[ewg] [PATCH-2.6.24 2/2 v2] [RFC] ib/cm: add basic performance counters
Sean Hefty
sean.hefty at intel.com
Fri Oct 5 14:31:00 PDT 2007
Add performance/debug counters to track sent/received messages, retries,
and duplicates. Counters are tracked per CM message type, per port.
The counters are always enabled, so intrusive state tracking is not done.
Counters are exported as:
/sys/class/infiniband_cm/device/port/counter_description/cm_attribute
for example:
/sys/class/infiniband_cm/mthca0/1/cm_tx_msgs/req
/sys/class/infiniband_cm/mthca0/1/cm_tx_retries/rep
Signed-off-by: Sean Hefty <sean.hefty at intel.com>
---
>From v1:
This moves the counters from debugfs to sysfs. Everything works fine for me,
but I'm not entirely sure if I'm using the kobject stuff in the best way.
This still depends on the ib_mad changes to export the number of retries.
There were no changes to that patch, so I'm not re-sending it at this time.
Is there still the possibility of getting this into OFED 1.3? Did feature
freeze occur on 10/3, or was it pushed out because of vacations?
drivers/infiniband/core/cm.c | 294 +++++++++++++++++++++++++++++++++++++++--
drivers/infiniband/core/ucm.c | 37 ++---
2 files changed, 296 insertions(+), 35 deletions(-)
diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c
index 2e39236..790149e 100644
--- a/drivers/infiniband/core/cm.c
+++ b/drivers/infiniband/core/cm.c
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2004-2006 Intel Corporation. All rights reserved.
+ * Copyright (c) 2004-2007 Intel Corporation. All rights reserved.
* Copyright (c) 2004 Topspin Corporation. All rights reserved.
* Copyright (c) 2004, 2005 Voltaire Corporation. All rights reserved.
* Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
@@ -37,12 +37,14 @@
#include <linux/completion.h>
#include <linux/dma-mapping.h>
+#include <linux/device.h>
#include <linux/err.h>
#include <linux/idr.h>
#include <linux/interrupt.h>
#include <linux/random.h>
#include <linux/rbtree.h>
#include <linux/spinlock.h>
+#include <linux/sysfs.h>
#include <linux/workqueue.h>
#include <rdma/ib_cache.h>
@@ -78,15 +80,92 @@ static struct ib_cm {
struct workqueue_struct *wq;
} cm;
+/* Counter indexes ordered by attribute ID */
+enum {
+ CM_REQ_COUNTER,
+ CM_MRA_COUNTER,
+ CM_REJ_COUNTER,
+ CM_REP_COUNTER,
+ CM_RTU_COUNTER,
+ CM_DREQ_COUNTER,
+ CM_DREP_COUNTER,
+ CM_SIDR_REQ_COUNTER,
+ CM_SIDR_REP_COUNTER,
+ CM_LAP_COUNTER,
+ CM_APR_COUNTER,
+ CM_ATTR_COUNT,
+ CM_ATTR_ID_OFFSET = 0x0010,
+};
+
+enum {
+ CM_XMIT,
+ CM_XMIT_RETRIES,
+ CM_RECV,
+ CM_RECV_DUPLICATES,
+ CM_COUNTER_GROUPS
+};
+
+static char const counter_group_names[CM_COUNTER_GROUPS]
+ [sizeof("cm_rx_duplicates")] = {
+ "cm_tx_msgs", "cm_tx_retries",
+ "cm_rx_msgs", "cm_rx_duplicates"
+};
+
+struct cm_counter_group {
+ struct kobject obj;
+ atomic_long_t counter[CM_ATTR_COUNT];
+};
+
+struct cm_counter_attribute {
+ struct attribute attr;
+ int index;
+};
+
+#define CM_COUNTER_ATTR(_name, _index) \
+struct cm_counter_attribute cm_##_name##_counter_attr = { \
+ .attr = {.name = __stringify(_name), .mode = 0444, .owner = THIS_MODULE}, \
+ .index = _index \
+}
+
+static CM_COUNTER_ATTR(req, CM_REQ_COUNTER);
+static CM_COUNTER_ATTR(mra, CM_MRA_COUNTER);
