diff mbox series

[rdma-next,10/10] RDMA/nldev: Add support to get current enabled optional counters

Message ID 20210818112428.209111-11-markzhang@nvidia.com
State New
Headers show
Series Optional counter statistics support | expand

Commit Message

Mark Zhang Aug. 18, 2021, 11:24 a.m. UTC 
From: Aharon Landau <aharonl@nvidia.com>

This patch adds the ability to show the added and supported optional
counters for each link through RDMA netlink.

Examples:

$ rdma statistic mode
link rocep8s0f0/1
    Optional-set: cc_rx_ce_pkts

$ rdma statistic mode supported
link rocep8s0f0/1
    Optional-set: cc_rx_ce_pkts cc_rx_cnp_pkts cc_tx_cnp_pkts
link rocep8s0f1/1
    Optional-set: cc_rx_ce_pkts cc_rx_cnp_pkts cc_tx_cnp_pkts

Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Signed-off-by: Neta Ostrovsky <netao@nvidia.com>
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
---
 drivers/infiniband/core/nldev.c  | 91 +++++++++++++++++++++++++++++++-
 include/uapi/rdma/rdma_netlink.h |  2 +
 2 files changed, 91 insertions(+), 2 deletions(-)

Comments

Jason Gunthorpe Aug. 23, 2021, 7:44 p.m. UTC | #1 
On Wed, Aug 18, 2021 at 02:24:28PM +0300, Mark Zhang wrote:

> diff --git a/include/uapi/rdma/rdma_netlink.h b/include/uapi/rdma/rdma_netlink.h

> index 79e6ca87d2e0..57f39d8fe434 100644

> +++ b/include/uapi/rdma/rdma_netlink.h

> @@ -557,6 +557,8 @@ enum rdma_nldev_attr {

>  	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY,	/* nested table */

>  	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_NAME,	/* string */

>  	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE,	/* u64 */

> +	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST,		/* u8 */

> +	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED,	/* u8 */


See, here - shouldn't manipulation of MODE_LIST be done by a normal
RDMA_NLDEV_CMD_STAT_SET with the new MODE_LIST array? This doesn't seem
netlinky at all..

Jason
Mark Zhang Aug. 24, 2021, 2:13 a.m. UTC | #2 
On 8/24/2021 3:44 AM, Jason Gunthorpe wrote:
> On Wed, Aug 18, 2021 at 02:24:28PM +0300, Mark Zhang wrote:

> 

>> diff --git a/include/uapi/rdma/rdma_netlink.h b/include/uapi/rdma/rdma_netlink.h

>> index 79e6ca87d2e0..57f39d8fe434 100644

>> +++ b/include/uapi/rdma/rdma_netlink.h

>> @@ -557,6 +557,8 @@ enum rdma_nldev_attr {

>>   	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY,	/* nested table */

>>   	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_NAME,	/* string */

>>   	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE,	/* u64 */

>> +	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST,		/* u8 */

>> +	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED,	/* u8 */

> 

> See, here - shouldn't manipulation of MODE_LIST be done by a normal

> RDMA_NLDEV_CMD_STAT_SET with the new MODE_LIST array? This doesn't seem

> netlinky at all..


Both of them are flags and this is a "get" operation; "MODE_LIST" asks 
kernel to return currently enabled op-counters, "MODE_LIST_SUPPORTED" 
asks kernel to return supported op-counters. Maybe the macro name are 
not good?
Jason Gunthorpe Aug. 24, 2021, 1:13 p.m. UTC | #3 
On Tue, Aug 24, 2021 at 10:13:34AM +0800, Mark Zhang wrote:
> On 8/24/2021 3:44 AM, Jason Gunthorpe wrote:

> > On Wed, Aug 18, 2021 at 02:24:28PM +0300, Mark Zhang wrote:

> > 

> > > diff --git a/include/uapi/rdma/rdma_netlink.h b/include/uapi/rdma/rdma_netlink.h

> > > index 79e6ca87d2e0..57f39d8fe434 100644

> > > +++ b/include/uapi/rdma/rdma_netlink.h

> > > @@ -557,6 +557,8 @@ enum rdma_nldev_attr {

> > >   	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY,	/* nested table */

> > >   	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_NAME,	/* string */

> > >   	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE,	/* u64 */

> > > +	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST,		/* u8 */

> > > +	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED,	/* u8 */

> > 

> > See, here - shouldn't manipulation of MODE_LIST be done by a normal

> > RDMA_NLDEV_CMD_STAT_SET with the new MODE_LIST array? This doesn't seem

> > netlinky at all..

> 

> Both of them are flags and this is a "get" operation; "MODE_LIST" asks

> kernel to return currently enabled op-counters, "MODE_LIST_SUPPORTED" asks

> kernel to return supported op-counters. Maybe the macro name are not good?


The marcors are fine, the protocol is just a bit wonky. The ADD/REMOVE
idea should only be used on top level objects, but this is a nested
sub so you should be using SET to manipulate it and it should provide
the entire current list, not a add/remove type operation.

