;
+			if (mlx5_rl_are_equal_raw(&table->rl_entry[i],

rl_in,

drivers/net/ethernet/mellanox/mlx5/core/rl.c +8 -8
+						  uid))
+				return &table->rl_entry[i];
+		} else if (!empty_found) {
  			empty_found = true;
  			ret_entry = &table->rl_entry[i];
  		}
@@ -126,18 +143,19 @@ static struct mlx5_rl_entry
 *find_rl_entry(struct mlx5_rl_table *table,
  }
 
  static int mlx5_set_pp_rate_limit_cmd(struct mlx5_core_dev *dev,
-				      u16 index,
-				      struct mlx5_rate_limit *rl)
+				      struct mlx5_rl_entry *entry, bool
 set)
  {
-	u32 in[MLX5_ST_SZ_DW(set_pp_rate_limit_in)]   = {0};
-	u32 out[MLX5_ST_SZ_DW(set_pp_rate_limit_out)] = {0};
+	u32 in[MLX5_ST_SZ_DW(set_pp_rate_limit_in)]   = {};
+	u32 out[MLX5_ST_SZ_DW(set_pp_rate_limit_out)] = {};
+	void *pp_context;
 
+	pp_context = MLX5_ADDR_OF(set_pp_rate_limit_in, in, ctx);
  	MLX5_SET(set_pp_rate_limit_in, in, opcode,
  		 MLX5_CMD_OP_SET_PP_RATE_LIMIT);
-	MLX5_SET(set_pp_rate_limit_in, in, rate_limit_index, index);
-	MLX5_SET(set_pp_rate_limit_in, in, rate_limit, rl->rate);
-	MLX5_SET(set_pp_rate_limit_in, in, burst_upper_bound, rl-
 > max_burst_sz);
-	MLX5_SET(set_pp_rate_limit_in, in, typical_packet_size, rl-
Quoted 1 line

typical_pkt_sz);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -0
+	MLX5_SET(set_pp_rate_limit_in, in, uid, entry->uid);
+	MLX5_SET(set_pp_rate_limit_in, in, rate_limit_index, entry-
Quoted 1 line

index);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -0
+	if (set)
+		memcpy(pp_context, entry->rl_raw, sizeof(entry-
Quoted 1 line

rl_raw));

return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
}

drivers/net/ethernet/mellanox/mlx5/core/rl.c +50 -10
@@ -158,23 +176,25 @@ bool mlx5_rl_are_equal(struct mlx5_rate_limit
 *rl_0,
  }
  EXPORT_SYMBOL(mlx5_rl_are_equal);
 
-int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
-		     struct mlx5_rate_limit *rl)
+int mlx5_rl_add_rate_raw(struct mlx5_core_dev *dev, void *rl_in, u16
 uid,
+			 bool dedicated_entry, u16 *index)
  {
  	struct mlx5_rl_table *table = &dev->priv.rl_table;
  	struct mlx5_rl_entry *entry;
  	int err = 0;
+	u32 rate;
 
+	rate = MLX5_GET(set_pp_rate_limit_context, rl_in, rate_limit);
  	mutex_lock(&table->rl_lock);
 
-	if (!rl->rate || !mlx5_rl_is_in_range(dev, rl->rate)) {
+	if (!rate || !mlx5_rl_is_in_range(dev, rate)) {
  		mlx5_core_err(dev, "Invalid rate: %u, should be %u to
 %u\n",
-			      rl->rate, table->min_rate, table-
 > max_rate);
+			      rate, table->min_rate, table->max_rate);
  		err = -EINVAL;
  		goto out;
  	}
 