+static CM_COUNTER_ATTR(rej, CM_REJ_COUNTER);
+static CM_COUNTER_ATTR(rep, CM_REP_COUNTER);
+static CM_COUNTER_ATTR(rtu, CM_RTU_COUNTER);
+static CM_COUNTER_ATTR(dreq, CM_DREQ_COUNTER);
+static CM_COUNTER_ATTR(drep, CM_DREP_COUNTER);
+static CM_COUNTER_ATTR(sidr_req, CM_SIDR_REQ_COUNTER);
+static CM_COUNTER_ATTR(sidr_rep, CM_SIDR_REP_COUNTER);
+static CM_COUNTER_ATTR(lap, CM_LAP_COUNTER);
+static CM_COUNTER_ATTR(apr, CM_APR_COUNTER);
+
+static struct attribute *cm_counter_default_attrs[] = {
+ &cm_req_counter_attr.attr,
+ &cm_mra_counter_attr.attr,
+ &cm_rej_counter_attr.attr,
+ &cm_rep_counter_attr.attr,
+ &cm_rtu_counter_attr.attr,
+ &cm_dreq_counter_attr.attr,
+ &cm_drep_counter_attr.attr,
+ &cm_sidr_req_counter_attr.attr,
+ &cm_sidr_rep_counter_attr.attr,
+ &cm_lap_counter_attr.attr,
+ &cm_apr_counter_attr.attr,
+ NULL
+};
+
struct cm_port {
struct cm_device *cm_dev;
struct ib_mad_agent *mad_agent;
+ struct kobject port_obj;
u8 port_num;
+ struct cm_counter_group counter_group[CM_COUNTER_GROUPS];
};
struct cm_device {
struct list_head list;
struct ib_device *device;
+ struct kobject dev_obj;
u8 ack_delay;
struct cm_port port[0];
};
@@ -1270,6 +1349,9 @@ static void cm_dup_req_handler(struct cm_work *work,
struct ib_mad_send_buf *msg = NULL;
int ret;
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_REQ_COUNTER]);
+
/* Quick state check to discard duplicate REQs. */
if (cm_id_priv->id.state == IB_CM_REQ_RCVD)
return;
@@ -1616,6 +1698,8 @@ static void cm_dup_rep_handler(struct cm_work *work)
if (!cm_id_priv)
return;
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_REP_COUNTER]);
ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg);
if (ret)
goto deref;
@@ -1781,6 +1865,8 @@ static int cm_rtu_handler(struct cm_work *work)
if (cm_id_priv->id.state != IB_CM_REP_SENT &&
cm_id_priv->id.state != IB_CM_MRA_REP_RCVD) {
spin_unlock_irq(&cm_id_priv->lock);
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_RTU_COUNTER]);
goto out;
}
cm_id_priv->id.state = IB_CM_ESTABLISHED;
@@ -1958,6 +2044,8 @@ static int cm_dreq_handler(struct cm_work *work)
cm_id_priv = cm_acquire_id(dreq_msg->remote_comm_id,
dreq_msg->local_comm_id);
if (!cm_id_priv) {
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_DREQ_COUNTER]);
cm_issue_drep(work->port, work->mad_recv_wc);
return -EINVAL;
}
@@ -1977,6 +2065,8 @@ static int cm_dreq_handler(struct cm_work *work)
case IB_CM_MRA_REP_RCVD:
break;
case IB_CM_TIMEWAIT:
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_DREQ_COUNTER]);
if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
goto unlock;
@@ -1988,6 +2078,10 @@ static int cm_dreq_handler(struct cm_work *work)
if (ib_post_send_mad(msg, NULL))
cm_free_msg(msg);
goto deref;
+ case IB_CM_DREQ_RCVD:
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_DREQ_COUNTER]);
+ goto unlock;
default:
goto unlock;
}
@@ -2339,10 +2433,20 @@ static int cm_mra_handler(struct cm_work *work)
if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_OTHER ||
cm_id_priv->id.lap_state != IB_CM_LAP_SENT ||
ib_modify_mad(cm_id_priv->av.port->mad_agent,
- cm_id_priv->msg, timeout))
+ cm_id_priv->msg, timeout)) {
+ if (cm_id_priv->id.lap_state == IB_CM_MRA_LAP_RCVD)
+ atomic_long_inc(&work->port->
+ counter_group[CM_RECV_DUPLICATES].
