From: Rongwei Liu <rongw...@nvidia.com>

Add a new devarg probe_opt_en to control probe optimization
in PMD.

By default, the value is 0 and no behavior changed.

Signed-off-by: Rongwei Liu <rongw...@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viachesl...@nvidia.com>
---
 doc/guides/nics/mlx5.rst                |  7 +++++++
 drivers/common/mlx5/linux/mlx5_nl.c     | 12 ++++++++----
 drivers/common/mlx5/mlx5_common.c       | 15 +++++++++++++++
 drivers/common/mlx5/mlx5_common.h       |  2 ++
 drivers/net/mlx5/linux/mlx5_ethdev_os.c |  5 ++++-
 drivers/net/mlx5/linux/mlx5_os.c        |  2 +-
 6 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 1dccdaad50..b4a4e57cde 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -1436,6 +1436,13 @@ for an additional list of options shared with other mlx5 
drivers.
 
   By default, the PMD will set this value to 1.
 
+- ``probe_opt_en`` parameter [int]
+
+  A non-zero value optimizes the probe process, especially for large scale.
+  PMD will hold the IB device information internally and reuse it.
+
+  By default, the PMD will set this value to 0.
+
 - ``lacp_by_user`` parameter [int]
 
   A nonzero value enables the control of LACP traffic by the user application.
diff --git a/drivers/common/mlx5/linux/mlx5_nl.c 
b/drivers/common/mlx5/linux/mlx5_nl.c
index e98073aafe..745e443f8f 100644
--- a/drivers/common/mlx5/linux/mlx5_nl.c
+++ b/drivers/common/mlx5/linux/mlx5_nl.c
@@ -1148,7 +1148,7 @@ mlx5_nl_ifindex(int nl, const char *name, uint32_t 
pindex, struct mlx5_dev_info
                        .flags = 0,
        };
 
-       if (!strcmp(name, dev_info->ibname)) {
+       if (dev_info->probe_opt && !strcmp(name, dev_info->ibname)) {
                if (dev_info->port_info && pindex <= dev_info->port_num &&
                    dev_info->port_info[pindex].valid) {
                        if (!dev_info->port_info[pindex].ifindex)
@@ -1161,7 +1161,7 @@ mlx5_nl_ifindex(int nl, const char *name, uint32_t 
pindex, struct mlx5_dev_info
 
        ret = mlx5_nl_port_info(nl, pindex, &data);
 
-       if (!strcmp(dev_info->ibname, name)) {
+       if (dev_info->probe_opt && !strcmp(dev_info->ibname, name)) {
                if ((!ret || ret == -ENODEV) && dev_info->port_info &&
                    pindex <= dev_info->port_num) {
                        if (!ret)
@@ -1201,7 +1201,8 @@ mlx5_nl_port_state(int nl, const char *name, uint32_t 
pindex, struct mlx5_dev_in
                        .ibindex = UINT32_MAX,
        };
 
-       if (dev_info && !strcmp(name, dev_info->ibname) && dev_info->port_num)
+       if (dev_info && dev_info->probe_opt &&
+           !strcmp(name, dev_info->ibname) && dev_info->port_num)
                data.ibindex = dev_info->ibindex;
        if (mlx5_nl_port_info(nl, pindex, &data) < 0)
                return -rte_errno;
@@ -1244,7 +1245,8 @@ mlx5_nl_portnum(int nl, const char *name, struct 
mlx5_dev_info *dev_info)
        uint32_t sn = MLX5_NL_SN_GENERATE;
        int ret, size;
 
