Skip to contentSkip to conversation

Statistics counter support

PATCH In review Leon Romanovsky 2 participants 24 messages Jul 2, 2019 – Jul 5, 2019

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

v5 cover Statistics counter support

From: Leon Romanovsky <[email protected]>

Changelog:
v4 -> v5:
* Patch #6 and #14 - consolidated many counter release functions, removed mutex lock protection from dealloc_counter() call and simplified kref_put/kref_get operations.
* Added Saeed's ACK tags.
v3 -> v4:
* Add counter_dealloc() callback function
* Moved to kref implementation
* Fixed lock during spinlock
v2 -> v3:
* We didn't change use of atomics over kref for management of unbind counter from QP. The reason to it that bind and unbind are non-symmetric in regards of put and get, so we need to count differently memory release flows of HW objects (restrack) and SW bind operations.
* Everything else was addressed.
v1 -> v2:
* Rebased to latest rdma-next
v0 -> v1:
* Changed wording of counter comment
* Removed unneeded assignments
* Added extra patch to present global counters

----------------------------------------------------

Hi,

This series from Mark provides dynamic statistics infrastructure. He uses netlink interface to configure and retrieve those counters.

This infrastructure allows to users monitor various objects by binding to them counters. As the beginning, we used QP object as target for those counters, but future patches will include ODP MR information too.

Two binding modes are supported:
- Auto: This allows a user to build automatic set of objects to a counter according to common criteria. For example in a per-type scheme, where in one process all QPs with same QP type are bound automatically to a single counter.
- Manual: This allows a user to manually bind objects on a counter.

Those two modes are mutual-exclusive with separation between processes, objects created by different processes cannot be bound to a same counter.

For objects which don't support counter binding, we will return pre-allocated counters.

$ rdma statistic qp set link mlx5_2/1 auto type on
$ rdma statistic qp set link mlx5_2/1 auto off
$ rdma statistic qp bind link mlx5_2/1 lqpn 178
$ rdma statistic qp unbind link mlx5_2/1 cntn 4 lqpn 178
$ rdma statistic show
$ rdma statistic qp mode

Thanks

Mark Zhang (17):
net/mlx5: Add rts2rts_qp_counters_set_id field in hca cap RDMA/restrack: Introduce statistic counter RDMA/restrack: Add an API to attach a task to a resource RDMA/restrack: Make is_visible_in_pid_ns() as an API RDMA/counter: Add set/clear per-port auto mode support RDMA/counter: Add "auto" configuration mode support IB/mlx5: Support set qp counter
IB/mlx5: Add counter set id as a parameter for

    mlx5_ib_query_q_counters()

IB/mlx5: Support statistic q counter configuration RDMA/nldev: Allow counter auto mode configration through RDMA netlink RDMA/netlink: Implement counter dumpit calback
IB/mlx5: Add counter_alloc_stats() and counter_update_stats() support RDMA/core: Get sum value of all counters when perform a sysfs stat
read
RDMA/counter: Allow manual mode configuration support RDMA/nldev: Allow counter manual mode configration through RDMA
netlink
RDMA/nldev: Allow get counter mode through RDMA netlink RDMA/nldev: Allow get default counter statistics through RDMA netlink

 drivers/infiniband/core/Makefile     |   2 +-
 drivers/infiniband/core/counters.c   | 634 +++++++++++++++++++++++++++
 drivers/infiniband/core/device.c     |  10 +
 drivers/infiniband/core/nldev.c      | 551 ++++++++++++++++++++++-
 drivers/infiniband/core/restrack.c   |  49 ++-
 drivers/infiniband/core/restrack.h   |   3 +
 drivers/infiniband/core/sysfs.c      |  16 +-
 drivers/infiniband/core/verbs.c      |   9 +
 drivers/infiniband/hw/mlx5/main.c    |  77 +++-
 drivers/infiniband/hw/mlx5/mlx5_ib.h |   6 +
 drivers/infiniband/hw/mlx5/qp.c      |  76 +++-
 include/linux/mlx5/mlx5_ifc.h        |   4 +-
 include/linux/mlx5/qp.h              |   1 +
 include/rdma/ib_verbs.h              |  31 ++
 include/rdma/rdma_counter.h          |  65 +++
 include/rdma/restrack.h              |   4 +
 include/uapi/rdma/rdma_netlink.h     |  52 ++-
 17 files changed, 1559 insertions(+), 31 deletions(-)
 create mode 100644 drivers/infiniband/core/counters.c
 create mode 100644 include/rdma/rdma_counter.h
#1 lore raw

PATCH v5 1/17 net/mlx5: Add rts2rts_qp_counters_set_id field in hca cap

From: Mark Zhang <[email protected]>

Add rts2rts_qp_counters_set_id field in hca cap so that RTS2RTS qp modification can be used to change the counter of a QP.

---
 include/linux/mlx5/mlx5_ifc.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
include/linux/mlx5/mlx5_ifc.h +3 -1
index e3c154b573a2..16348528fef6 100644
@@ -1028,7 +1028,9 @@ struct mlx5_ifc_cmd_hca_cap_bits {
 	u8         cc_modify_allowed[0x1];
 	u8         start_pad[0x1];
 	u8         cache_line_128byte[0x1];
-	u8         reserved_at_165[0xa];
+	u8         reserved_at_165[0x4];
+	u8         rts2rts_qp_counters_set_id[0x1];
+	u8         reserved_at_16a[0x5];
 	u8         qcam_reg[0x1];
 	u8         gid_table_size[0x10];
 
Signature · 1 line
2.20.1
#2 lore raw

PATCH v5 2/17 RDMA/restrack: Introduce statistic counter

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Introduce statistic counter as a new resource. It allows a user to monitor specific objects (e.g., QPs) by binding to a counter.

In some cases a user counter resource is created with task other then "current", because its creation is done as part of rdmatool call.

---
 drivers/infiniband/core/restrack.c | 22 +++++++++++++++++-----
 include/rdma/rdma_counter.h        | 18 ++++++++++++++++++
 include/rdma/restrack.h            |  4 ++++
 3 files changed, 39 insertions(+), 5 deletions(-)
 create mode 100644 include/rdma/rdma_counter.h
drivers/infiniband/core/restrack.c +17 -5
index 3b5ff2f7b5f8..95573f292aae 100644
@@ -6,6 +6,7 @@
 #include <rdma/rdma_cm.h>
 #include <rdma/ib_verbs.h>
 #include <rdma/restrack.h>
+#include <rdma/rdma_counter.h>
 #include <linux/mutex.h>
 #include <linux/sched/task.h>
 #include <linux/pid_namespace.h>
@@ -45,6 +46,7 @@ static const char *type2str(enum rdma_restrack_type type)
 		[RDMA_RESTRACK_CM_ID] = "CM_ID",
 		[RDMA_RESTRACK_MR] = "MR",
 		[RDMA_RESTRACK_CTX] = "CTX",
+		[RDMA_RESTRACK_COUNTER] = "COUNTER",
 	};
 
 	return names[type];
@@ -169,6 +171,8 @@ static struct ib_device *res_to_dev(struct rdma_restrack_entry *res)
 		return container_of(res, struct ib_mr, res)->device;
 	case RDMA_RESTRACK_CTX:
 		return container_of(res, struct ib_ucontext, res)->device;
+	case RDMA_RESTRACK_COUNTER:
+		return container_of(res, struct rdma_counter, res)->device;
 	default:
 		WARN_ONCE(true, "Wrong resource tracking type %u\n", res->type);
 		return NULL;
@@ -203,15 +207,22 @@ static void rdma_restrack_add(struct rdma_restrack_entry *res)
 
 	kref_init(&res->kref);
 	init_completion(&res->comp);
-	if (res->type != RDMA_RESTRACK_QP)
-		ret = xa_alloc_cyclic(&rt->xa, &res->id, res, xa_limit_32b,
-				&rt->next_id, GFP_KERNEL);
-	else {
+	if (res->type == RDMA_RESTRACK_QP) {
 		/* Special case to ensure that LQPN points to right QP */
 		struct ib_qp *qp = container_of(res, struct ib_qp, res);
 
 		ret = xa_insert(&rt->xa, qp->qp_num, res, GFP_KERNEL);
 		res->id = ret ? 0 : qp->qp_num;
+	} else if (res->type == RDMA_RESTRACK_COUNTER) {
+		/* Special case to ensure that cntn points to right counter */
+		struct rdma_counter *counter;
+
+		counter = container_of(res, struct rdma_counter, res);
+		ret = xa_insert(&rt->xa, counter->id, res, GFP_KERNEL);
+		res->id = ret ? 0 : counter->id;
+	} else {
+		ret = xa_alloc_cyclic(&rt->xa, &res->id, res, xa_limit_32b,
+				      &rt->next_id, GFP_KERNEL);
 	}
 
 	if (!ret)
@@ -237,7 +248,8 @@ EXPORT_SYMBOL(rdma_restrack_kadd);
  */
 void rdma_restrack_uadd(struct rdma_restrack_entry *res)
 {
-	if (res->type != RDMA_RESTRACK_CM_ID)
+	if ((res->type != RDMA_RESTRACK_CM_ID) &&
+	    (res->type != RDMA_RESTRACK_COUNTER))
 		res->task = NULL;
 
 	if (!res->task)
include/rdma/rdma_counter.h new file +18 -0
new file mode 100644
index 000000000000..283ac1a0cdb7
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/*
+ * Copyright (c) 2019 Mellanox Technologies. All rights reserved.
+ */
+
+#ifndef _RDMA_COUNTER_H_
+#define _RDMA_COUNTER_H_
+
+#include <rdma/ib_verbs.h>
+#include <rdma/restrack.h>
+
+struct rdma_counter {
+	struct rdma_restrack_entry	res;
+	struct ib_device		*device;
+	uint32_t			id;
+	u8				port;
+};
+#endif /* _RDMA_COUNTER_H_ */
include/rdma/restrack.h +4 -0
index ecf3c7702a4f..4041a4d96524 100644
@@ -42,6 +42,10 @@ enum rdma_restrack_type {
 	 * @RDMA_RESTRACK_CTX: Verbs contexts (CTX)
 	 */
 	RDMA_RESTRACK_CTX,
+	/**
+	 * @RDMA_RESTRACK_COUNTER: Statistic Counter
+	 */
+	RDMA_RESTRACK_COUNTER,
 	/**
 	 * @RDMA_RESTRACK_MAX: Last entry, used for array dclarations
 	 */
Signature · 1 line
2.20.1
#3 lore raw

PATCH v5 3/17 RDMA/restrack: Add an API to attach a task to a resource

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Add rdma_restrack_attach_task() which is able to attach a task other then "current" to a resource.

---
 drivers/infiniband/core/restrack.c | 14 ++++++++++++++
 drivers/infiniband/core/restrack.h |  2 ++
 2 files changed, 16 insertions(+)
drivers/infiniband/core/restrack.c +14 -0
index 95573f292aae..3714634ae296 100644
@@ -194,6 +194,20 @@ void rdma_restrack_set_task(struct rdma_restrack_entry *res,
 }
 EXPORT_SYMBOL(rdma_restrack_set_task);
 
+/**
+ * rdma_restrack_attach_task() - attach the task onto this resource
+ * @res:  resource entry
+ * @task: the task to attach, the current task will be used if it is NULL.
+ */
+void rdma_restrack_attach_task(struct rdma_restrack_entry *res,
+			       struct task_struct *task)
+{
+	if (res->task)
+		put_task_struct(res->task);
+	get_task_struct(task);
+	res->task = task;
+}
+
 static void rdma_restrack_add(struct rdma_restrack_entry *res)
 {
 	struct ib_device *dev = res_to_dev(res);
drivers/infiniband/core/restrack.h +2 -0
index 09a1fbdf578e..d084e5f89849 100644
@@ -25,4 +25,6 @@ struct rdma_restrack_root {
 
 int rdma_restrack_init(struct ib_device *dev);
 void rdma_restrack_clean(struct ib_device *dev);
+void rdma_restrack_attach_task(struct rdma_restrack_entry *res,
+			       struct task_struct *task);
 #endif /* _RDMA_CORE_RESTRACK_H_ */
Signature · 1 line
2.20.1
#4 lore raw

PATCH v5 4/17 RDMA/restrack: Make is_visible_in_pid_ns() as an API

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Remove is_visible_in_pid_ns() from nldev.c and make it as a restrack API, so that it can be taken advantage by other parts like counter.