+ counter[CM_MRA_COUNTER]);
goto out;
+ }
cm_id_priv->id.lap_state = IB_CM_MRA_LAP_RCVD;
break;
+ case IB_CM_MRA_REQ_RCVD:
+ case IB_CM_MRA_REP_RCVD:
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_MRA_COUNTER]);
+ /* fall through */
default:
goto out;
}
@@ -2502,6 +2606,8 @@ static int cm_lap_handler(struct cm_work *work)
case IB_CM_LAP_IDLE:
break;
case IB_CM_MRA_LAP_SENT:
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_LAP_COUNTER]);
if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
goto unlock;
@@ -2515,6 +2621,10 @@ static int cm_lap_handler(struct cm_work *work)
if (ib_post_send_mad(msg, NULL))
cm_free_msg(msg);
goto deref;
+ case IB_CM_LAP_RCVD:
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_LAP_COUNTER]);
+ goto unlock;
default:
goto unlock;
}
@@ -2796,6 +2906,8 @@ static int cm_sidr_req_handler(struct cm_work *work)
cur_cm_id_priv = cm_insert_remote_sidr(cm_id_priv);
if (cur_cm_id_priv) {
spin_unlock_irq(&cm.lock);
+ atomic_long_inc(&work->port->counter_group[CM_RECV_DUPLICATES].
+ counter[CM_SIDR_REQ_COUNTER]);
goto out; /* Duplicate message. */
}
cm_id_priv->id.state = IB_CM_SIDR_REQ_RCVD;
@@ -2990,6 +3102,27 @@ static void cm_send_handler(struct ib_mad_agent *mad_agent,
struct ib_mad_send_wc *mad_send_wc)
{
struct ib_mad_send_buf *msg = mad_send_wc->send_buf;
+ struct cm_port *port;
+ u16 attr_index;
+
+ port = mad_agent->context;
+ attr_index = be16_to_cpu(((struct ib_mad_hdr *)
+ msg->mad)->attr_id) - CM_ATTR_ID_OFFSET;
+
+ /*
+ * If the send was in response to a received message (context[0] is not
+ * set to a cm_id), and is not a REJ, then it is a send that was
+ * manually retried.
+ */
+ if (!msg->context[0] && (attr_index != CM_REJ_COUNTER))
+ msg->retries = 1;
+
+ atomic_long_add(1 + msg->retries,
+ &port->counter_group[CM_XMIT].counter[attr_index]);
+ if (msg->retries)
+ atomic_long_add(msg->retries,
+ &port->counter_group[CM_XMIT_RETRIES].
+ counter[attr_index]);
switch (mad_send_wc->status) {
case IB_WC_SUCCESS:
@@ -3148,8 +3281,10 @@ EXPORT_SYMBOL(ib_cm_notify);
static void cm_recv_handler(struct ib_mad_agent *mad_agent,
struct ib_mad_recv_wc *mad_recv_wc)
{
+ struct cm_port *port = mad_agent->context;
struct cm_work *work;
enum ib_cm_event_type event;
+ u16 attr_id;
int paths = 0;
switch (mad_recv_wc->recv_buf.mad->mad_hdr.attr_id) {
@@ -3194,6 +3329,10 @@ static void cm_recv_handler(struct ib_mad_agent *mad_agent,
return;
}
+ attr_id = be16_to_cpu(mad_recv_wc->recv_buf.mad->mad_hdr.attr_id);
+ atomic_long_inc(&port->counter_group[CM_RECV].
