From: "Minggang Li(Gavin)" <gavinl@nvidia.com>
To: <matan@nvidia.com>, <viacheslavo@nvidia.com>, <orika@nvidia.com>,
<thomas@monjalon.net>, Dariusz Sosnowski <dsosnowski@nvidia.com>,
Bing Zhao <bingz@nvidia.com>, Suanming Mou <suanmingm@nvidia.com>
Cc: <dev@dpdk.org>, <rasland@nvidia.com>
Subject: [PATCH 7/7] mlx5: add backward compatibility for RDMA monitor
Date: Mon, 23 Dec 2024 12:11:01 +0200 [thread overview]
Message-ID: <20241223101101.677449-8-gavinl@nvidia.com> (raw)
In-Reply-To: <20241223101101.677449-1-gavinl@nvidia.com>
Fallback to the old way to update port information if the kernel driver
does not support RDMA monitor.
Signed-off-by: Minggang Li(Gavin) <gavinl@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
doc/guides/rel_notes/release_24_11.rst | 14 +++++
drivers/common/mlx5/linux/mlx5_nl.c | 73 +++++++++++++++++++++++++
drivers/common/mlx5/version.map | 1 +
drivers/net/mlx5/linux/mlx5_ethdev_os.c | 2 +-
drivers/net/mlx5/linux/mlx5_os.c | 27 +++++++--
drivers/net/mlx5/mlx5.h | 1 +
6 files changed, 111 insertions(+), 7 deletions(-)
diff --git a/doc/guides/rel_notes/release_24_11.rst b/doc/guides/rel_notes/release_24_11.rst
index 8486cd986f..567ac42663 100644
--- a/doc/guides/rel_notes/release_24_11.rst
+++ b/doc/guides/rel_notes/release_24_11.rst
@@ -288,6 +288,20 @@ New Features
Added ability for a node to advertise and update multiple xstat counters,
that can be retrieved using ``rte_graph_cluster_stats_get``.
+* **Updated NVIDIA mlx5 driver.**
+
+ Optimized port probe in large scale.
+ This feature enhances the efficiency of probing VF/SFs on a large scale
+ by significantly reducing the probing time. To activate this feature,
+ set ``probe_opt_en`` to a non-zero value during device probing. It
+ leverages a capability from the RDMA driver, expected to be released in
+ the upcoming kernel version 6.12 or its equivalent in OFED 24.10,
+ specifically the RDMA monitor. For additional details on the limitations
+ of devargs, refer to "doc/guides/nics/mlx5.rst".
+
+ If there are lots of VFs/SFs to be probed by the application, eg, 300
+ VFs/SFs, the option should be enabled to save probing time.
+
Removed Items
-------------
diff --git a/drivers/common/mlx5/linux/mlx5_nl.c b/drivers/common/mlx5/linux/mlx5_nl.c
index ce1c2a8e75..12f1a620f3 100644
--- a/drivers/common/mlx5/linux/mlx5_nl.c
+++ b/drivers/common/mlx5/linux/mlx5_nl.c
@@ -2152,3 +2152,76 @@ mlx5_nl_rdma_monitor_info_get(struct nlmsghdr *hdr, struct mlx5_nl_port_info *da
error:
rte_errno = EINVAL;
}
+
+static int
+mlx5_nl_rdma_monitor_cap_get_cb(struct nlmsghdr *hdr, void *arg)
+{
+ size_t off = NLMSG_HDRLEN;
+ uint8_t *cap = arg;
+
+ if (hdr->nlmsg_type != RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_SYS_GET))
+ goto error;
+
+ *cap = 0;
+ while (off < hdr->nlmsg_len) {
+ struct nlattr *na = (void *)((uintptr_t)hdr + off);
+ void *payload = (void *)((uintptr_t)na + NLA_HDRLEN);
+
+ if (na->nla_len > hdr->nlmsg_len - off)
+ goto error;
+ switch (na->nla_type) {
+ case RDMA_NLDEV_SYS_ATTR_MONITOR_MODE:
+ *cap = *(uint8_t *)payload;
+ return 0;
+ default:
+ break;
+ }
+ off += NLA_ALIGN(na->nla_len);
+ }
+
+ return 0;
+
+error:
+ return -EINVAL;
+}
+
+/**
+ * Get RDMA monitor support in driver.
+ *
+ *
+ * @param nl
+ * Netlink socket of the RDMA kind (NETLINK_RDMA).
+ * @param[out] cap
+ * Pointer to port info.
+ * @return
+ * 0 on success, negative on error and rte_errno is set.