---
 drivers/infiniband/core/nldev.c    | 15 ++-------------
 drivers/infiniband/core/restrack.c | 13 +++++++++++++
 drivers/infiniband/core/restrack.h |  1 +
 3 files changed, 16 insertions(+), 13 deletions(-)
drivers/infiniband/core/nldev.c +2 -13
index 5499f5629dc2..d9ebfb50962b 100644
@@ -1002,17 +1002,6 @@ static const struct nldev_fill_res_entry fill_entries[RDMA_RESTRACK_MAX] = {
 	},
 };
 
-static bool is_visible_in_pid_ns(struct rdma_restrack_entry *res)
-{
-	/*
-	 * 1. Kern resources should be visible in init name space only
-	 * 2. Present only resources visible in the current namespace
-	 */
-	if (rdma_is_kernel_res(res))
-		return task_active_pid_ns(current) == &init_pid_ns;
-	return task_active_pid_ns(current) == task_active_pid_ns(res->task);
-}
-
 static int res_get_common_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 			       struct netlink_ext_ack *extack,
 			       enum rdma_restrack_type res_type)
@@ -1057,7 +1046,7 @@ static int res_get_common_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 		goto err;
 	}
 
-	if (!is_visible_in_pid_ns(res)) {
+	if (!rdma_is_visible_in_pid_ns(res)) {
 		ret = -ENOENT;
 		goto err_get;
 	}
@@ -1169,7 +1158,7 @@ static int res_get_common_dumpit(struct sk_buff *skb,
 	 * objects.
 	 */
 	xa_for_each(&rt->xa, id, res) {
-		if (!is_visible_in_pid_ns(res))
+		if (!rdma_is_visible_in_pid_ns(res))
 			continue;
 
 		if (idx < start || !rdma_restrack_get(res))
drivers/infiniband/core/restrack.c +13 -0
index 3714634ae296..bddff426ee0f 100644
@@ -349,3 +349,16 @@ void rdma_restrack_del(struct rdma_restrack_entry *res)
 	}
 }
 EXPORT_SYMBOL(rdma_restrack_del);
+
+bool rdma_is_visible_in_pid_ns(struct rdma_restrack_entry *res)
+{
+	/*
+	 * 1. Kern resources should be visible in init
+	 *    namespace only
+	 * 2. Present only resources visible in the current
+	 *     namespace
+	 */
+	if (rdma_is_kernel_res(res))
+		return task_active_pid_ns(current) == &init_pid_ns;
+	return task_active_pid_ns(current) == task_active_pid_ns(res->task);
+}
drivers/infiniband/core/restrack.h +1 -0
index d084e5f89849..7bd177cc0a61 100644
@@ -27,4 +27,5 @@ int rdma_restrack_init(struct ib_device *dev);
 void rdma_restrack_clean(struct ib_device *dev);
 void rdma_restrack_attach_task(struct rdma_restrack_entry *res,
 			       struct task_struct *task);
+bool rdma_is_visible_in_pid_ns(struct rdma_restrack_entry *res);
 #endif /* _RDMA_CORE_RESTRACK_H_ */
Signature · 1 line
2.20.1
#5 lore raw

PATCH v5 5/17 RDMA/counter: Add set/clear per-port auto mode support

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Add an API to support set/clear per-port auto mode.

---
 drivers/infiniband/core/Makefile   |  2 +-
 drivers/infiniband/core/counters.c | 74 ++++++++++++++++++++++++++++++
 drivers/infiniband/core/device.c   |  5 ++
 include/rdma/ib_verbs.h            |  2 +
 include/rdma/rdma_counter.h        | 24 ++++++++++
 include/uapi/rdma/rdma_netlink.h   | 26 +++++++++++
 6 files changed, 132 insertions(+), 1 deletion(-)
 create mode 100644 drivers/infiniband/core/counters.c
drivers/infiniband/core/Makefile +1 -1
index 42f1b2a4f746..09881bd5f12d 100644
@@ -11,7 +11,7 @@ ib_core-y :=			packer.o ud_header.o verbs.o cq.o rw.o sysfs.o \
 				device.o fmr_pool.o cache.o netlink.o \
 				roce_gid_mgmt.o mr_pool.o addr.o sa_query.o \
 				multicast.o mad.o smi.o agent.o mad_rmpp.o \
-				nldev.o restrack.o
+				nldev.o restrack.o counters.o
 
 ib_core-$(CONFIG_SECURITY_INFINIBAND) += security.o
 ib_core-$(CONFIG_CGROUP_RDMA) += cgroup.o
drivers/infiniband/core/counters.c new file +74 -0

74 lines not on this page · raw message

drivers/infiniband/core/device.c +5 -0
index 8a6ccb936dfe..6579865e4866 100644
@@ -46,6 +46,7 @@
 #include <rdma/rdma_netlink.h>
 #include <rdma/ib_addr.h>
 #include <rdma/ib_cache.h>
+#include <rdma/rdma_counter.h>
 
 #include "core_priv.h"
 #include "restrack.h"
@@ -492,10 +493,12 @@ static void ib_device_release(struct device *device)
 	if (dev->port_data) {
 		ib_cache_release_one(dev);
 		ib_security_release_port_pkey_list(dev);
+		rdma_counter_release(dev);
 		kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu,
 				       pdata[0]),
 			  rcu_head);
 	}
+
 	xa_destroy(&dev->compat_devs);
 	xa_destroy(&dev->client_data);
 	kfree_rcu(dev, rcu_head);
@@ -1316,6 +1319,8 @@ int ib_register_device(struct ib_device *device, const char *name)
 
 	ib_device_register_rdmacg(device);
 
+	rdma_counter_init(device);
+
 	/*
 	 * Ensure that ADD uevent is not fired because it
 	 * is too early amd device is not initialized yet.
include/rdma/ib_verbs.h +2 -0
index 26e9c2594913..3d19c056fbc0 100644
@@ -62,6 +62,7 @@
 #include <linux/irqflags.h>
 #include <linux/preempt.h>
 #include <uapi/rdma/ib_user_verbs.h>
+#include <rdma/rdma_counter.h>
 #include <rdma/restrack.h>
 #include <rdma/signature.h>
 #include <uapi/rdma/rdma_user_ioctl.h>
@@ -2119,6 +2120,7 @@ struct ib_port_data {
 	spinlock_t netdev_lock;
 	struct net_device __rcu *netdev;
 	struct hlist_node ndev_hash_link;
+	struct rdma_port_counter port_counter;
 };
 
 /* rdma netdev type - specifies protocol type */
include/rdma/rdma_counter.h +24 -0
index 283ac1a0cdb7..8dd2619c015d 100644
@@ -6,8 +6,26 @@
 #ifndef _RDMA_COUNTER_H_
 #define _RDMA_COUNTER_H_
 
+#include <linux/mutex.h>
+
 #include <rdma/ib_verbs.h>
 #include <rdma/restrack.h>
+#include <rdma/rdma_netlink.h>
+
+struct auto_mode_param {
+	int qp_type;
+};
+
+struct rdma_counter_mode {
+	enum rdma_nl_counter_mode mode;
+	enum rdma_nl_counter_mask mask;
+	struct auto_mode_param param;
+};
+
+struct rdma_port_counter {
+	struct rdma_counter_mode mode;
+	struct mutex lock;
+};
 
 struct rdma_counter {
 	struct rdma_restrack_entry	res;
@@ -15,4 +33,10 @@ struct rdma_counter {
 	uint32_t			id;
 	u8				port;
 };
+
+void rdma_counter_init(struct ib_device *dev);
+void rdma_counter_release(struct ib_device *dev);
+int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
+			       bool on, enum rdma_nl_counter_mask mask);
+
 #endif /* _RDMA_COUNTER_H_ */
include/uapi/rdma/rdma_netlink.h +26 -0
index 650cee8c4bf1..e3cd912e9cef 100644
@@ -507,4 +507,30 @@ enum rdma_nldev_attr {
 	 */
 	RDMA_NLDEV_ATTR_MAX
 };
+
+/*
+ * Supported counter bind modes. All modes are mutual-exclusive.
+ */
+enum rdma_nl_counter_mode {
+	RDMA_COUNTER_MODE_NONE,
+
+	/*
+	 * A qp is bound with a counter automatically during initialization
+	 * based on the auto mode (e.g., qp type, ...)
+	 */
+	RDMA_COUNTER_MODE_AUTO,
+
+	/*
+	 * Always the end
+	 */
+	RDMA_COUNTER_MODE_MAX,
+};
+
+/*
+ * Supported criteria in counter auto mode.
+ * Currently only "qp type" is supported
+ */
+enum rdma_nl_counter_mask {
+	RDMA_COUNTER_MASK_QP_TYPE = 1,
+};
 #endif /* _UAPI_RDMA_NETLINK_H */
Signature · 1 line
2.20.1
#6 lore raw

PATCH v5 6/17 RDMA/counter: Add "auto" configuration mode support

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

In auto mode all QPs belong to one category are bind automatically to a single counter set. Currently only "qp type" is supported.

In this mode the qp counter is set in RST2INIT modification, and when a qp is destroyed the counter is unbound.

---
 drivers/infiniband/core/counters.c | 221 +++++++++++++++++++++++++++++
 drivers/infiniband/core/device.c   |   3 +
 drivers/infiniband/core/verbs.c    |   9 ++
 include/rdma/ib_verbs.h            |  18 +++
 include/rdma/rdma_counter.h        |   8 ++
 5 files changed, 259 insertions(+)
drivers/infiniband/core/counters.c +221 -0
index 6167914fba06..60639452669c 100644
@@ -54,6 +54,227 @@ int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
 	return ret;
 }
 