+ counter[attr_id - CM_ATTR_ID_OFFSET]);
+
work = kmalloc(sizeof *work + sizeof(struct ib_sa_path_rec) * paths,
GFP_KERNEL);
if (!work) {
@@ -3204,7 +3343,7 @@ static void cm_recv_handler(struct ib_mad_agent *mad_agent,
INIT_DELAYED_WORK(&work->work, cm_work_handler);
work->cm_event.event = event;
work->mad_recv_wc = mad_recv_wc;
- work->port = (struct cm_port *)mad_agent->context;
+ work->port = port;
queue_delayed_work(cm.wq, &work->work, 0);
}
@@ -3379,6 +3518,110 @@ static void cm_get_ack_delay(struct cm_device *cm_dev)
cm_dev->ack_delay = attr.local_ca_ack_delay;
}
+static ssize_t cm_show_counter(struct kobject *obj, struct attribute *attr,
+ char *buf)
+{
+ struct cm_counter_group *group;
+ struct cm_counter_attribute *cm_attr;
+
+ group = container_of(obj, struct cm_counter_group, obj);
+ cm_attr = container_of(attr, struct cm_counter_attribute, attr);
+
+ return sprintf(buf, "%ld\n",
+ atomic_long_read(&group->counter[cm_attr->index]));
+}
+
+static struct sysfs_ops cm_counter_ops = {
+ .show = cm_show_counter
+};
+
+static struct kobj_type cm_counter_obj_type = {
+ .sysfs_ops = &cm_counter_ops,
+ .default_attrs = cm_counter_default_attrs
+};
+
+static void cm_release_dev_obj(struct kobject *obj)
+{
+ struct cm_device *cm_dev;
+
+ cm_dev = container_of(obj, struct cm_device, dev_obj);
+ kfree(cm_dev);
+}
+
+static struct kobj_type cm_dev_obj_type = {
+ .release = cm_release_dev_obj
+};
+
+static struct class cm_class = {
+ .name = "infiniband_cm",
+};
+EXPORT_SYMBOL(cm_class);
+
+static int cm_add_fs_obj(struct kobject *obj, struct kobject *parent,
+ struct kobj_type *type, const char *name)
+{
+ int ret;
+
+ ret = kobject_set_name(obj, "%s", name);
+ if (ret)
+ return ret;
+
+ obj->ktype = type;
+ obj->parent = kobject_get(parent);
+ if (!obj->parent)
+ return -EBUSY;
+
+ ret = kobject_register(obj);
+ if (ret)
+ kobject_put(parent);
+
+ return ret;
+}
+
+static void cm_remove_fs_obj(struct kobject *obj)
+{
+ kobject_put(obj->parent);
+ kobject_unregister(obj);
+}
+
+static int cm_create_port_fs(struct cm_port *port)
+{
+ char port_name[8];
+ int i, ret;
+
+ snprintf(port_name, sizeof port_name, "%d", port->port_num);
+ ret = cm_add_fs_obj(&port->port_obj, &port->cm_dev->dev_obj,
+ NULL, port_name);
+ if (ret)
+ return ret;
+
+ for (i = 0; i < CM_COUNTER_GROUPS; i++) {
+ ret = cm_add_fs_obj(&port->counter_group[i].obj, &port->port_obj,
+ &cm_counter_obj_type, counter_group_names[i]);
+ if (ret)
+ goto error;
+ }
+
+ return 0;
+
+error:
+ while (i--)
+ cm_remove_fs_obj(&port->counter_group[i].obj);
+ cm_remove_fs_obj(&port->port_obj);
+ return ret;
+
+}
+
+static void cm_remove_port_fs(struct cm_port *port)
+{
+ int i;
+
+ for (i = 0; i < CM_COUNTER_GROUPS; i++)
+ cm_remove_fs_obj(&port->counter_group[i].obj);
+
+ cm_remove_fs_obj(&port->port_obj);