-	entry = find_rl_entry(table, rl);
+	entry = find_rl_entry(table, rl_in, uid, dedicated_entry);
  	if (!entry) {
  		mlx5_core_err(dev, "Max number of %u rates reached\n",
  			      table->max_size);
@@ -185,16 +205,24 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev,
 u16 *index,
  		/* rate already configured */
  		entry->refcount++;
  	} else {
+		memcpy(entry->rl_raw, rl_in, sizeof(entry->rl_raw));
+		entry->uid = uid;
  		/* new rate limit */
-		err = mlx5_set_pp_rate_limit_cmd(dev, entry->index,
 rl);
+		err = mlx5_set_pp_rate_limit_cmd(dev, entry, true);
  		if (err) {
-			mlx5_core_err(dev, "Failed configuring rate
 limit(err %d): rate %u, max_burst_sz %u, typical_pkt_sz %u\n",
-				      err, rl->rate, rl->max_burst_sz,
-				      rl->typical_pkt_sz);
+			mlx5_core_err(
+				dev,
+				"Failed configuring rate limit(err %d):
 rate %u, max_burst_sz %u, typical_pkt_sz %u\n",
+				err, rate,
+				MLX5_GET(set_pp_rate_limit_context,
 rl_in,
+					 burst_upper_bound),
+				MLX5_GET(set_pp_rate_limit_context,
 rl_in,
+					 typical_packet_size));
  			goto out;
  		}
-		entry->rl = *rl;
+
  		entry->refcount = 1;
+		entry->dedicated = dedicated_entry;
  	}
  	*index = entry->index;
@@ -202,20 +230,61 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev,
 u16 *index,
  	mutex_unlock(&table->rl_lock);
  	return err;
  }
+EXPORT_SYMBOL(mlx5_rl_add_rate_raw);
+
+void mlx5_rl_remove_rate_raw(struct mlx5_core_dev *dev, u16 index)
+{
+	struct mlx5_rl_table *table = &dev->priv.rl_table;
+	struct mlx5_rl_entry *entry;
+
+	mutex_lock(&table->rl_lock);
+	entry = &table->rl_entry[index - 1];
+	entry->refcount--;
+	if (!entry->refcount)
+		/* need to remove rate */
+		mlx5_set_pp_rate_limit_cmd(dev, entry, false);
+	mutex_unlock(&table->rl_lock);
+}
+EXPORT_SYMBOL(mlx5_rl_remove_rate_raw);
+
+int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u16 *index,
+		     struct mlx5_rate_limit *rl)
+{
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)] = {};
+
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, rate_limit, rl-
 > rate);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, burst_upper_bound,
+		 rl->max_burst_sz);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw,
 typical_packet_size,
+		 rl->typical_pkt_sz);
+
+	return mlx5_rl_add_rate_raw(dev, rl_raw,
+				    MLX5_CAP_QOS(dev,

packet_pacing_uid) ?

drivers/net/ethernet/mellanox/mlx5/core/rl.c +3 -0
+					MLX5_SHARED_RESOURCE_UID : 0,
+				    false, index);
+}
  EXPORT_SYMBOL(mlx5_rl_add_rate);
 
  void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, struct

mlx5_rate_limit *rl)
{

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -1
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)] = {};
  	struct mlx5_rl_table *table = &dev->priv.rl_table;
  	struct mlx5_rl_entry *entry = NULL;
-	struct mlx5_rate_limit reset_rl = {0};
 
  	/* 0 is a reserved value for unlimited rate */
  	if (rl->rate == 0)
  		return;
 
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, rate_limit, rl-
Quoted 1 line

rate);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +3 -0
+	MLX5_SET(set_pp_rate_limit_context, rl_raw, burst_upper_bound,
+		 rl->max_burst_sz);
+	MLX5_SET(set_pp_rate_limit_context, rl_raw,

typical_packet_size,

drivers/net/ethernet/mellanox/mlx5/core/rl.c +5 -1
+		 rl->typical_pkt_sz);
+
  	mutex_lock(&table->rl_lock);
-	entry = find_rl_entry(table, rl);
+	entry = find_rl_entry(table, rl_raw,
+			      MLX5_CAP_QOS(dev, packet_pacing_uid) ?
+				MLX5_SHARED_RESOURCE_UID : 0, false);
  	if (!entry || !entry->refcount) {
  		mlx5_core_warn(dev, "Rate %u, max_burst_sz %u

typical_pkt_sz %u are not configured\n",
rl->rate, rl->max_burst_sz, rl-

Quoted 1 line

typical_pkt_sz);

drivers/net/ethernet/mellanox/mlx5/core/rl.c +2 -4
@@ -223,11 +292,9 @@ void mlx5_rl_remove_rate(struct mlx5_core_dev
 *dev, struct mlx5_rate_limit *rl)
  	}
 
  	entry->refcount--;
-	if (!entry->refcount) {
+	if (!entry->refcount)
  		/* need to remove rate */
-		mlx5_set_pp_rate_limit_cmd(dev, entry->index,
 &reset_rl);
-		entry->rl = reset_rl;
-	}
+		mlx5_set_pp_rate_limit_cmd(dev, entry, false);
 
  out:
  	mutex_unlock(&table->rl_lock);
drivers/net/ethernet/mellanox/mlx5/core/rl.c +3 -4
@@ -273,14 +340,13 @@ int mlx5_init_rl_table(struct mlx5_core_dev
 *dev)
  void mlx5_cleanup_rl_table(struct mlx5_core_dev *dev)
  {
  	struct mlx5_rl_table *table = &dev->priv.rl_table;
-	struct mlx5_rate_limit rl = {0};
  	int i;
 