+static struct rdma_counter *rdma_counter_alloc(struct ib_device *dev, u8 port,
+					       enum rdma_nl_counter_mode mode)
+{
+	struct rdma_counter *counter;
+
+	if (!dev->ops.counter_dealloc)
+		return NULL;
+
+	counter = kzalloc(sizeof(*counter), GFP_KERNEL);
+	if (!counter)
+		return NULL;
+
+	counter->device    = dev;
+	counter->port      = port;
+	counter->res.type  = RDMA_RESTRACK_COUNTER;
+	counter->mode.mode = mode;
+	kref_init(&counter->kref);
+	mutex_init(&counter->lock);
+
+	return counter;
+}
+
+static void rdma_counter_free(struct rdma_counter *counter)
+{
+	rdma_restrack_del(&counter->res);
+	kfree(counter);
+}
+
+static void auto_mode_init_counter(struct rdma_counter *counter,
+				   const struct ib_qp *qp,
+				   enum rdma_nl_counter_mask new_mask)
+{
+	struct auto_mode_param *param = &counter->mode.param;
+
+	counter->mode.mode = RDMA_COUNTER_MODE_AUTO;
+	counter->mode.mask = new_mask;
+
+	if (new_mask & RDMA_COUNTER_MASK_QP_TYPE)
+		param->qp_type = qp->qp_type;
+}
+
+static bool auto_mode_match(struct ib_qp *qp, struct rdma_counter *counter,
+			    enum rdma_nl_counter_mask auto_mask)
+{
+	struct auto_mode_param *param = &counter->mode.param;
+	bool match = true;
+
+	if (rdma_is_kernel_res(&counter->res) != rdma_is_kernel_res(&qp->res))
+		return false;
+
+	/* Ensure that counter belong to right PID */
+	if (!rdma_is_kernel_res(&counter->res) &&
+	    !rdma_is_kernel_res(&qp->res) &&
+	    (task_pid_vnr(counter->res.task) != current->pid))
+		return false;
+
+	if (auto_mask & RDMA_COUNTER_MASK_QP_TYPE)
+		match &= (param->qp_type == qp->qp_type);
+
+	return match;
+}
+
+static int __rdma_counter_bind_qp(struct rdma_counter *counter,
+				  struct ib_qp *qp)
+{
+	int ret;
+
+	if (qp->counter)
+		return -EINVAL;
+
+	if (!qp->device->ops.counter_bind_qp)
+		return -EOPNOTSUPP;
+
+	mutex_lock(&counter->lock);
+	ret = qp->device->ops.counter_bind_qp(counter, qp);
+	mutex_unlock(&counter->lock);
+
+	return ret;
+}
+
+static int __rdma_counter_unbind_qp(struct ib_qp *qp)
+{
+	struct rdma_counter *counter = qp->counter;
+	int ret;
+
+	if (!qp->device->ops.counter_unbind_qp)
+		return -EOPNOTSUPP;
+
+	mutex_lock(&counter->lock);
+	ret = qp->device->ops.counter_unbind_qp(qp);
+	mutex_unlock(&counter->lock);
+
+	return ret;
+}
+
+/**
+ * rdma_get_counter_auto_mode - Find the counter that @qp should be bound
+ *     with in auto mode
+ *
+ * Return: The counter (with ref-count increased) if found
+ */
+static struct rdma_counter *rdma_get_counter_auto_mode(struct ib_qp *qp,
+						       u8 port)
+{
+	struct rdma_port_counter *port_counter;
+	struct rdma_counter *counter = NULL;
+	struct ib_device *dev = qp->device;
+	struct rdma_restrack_entry *res;
+	struct rdma_restrack_root *rt;
+	unsigned long id = 0;
+
+	port_counter = &dev->port_data[port].port_counter;
+	rt = &dev->res[RDMA_RESTRACK_COUNTER];
+	xa_lock(&rt->xa);
+	xa_for_each(&rt->xa, id, res) {
+		if (!rdma_is_visible_in_pid_ns(res))
+			continue;
+
+		counter = container_of(res, struct rdma_counter, res);
+		if ((counter->device != qp->device) || (counter->port != port))
+			goto next;
+
+		if (auto_mode_match(qp, counter, port_counter->mode.mask))
+			break;
+next:
+		counter = NULL;
+	}
+
+	if (counter)
+		kref_get(&counter->kref);
+
+	xa_unlock(&rt->xa);
+	return counter;
+}
+
+static void rdma_counter_res_add(struct rdma_counter *counter,
+				 struct ib_qp *qp)
+{
+	if (rdma_is_kernel_res(&qp->res)) {
+		rdma_restrack_set_task(&counter->res, qp->res.kern_name);
+		rdma_restrack_kadd(&counter->res);
+	} else {
+		rdma_restrack_attach_task(&counter->res, qp->res.task);
+		rdma_restrack_uadd(&counter->res);
+	}
+}
+
+static void counter_release(struct kref *kref)
+{
+	struct rdma_counter *counter;
+
+	counter = container_of(kref, struct rdma_counter, kref);
+	counter->device->ops.counter_dealloc(counter);
+	rdma_counter_free(counter);
+}
+
+/**
+ * rdma_counter_bind_qp_auto - Check and bind the QP to a counter base on
+ *   the auto-mode rule
+ */
+int rdma_counter_bind_qp_auto(struct ib_qp *qp, u8 port)
+{
+	struct rdma_port_counter *port_counter;
+	struct ib_device *dev = qp->device;
+	struct rdma_counter *counter;
+	int ret;
+
+	if (!rdma_is_port_valid(dev, port))
+		return -EINVAL;
+
+	port_counter = &dev->port_data[port].port_counter;
+	if (port_counter->mode.mode != RDMA_COUNTER_MODE_AUTO)
+		return 0;
+
+	counter = rdma_get_counter_auto_mode(qp, port);
+	if (counter) {
+		ret = __rdma_counter_bind_qp(counter, qp);
+		if (ret) {
+			kref_put(&counter->kref, counter_release);
+			return ret;
+		}
+	} else {
+		counter = rdma_counter_alloc(dev, port, RDMA_COUNTER_MODE_AUTO);
+		if (!counter)
+			return -ENOMEM;
+
+		auto_mode_init_counter(counter, qp, port_counter->mode.mask);
+
+		ret = __rdma_counter_bind_qp(counter, qp);
+		if (ret) {
+			rdma_counter_free(counter);
+			return ret;
+		}
+
+		rdma_counter_res_add(counter, qp);
+	}
+
+	return 0;
+}
+
+/**
+ * rdma_counter_unbind_qp - Unbind a qp from a counter
+ * @force:
+ *   true - Decrease the counter ref-count anyway (e.g., qp destroy)
+ */
+int rdma_counter_unbind_qp(struct ib_qp *qp, bool force)
+{
+	struct rdma_counter *counter = qp->counter;
+	int ret;
+
+	if (!counter)
+		return -EINVAL;
+
+	ret = __rdma_counter_unbind_qp(qp);
+	if (ret && !force)
+		return ret;
+
+	kref_put(&counter->kref, counter_release);
+	return 0;
+}
+
 void rdma_counter_init(struct ib_device *dev)
 {
 	struct rdma_port_counter *port_counter;
drivers/infiniband/core/device.c +3 -0
index 6579865e4866..f3181b74c863 100644
@@ -2471,6 +2471,9 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops)
 	SET_DEVICE_OP(dev_ops, alloc_xrcd);
 	SET_DEVICE_OP(dev_ops, attach_mcast);
 	SET_DEVICE_OP(dev_ops, check_mr_status);
+	SET_DEVICE_OP(dev_ops, counter_bind_qp);
+	SET_DEVICE_OP(dev_ops, counter_dealloc);
+	SET_DEVICE_OP(dev_ops, counter_unbind_qp);
 	SET_DEVICE_OP(dev_ops, create_ah);
 	SET_DEVICE_OP(dev_ops, create_counters);
 	SET_DEVICE_OP(dev_ops, create_cq);
drivers/infiniband/core/verbs.c +9 -0
index 4a04e94a72db..92349bf37589 100644
@@ -1690,6 +1690,14 @@ static int _ib_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
 		}
 	}
 
+	/*
+	 * Bind this qp to a counter automatically based on the rdma counter
+	 * rules. This only set in RST2INIT with port specified
+	 */
+	if (!qp->counter && (attr_mask & IB_QP_PORT) &&
+	    ((attr_mask & IB_QP_STATE) && attr->qp_state == IB_QPS_INIT))
+		rdma_counter_bind_qp_auto(qp, attr->port_num);
+
 	ret = ib_security_modify_qp(qp, attr, attr_mask, udata);
 	if (ret)
 		goto out;
@@ -1885,6 +1893,7 @@ int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata)
 	if (!qp->uobject)
 		rdma_rw_cleanup_mrs(qp);
 
+	rdma_counter_unbind_qp(qp, true);
 	rdma_restrack_del(&qp->res);
 	ret = qp->device->ops.destroy_qp(qp, udata);
 	if (!ret) {
include/rdma/ib_verbs.h +18 -0
index 3d19c056fbc0..0205472eb73a 100644
@@ -1698,6 +1698,9 @@ struct ib_qp {
 	 * Implementation details of the RDMA core, don't use in drivers:
 	 */
 	struct rdma_restrack_entry     res;
+
+	/* The counter the qp is bind to */
+	struct rdma_counter    *counter;
 };
 
 struct ib_dm {
@@ -2485,6 +2488,21 @@ struct ib_device_ops {
 			 u8 pdata_len);
 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
+	/**
+	 * counter_bind_qp - Bind a QP to a counter.
+	 * @counter - The counter to be bound. If counter->id is zero then
+	 *   the driver needs to allocate a new counter and set counter->id
+	 */
+	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
+	/**
+	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
+	 *   counter and bind it onto the default one
+	 */
+	int (*counter_unbind_qp)(struct ib_qp *qp);
+	/**
+	 * counter_dealloc -De-allocate the hw counter
+	 */
+	int (*counter_dealloc)(struct rdma_counter *counter);
 
 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
include/rdma/rdma_counter.h +8 -0
index 8dd2619c015d..9f93a2403c9c 100644
@@ -7,11 +7,14 @@
 #define _RDMA_COUNTER_H_
 
 #include <linux/mutex.h>
+#include <linux/pid_namespace.h>
 
 #include <rdma/ib_verbs.h>
 #include <rdma/restrack.h>
 #include <rdma/rdma_netlink.h>
 
+struct ib_qp;
+
 struct auto_mode_param {
 	int qp_type;
 };
@@ -31,6 +34,9 @@ struct rdma_counter {
 	struct rdma_restrack_entry	res;
 	struct ib_device		*device;
 	uint32_t			id;
+	struct kref			kref;
+	struct rdma_counter_mode	mode;
+	struct mutex			lock;
 	u8				port;
 };
 
@@ -38,5 +44,7 @@ void rdma_counter_init(struct ib_device *dev);
 void rdma_counter_release(struct ib_device *dev);
 int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
 			       bool on, enum rdma_nl_counter_mask mask);
+int rdma_counter_bind_qp_auto(struct ib_qp *qp, u8 port);
+int rdma_counter_unbind_qp(struct ib_qp *qp, bool force);
 
 #endif /* _RDMA_COUNTER_H_ */
Signature · 1 line
2.20.1
#7 lore raw

Re: [PATCH rdma-next v5 06/17] RDMA/counter: Add "auto" configuration mode support

Quoted 145 lines · Leon Romanovsky

From: Mark Zhang <[email protected]>

In auto mode all QPs belong to one category are bind automatically to a single counter set. Currently only "qp type" is supported.

In this mode the qp counter is set in RST2INIT modification, and when a qp is destroyed the counter is unbound.

 drivers/infiniband/core/counters.c | 221 +++++++++++++++++++++++++++++
 drivers/infiniband/core/device.c   |   3 +
 drivers/infiniband/core/verbs.c    |   9 ++
 include/rdma/ib_verbs.h            |  18 +++
 include/rdma/rdma_counter.h        |   8 ++
 5 files changed, 259 insertions(+)
drivers/infiniband/core/counters.c +121 -0
index 6167914fba06..60639452669c 100644
+++ b/drivers/infiniband/core/counters.c
@@ -54,6 +54,227 @@ int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
 	return ret;
 }
 
+static struct rdma_counter *rdma_counter_alloc(struct ib_device *dev, u8 port,
+					       enum rdma_nl_counter_mode mode)
+{
+	struct rdma_counter *counter;
+
+	if (!dev->ops.counter_dealloc)
+		return NULL;
+
+	counter = kzalloc(sizeof(*counter), GFP_KERNEL);
+	if (!counter)
+		return NULL;
+
+	counter->device    = dev;
+	counter->port      = port;
+	counter->res.type  = RDMA_RESTRACK_COUNTER;
+	counter->mode.mode = mode;
+	kref_init(&counter->kref);
+	mutex_init(&counter->lock);
+
+	return counter;
+}
+
+static void rdma_counter_free(struct rdma_counter *counter)
+{
+	rdma_restrack_del(&counter->res);
+	kfree(counter);
+}
+
+static void auto_mode_init_counter(struct rdma_counter *counter,
+				   const struct ib_qp *qp,
+				   enum rdma_nl_counter_mask new_mask)
+{
+	struct auto_mode_param *param = &counter->mode.param;
+
+	counter->mode.mode = RDMA_COUNTER_MODE_AUTO;
+	counter->mode.mask = new_mask;
+
+	if (new_mask & RDMA_COUNTER_MASK_QP_TYPE)
+		param->qp_type = qp->qp_type;
+}
+
+static bool auto_mode_match(struct ib_qp *qp, struct rdma_counter *counter,
+			    enum rdma_nl_counter_mask auto_mask)
+{
+	struct auto_mode_param *param = &counter->mode.param;
+	bool match = true;
+
+	if (rdma_is_kernel_res(&counter->res) != rdma_is_kernel_res(&qp->res))
+		return false;
+
+	/* Ensure that counter belong to right PID */
+	if (!rdma_is_kernel_res(&counter->res) &&
+	    !rdma_is_kernel_res(&qp->res) &&
+	    (task_pid_vnr(counter->res.task) != current->pid))
+		return false;
+
+	if (auto_mask & RDMA_COUNTER_MASK_QP_TYPE)
+		match &= (param->qp_type == qp->qp_type);
+
+	return match;
+}
+
+static int __rdma_counter_bind_qp(struct rdma_counter *counter,
+				  struct ib_qp *qp)
+{
+	int ret;
+
+	if (qp->counter)
+		return -EINVAL;
+
+	if (!qp->device->ops.counter_bind_qp)
+		return -EOPNOTSUPP;
+
+	mutex_lock(&counter->lock);
+	ret = qp->device->ops.counter_bind_qp(counter, qp);
+	mutex_unlock(&counter->lock);
+
+	return ret;
+}
+
+static int __rdma_counter_unbind_qp(struct ib_qp *qp)
+{
+	struct rdma_counter *counter = qp->counter;
+	int ret;
+
+	if (!qp->device->ops.counter_unbind_qp)
+		return -EOPNOTSUPP;
+
+	mutex_lock(&counter->lock);
+	ret = qp->device->ops.counter_unbind_qp(qp);
+	mutex_unlock(&counter->lock);
+
+	return ret;
+}
+
+/**
+ * rdma_get_counter_auto_mode - Find the counter that @qp should be bound
+ *     with in auto mode
+ *
+ * Return: The counter (with ref-count increased) if found
+ */
+static struct rdma_counter *rdma_get_counter_auto_mode(struct ib_qp *qp,
+						       u8 port)
+{
+	struct rdma_port_counter *port_counter;
+	struct rdma_counter *counter = NULL;
+	struct ib_device *dev = qp->device;
+	struct rdma_restrack_entry *res;
+	struct rdma_restrack_root *rt;
+	unsigned long id = 0;
+
+	port_counter = &dev->port_data[port].port_counter;
+	rt = &dev->res[RDMA_RESTRACK_COUNTER];
+	xa_lock(&rt->xa);
+	xa_for_each(&rt->xa, id, res) {
+		if (!rdma_is_visible_in_pid_ns(res))
+			continue;
+
+		counter = container_of(res, struct rdma_counter, res);
+		if ((counter->device != qp->device) || (counter->port != port))

drivers/infiniband/core/counters.c int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port, · line 186

Quoted 10 lines
drivers/infiniband/core/counters.c +10 -0
+			goto next;
+
+		if (auto_mode_match(qp, counter, port_counter->mode.mask))
+			break;
+next:
+		counter = NULL;
+	}
+
+	if (counter)
+		kref_get(&counter->kref);

This still needs to be kref_get_unless_zero:

	if (counter && !kref_get_unless_zero(&counter->kref))
		counter = NULL;

Jason

#8 lore raw

PATCH v5 7/17 IB/mlx5: Support set qp counter

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Support bind a qp with counter. If counter is null then bind the qp to the default counter. Different QP state has different operation:
- RESET: Set the counter field so that it will take effective during RST2INIT change;
- RTS: Issue an RTS2RTS change to update the QP counter;
- Other: Set the counter field and mark the counter_pending flag, when QP is moved to RTS state and this flag is set, then issue an RTS2RTS modification to update the counter.