+}
+
static void cm_add_one(struct ib_device *device)
{
struct cm_device *cm_dev;
@@ -3397,7 +3640,7 @@ static void cm_add_one(struct ib_device *device)
if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
return;
- cm_dev = kmalloc(sizeof(*cm_dev) + sizeof(*port) *
+ cm_dev = kzalloc(sizeof(*cm_dev) + sizeof(*port) *
device->phys_port_cnt, GFP_KERNEL);
if (!cm_dev)
return;
@@ -3405,11 +3648,23 @@ static void cm_add_one(struct ib_device *device)
cm_dev->device = device;
cm_get_ack_delay(cm_dev);
+ ret = cm_add_fs_obj(&cm_dev->dev_obj, &cm_class.subsys.kobj,
+ &cm_dev_obj_type, device->name);
+ if (ret) {
+ kfree(cm_dev);
+ return;
+ }
+
set_bit(IB_MGMT_METHOD_SEND, reg_req.method_mask);
for (i = 1; i <= device->phys_port_cnt; i++) {
port = &cm_dev->port[i-1];
port->cm_dev = cm_dev;
port->port_num = i;
+
+ ret = cm_create_port_fs(port);
+ if (ret)
+ goto error1;
+
port->mad_agent = ib_register_mad_agent(device, i,
IB_QPT_GSI,
®_req,
@@ -3418,11 +3673,11 @@ static void cm_add_one(struct ib_device *device)
cm_recv_handler,
port);
if (IS_ERR(port->mad_agent))
- goto error1;
+ goto error2;
ret = ib_modify_port(device, i, 0, &port_modify);
if (ret)
- goto error2;
+ goto error3;
}
ib_set_client_data(device, &cm_client, cm_dev);
@@ -3431,8 +3686,10 @@ static void cm_add_one(struct ib_device *device)
write_unlock_irqrestore(&cm.device_lock, flags);
return;
-error2:
+error3:
ib_unregister_mad_agent(port->mad_agent);
+error2:
+ cm_remove_port_fs(port);
error1:
port_modify.set_port_cap_mask = 0;
port_modify.clr_port_cap_mask = IB_PORT_CM_SUP;
@@ -3440,8 +3697,9 @@ error1:
port = &cm_dev->port[i-1];
ib_modify_port(device, port->port_num, 0, &port_modify);
ib_unregister_mad_agent(port->mad_agent);
+ cm_remove_port_fs(port);
}
- kfree(cm_dev);
+ cm_remove_fs_obj(&cm_dev->dev_obj);
}
static void cm_remove_one(struct ib_device *device)
@@ -3466,8 +3724,9 @@ static void cm_remove_one(struct ib_device *device)
port = &cm_dev->port[i-1];
ib_modify_port(device, port->port_num, 0, &port_modify);
ib_unregister_mad_agent(port->mad_agent);
+ cm_remove_port_fs(port);
}
- kfree(cm_dev);
+ cm_remove_fs_obj(&cm_dev->dev_obj);
}
static int __init ib_cm_init(void)
@@ -3488,17 +3747,25 @@ static int __init ib_cm_init(void)
idr_pre_get(&cm.local_id_table, GFP_KERNEL);
INIT_LIST_HEAD(&cm.timewait_list);
- cm.wq = create_workqueue("ib_cm");
- if (!cm.wq)
+ ret = class_register(&cm_class);
+ if (ret)
return -ENOMEM;
+ cm.wq = create_workqueue("ib_cm");
+ if (!cm.wq) {
+ ret = -ENOMEM;
+ goto error1;
+ }
+
ret = ib_register_client(&cm_client);
if (ret)
- goto error;
+ goto error2;
return 0;
-error:
+error2:
destroy_workqueue(cm.wq);
+error1:
+ class_unregister(&cm_class);
return ret;
}
@@ -3519,6 +3786,7 @@ static void __exit ib_cm_cleanup(void)
}
ib_unregister_client(&cm_client);