  	/* Clear all configured rates */
  	for (i = 0; i < table->max_size; i++)
-		if (table->rl_entry[i].rl.rate)
-			mlx5_set_pp_rate_limit_cmd(dev, table-
 > rl_entry[i].index,
-						   &rl);
+		if (table->rl_entry[i].refcount)
+			mlx5_set_pp_rate_limit_cmd(dev, &table-
 > rl_entry[i],
+						   false);
  
  	kfree(dev->priv.rl_table.rl_entry);
  }
include/linux/mlx5/driver.h +8 -3
index 277a51d3ec40..f2b4225ed650 100644
@@ -518,9 +518,11 @@ struct mlx5_rate_limit {
  };
 
  struct mlx5_rl_entry {
-	struct mlx5_rate_limit	rl;
-	u16                     index;
-	u16                     refcount;
+	u8 rl_raw[MLX5_ST_SZ_BYTES(set_pp_rate_limit_context)];
+	u16 index;
+	u64 refcount;
+	u16 uid;
+	u8 dedicated : 1;
  };
 
  struct mlx5_rl_table {
@@ -1007,6 +1009,9 @@ int mlx5_rl_add_rate(struct mlx5_core_dev *dev,
 u16 *index,
  		     struct mlx5_rate_limit *rl);
  void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, struct
 mlx5_rate_limit *rl);
  bool mlx5_rl_is_in_range(struct mlx5_core_dev *dev, u32 rate);
+int mlx5_rl_add_rate_raw(struct mlx5_core_dev *dev, void *rl_in, u16
 uid,
+			 bool dedicated_entry, u16 *index);
+void mlx5_rl_remove_rate_raw(struct mlx5_core_dev *dev, u16 index);

bool mlx5_rl_are_equal(struct mlx5_rate_limit *rl_0,
struct mlx5_rate_limit *rl_1);
int mlx5_alloc_bfreg(struct mlx5_core_dev *mdev, struct
mlx5_sq_bfreg *bfreg,

include/linux/mlx5/mlx5_ifc.h +16 -10
index ff8c9d527bb4..7d89ab64b372 100644
@@ -810,7 +810,9 @@ struct mlx5_ifc_qos_cap_bits {
  	u8         reserved_at_4[0x1];
  	u8         packet_pacing_burst_bound[0x1];
  	u8         packet_pacing_typical_size[0x1];
-	u8         reserved_at_7[0x19];
+	u8         reserved_at_7[0x4];
+	u8         packet_pacing_uid[0x1];
+	u8         reserved_at_c[0x14];
 
  	u8         reserved_at_20[0x20];
 
@@ -8262,9 +8264,20 @@ struct mlx5_ifc_set_pp_rate_limit_out_bits {
  	u8         reserved_at_40[0x40];
  };
 
+struct mlx5_ifc_set_pp_rate_limit_context_bits {
+	u8         rate_limit[0x20];
+
+	u8	   burst_upper_bound[0x20];
+
+	u8         reserved_at_40[0x10];
+	u8	   typical_packet_size[0x10];
+
+	u8         reserved_at_60[0x120];
+};
+
  struct mlx5_ifc_set_pp_rate_limit_in_bits {
  	u8         opcode[0x10];
-	u8         reserved_at_10[0x10];
+	u8         uid[0x10];
 
  	u8         reserved_at_20[0x10];
  	u8         op_mod[0x10];
@@ -8274,14 +8287,7 @@ struct mlx5_ifc_set_pp_rate_limit_in_bits {
 
  	u8         reserved_at_60[0x20];
 
-	u8         rate_limit[0x20];
-
-	u8	   burst_upper_bound[0x20];
-
-	u8         reserved_at_c0[0x10];
-	u8	   typical_packet_size[0x10];
-
-	u8         reserved_at_e0[0x120];
+	struct mlx5_ifc_set_pp_rate_limit_context_bits ctx;
  };
 
  struct mlx5_ifc_access_register_out_bits {
#4 lore raw

Re: [PATCH mlx5-next 1/2] net/mlx5: Expose raw packet pacing APIs

Quoted 38 lines · Yishai Hadas
Quoted 29 lines · Saeed Mahameed
Quoted 20 lines · Leon Romanovsky

From: Yishai Hadas <[email protected]>

Expose raw packet pacing APIs to be used by DEVX based applications.
The existing code was refactored to have a single flow with the new
raw
APIs.