---
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  6 +++
 drivers/infiniband/hw/mlx5/qp.c      | 76 +++++++++++++++++++++++++++-
 include/linux/mlx5/qp.h              |  1 +
 3 files changed, 81 insertions(+), 2 deletions(-)
drivers/infiniband/hw/mlx5/mlx5_ib.h +6 -0
index bdb83fc85f94..c0f4327bd1a5 100644
@@ -439,6 +439,10 @@ struct mlx5_ib_qp {
 	u32			flags_en;
 	/* storage for qp sub type when core qp type is IB_QPT_DRIVER */
 	enum ib_qp_type		qp_sub_type;
+	/* A flag to indicate if there's a new counter is configured
+	 * but not take effective
+	 */
+	u32                     counter_pending;
 };
 
 struct mlx5_ib_cq_buf {
@@ -1456,4 +1460,6 @@ void mlx5_ib_put_xlt_emergency_page(void);
 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
 			bool dyn_bfreg);
+
+int mlx5_ib_qp_set_counter(struct ib_qp *qp, struct rdma_counter *counter);
 #endif /* MLX5_IB_H */
drivers/infiniband/hw/mlx5/qp.c +74 -2

116 lines not on this page · raw message

include/linux/mlx5/qp.h +1 -0
index 08e43cd9e742..a8270869f0b6 100644
@@ -71,6 +71,7 @@ enum mlx5_qp_optpar {
 	MLX5_QP_OPTPAR_CQN_RCV			= 1 << 19,
 	MLX5_QP_OPTPAR_DC_HS			= 1 << 20,
 	MLX5_QP_OPTPAR_DC_KEY			= 1 << 21,
+	MLX5_QP_OPTPAR_COUNTER_SET_ID		= 1 << 25,
 };
 
 enum mlx5_qp_state {
Signature · 1 line
2.20.1
#9 lore raw

PATCH v5 8/17 IB/mlx5: Add counter set id as a parameter for mlx5_ib_query_q_counters()

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Add counter set id as a parameter so that this API can be used for querying any q counter.

---
 drivers/infiniband/hw/mlx5/main.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/infiniband/hw/mlx5/main.c +5 -5
index 05d2bfcb3d60..4740cfda5b17 100644
@@ -5443,7 +5443,8 @@ static struct rdma_hw_stats *mlx5_ib_alloc_hw_stats(struct ib_device *ibdev,
 
 static int mlx5_ib_query_q_counters(struct mlx5_core_dev *mdev,
 				    struct mlx5_ib_port *port,
-				    struct rdma_hw_stats *stats)
+				    struct rdma_hw_stats *stats,
+				    u16 set_id)
 {
 	int outlen = MLX5_ST_SZ_BYTES(query_q_counter_out);
 	void *out;
@@ -5454,9 +5455,7 @@ static int mlx5_ib_query_q_counters(struct mlx5_core_dev *mdev,
 	if (!out)
 		return -ENOMEM;
 
-	ret = mlx5_core_query_q_counter(mdev,
-					port->cnts.set_id, 0,
-					out, outlen);
+	ret = mlx5_core_query_q_counter(mdev, set_id, 0, out, outlen);
 	if (ret)
 		goto free;
 
@@ -5516,7 +5515,8 @@ static int mlx5_ib_get_hw_stats(struct ib_device *ibdev,
 		       port->cnts.num_ext_ppcnt_counters;
 
 	/* q_counters are per IB device, query the master mdev */
-	ret = mlx5_ib_query_q_counters(dev->mdev, port, stats);
+	ret = mlx5_ib_query_q_counters(dev->mdev, port, stats,
+				       port->cnts.set_id);
 	if (ret)
 		return ret;
 
Signature · 1 line
2.20.1
#10 lore raw

PATCH v5 9/17 IB/mlx5: Support statistic q counter configuration

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Add support for ib callbacks counter_bind_qp(), counter_unbind_qp() and counter_dealloc().

---
 drivers/infiniband/hw/mlx5/main.c | 44 +++++++++++++++++++++++++++++++
 1 file changed, 44 insertions(+)
drivers/infiniband/hw/mlx5/main.c +44 -0
index 4740cfda5b17..ffd6f16d3c37 100644
@@ -5552,6 +5552,47 @@ static int mlx5_ib_get_hw_stats(struct ib_device *ibdev,
 	return num_counters;
 }
 
+static int mlx5_ib_counter_bind_qp(struct rdma_counter *counter,
+				   struct ib_qp *qp)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->device);
+	u16 cnt_set_id = 0;
+	int err;
+
+	if (!counter->id) {
+		err = mlx5_cmd_alloc_q_counter(dev->mdev,
+					       &cnt_set_id,
+					       MLX5_SHARED_RESOURCE_UID);
+		if (err)
+			return err;
+		counter->id = cnt_set_id;
+	}
+
+	err = mlx5_ib_qp_set_counter(qp, counter);
+	if (err)
+		goto fail_set_counter;
+
+	return 0;
+
+fail_set_counter:
+	mlx5_core_dealloc_q_counter(dev->mdev, cnt_set_id);
+	counter->id = 0;
+
+	return err;
+}
+
+static int mlx5_ib_counter_unbind_qp(struct ib_qp *qp)
+{
+	return mlx5_ib_qp_set_counter(qp, NULL);
+}
+
+static int mlx5_ib_counter_dealloc(struct rdma_counter *counter)
+{
+	struct mlx5_ib_dev *dev = to_mdev(counter->device);
+
+	return mlx5_core_dealloc_q_counter(dev->mdev, counter->id);
+}
+
 static int mlx5_ib_rn_get_params(struct ib_device *device, u8 port_num,
 				 enum rdma_netdev_t type,
 				 struct rdma_netdev_alloc_params *params)
@@ -6475,6 +6516,9 @@ static void mlx5_ib_stage_odp_cleanup(struct mlx5_ib_dev *dev)
 static const struct ib_device_ops mlx5_ib_dev_hw_stats_ops = {
 	.alloc_hw_stats = mlx5_ib_alloc_hw_stats,
 	.get_hw_stats = mlx5_ib_get_hw_stats,
+	.counter_bind_qp = mlx5_ib_counter_bind_qp,
+	.counter_unbind_qp = mlx5_ib_counter_unbind_qp,
+	.counter_dealloc = mlx5_ib_counter_dealloc,
 };
 
 static int mlx5_ib_stage_counters_init(struct mlx5_ib_dev *dev)
Signature · 1 line
2.20.1
#11 lore raw

PATCH v5 10/17 RDMA/nldev: Allow counter auto mode configration through RDMA netlink

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Provide an option to enable/disable per-port counter auto mode through RDMA netlink. Limit it to users with ADMIN capability only.

---
 drivers/infiniband/core/nldev.c  | 78 ++++++++++++++++++++++++++++++++
 include/uapi/rdma/rdma_netlink.h |  8 ++++
 2 files changed, 86 insertions(+)
drivers/infiniband/core/nldev.c +78 -0
index d9ebfb50962b..9a4cf285f447 100644
@@ -126,6 +126,9 @@ static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
 	[RDMA_NLDEV_ATTR_RES_USECNT]		= { .type = NLA_U64 },
 	[RDMA_NLDEV_ATTR_SM_LID]		= { .type = NLA_U32 },
 	[RDMA_NLDEV_ATTR_SUBNET_PREFIX]		= { .type = NLA_U64 },
+	[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK]	= { .type = NLA_U32 },
+	[RDMA_NLDEV_ATTR_STAT_MODE]		= { .type = NLA_U32 },
+	[RDMA_NLDEV_ATTR_STAT_RES]		= { .type = NLA_U32 },
 	[RDMA_NLDEV_ATTR_SYS_IMAGE_GUID]	= { .type = NLA_U64 },
 	[RDMA_NLDEV_ATTR_UVERBS_DRIVER_ID]	= { .type = NLA_U32 },
 	[RDMA_NLDEV_NET_NS_FD]			= { .type = NLA_U32 },
@@ -1482,6 +1485,78 @@ static int nldev_set_sys_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 	return err;
 }
 
+static int nldev_stat_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
+			       struct netlink_ext_ack *extack)
+{
+	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
+	u32 index, port, mode, mask = 0;
+	struct ib_device *device;
+	struct sk_buff *msg;
+	int ret;
+
+	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
+			  nldev_policy, extack);
+	/* Currently only counter for QP is supported */
+	if (ret || !tb[RDMA_NLDEV_ATTR_STAT_RES] ||
+	    !tb[RDMA_NLDEV_ATTR_DEV_INDEX] ||
+	    !tb[RDMA_NLDEV_ATTR_PORT_INDEX] || !tb[RDMA_NLDEV_ATTR_STAT_MODE])
+		return -EINVAL;
+
+	if (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES]) != RDMA_NLDEV_ATTR_RES_QP)
+		return -EINVAL;
+
+	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
+	device = ib_device_get_by_index(sock_net(skb->sk), index);
+	if (!device)
+		return -EINVAL;
+
+	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
+	if (!rdma_is_port_valid(device, port)) {
+		ret = -EINVAL;
+		goto err;
+	}
+
+	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+	if (!msg) {
+		ret = -ENOMEM;
+		goto err;
+	}
+	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
+			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
+					 RDMA_NLDEV_CMD_STAT_SET),
+			0, 0);
+
+	mode = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_MODE]);
+	if (mode != RDMA_COUNTER_MODE_AUTO) {
+		ret = -EMSGSIZE;
+		goto err_msg;
+	}
+
+	if (tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK])
+		mask = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK]);
+
+	ret = rdma_counter_set_auto_mode(device, port,
+					 mask ? true : false, mask);
+	if (ret)
+		goto err_msg;
+
+	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_MODE, mode) ||
+	    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK, mask)) {
+		ret = -EMSGSIZE;
+		goto err_msg;
+	}
+
+	nlmsg_end(msg, nlh);
+	ib_device_put(device);
+	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
+
+err_msg:
+	nlmsg_free(msg);
+err:
+	ib_device_put(device);
+	return ret;
+}
+
 static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = {
 	[RDMA_NLDEV_CMD_GET] = {
 		.doit = nldev_get_doit,
@@ -1535,6 +1610,9 @@ static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = {
 	},
 	[RDMA_NLDEV_CMD_SYS_SET] = {
 		.doit = nldev_set_sys_set_doit,
+	},
+	[RDMA_NLDEV_CMD_STAT_SET] = {
+		.doit = nldev_stat_set_doit,
 		.flags = RDMA_NL_ADMIN_PERM,
 	},
 };
include/uapi/rdma/rdma_netlink.h +8 -0
index e3cd912e9cef..0cb47d23fd86 100644
@@ -281,6 +281,8 @@ enum rdma_nldev_command {
 
 	RDMA_NLDEV_CMD_GET_CHARDEV,
 
+	RDMA_NLDEV_CMD_STAT_SET,
+
 	RDMA_NLDEV_NUM_OPS
 };
 
@@ -488,6 +490,12 @@ enum rdma_nldev_attr {
 	 * File descriptor handle of the net namespace object
 	 */
 	RDMA_NLDEV_NET_NS_FD,			/* u32 */
+	/*
+	 * Counter-specific attributes.
+	 */
+	RDMA_NLDEV_ATTR_STAT_MODE,		/* u32 */
+	RDMA_NLDEV_ATTR_STAT_RES,		/* u32 */
+	RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK,	/* u32 */
 
 	/*
 	 * Information about a chardev.
Signature · 1 line
2.20.1
#12 lore raw

PATCH v5 11/17 RDMA/netlink: Implement counter dumpit calback

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

This patch adds the ability to return all available counters together with their properties and hwstats.