+ */
+int
+mlx5_nl_rdma_monitor_cap_get(int nl, uint8_t *cap)
+{
+ union {
+ struct nlmsghdr nh;
+ uint8_t buf[NLMSG_HDRLEN];
+ } req = {
+ .nh = {
+ .nlmsg_len = NLMSG_LENGTH(0),
+ .nlmsg_type = RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
+ RDMA_NLDEV_CMD_SYS_GET),
+ .nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK,
+ },
+ };
+ uint32_t sn = MLX5_NL_SN_GENERATE;
+ int ret;
+
+ ret = mlx5_nl_send(nl, &req.nh, sn);
+ if (ret < 0) {
+ rte_errno = -ret;
+ return ret;
+ }
+ ret = mlx5_nl_recv(nl, sn, mlx5_nl_rdma_monitor_cap_get_cb, cap);
+ if (ret < 0) {
+ rte_errno = -ret;
+ return ret;
+ }
+ return 0;
+}
diff --git a/drivers/common/mlx5/version.map b/drivers/common/mlx5/version.map
index 5230576006..8301485839 100644
--- a/drivers/common/mlx5/version.map
+++ b/drivers/common/mlx5/version.map
@@ -148,6 +148,7 @@ INTERNAL {
mlx5_nl_vlan_vmwa_delete; # WINDOWS_NO_EXPORT
mlx5_nl_rdma_monitor_init; # WINDOWS_NO_EXPORT
mlx5_nl_rdma_monitor_info_get; # WINDOWS_NO_EXPORT
+ mlx5_nl_rdma_monitor_cap_get; # WINDOWS_NO_EXPORT
mlx5_os_umem_dereg;
mlx5_os_umem_reg;
diff --git a/drivers/net/mlx5/linux/mlx5_ethdev_os.c b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
index 5156d96b3a..6b2c25a7c2 100644
--- a/drivers/net/mlx5/linux/mlx5_ethdev_os.c
+++ b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
@@ -736,7 +736,7 @@ mlx5_dev_interrupt_nl_cb(struct nlmsghdr *hdr, void *cb_arg)
if (mlx5_nl_parse_link_status_update(hdr, &if_index) < 0)
return;
- if (sh->cdev->config.probe_opt && sh->cdev->dev_info.port_num > 1)
+ if (sh->cdev->config.probe_opt && sh->cdev->dev_info.port_num > 1 && !sh->rdma_monitor_supp)
mlx5_handle_port_info_update(&sh->cdev->dev_info, if_index, hdr->nlmsg_type);
for (i = 0; i < sh->max_port; i++) {
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 16b275c71e..d3fd77af58 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -3017,6 +3017,7 @@ mlx5_os_dev_shared_handler_install(struct mlx5_dev_ctx_shared *sh)
{
struct ibv_context *ctx = sh->cdev->ctx;
int nlsk_fd;
+ uint8_t rdma_monitor_supp = 0;
sh->intr_handle = mlx5_os_interrupt_handler_create
(RTE_INTR_INSTANCE_F_SHARED, true,
@@ -3025,20 +3026,34 @@ mlx5_os_dev_shared_handler_install(struct mlx5_dev_ctx_shared *sh)
DRV_LOG(ERR, "Failed to allocate intr_handle.");
return;
}
- if (sh->cdev->config.probe_opt && sh->cdev->dev_info.port_num > 1) {
+ if (sh->cdev->config.probe_opt &&
+ sh->cdev->dev_info.port_num > 1 &&
+ !sh->rdma_monitor_supp) {
nlsk_fd = mlx5_nl_rdma_monitor_init();
if (nlsk_fd < 0) {
DRV_LOG(ERR, "Failed to create a socket for RDMA Netlink events: %s",
rte_strerror(rte_errno));
return;
}
- sh->intr_handle_ib = mlx5_os_interrupt_handler_create
- (RTE_INTR_INSTANCE_F_SHARED, true,
- nlsk_fd, mlx5_dev_interrupt_handler_ib, sh);
- if (sh->intr_handle_ib == NULL) {
- DRV_LOG(ERR, "Fail to allocate intr_handle");
+ if (mlx5_nl_rdma_monitor_cap_get(nlsk_fd, &rdma_monitor_supp)) {
+ DRV_LOG(ERR, "Failed to query RDMA monitor support: %s",
+ rte_strerror(rte_errno));
+ close(nlsk_fd);
return;
}
+ sh->rdma_monitor_supp = rdma_monitor_supp;
+ if (sh->rdma_monitor_supp) {
+ sh->intr_handle_ib = mlx5_os_interrupt_handler_create
+ (RTE_INTR_INSTANCE_F_SHARED, true,
+ nlsk_fd, mlx5_dev_interrupt_handler_ib, sh);
+ if (sh->intr_handle_ib == NULL) {
+ DRV_LOG(ERR, "Fail to allocate intr_handle");
+ close(nlsk_fd);
+ return;
+ }
+ } else {
+ close(nlsk_fd);
+ }
}
nlsk_fd = mlx5_nl_init(NETLINK_ROUTE, RTMGRP_LINK);
if (nlsk_fd < 0) {
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 688a7270ca..ab604042b9 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -1545,6 +1545,7 @@ struct mlx5_dev_ctx_shared {
uint32_t lag_rx_port_affinity_en:1;
/* lag_rx_port_affinity is supported. */
uint32_t hws_max_log_bulk_sz:5;
+ uint32_t rdma_monitor_supp:1;
/* Log of minimal HWS counters created hard coded. */
uint32_t hws_max_nb_counters; /* Maximal number for HWS counters. */
uint32_t max_port; /* Maximal IB device port index. */
--
2.34.1
prev parent reply other threads:[~2024-12-23 10:12 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-12-23 10:10 [PATCH V1 0/7] port probe time optimization Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 1/7] mailmap: update user name Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 2/7] net/mlx5: optimize device probing Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 3/7] net/mlx5: add new devargs to control probe optimization Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 4/7] common/mlx5: fix Netlink socket leak Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 5/7] common/mlx5: add RDMA monitor event awareness Minggang Li(Gavin)
2024-12-23 10:11 ` [PATCH 6/7] mlx5: use RDMA Netlink to update port information Minggang Li(Gavin)
2024-12-23 10:11 ` Minggang Li(Gavin) [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20241223101101.677449-8-gavinl@nvidia.com \
--to=gavinl@nvidia.com \
--cc=bingz@nvidia.com \
--cc=dev@dpdk.org \
--cc=dsosnowski@nvidia.com \
--cc=matan@nvidia.com \
--cc=orika@nvidia.com \
--cc=rasland@nvidia.com \
--cc=suanmingm@nvidia.com \
--cc=thomas@monjalon.net \
--cc=viacheslavo@nvidia.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).