The new raw APIs considered the input of 'pp_rate_limit_context', uid,
'dedicated', upon looking for an existing entry.

This raw mode enables future device specification data in the raw context without changing the existing logic and code.

The ability to ask for a dedicated entry gives control for application
to allocate entries according to its needs.

Quoted 7 lines

A dedicated entry may not be used by some other process and it also
enables the process spreading its resources to some different entries
for use different hardware resources as part of enforcing the rate.

Quoted 7 lines

It sounds like the dedicated means "no sharing" which means you don't
need to use the mlx5_core API and you can go directly to FW.. The problem is that the entry indices are managed by driver, and i guess
this is the reason why you had to expand the mlx5_core API..

Quoted 11 lines

The main reason for introducing the new mlx5_core APIs was the need to
support the "shared mode" in a "raw data" format to prevent future touching the kernel once PRM will support extra fields. As the RL indices are managed by the driver (mlx5_core) including the
sharing, we couldn’t go directly to FW, the legacy API was refactored
inside the core to have one flow with the new raw APIs. So we may need those APIs regardless the dedicated mode.

I not a fan of legacy APIs, all of the APIs are mlx5 internals and i would like to keep one API which is only PRM dependent as much as possible.

Anyway thanks for the clarification, i think the patch is good as is, we can improve and remove the legacy API in the future and keep the raw API.

Quoted 42 lines
Quoted 12 lines

I would like to suggest some alternatives to simplify the approach and
allow using RAW PRM for DEVX properly.

1. preserve RL entries for DEVX and let DEVX access FW directly with
PRM commands.
2. keep mlx5_core API simple and instead of adding this raw/non raw api
and complicating the RL API with this dedicated bit:

Quoted 7 lines

just add mlx5_rl_{alloc/free}_index(), this will dedicate for you the
RL index form the end of the RL indices database and you are free to
access the FW with this index the way you like via direct PRM commands.

As mentioned above, we may still need the new mlx5_core raw APIs for the
shared mode which is the main usage of the API, we found it reasonable
to have the dedicate flag in the new raw alloc API instead of exposing
more two new APIs only for that.

Please note that even if we'll go with those 2 extra APIs for the dedicated mode, we may still need to maintain in the core this information to prevent returning this entry for other cases.

Also the idea to preserve some entries at the end might be wasteful as
there is no guarantee that DEVX will really be used, and even so it may
not ask for entries in a dedicated mode.

Presering them for this optional use case might prevent using them for
all other cases.

Quoted 6 lines
Quoted 3 lines
Quoted 2 lines

The counter per entry mas changed to be u64 to prevent any option to

typo ^^^ was

Sure, thanks.

Leon, Other than the typo i am good with this patch.
you can fix up the patch prior to pulling into mlx5-next, no need for v2.

thanks,
Saeed.

#5 lore raw

Re: [PATCH mlx5-next 1/2] net/mlx5: Expose raw packet pacing APIs

Quoted 103 lines · Saeed Mahameed
Quoted 38 lines · Yishai Hadas
Quoted 29 lines · Saeed Mahameed
Quoted 20 lines · Leon Romanovsky

From: Yishai Hadas <[email protected]>

Expose raw packet pacing APIs to be used by DEVX based applications.
The existing code was refactored to have a single flow with the new
raw
APIs.

The new raw APIs considered the input of 'pp_rate_limit_context', uid,
'dedicated', upon looking for an existing entry.

This raw mode enables future device specification data in the raw context without changing the existing logic and code.

The ability to ask for a dedicated entry gives control for application
to allocate entries according to its needs.

Quoted 7 lines

A dedicated entry may not be used by some other process and it also
enables the process spreading its resources to some different entries
for use different hardware resources as part of enforcing the rate.

Quoted 7 lines

It sounds like the dedicated means "no sharing" which means you don't
need to use the mlx5_core API and you can go directly to FW.. The problem is that the entry indices are managed by driver, and i guess
this is the reason why you had to expand the mlx5_core API..

Quoted 11 lines

The main reason for introducing the new mlx5_core APIs was the need to
support the "shared mode" in a "raw data" format to prevent future touching the kernel once PRM will support extra fields. As the RL indices are managed by the driver (mlx5_core) including the
sharing, we couldn’t go directly to FW, the legacy API was refactored
inside the core to have one flow with the new raw APIs. So we may need those APIs regardless the dedicated mode.