---
 drivers/infiniband/core/counters.c |  26 +++-
 drivers/infiniband/core/device.c   |   2 +
 drivers/infiniband/core/nldev.c    | 213 +++++++++++++++++++++++++++++
 include/rdma/ib_verbs.h            |  10 ++
 include/rdma/rdma_counter.h        |   3 +
 include/uapi/rdma/rdma_netlink.h   |  10 +-
 6 files changed, 262 insertions(+), 2 deletions(-)
drivers/infiniband/core/counters.c +25 -1
index 60639452669c..ca9adee19159 100644
@@ -59,7 +59,7 @@ static struct rdma_counter *rdma_counter_alloc(struct ib_device *dev, u8 port,
 {
 	struct rdma_counter *counter;
 
-	if (!dev->ops.counter_dealloc)
+	if (!dev->ops.counter_dealloc || !dev->ops.counter_alloc_stats)
 		return NULL;
 
 	counter = kzalloc(sizeof(*counter), GFP_KERNEL);
@@ -69,16 +69,25 @@ static struct rdma_counter *rdma_counter_alloc(struct ib_device *dev, u8 port,
 	counter->device    = dev;
 	counter->port      = port;
 	counter->res.type  = RDMA_RESTRACK_COUNTER;
+	counter->stats     = dev->ops.counter_alloc_stats(counter);
+	if (!counter->stats)
+		goto err_stats;
+
 	counter->mode.mode = mode;
 	kref_init(&counter->kref);
 	mutex_init(&counter->lock);
 
 	return counter;
+
+err_stats:
+	kfree(counter);
+	return NULL;
 }
 
 static void rdma_counter_free(struct rdma_counter *counter)
 {
 	rdma_restrack_del(&counter->res);
+	kfree(counter->stats);
 	kfree(counter);
 }
 
@@ -275,6 +284,21 @@ int rdma_counter_unbind_qp(struct ib_qp *qp, bool force)
 	return 0;
 }
 
+int rdma_counter_query_stats(struct rdma_counter *counter)
+{
+	struct ib_device *dev = counter->device;
+	int ret;
+
+	if (!dev->ops.counter_update_stats)
+		return -EINVAL;
+
+	mutex_lock(&counter->lock);
+	ret = dev->ops.counter_update_stats(counter);
+	mutex_unlock(&counter->lock);
+
+	return ret;
+}
+
 void rdma_counter_init(struct ib_device *dev)
 {
 	struct rdma_port_counter *port_counter;
drivers/infiniband/core/device.c +2 -0
index f3181b74c863..bdf61499e6d5 100644
@@ -2471,9 +2471,11 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops)
 	SET_DEVICE_OP(dev_ops, alloc_xrcd);
 	SET_DEVICE_OP(dev_ops, attach_mcast);
 	SET_DEVICE_OP(dev_ops, check_mr_status);
+	SET_DEVICE_OP(dev_ops, counter_alloc_stats);
 	SET_DEVICE_OP(dev_ops, counter_bind_qp);
 	SET_DEVICE_OP(dev_ops, counter_dealloc);
 	SET_DEVICE_OP(dev_ops, counter_unbind_qp);
+	SET_DEVICE_OP(dev_ops, counter_update_stats);
 	SET_DEVICE_OP(dev_ops, create_ah);
 	SET_DEVICE_OP(dev_ops, create_counters);
 	SET_DEVICE_OP(dev_ops, create_cq);
drivers/infiniband/core/nldev.c +213 -0

249 lines not on this page · raw message

include/rdma/ib_verbs.h +10 -0
index 0205472eb73a..0c5151a12ae4 100644
@@ -2503,6 +2503,16 @@ struct ib_device_ops {
 	 * counter_dealloc -De-allocate the hw counter
 	 */
 	int (*counter_dealloc)(struct rdma_counter *counter);
+	/**
+	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
+	 * the driver initialized data.
+	 */
+	struct rdma_hw_stats *(*counter_alloc_stats)(
+		struct rdma_counter *counter);
+	/**
+	 * counter_update_stats - Query the stats value of this counter
+	 */
+	int (*counter_update_stats)(struct rdma_counter *counter);
 
 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
include/rdma/rdma_counter.h +3 -0
index 9f93a2403c9c..f2a5c8efc404 100644
@@ -37,6 +37,7 @@ struct rdma_counter {
 	struct kref			kref;
 	struct rdma_counter_mode	mode;
 	struct mutex			lock;
+	struct rdma_hw_stats		*stats;
 	u8				port;
 };
 
@@ -47,4 +48,6 @@ int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
 int rdma_counter_bind_qp_auto(struct ib_qp *qp, u8 port);
 int rdma_counter_unbind_qp(struct ib_qp *qp, bool force);
 
+int rdma_counter_query_stats(struct rdma_counter *counter);
+
 #endif /* _RDMA_COUNTER_H_ */
include/uapi/rdma/rdma_netlink.h +9 -1
index 0cb47d23fd86..22c5bc7a82dd 100644
@@ -283,6 +283,8 @@ enum rdma_nldev_command {
 
 	RDMA_NLDEV_CMD_STAT_SET,
 
+	RDMA_NLDEV_CMD_STAT_GET, /* can dump */
+
 	RDMA_NLDEV_NUM_OPS
 };
 
@@ -496,7 +498,13 @@ enum rdma_nldev_attr {
 	RDMA_NLDEV_ATTR_STAT_MODE,		/* u32 */
 	RDMA_NLDEV_ATTR_STAT_RES,		/* u32 */
 	RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK,	/* u32 */
-
+	RDMA_NLDEV_ATTR_STAT_COUNTER,		/* nested table */
+	RDMA_NLDEV_ATTR_STAT_COUNTER_ENTRY,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_COUNTER_ID,	/* u32 */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTERS,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME,	/* string */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE,	/* u64 */
 	/*
 	 * Information about a chardev.
 	 * CHARDEV_TYPE is the name of the chardev ABI (ie uverbs, umad, etc)
Signature · 1 line
2.20.1
#13 lore raw

Re: [PATCH rdma-next v5 11/17] RDMA/netlink: Implement counter dumpit calback

Quoted 17 lines · Leon Romanovsky
include/uapi/rdma/rdma_netlink.h +3 -1
index 0cb47d23fd86..22c5bc7a82dd 100644
+++ b/include/uapi/rdma/rdma_netlink.h
@@ -283,6 +283,8 @@ enum rdma_nldev_command {
 
 	RDMA_NLDEV_CMD_STAT_SET,
 
+	RDMA_NLDEV_CMD_STAT_GET, /* can dump */
+
 	RDMA_NLDEV_NUM_OPS
 };
 
@@ -496,7 +498,13 @@ enum rdma_nldev_attr {
 	RDMA_NLDEV_ATTR_STAT_MODE,		/* u32 */
 	RDMA_NLDEV_ATTR_STAT_RES,		/* u32 */
 	RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK,	/* u32 */
-

include/uapi/rdma/rdma_netlink.h enum rdma_nldev_attr { · line 510

Quoted 10 lines
include/uapi/rdma/rdma_netlink.h +7 -0
+	RDMA_NLDEV_ATTR_STAT_COUNTER,		/* nested table */
+	RDMA_NLDEV_ATTR_STAT_COUNTER_ENTRY,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_COUNTER_ID,	/* u32 */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTERS,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME,	/* string */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE,	/* u64 */
 	/*
 	 * Information about a chardev.
 	 * CHARDEV_TYPE is the name of the chardev ABI (ie uverbs, umad, etc)

This is in the wrong place, needs to be at the end.

Jason

#14 lore raw

Re: [PATCH rdma-next v5 11/17] RDMA/netlink: Implement counter dumpit calback

Quoted 22 lines · Jason Gunthorpe
Quoted 21 lines · Leon Romanovsky
include/uapi/rdma/rdma_netlink.h +7 -1
index 0cb47d23fd86..22c5bc7a82dd 100644
+++ b/include/uapi/rdma/rdma_netlink.h
@@ -283,6 +283,8 @@ enum rdma_nldev_command {
 
 	RDMA_NLDEV_CMD_STAT_SET,
 
+	RDMA_NLDEV_CMD_STAT_GET, /* can dump */
+
 	RDMA_NLDEV_NUM_OPS
 };
 
@@ -496,7 +498,13 @@ enum rdma_nldev_attr {
 	RDMA_NLDEV_ATTR_STAT_MODE,		/* u32 */
 	RDMA_NLDEV_ATTR_STAT_RES,		/* u32 */
 	RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK,	/* u32 */
-
+	RDMA_NLDEV_ATTR_STAT_COUNTER,		/* nested table */
+	RDMA_NLDEV_ATTR_STAT_COUNTER_ENTRY,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_COUNTER_ID,	/* u32 */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTERS,	/* nested table */
Quoted 8 lines

include/uapi/rdma/rdma_netlink.h enum rdma_nldev_attr { · line 510

Quoted 6 lines
include/uapi/rdma/rdma_netlink.h +3 -0
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY,	/* nested table */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME,	/* string */
+	RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE,	/* u64 */
 	/*
 	 * Information about a chardev.
 	 * CHARDEV_TYPE is the name of the chardev ABI (ie uverbs, umad, etc)

This is in the wrong place, needs to be at the end.

Yes, it is rebase error.

Thanks

Quoted 2 lines

Jason

#15 lore raw

PATCH v5 12/17 IB/mlx5: Add counter_alloc_stats() and counter_update_stats() support

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Add support for ib callback counter_alloc_stats() and counter_update_stats().

---
 drivers/infiniband/hw/mlx5/main.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
drivers/infiniband/hw/mlx5/main.c +23 -0
index ffd6f16d3c37..d6751b2cfa1c 100644
@@ -5552,6 +5552,27 @@ static int mlx5_ib_get_hw_stats(struct ib_device *ibdev,
 	return num_counters;
 }
 
+static struct rdma_hw_stats *
+mlx5_ib_counter_alloc_stats(struct rdma_counter *counter)
+{
+	struct mlx5_ib_dev *dev = to_mdev(counter->device);
+	struct mlx5_ib_port *port = &dev->port[counter->port - 1];
+
+	/* Q counters are in the beginning of all counters */
+	return rdma_alloc_hw_stats_struct(port->cnts.names,
+					  port->cnts.num_q_counters,
+					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
+}
+
+static int mlx5_ib_counter_update_stats(struct rdma_counter *counter)
+{
+	struct mlx5_ib_dev *dev = to_mdev(counter->device);
+	struct mlx5_ib_port *port = &dev->port[counter->port - 1];
+
+	return mlx5_ib_query_q_counters(dev->mdev, port,
+					counter->stats, counter->id);
+}
+
 static int mlx5_ib_counter_bind_qp(struct rdma_counter *counter,
 				   struct ib_qp *qp)
 {
@@ -6519,6 +6540,8 @@ static const struct ib_device_ops mlx5_ib_dev_hw_stats_ops = {
 	.counter_bind_qp = mlx5_ib_counter_bind_qp,
 	.counter_unbind_qp = mlx5_ib_counter_unbind_qp,
 	.counter_dealloc = mlx5_ib_counter_dealloc,
+	.counter_alloc_stats = mlx5_ib_counter_alloc_stats,
+	.counter_update_stats = mlx5_ib_counter_update_stats,
 };
 
 static int mlx5_ib_stage_counters_init(struct mlx5_ib_dev *dev)
Signature · 1 line
2.20.1
#16 lore raw

PATCH v5 13/17 RDMA/core: Get sum value of all counters when perform a sysfs stat read

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Since a QP can only be bound to one counter, then if it is bound to a separate counter, for backward compatibility purpose, the statistic value must be:
* stat of default counter
+ stat of all running allocated counters
+ stat of all deallocated counters (history stats)

---
 drivers/infiniband/core/counters.c | 89 ++++++++++++++++++++++++++++++
 drivers/infiniband/core/sysfs.c    | 10 +++-
 include/rdma/rdma_counter.h        |  2 +
 3 files changed, 98 insertions(+), 3 deletions(-)
drivers/infiniband/core/counters.c +89 -0
index ca9adee19159..2de4c555eba9 100644
@@ -158,6 +158,20 @@ static int __rdma_counter_unbind_qp(struct ib_qp *qp)
 	return ret;
 }
 
+static void counter_history_stat_update(const struct rdma_counter *counter)
+{
+	struct ib_device *dev = counter->device;
+	struct rdma_port_counter *port_counter;
+	int i;
+
+	port_counter = &dev->port_data[counter->port].port_counter;
+	if (!port_counter->hstats)
+		return;
+
+	for (i = 0; i < counter->stats->num_counters; i++)
+		port_counter->hstats->value[i] += counter->stats->value[i];
+}
+
 /**
  * rdma_get_counter_auto_mode - Find the counter that @qp should be bound
  *     with in auto mode
@@ -215,6 +229,7 @@ static void counter_release(struct kref *kref)
 	struct rdma_counter *counter;
 
 	counter = container_of(kref, struct rdma_counter, kref);
+	counter_history_stat_update(counter);
 	counter->device->ops.counter_dealloc(counter);
 	rdma_counter_free(counter);
 }
@@ -299,6 +314,55 @@ int rdma_counter_query_stats(struct rdma_counter *counter)
 	return ret;
 }
 