Jason
diff mbox series

Patch

diff --git a/drivers/infiniband/core/nldev.c b/drivers/infiniband/core/nldev.c
index b665651dfb1d..611e4fe6a244 100644
--- a/drivers/infiniband/core/nldev.c
+++ b/drivers/infiniband/core/nldev.c
@@ -159,6 +159,8 @@  static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
 	[RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_NAME] = { .type = NLA_NUL_STRING,
 				  .len = RDMA_NLDEV_ATTR_OPCOUNTER_NAME_SIZE },
 	[RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE] = { .type = NLA_U64 },
+	[RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST]	= { .type = NLA_U8 },
+	[RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED] = { .type = NLA_U8 },
 };
 
 static int put_driver_name_print_type(struct sk_buff *msg, const char *name,
@@ -946,7 +948,7 @@  int rdma_nl_stat_hwcounter_entry(struct sk_buff *msg, const char *name,
 EXPORT_SYMBOL(rdma_nl_stat_hwcounter_entry);
 
 static int rdma_nl_stat_opcounter_entry(struct sk_buff *msg, const char *name,
-					u64 value)
+					u64 value, bool send_value)
 {
 	struct nlattr *entry_attr;
 
@@ -960,6 +962,12 @@  static int rdma_nl_stat_opcounter_entry(struct sk_buff *msg, const char *name,
 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE,
 			      value, RDMA_NLDEV_ATTR_PAD))
 		goto err;
+	if (send_value) {
+		if (nla_put_u64_64bit(msg,
+				      RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE,
+				      value, RDMA_NLDEV_ATTR_PAD))
+			goto err;
+	}
 
 	nla_nest_end(msg, entry_attr);
 	return 0;
@@ -2148,6 +2156,79 @@  static int nldev_stat_del_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
 	return ret;
 }
 
+static int stat_get_doit_list_opstats(struct sk_buff *skb,
+				      struct nlmsghdr *nlh,
+				      struct netlink_ext_ack *extack,
+				      struct nlattr *tb[], u32 index,
+				      struct ib_device *device, u32 port)
+{
+	struct rdma_op_stats *opstats;
+	struct nlattr *opstats_list;
+	bool list_supported = false;
+	struct sk_buff *msg;
+	int i, ret;
+
+	if (tb[RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED])
+		list_supported = true;
+
+	opstats = device->port_data[port].port_counter.opstats;
+	if (!opstats)
+		return -EOPNOTSUPP;
+
+	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;
+	}
+
+	opstats_list =
+		nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_OPCOUNTERS);
+
+	if (!opstats_list) {
+		ret = -EMSGSIZE;
+		goto err_msg;
+	}
+
+	mutex_lock(&opstats->lock);
+
+	for (i = 0; i < opstats->num_opcounters; i++) {
+		if (!opstats->opcounters[i].enabled && !list_supported)
+			continue;
+		ret = rdma_nl_stat_opcounter_entry(msg,
+						   opstats->opcounters[i].name,
+						   0, false);
+		if (ret)
+			goto err_opstats_msg;
+	}
+	nla_nest_end(msg, opstats_list);
+
+	mutex_unlock(&opstats->lock);
+
+	nlmsg_end(msg, nlh);
+	ib_device_put(device);
+	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
+
+err_opstats_msg:
+	nla_nest_cancel(msg, opstats_list);
+	mutex_unlock(&opstats->lock);
+err_msg:
+	nlmsg_free(msg);
+err:
+	ib_device_put(device);
+	return ret;
+}
+
 static int stat_get_optional_counter(struct sk_buff *msg,
 				     struct ib_device *device, u32 port)
 {
@@ -2172,7 +2253,8 @@  static int stat_get_optional_counter(struct sk_buff *msg,
 			continue;
 		ret = rdma_nl_stat_opcounter_entry(msg,
 						   opstats->opcounters[i].name,
-						   opstats->opcounters[i].value);
+						   opstats->opcounters[i].value,
+						   true);
 		if (ret)
 			goto err;
 	}
@@ -2212,6 +2294,11 @@  static int stat_get_doit_default_counter(struct sk_buff *skb,
 		goto err;
 	}
 
+	if (tb[RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST] ||
+	    tb[RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED])
+		return stat_get_doit_list_opstats(skb, nlh, extack, tb,
+						  index, device, port);
+
 	if (!device->ops.alloc_hw_port_stats || !device->ops.get_hw_stats) {
 		ret = -EINVAL;
 		goto err;
diff --git a/include/uapi/rdma/rdma_netlink.h b/include/uapi/rdma/rdma_netlink.h
index 79e6ca87d2e0..57f39d8fe434 100644
--- a/include/uapi/rdma/rdma_netlink.h
+++ b/include/uapi/rdma/rdma_netlink.h
@@ -557,6 +557,8 @@  enum rdma_nldev_attr {
 	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY,	/* nested table */
 	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_NAME,	/* string */
 	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE,	/* u64 */
+	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST,		/* u8 */
+	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED,	/* u8 */
 
 	/*
 	 * Always the end