+ class_unregister(&cm_class);
idr_destroy(&cm.local_id_table);
}
diff --git a/drivers/infiniband/core/ucm.c b/drivers/infiniband/core/ucm.c
index 424983f..4291ab4 100644
--- a/drivers/infiniband/core/ucm.c
+++ b/drivers/infiniband/core/ucm.c
@@ -106,6 +106,9 @@ enum {
IB_UCM_MAX_DEVICES = 32
};
+/* ib_cm and ib_user_cm modules share /sys/class/infiniband_cm */
+extern struct class cm_class;
+
#define IB_UCM_BASE_DEV MKDEV(IB_UCM_MAJOR, IB_UCM_BASE_MINOR)
static void ib_ucm_add_one(struct ib_device *device);
@@ -1199,7 +1202,7 @@ static int ib_ucm_close(struct inode *inode, struct file *filp)
return 0;
}
-static void ib_ucm_release_class_dev(struct class_device *class_dev)
+static void ucm_release_class_dev(struct class_device *class_dev)
{
struct ib_ucm_device *dev;
@@ -1217,11 +1220,6 @@ static const struct file_operations ucm_fops = {
.poll = ib_ucm_poll,
};
-static struct class ucm_class = {
- .name = "infiniband_cm",
- .release = ib_ucm_release_class_dev
-};
-
static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
{
struct ib_ucm_device *dev;
@@ -1257,9 +1255,10 @@ static void ib_ucm_add_one(struct ib_device *device)
if (cdev_add(&ucm_dev->dev, IB_UCM_BASE_DEV + ucm_dev->devnum, 1))
goto err;
- ucm_dev->class_dev.class = &ucm_class;
+ ucm_dev->class_dev.class = &cm_class;
ucm_dev->class_dev.dev = device->dma_device;
ucm_dev->class_dev.devt = ucm_dev->dev.dev;
+ ucm_dev->class_dev.release = ucm_release_class_dev;
snprintf(ucm_dev->class_dev.class_id, BUS_ID_SIZE, "ucm%d",
ucm_dev->devnum);
if (class_device_register(&ucm_dev->class_dev))
@@ -1306,40 +1305,34 @@ static int __init ib_ucm_init(void)
"infiniband_cm");
if (ret) {
printk(KERN_ERR "ucm: couldn't register device number\n");
- goto err;
+ goto error1;
}
- ret = class_register(&ucm_class);
- if (ret) {
- printk(KERN_ERR "ucm: couldn't create class infiniband_cm\n");
- goto err_chrdev;
- }
-
- ret = class_create_file(&ucm_class, &class_attr_abi_version);
+ ret = class_create_file(&cm_class, &class_attr_abi_version);
if (ret) {
printk(KERN_ERR "ucm: couldn't create abi_version attribute\n");
- goto err_class;
+ goto error2;
}
ret = ib_register_client(&ucm_client);
if (ret) {
printk(KERN_ERR "ucm: couldn't register client\n");
- goto err_class;
+ goto error3;
}
return 0;
-err_class:
- class_unregister(&ucm_class);
-err_chrdev:
+error3:
+ class_remove_file(&cm_class, &class_attr_abi_version);
+error2:
unregister_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES);
-err:
+error1:
return ret;
}
static void __exit ib_ucm_cleanup(void)
{
ib_unregister_client(&ucm_client);
- class_unregister(&ucm_class);
+ class_remove_file(&cm_class, &class_attr_abi_version);
unregister_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES);
idr_destroy(&ctx_id_table);
}
More information about the ewg
mailing list