+static u64 get_running_counters_hwstat_sum(struct ib_device *dev,
+					   u8 port, u32 index)
+{
+	struct rdma_restrack_entry *res;
+	struct rdma_restrack_root *rt;
+	struct rdma_counter *counter;
+	unsigned long id = 0;
+	u64 sum = 0;
+
+	rt = &dev->res[RDMA_RESTRACK_COUNTER];
+	xa_lock(&rt->xa);
+	xa_for_each(&rt->xa, id, res) {
+		if (!rdma_restrack_get(res))
+			continue;
+
+		xa_unlock(&rt->xa);
+
+		counter = container_of(res, struct rdma_counter, res);
+		if ((counter->device != dev) || (counter->port != port) ||
+		    rdma_counter_query_stats(counter))
+			goto next;
+
+		sum += counter->stats->value[index];
+
+next:
+		xa_lock(&rt->xa);
+		rdma_restrack_put(res);
+	}
+
+	xa_unlock(&rt->xa);
+	return sum;
+}
+
+/**
+ * rdma_counter_get_hwstat_value() - Get the sum value of all counters on a
+ *   specific port, including the running ones and history data
+ */
+u64 rdma_counter_get_hwstat_value(struct ib_device *dev, u8 port, u32 index)
+{
+	struct rdma_port_counter *port_counter;
+	u64 sum;
+
+	port_counter = &dev->port_data[port].port_counter;
+	sum = get_running_counters_hwstat_sum(dev, port, index);
+	sum += port_counter->hstats->value[index];
+
+	return sum;
+}
+
 void rdma_counter_init(struct ib_device *dev)
 {
 	struct rdma_port_counter *port_counter;
@@ -311,9 +375,34 @@ void rdma_counter_init(struct ib_device *dev)
 		port_counter = &dev->port_data[port].port_counter;
 		port_counter->mode.mode = RDMA_COUNTER_MODE_NONE;
 		mutex_init(&port_counter->lock);
+
+		port_counter->hstats = dev->ops.alloc_hw_stats(dev, port);
+		if (!port_counter->hstats)
+			goto fail;
 	}
+
+	return;
+
+fail:
+	rdma_for_each_port(dev, port) {
+		port_counter = &dev->port_data[port].port_counter;
+		kfree(port_counter->hstats);
+		port_counter->hstats = NULL;
+	}
+
+	return;
 }
 
 void rdma_counter_release(struct ib_device *dev)
 {
+	struct rdma_port_counter *port_counter;
+	u32 port;
+
+	if (!dev->ops.alloc_hw_stats)
+		return;
+
+	rdma_for_each_port(dev, port) {
+		port_counter = &dev->port_data[port].port_counter;
+		kfree(port_counter->hstats);
+	}
 }
drivers/infiniband/core/sysfs.c +7 -3
index c78d0c9646ae..c59b80e0a740 100644
@@ -43,6 +43,7 @@
 #include <rdma/ib_mad.h>
 #include <rdma/ib_pma.h>
 #include <rdma/ib_cache.h>
+#include <rdma/rdma_counter.h>
 
 struct ib_port;
 
@@ -800,9 +801,12 @@ static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
 	return 0;
 }
 
-static ssize_t print_hw_stat(struct rdma_hw_stats *stats, int index, char *buf)
+static ssize_t print_hw_stat(struct ib_device *dev, int port_num,
+			     struct rdma_hw_stats *stats, int index, char *buf)
 {
-	return sprintf(buf, "%llu\n", stats->value[index]);
+	u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
+
+	return sprintf(buf, "%llu\n", stats->value[index] + v);
 }
 
 static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
@@ -828,7 +832,7 @@ static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
 	ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
 	if (ret)
 		goto unlock;
-	ret = print_hw_stat(stats, hsa->index, buf);
+	ret = print_hw_stat(dev, hsa->port_num, stats, hsa->index, buf);
 unlock:
 	mutex_unlock(&stats->lock);
 
include/rdma/rdma_counter.h +2 -0
index f2a5c8efc404..bf2c3578768f 100644
@@ -27,6 +27,7 @@ struct rdma_counter_mode {
 
 struct rdma_port_counter {
 	struct rdma_counter_mode mode;
+	struct rdma_hw_stats *hstats;
 	struct mutex lock;
 };
 
@@ -49,5 +50,6 @@ int rdma_counter_bind_qp_auto(struct ib_qp *qp, u8 port);
 int rdma_counter_unbind_qp(struct ib_qp *qp, bool force);
 
 int rdma_counter_query_stats(struct rdma_counter *counter);
+u64 rdma_counter_get_hwstat_value(struct ib_device *dev, u8 port, u32 index);
 
 #endif /* _RDMA_COUNTER_H_ */
Signature · 1 line
2.20.1
#17 lore raw

PATCH v5 14/17 RDMA/counter: Allow manual mode configuration support

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

In manual mode a QP is bound to a counter manually. If counter is not specified then a new one will be allocated.
Manually mode is enabled when user binds a QP, and disabled when the last manually bound QP is unbound.
When auto-mode is turned off and there are counters left, manual mode is enabled so that the user is able to access these counters.

---
 drivers/infiniband/core/counters.c | 219 ++++++++++++++++++++++++++++-
 include/rdma/rdma_counter.h        |   7 +
 include/uapi/rdma/rdma_netlink.h   |   6 +
 3 files changed, 229 insertions(+), 3 deletions(-)
drivers/infiniband/core/counters.c +216 -3
index 2de4c555eba9..6ac8bf1240de 100644
@@ -27,7 +27,9 @@ static int __counter_set_mode(struct rdma_counter_mode *curr,
 /**
  * rdma_counter_set_auto_mode() - Turn on/off per-port auto mode
  *
- * When @on is true, the @mask must be set
+ * When @on is true, the @mask must be set; When @on is false, it goes
+ * into manual mode if there's any counter, so that the user is able to
+ * manually access them.
  */
 int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
 			       bool on, enum rdma_nl_counter_mask mask)
@@ -45,8 +47,13 @@ int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
 			ret = -EINVAL;
 			goto out;
 		}
-		ret = __counter_set_mode(&port_counter->mode,
-					 RDMA_COUNTER_MODE_NONE, 0);
+
+		if (port_counter->num_counters)
+			ret = __counter_set_mode(&port_counter->mode,
+						 RDMA_COUNTER_MODE_MANUAL, 0);
+		else
+			ret = __counter_set_mode(&port_counter->mode,
+						 RDMA_COUNTER_MODE_NONE, 0);
 	}
 
 out:
@@ -57,7 +64,9 @@ int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
 static struct rdma_counter *rdma_counter_alloc(struct ib_device *dev, u8 port,
 					       enum rdma_nl_counter_mode mode)
 {
+	struct rdma_port_counter *port_counter;
 	struct rdma_counter *counter;
+	int ret;
 
 	if (!dev->ops.counter_dealloc || !dev->ops.counter_alloc_stats)
 		return NULL;
@@ -73,12 +82,27 @@ static struct rdma_counter *rdma_counter_alloc(struct ib_device *dev, u8 port,
 	if (!counter->stats)
 		goto err_stats;
 
+	port_counter = &dev->port_data[port].port_counter;
+	mutex_lock(&port_counter->lock);
+	if (mode == RDMA_COUNTER_MODE_MANUAL) {
+		ret = __counter_set_mode(&port_counter->mode,
+					 RDMA_COUNTER_MODE_MANUAL, 0);
+		if (ret)
+			goto err_mode;
+	}
+
+	port_counter->num_counters++;
+	mutex_unlock(&port_counter->lock);
+
 	counter->mode.mode = mode;
 	kref_init(&counter->kref);
 	mutex_init(&counter->lock);
 
 	return counter;
 
+err_mode:
+	mutex_unlock(&port_counter->lock);
+	kfree(counter->stats);
 err_stats:
 	kfree(counter);
 	return NULL;
@@ -86,6 +110,18 @@ static struct rdma_counter *rdma_counter_alloc(struct ib_device *dev, u8 port,
 
 static void rdma_counter_free(struct rdma_counter *counter)
 {
+	struct rdma_port_counter *port_counter;
+
+	port_counter = &counter->device->port_data[counter->port].port_counter;
+	mutex_lock(&port_counter->lock);
+	port_counter->num_counters--;
+	if (!port_counter->num_counters &&
+	    (port_counter->mode.mode == RDMA_COUNTER_MODE_MANUAL))
+		__counter_set_mode(&port_counter->mode, RDMA_COUNTER_MODE_NONE,
+				   0);
+
+	mutex_unlock(&port_counter->lock);
+
 	rdma_restrack_del(&counter->res);
 	kfree(counter->stats);
 	kfree(counter);
@@ -363,6 +399,183 @@ u64 rdma_counter_get_hwstat_value(struct ib_device *dev, u8 port, u32 index)
 	return sum;
 }
 
+static struct ib_qp *rdma_counter_get_qp(struct ib_device *dev, u32 qp_num)
+{
+	struct rdma_restrack_entry *res = NULL;
+	struct ib_qp *qp = NULL;
+
+	res = rdma_restrack_get_byid(dev, RDMA_RESTRACK_QP, qp_num);
+	if (IS_ERR(res))
+		return NULL;
+
+	if (!rdma_is_visible_in_pid_ns(res))
+		goto err;
+
+	qp = container_of(res, struct ib_qp, res);
+	if (qp->qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
+		goto err;
+
+	return qp;
+
+err:
+	rdma_restrack_put(&qp->res);
+	return NULL;
+}
+
+static int rdma_counter_bind_qp_manual(struct rdma_counter *counter,
+				       struct ib_qp *qp)
+{
+	if ((counter->device != qp->device) || (counter->port != qp->port))
+		return -EINVAL;
+
+	return __rdma_counter_bind_qp(counter, qp);
+}
+
+static struct rdma_counter *rdma_get_counter_by_id(struct ib_device *dev,
+						   u32 counter_id)
+{
+	struct rdma_restrack_entry *res;
+	struct rdma_counter *counter;
+
+	res = rdma_restrack_get_byid(dev, RDMA_RESTRACK_COUNTER, counter_id);
+	if (IS_ERR(res))
+		return NULL;
+
+	if (!rdma_is_visible_in_pid_ns(res)) {
+		rdma_restrack_put(res);
+		return NULL;
+	}
+
+	counter = container_of(res, struct rdma_counter, res);
+	kref_get(&counter->kref);
+	rdma_restrack_put(res);
+
+	return counter;
+}
+
+/**
+ * rdma_counter_bind_qpn() - Bind QP @qp_num to counter @counter_id
+ */
+int rdma_counter_bind_qpn(struct ib_device *dev, u8 port,
+			  u32 qp_num, u32 counter_id)
+{
+	struct rdma_counter *counter;
+	struct ib_qp *qp;
+	int ret;
+
+	qp = rdma_counter_get_qp(dev, qp_num);
+	if (!qp)
+		return -ENOENT;
+
+	counter = rdma_get_counter_by_id(dev, counter_id);
+	if (!counter) {
+		ret = -ENOENT;
+		goto err;
+	}
+
+	if (counter->res.task != qp->res.task) {
+		ret = -EINVAL;
+		goto err_task;
+	}
+
+	ret = rdma_counter_bind_qp_manual(counter, qp);
+	if (ret)
+		goto err_task;
+
+	rdma_restrack_put(&qp->res);
+	return 0;
+
+err_task:
+	kref_put(&counter->kref, counter_release);
+err:
+	rdma_restrack_put(&qp->res);
+	return ret;
+}
+
+/**
+ * rdma_counter_bind_qpn_alloc() - Alloc a counter and bind QP @qp_num to it
+ *   The id of new counter is returned in @counter_id
+ */
+int rdma_counter_bind_qpn_alloc(struct ib_device *dev, u8 port,
+				u32 qp_num, u32 *counter_id)
+{
+	struct rdma_counter *counter;
+	struct ib_qp *qp;
+	int ret;
+
+	if (!rdma_is_port_valid(dev, port))
+		return -EINVAL;
+
+	qp = rdma_counter_get_qp(dev, qp_num);
+	if (!qp)
+		return -ENOENT;
+
+	if (rdma_is_port_valid(dev, qp->port) && (qp->port != port)) {
+		ret = -EINVAL;
+		goto err;
+	}
+
+	counter = rdma_counter_alloc(dev, port, RDMA_COUNTER_MODE_MANUAL);
+	if (!counter) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	ret = rdma_counter_bind_qp_manual(counter, qp);
+	if (ret)
+		goto err_bind;
+
+	if (counter_id)
+		*counter_id = counter->id;
+
+	rdma_counter_res_add(counter, qp);
+
+	rdma_restrack_put(&qp->res);
+	return ret;
+
+err_bind:
+	rdma_counter_free(counter);
+err:
+	rdma_restrack_put(&qp->res);
+	return ret;
+}
+
+/**
+ * rdma_counter_unbind_qpn() - Unbind QP @qp_num from a counter
+ */
+int rdma_counter_unbind_qpn(struct ib_device *dev, u8 port,
+			    u32 qp_num, u32 counter_id)
+{
+	struct rdma_port_counter *port_counter;
+	struct ib_qp *qp;
+	int ret;
+
+	if (!rdma_is_port_valid(dev, port))
+		return -EINVAL;
+
+	qp = rdma_counter_get_qp(dev, qp_num);
+	if (!qp)
+		return -ENOENT;
+
+	if (rdma_is_port_valid(dev, qp->port) && (qp->port != port)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	port_counter = &dev->port_data[port].port_counter;
+	if (!qp->counter || qp->counter->id != counter_id ||
+	    port_counter->mode.mode != RDMA_COUNTER_MODE_MANUAL) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = rdma_counter_unbind_qp(qp, false);
+
+out:
+	rdma_restrack_put(&qp->res);
+	return ret;
+}
+
 void rdma_counter_init(struct ib_device *dev)
 {
 	struct rdma_port_counter *port_counter;
include/rdma/rdma_counter.h +7 -0
index bf2c3578768f..6603e10eb352 100644
@@ -28,6 +28,7 @@ struct rdma_counter_mode {
 struct rdma_port_counter {
 	struct rdma_counter_mode mode;
 	struct rdma_hw_stats *hstats;
+	unsigned int num_counters;
 	struct mutex lock;
 };
 