I not a fan of legacy APIs, all of the APIs are mlx5 internals and i would like to keep one API which is only PRM dependent as much as possible.

Anyway thanks for the clarification, i think the patch is good as is, we can improve and remove the legacy API in the future and keep the raw API.

Quoted 44 lines
Quoted 12 lines

I would like to suggest some alternatives to simplify the approach and
allow using RAW PRM for DEVX properly.

1. preserve RL entries for DEVX and let DEVX access FW directly with
PRM commands.
2. keep mlx5_core API simple and instead of adding this raw/non raw api
and complicating the RL API with this dedicated bit:

Quoted 7 lines

just add mlx5_rl_{alloc/free}_index(), this will dedicate for you the
RL index form the end of the RL indices database and you are free to
access the FW with this index the way you like via direct PRM commands.

As mentioned above, we may still need the new mlx5_core raw APIs for the
shared mode which is the main usage of the API, we found it reasonable
to have the dedicate flag in the new raw alloc API instead of exposing
more two new APIs only for that.

Please note that even if we'll go with those 2 extra APIs for the dedicated mode, we may still need to maintain in the core this information to prevent returning this entry for other cases.

Also the idea to preserve some entries at the end might be wasteful as
there is no guarantee that DEVX will really be used, and even so it may
not ask for entries in a dedicated mode.

Presering them for this optional use case might prevent using them for
all other cases.

Quoted 2 lines
Quoted 2 lines

The counter per entry mas changed to be u64 to prevent any option to

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

typo ^^^ was

Sure, thanks.

Leon, Other than the typo i am good with this patch.
you can fix up the patch prior to pulling into mlx5-next, no need for v2.

Thanks Saeed, I'll apply it once Doug/Jason ack RDMA part of the series.

Quoted 7 lines

thanks,
Saeed.

#6 lore raw

PATCH 2/2 IB/mlx5: Introduce UAPIs to manage packet pacing

In reply to Leon Romanovsky

From: Yishai Hadas <[email protected]>

Introduce packet pacing uobject and its alloc and destroy methods.

This uobject holds mlx5 packet pacing context according to the device specification and enables managing packet pacing device entries that are needed by DEVX applications.

---
 drivers/infiniband/hw/mlx5/Makefile       |   1 +
 drivers/infiniband/hw/mlx5/main.c         |   1 +
 drivers/infiniband/hw/mlx5/mlx5_ib.h      |   6 +
 drivers/infiniband/hw/mlx5/qos.c          | 136 ++++++++++++++++++++++
 include/uapi/rdma/mlx5_user_ioctl_cmds.h  |  17 +++
 include/uapi/rdma/mlx5_user_ioctl_verbs.h |   4 +
 6 files changed, 165 insertions(+)
 create mode 100644 drivers/infiniband/hw/mlx5/qos.c
drivers/infiniband/hw/mlx5/Makefile +1 -0
index d0a043ccbe58..2a334800f109 100644
@@ -8,3 +8,4 @@ mlx5_ib-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += odp.o
 mlx5_ib-$(CONFIG_MLX5_ESWITCH) += ib_rep.o
 mlx5_ib-$(CONFIG_INFINIBAND_USER_ACCESS) += devx.o
 mlx5_ib-$(CONFIG_INFINIBAND_USER_ACCESS) += flow.o
+mlx5_ib-$(CONFIG_INFINIBAND_USER_ACCESS) += qos.o
drivers/infiniband/hw/mlx5/main.c +1 -0
index 398e9f32746e..01bddd584405 100644
@@ -6250,6 +6250,7 @@ ADD_UVERBS_ATTRIBUTES_SIMPLE(
 static const struct uapi_definition mlx5_ib_defs[] = {
 	UAPI_DEF_CHAIN(mlx5_ib_devx_defs),
 	UAPI_DEF_CHAIN(mlx5_ib_flow_defs),
+	UAPI_DEF_CHAIN(mlx5_ib_qos_defs),
 
 	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_FLOW_ACTION,
 				&mlx5_ib_flow_action),
drivers/infiniband/hw/mlx5/mlx5_ib.h +6 -0
index 2d65df34a6d0..6c85ca766678 100644
@@ -201,6 +201,11 @@ struct mlx5_ib_flow_matcher {
 	u8			match_criteria_enable;
 };
 