@@ -51,5 +52,11 @@ int rdma_counter_unbind_qp(struct ib_qp *qp, bool force);
 
 int rdma_counter_query_stats(struct rdma_counter *counter);
 u64 rdma_counter_get_hwstat_value(struct ib_device *dev, u8 port, u32 index);
+int rdma_counter_bind_qpn(struct ib_device *dev, u8 port,
+			  u32 qp_num, u32 counter_id);
+int rdma_counter_bind_qpn_alloc(struct ib_device *dev, u8 port,
+				u32 qp_num, u32 *counter_id);
+int rdma_counter_unbind_qpn(struct ib_device *dev, u8 port,
+			    u32 qp_num, u32 counter_id);
 
 #endif /* _RDMA_COUNTER_H_ */
include/uapi/rdma/rdma_netlink.h +6 -0
index 22c5bc7a82dd..ebd728f9e351 100644
@@ -536,6 +536,12 @@ enum rdma_nl_counter_mode {
 	 */
 	RDMA_COUNTER_MODE_AUTO,
 
+	/*
+	 * Which qp are bound with which counter is explicitly specified
+	 * by the user
+	 */
+	RDMA_COUNTER_MODE_MANUAL,
+
 	/*
 	 * Always the end
 	 */
Signature · 1 line
2.20.1
#18 lore raw

PATCH v5 15/17 RDMA/nldev: Allow counter manual mode configration through RDMA netlink

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Provide an option to allow users to manually bind a qp with a counter through RDMA netlink. Limit it to users with ADMIN capability only.

---
 drivers/infiniband/core/nldev.c  | 111 +++++++++++++++++++++++++++----
 include/rdma/rdma_counter.h      |   3 +
 include/uapi/rdma/rdma_netlink.h |   2 +
 3 files changed, 103 insertions(+), 13 deletions(-)
drivers/infiniband/core/nldev.c +98 -13
index cebc15b23b15..3d750eca53d5 100644
@@ -1649,8 +1649,8 @@ static int nldev_set_sys_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 static int nldev_stat_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 			       struct netlink_ext_ack *extack)
 {
+	u32 index, port, mode, mask = 0, qpn, cntn = 0;
 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
-	u32 index, port, mode, mask = 0;
 	struct ib_device *device;
 	struct sk_buff *msg;
 	int ret;
@@ -1688,30 +1688,111 @@ static int nldev_stat_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 			0, 0);
 
 	mode = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_MODE]);
-	if (mode != RDMA_COUNTER_MODE_AUTO) {
-		ret = -EMSGSIZE;
-		goto err_msg;
+	if (mode == RDMA_COUNTER_MODE_AUTO) {
+		if (tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK])
+			mask = nla_get_u32(
+				tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK]);
+
+		ret = rdma_counter_set_auto_mode(device, port,
+						 mask ? true : false, mask);
+		if (ret)
+			goto err_msg;
+	} else {
+		qpn = nla_get_u32(tb[RDMA_NLDEV_ATTR_RES_LQPN]);
+		if (tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]) {
+			cntn = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]);
+			ret = rdma_counter_bind_qpn(device, port, qpn, cntn);
+		} else {
+			ret = rdma_counter_bind_qpn_alloc(device, port,
+							  qpn, &cntn);
+		}
+		if (ret)
+			goto err_msg;
+
+		if (fill_nldev_handle(msg, device) ||
+		    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port) ||
+		    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, cntn) ||
+		    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn)) {
+			ret = -EMSGSIZE;
+			goto err_fill;
+		}
 	}
 
-	if (tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK])
-		mask = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK]);
+	nlmsg_end(msg, nlh);
+	ib_device_put(device);
+	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
+
+err_fill:
+	rdma_counter_unbind_qpn(device, port, qpn, cntn);
+err_msg:
+	nlmsg_free(msg);
+err:
+	ib_device_put(device);
+	return ret;
+}
+
+static int nldev_stat_del_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
+			       struct netlink_ext_ack *extack)
+{
+	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
+	struct ib_device *device;
+	struct sk_buff *msg;
+	u32 index, port, qpn, cntn;
+	int ret;
 
-	ret = rdma_counter_set_auto_mode(device, port,
-					 mask ? true : false, mask);
+	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
+			  nldev_policy, extack);
+	if (ret || !tb[RDMA_NLDEV_ATTR_STAT_RES] ||
+	    !tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !tb[RDMA_NLDEV_ATTR_PORT_INDEX] ||
+	    !tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID] ||
+	    !tb[RDMA_NLDEV_ATTR_RES_LQPN])
+		return -EINVAL;
+
+	if (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES]) != RDMA_NLDEV_ATTR_RES_QP)
+		return -EINVAL;
+
+	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
+	device = ib_device_get_by_index(sock_net(skb->sk), index);
+	if (!device)
+		return -EINVAL;
+
+	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
+	if (!rdma_is_port_valid(device, port)) {
+		ret = -EINVAL;
+		goto err;
+	}
+
+	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+	if (!msg) {
+		ret = -ENOMEM;
+		goto err;
+	}
+	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
+			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
+					 RDMA_NLDEV_CMD_STAT_SET),
+			0, 0);
+
+	cntn = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]);
+	qpn = nla_get_u32(tb[RDMA_NLDEV_ATTR_RES_LQPN]);
+	ret = rdma_counter_unbind_qpn(device, port, qpn, cntn);
 	if (ret)
-		goto err_msg;
+		goto err_unbind;
 
-	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_MODE, mode) ||
-	    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK, mask)) {
+	if (fill_nldev_handle(msg, device) ||
+	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port) ||
+	    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, cntn) ||
+	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn)) {
 		ret = -EMSGSIZE;
-		goto err_msg;
+		goto err_fill;
 	}
 
 	nlmsg_end(msg, nlh);
 	ib_device_put(device);
 	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
 
-err_msg:
+err_fill:
+	rdma_counter_bind_qpn(device, port, qpn, cntn);
+err_unbind:
 	nlmsg_free(msg);
 err:
 	ib_device_put(device);
@@ -1828,6 +1909,10 @@ static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = {
 		.doit = nldev_stat_get_doit,
 		.dump = nldev_stat_get_dumpit,
 	},
+	[RDMA_NLDEV_CMD_STAT_DEL] = {
+		.doit = nldev_stat_del_doit,
+		.flags = RDMA_NL_ADMIN_PERM,
+	},
 };
 
 void __init nldev_init(void)
include/rdma/rdma_counter.h +3 -0
index 6603e10eb352..68827700ba95 100644
@@ -58,5 +58,8 @@ int rdma_counter_bind_qpn_alloc(struct ib_device *dev, u8 port,
 				u32 qp_num, u32 *counter_id);
 int rdma_counter_unbind_qpn(struct ib_device *dev, u8 port,
 			    u32 qp_num, u32 counter_id);
+int rdma_counter_get_mode(struct ib_device *dev, u8 port,
+			  enum rdma_nl_counter_mode *mode,
+			  enum rdma_nl_counter_mask *mask);
 
 #endif /* _RDMA_COUNTER_H_ */
include/uapi/rdma/rdma_netlink.h +2 -0
index ebd728f9e351..d770bd21e873 100644
@@ -285,6 +285,8 @@ enum rdma_nldev_command {
 
 	RDMA_NLDEV_CMD_STAT_GET, /* can dump */
 
+	RDMA_NLDEV_CMD_STAT_DEL,
+
 	RDMA_NLDEV_NUM_OPS
 };
 
Signature · 1 line
2.20.1
#19 lore raw

PATCH v5 16/17 RDMA/nldev: Allow get counter mode through RDMA netlink

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

Provide an option to get current counter mode through RDMA netlink.

---
 drivers/infiniband/core/counters.c | 13 ++++++
 drivers/infiniband/core/nldev.c    | 66 +++++++++++++++++++++++++++++-
 2 files changed, 78 insertions(+), 1 deletion(-)
drivers/infiniband/core/counters.c +13 -0
index 6ac8bf1240de..e924e9f7956d 100644
@@ -576,6 +576,19 @@ int rdma_counter_unbind_qpn(struct ib_device *dev, u8 port,
 	return ret;
 }
 
+int rdma_counter_get_mode(struct ib_device *dev, u8 port,
+			  enum rdma_nl_counter_mode *mode,
+			  enum rdma_nl_counter_mask *mask)
+{
+	struct rdma_port_counter *port_counter;
+
+	port_counter = &dev->port_data[port].port_counter;
+	*mode = port_counter->mode.mode;
+	*mask = port_counter->mode.mask;
+
+	return 0;
+}
+
 void rdma_counter_init(struct ib_device *dev)
 {
 	struct rdma_port_counter *port_counter;
drivers/infiniband/core/nldev.c +65 -1

78 lines not on this page · raw message

Signature · 1 line
2.20.1
#20 lore raw

PATCH v5 17/17 RDMA/nldev: Allow get default counter statistics through RDMA netlink

In reply to Leon Romanovsky

From: Mark Zhang <[email protected]>

This patch adds the ability to return the hwstats of per-port default counters (which can also be queried through sysfs nodes).