+struct mlx5_ib_pp {
+	u16 index;
+	struct mlx5_core_dev *mdev;
+};
+
 struct mlx5_ib_flow_db {
 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
 	struct mlx5_ib_flow_prio	egress_prios[MLX5_IB_NUM_FLOW_FT];
@@ -1379,6 +1384,7 @@ int mlx5_ib_fill_stat_entry(struct sk_buff *msg,
 
 extern const struct uapi_definition mlx5_ib_devx_defs[];
 extern const struct uapi_definition mlx5_ib_flow_defs[];
+extern const struct uapi_definition mlx5_ib_qos_defs[];
 
 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
 int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user);
drivers/infiniband/hw/mlx5/qos.c new file +136 -0

136 lines not on this page · raw message

include/uapi/rdma/mlx5_user_ioctl_cmds.h +17 -0
index afe7da6f2b8e..8f4a417fc70a 100644
@@ -143,6 +143,22 @@ enum mlx5_ib_devx_umem_dereg_attrs {
 	MLX5_IB_ATTR_DEVX_UMEM_DEREG_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
 };
 
+enum mlx5_ib_pp_obj_methods {
+	MLX5_IB_METHOD_PP_OBJ_ALLOC = (1U << UVERBS_ID_NS_SHIFT),
+	MLX5_IB_METHOD_PP_OBJ_DESTROY,
+};
+
+enum mlx5_ib_pp_alloc_attrs {
+	MLX5_IB_ATTR_PP_OBJ_ALLOC_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	MLX5_IB_ATTR_PP_OBJ_ALLOC_CTX,
+	MLX5_IB_ATTR_PP_OBJ_ALLOC_FLAGS,
+	MLX5_IB_ATTR_PP_OBJ_ALLOC_INDEX,
+};
+
+enum mlx5_ib_pp_obj_destroy_attrs {
+	MLX5_IB_ATTR_PP_OBJ_DESTROY_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+};
+
 enum mlx5_ib_devx_umem_methods {
 	MLX5_IB_METHOD_DEVX_UMEM_REG = (1U << UVERBS_ID_NS_SHIFT),
 	MLX5_IB_METHOD_DEVX_UMEM_DEREG,
@@ -173,6 +189,7 @@ enum mlx5_ib_objects {
 	MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD,
 	MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
 	MLX5_IB_OBJECT_VAR,
+	MLX5_IB_OBJECT_PP,
 };
 
 enum mlx5_ib_flow_matcher_create_attrs {
include/uapi/rdma/mlx5_user_ioctl_verbs.h +4 -0
index 88b6ca70c2fe..b4641a7865f7 100644
@@ -73,5 +73,9 @@ struct mlx5_ib_uapi_devx_async_event_hdr {
 	__u8		out_data[];
 };
 
+enum mlx5_ib_uapi_pp_alloc_flags {
+	MLX5_IB_UAPI_PP_ALLOC_FLAGS_DEDICATED_INDEX = 1 << 0,
+};
+
 #endif
 
Signature · 1 line
2.24.1
#7 lore raw
In reply to Leon Romanovsky
Quoted 8 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Hi,

This series from Yishai extends packet pacing to work over DEVX interface. In first patch, he refactors the mlx5_core internal logic. In second patch, the RDMA APIs are added.

Quoted 5 lines

Thanks

Yishai Hadas (2):
net/mlx5: Expose raw packet pacing APIs IB/mlx5: Introduce UAPIs to manage packet pacing

It looks Ok, can you apply this to the shared branch?

Thanks,
Jason

author Applied
#8 lore raw
Quoted 9 lines · Jason Gunthorpe
Quoted 8 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Hi,

This series from Yishai extends packet pacing to work over DEVX interface. In first patch, he refactors the mlx5_core internal logic. In second patch, the RDMA APIs are added.

Quoted 7 lines
Quoted 5 lines

Thanks

Yishai Hadas (2):
net/mlx5: Expose raw packet pacing APIs IB/mlx5: Introduce UAPIs to manage packet pacing

It looks Ok, can you apply this to the shared branch?

Applied to mlx5-next
1326034b3ce7 net/mlx5: Expose raw packet pacing APIs

Quoted 3 lines

Thanks,
Jason

#9 lore raw
In reply to Leon Romanovsky
Quoted 8 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Hi,

This series from Yishai extends packet pacing to work over DEVX interface. In first patch, he refactors the mlx5_core internal logic. In second patch, the RDMA APIs are added.

Quoted 5 lines

Thanks

Yishai Hadas (2):
net/mlx5: Expose raw packet pacing APIs IB/mlx5: Introduce UAPIs to manage packet pacing

Applied to for-next, thanks

Jason