---
 drivers/infiniband/core/nldev.c | 98 ++++++++++++++++++++++++++++++++-
 drivers/infiniband/core/sysfs.c |  6 ++
 include/rdma/ib_verbs.h         |  1 +
 3 files changed, 104 insertions(+), 1 deletion(-)
drivers/infiniband/core/nldev.c +97 -1
index 4993f47b0731..a4431ed566b6 100644
@@ -1799,6 +1799,99 @@ static int nldev_stat_del_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 	return ret;
 }
 
+static int stat_get_doit_default_counter(struct sk_buff *skb,
+					 struct nlmsghdr *nlh,
+					 struct netlink_ext_ack *extack,
+					 struct nlattr *tb[])
+{
+	struct rdma_hw_stats *stats;
+	struct nlattr *table_attr;
+	struct ib_device *device;
+	int ret, num_cnts, i;
+	struct sk_buff *msg;
+	u32 index, port;
+	u64 v;
+
+	if (!tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
+		return -EINVAL;
+
+	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
+	device = ib_device_get_by_index(sock_net(skb->sk), index);
+	if (!device)
+		return -EINVAL;
+
+	if (!device->ops.alloc_hw_stats || !device->ops.get_hw_stats) {
+		ret = -EINVAL;
+		goto err;
+	}
+
+	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
+	if (!rdma_is_port_valid(device, port)) {
+		ret = -EINVAL;
+		goto err;
+	}
+
+	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+	if (!msg) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
+			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
+					 RDMA_NLDEV_CMD_STAT_GET),
+			0, 0);
+
+	if (fill_nldev_handle(msg, device) ||
+	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port)) {
+		ret = -EMSGSIZE;
+		goto err_msg;
+	}
+
+	stats = device->port_data ? device->port_data[port].hw_stats : NULL;
+	if (stats == NULL) {
+		ret = -EINVAL;
+		goto err_msg;
+	}
+	mutex_lock(&stats->lock);
+
+	num_cnts = device->ops.get_hw_stats(device, stats, port, 0);
+	if (num_cnts < 0) {
+		ret = -EINVAL;
+		goto err_stats;
+	}
+
+	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTERS);
+	if (!table_attr) {
+		ret = -EMSGSIZE;
+		goto err_stats;
+	}
+	for (i = 0; i < num_cnts; i++) {
+		v = stats->value[i] +
+			rdma_counter_get_hwstat_value(device, port, i);
+		if (fill_stat_hwcounter_entry(msg, stats->names[i], v)) {
+			ret = -EMSGSIZE;
+			goto err_table;
+		}
+	}
+	nla_nest_end(msg, table_attr);
+
+	mutex_unlock(&stats->lock);
+	nlmsg_end(msg, nlh);
+	ib_device_put(device);
+	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
+
+err_table:
+	nla_nest_cancel(msg, table_attr);
+err_stats:
+	mutex_unlock(&stats->lock);
+err_msg:
+	nlmsg_free(msg);
+err:
+	ib_device_put(device);
+	return ret;
+}
+
 static int stat_get_doit_qp(struct sk_buff *skb, struct nlmsghdr *nlh,
 			    struct netlink_ext_ack *extack, struct nlattr *tb[])
 
@@ -1871,9 +1964,12 @@ static int nldev_stat_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 
 	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
 			  nldev_policy, extack);
-	if (ret || !tb[RDMA_NLDEV_ATTR_STAT_RES])
+	if (ret)
 		return -EINVAL;
 
+	if (!tb[RDMA_NLDEV_ATTR_STAT_RES])
+		return stat_get_doit_default_counter(skb, nlh, extack, tb);
+
 	switch (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES])) {
 	case RDMA_NLDEV_ATTR_RES_QP:
 		ret = stat_get_doit_qp(skb, nlh, extack, tb);
drivers/infiniband/core/sysfs.c +6 -0
index c59b80e0a740..b477295a96c2 100644
@@ -1003,6 +1003,8 @@ static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
 			goto err;
 		port->hw_stats_ag = hsag;
 		port->hw_stats = stats;
+		if (device->port_data)
+			device->port_data[port_num].hw_stats = stats;
 	} else {
 		struct kobject *kobj = &device->dev.kobj;
 		ret = sysfs_create_group(kobj, hsag);
@@ -1293,6 +1295,8 @@ const struct attribute_group ib_dev_attr_group = {
 
 void ib_free_port_attrs(struct ib_core_device *coredev)
 {
+	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
+	bool is_full_dev = &device->coredev == coredev;
 	struct kobject *p, *t;
 
 	list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
@@ -1302,6 +1306,8 @@ void ib_free_port_attrs(struct ib_core_device *coredev)
 		if (port->hw_stats_ag)
 			free_hsag(&port->kobj, port->hw_stats_ag);
 		kfree(port->hw_stats);
+		if (device->port_data && is_full_dev)
+			device->port_data[port->port_num].hw_stats = NULL;
 
 		if (port->pma_table)
 			sysfs_remove_group(p, port->pma_table);
include/rdma/ib_verbs.h +1 -0
index 0c5151a12ae4..50806bef9f20 100644
@@ -2124,6 +2124,7 @@ struct ib_port_data {
 	struct net_device __rcu *netdev;
 	struct hlist_node ndev_hash_link;
 	struct rdma_port_counter port_counter;
+	struct rdma_hw_stats *hw_stats;
 };
 
 /* rdma netdev type - specifies protocol type */
Signature · 1 line
2.20.1
#21 lore raw
In reply to Leon Romanovsky
Quoted 68 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Changelog:
v4 -> v5:
* Patch #6 and #14 - consolidated many counter release functions, removed mutex lock protection from dealloc_counter() call and simplified kref_put/kref_get operations.
* Added Saeed's ACK tags.
v3 -> v4:
* Add counter_dealloc() callback function
* Moved to kref implementation
* Fixed lock during spinlock
v2 -> v3:
* We didn't change use of atomics over kref for management of unbind counter from QP. The reason to it that bind and unbind are non-symmetric in regards of put and get, so we need to count differently memory release flows of HW objects (restrack) and SW bind operations.
* Everything else was addressed.
v1 -> v2:
* Rebased to latest rdma-next
v0 -> v1:
* Changed wording of counter comment
* Removed unneeded assignments
* Added extra patch to present global counters

Hi,

This series from Mark provides dynamic statistics infrastructure. He uses netlink interface to configure and retrieve those counters.

This infrastructure allows to users monitor various objects by binding to them counters. As the beginning, we used QP object as target for those counters, but future patches will include ODP MR information too.

Two binding modes are supported:
- Auto: This allows a user to build automatic set of objects to a counter according to common criteria. For example in a per-type scheme, where in one process all QPs with same QP type are bound automatically to a single counter.
- Manual: This allows a user to manually bind objects on a counter.

Those two modes are mutual-exclusive with separation between processes, objects created by different processes cannot be bound to a same counter.

For objects which don't support counter binding, we will return pre-allocated counters.

$ rdma statistic qp set link mlx5_2/1 auto type on
$ rdma statistic qp set link mlx5_2/1 auto off
$ rdma statistic qp bind link mlx5_2/1 lqpn 178
$ rdma statistic qp unbind link mlx5_2/1 cntn 4 lqpn 178
$ rdma statistic show
$ rdma statistic qp mode

Thanks

Mark Zhang (17):
net/mlx5: Add rts2rts_qp_counters_set_id field in hca cap RDMA/restrack: Introduce statistic counter RDMA/restrack: Add an API to attach a task to a resource RDMA/restrack: Make is_visible_in_pid_ns() as an API RDMA/counter: Add set/clear per-port auto mode support RDMA/counter: Add "auto" configuration mode support IB/mlx5: Support set qp counter
IB/mlx5: Add counter set id as a parameter for

    mlx5_ib_query_q_counters()
Quoted 11 lines

IB/mlx5: Support statistic q counter configuration RDMA/nldev: Allow counter auto mode configration through RDMA netlink RDMA/netlink: Implement counter dumpit calback
IB/mlx5: Add counter_alloc_stats() and counter_update_stats() support RDMA/core: Get sum value of all counters when perform a sysfs stat
read
RDMA/counter: Allow manual mode configuration support RDMA/nldev: Allow counter manual mode configration through RDMA
netlink
RDMA/nldev: Allow get counter mode through RDMA netlink RDMA/nldev: Allow get default counter statistics through RDMA netlink

Well, I can made the needed edits, can you apply the the first patch to the shared branch?

Thanks,
Jason

#22 lore raw
Quoted 81 lines · Jason Gunthorpe
Quoted 68 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Changelog:
v4 -> v5:
* Patch #6 and #14 - consolidated many counter release functions, removed mutex lock protection from dealloc_counter() call and simplified kref_put/kref_get operations.
* Added Saeed's ACK tags.
v3 -> v4:
* Add counter_dealloc() callback function
* Moved to kref implementation
* Fixed lock during spinlock
v2 -> v3:
* We didn't change use of atomics over kref for management of unbind counter from QP. The reason to it that bind and unbind are non-symmetric in regards of put and get, so we need to count differently memory release flows of HW objects (restrack) and SW bind operations.
* Everything else was addressed.
v1 -> v2:
* Rebased to latest rdma-next
v0 -> v1:
* Changed wording of counter comment
* Removed unneeded assignments
* Added extra patch to present global counters

Hi,

This series from Mark provides dynamic statistics infrastructure. He uses netlink interface to configure and retrieve those counters.

This infrastructure allows to users monitor various objects by binding to them counters. As the beginning, we used QP object as target for those counters, but future patches will include ODP MR information too.

Two binding modes are supported:
- Auto: This allows a user to build automatic set of objects to a counter according to common criteria. For example in a per-type scheme, where in one process all QPs with same QP type are bound automatically to a single counter.
- Manual: This allows a user to manually bind objects on a counter.

Those two modes are mutual-exclusive with separation between processes, objects created by different processes cannot be bound to a same counter.

For objects which don't support counter binding, we will return pre-allocated counters.

$ rdma statistic qp set link mlx5_2/1 auto type on
$ rdma statistic qp set link mlx5_2/1 auto off
$ rdma statistic qp bind link mlx5_2/1 lqpn 178
$ rdma statistic qp unbind link mlx5_2/1 cntn 4 lqpn 178
$ rdma statistic show
$ rdma statistic qp mode

Thanks

Mark Zhang (17):
net/mlx5: Add rts2rts_qp_counters_set_id field in hca cap RDMA/restrack: Introduce statistic counter RDMA/restrack: Add an API to attach a task to a resource RDMA/restrack: Make is_visible_in_pid_ns() as an API RDMA/counter: Add set/clear per-port auto mode support RDMA/counter: Add "auto" configuration mode support IB/mlx5: Support set qp counter
IB/mlx5: Add counter set id as a parameter for

    mlx5_ib_query_q_counters()
Quoted 11 lines

IB/mlx5: Support statistic q counter configuration RDMA/nldev: Allow counter auto mode configration through RDMA netlink RDMA/netlink: Implement counter dumpit calback
IB/mlx5: Add counter_alloc_stats() and counter_update_stats() support RDMA/core: Get sum value of all counters when perform a sysfs stat
read
RDMA/counter: Allow manual mode configuration support RDMA/nldev: Allow counter manual mode configration through RDMA
netlink
RDMA/nldev: Allow get counter mode through RDMA netlink RDMA/nldev: Allow get default counter statistics through RDMA netlink

Quoted 2 lines

Well, I can made the needed edits, can you apply the the first patch to the shared branch?

Thanks, pushed
f8efee08dd9d net/mlx5: Add rts2rts_qp_counters_set_id field in hca cap

Quoted 3 lines

Thanks,
Jason

#23 lore raw
In reply to Leon Romanovsky
Quoted 68 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Changelog:
v4 -> v5:
* Patch #6 and #14 - consolidated many counter release functions, removed mutex lock protection from dealloc_counter() call and simplified kref_put/kref_get operations.
* Added Saeed's ACK tags.
v3 -> v4:
* Add counter_dealloc() callback function
* Moved to kref implementation
* Fixed lock during spinlock
v2 -> v3:
* We didn't change use of atomics over kref for management of unbind counter from QP. The reason to it that bind and unbind are non-symmetric in regards of put and get, so we need to count differently memory release flows of HW objects (restrack) and SW bind operations.
* Everything else was addressed.
v1 -> v2:
* Rebased to latest rdma-next
v0 -> v1:
* Changed wording of counter comment
* Removed unneeded assignments
* Added extra patch to present global counters

Hi,

This series from Mark provides dynamic statistics infrastructure. He uses netlink interface to configure and retrieve those counters.

This infrastructure allows to users monitor various objects by binding to them counters. As the beginning, we used QP object as target for those counters, but future patches will include ODP MR information too.

Two binding modes are supported:
- Auto: This allows a user to build automatic set of objects to a counter according to common criteria. For example in a per-type scheme, where in one process all QPs with same QP type are bound automatically to a single counter.
- Manual: This allows a user to manually bind objects on a counter.

Those two modes are mutual-exclusive with separation between processes, objects created by different processes cannot be bound to a same counter.

For objects which don't support counter binding, we will return pre-allocated counters.

$ rdma statistic qp set link mlx5_2/1 auto type on
$ rdma statistic qp set link mlx5_2/1 auto off
$ rdma statistic qp bind link mlx5_2/1 lqpn 178
$ rdma statistic qp unbind link mlx5_2/1 cntn 4 lqpn 178
$ rdma statistic show
$ rdma statistic qp mode

Thanks

Mark Zhang (17):
net/mlx5: Add rts2rts_qp_counters_set_id field in hca cap RDMA/restrack: Introduce statistic counter RDMA/restrack: Add an API to attach a task to a resource RDMA/restrack: Make is_visible_in_pid_ns() as an API RDMA/counter: Add set/clear per-port auto mode support RDMA/counter: Add "auto" configuration mode support IB/mlx5: Support set qp counter
IB/mlx5: Add counter set id as a parameter for

    mlx5_ib_query_q_counters()
Quoted 11 lines

IB/mlx5: Support statistic q counter configuration RDMA/nldev: Allow counter auto mode configration through RDMA netlink RDMA/netlink: Implement counter dumpit calback
IB/mlx5: Add counter_alloc_stats() and counter_update_stats() support RDMA/core: Get sum value of all counters when perform a sysfs stat
read
RDMA/counter: Allow manual mode configuration support RDMA/nldev: Allow counter manual mode configration through RDMA
netlink
RDMA/nldev: Allow get counter mode through RDMA netlink RDMA/nldev: Allow get default counter statistics through RDMA netlink

Okay, applied to for-next

Thanks,
Jason