-       if (dev_info->port_num && !strcmp(name, dev_info->ibname))
+       if (dev_info->probe_opt && dev_info->port_num &&
+           !strcmp(name, dev_info->ibname))
                return dev_info->port_num;
 
        ret = mlx5_nl_send(nl, &req, sn);
@@ -1263,6 +1265,8 @@ mlx5_nl_portnum(int nl, const char *name, struct 
mlx5_dev_info *dev_info)
                rte_errno = EINVAL;
                return 0;
        }
+       if (!dev_info->probe_opt)
+               return data.portnum;
        MLX5_ASSERT(!strlen(dev_info->ibname));
        dev_info->port_num = data.portnum;
        dev_info->ibindex = data.ibindex;
diff --git a/drivers/common/mlx5/mlx5_common.c 
b/drivers/common/mlx5/mlx5_common.c
index 0aaae91c31..9abae4a374 100644
--- a/drivers/common/mlx5/mlx5_common.c
+++ b/drivers/common/mlx5/mlx5_common.c
@@ -40,6 +40,9 @@ uint8_t haswell_broadwell_cpu;
 /* The default memory allocator used in PMD. */
 #define MLX5_SYS_MEM_EN "sys_mem_en"
 
+/* Probe optimization in PMD. */
+#define MLX5_PROBE_OPT "probe_opt_en"
+
 /*
  * Device parameter to force doorbell register mapping
  * to non-cached region eliminating the extra write memory barrier.
@@ -295,6 +298,8 @@ mlx5_common_args_check_handler(const char *key, const char 
*val, void *opaque)
                config->device_fd = tmp;
        } else if (strcmp(key, MLX5_PD_HANDLE) == 0) {
                config->pd_handle = tmp;
+       } else if (strcmp(key, MLX5_PROBE_OPT) == 0) {
+               config->probe_opt = !!tmp;
        }
        return 0;
 }
@@ -324,6 +329,7 @@ mlx5_common_config_get(struct mlx5_kvargs_ctrl *mkvlist,
                MLX5_MR_MEMPOOL_REG_EN,
                MLX5_DEVICE_FD,
                MLX5_PD_HANDLE,
+               MLX5_PROBE_OPT,
                NULL,
        };
        int ret = 0;
@@ -332,6 +338,7 @@ mlx5_common_config_get(struct mlx5_kvargs_ctrl *mkvlist,
        config->mr_ext_memseg_en = 1;
        config->mr_mempool_reg_en = 1;
        config->sys_mem_en = 0;
+       config->probe_opt = 0;
        config->dbnc = MLX5_ARG_UNSET;
        config->device_fd = MLX5_ARG_UNSET;
        config->pd_handle = MLX5_ARG_UNSET;
@@ -351,6 +358,7 @@ mlx5_common_config_get(struct mlx5_kvargs_ctrl *mkvlist,
        DRV_LOG(DEBUG, "mr_ext_memseg_en is %u.", config->mr_ext_memseg_en);
        DRV_LOG(DEBUG, "mr_mempool_reg_en is %u.", config->mr_mempool_reg_en);
        DRV_LOG(DEBUG, "sys_mem_en is %u.", config->sys_mem_en);
+       DRV_LOG(DEBUG, "probe_opt_en is %u.", config->probe_opt);
        DRV_LOG(DEBUG, "Send Queue doorbell mapping parameter is %d.",
                config->dbnc);
        return ret;
@@ -791,6 +799,7 @@ mlx5_common_dev_create(struct rte_device *eal_dev, uint32_t 
classes,
        if (TAILQ_EMPTY(&devices_list))
                rte_mem_event_callback_register("MLX5_MEM_EVENT_CB",
                                                mlx5_mr_mem_event_cb, NULL);
+       cdev->dev_info.probe_opt = cdev->config.probe_opt;
 exit:
        pthread_mutex_lock(&devices_list_lock);
        TAILQ_INSERT_HEAD(&devices_list, cdev, next);
@@ -880,6 +889,12 @@ mlx5_common_probe_again_args_validate(struct 
mlx5_common_device *cdev,
                        cdev->dev->name);
                goto error;
        }
+       if (cdev->config.probe_opt != config->probe_opt) {
+               DRV_LOG(ERR, "\"" MLX5_PROBE_OPT"\" "
+                       "configuration mismatch for device %s.",
+                       cdev->dev->name);
+               goto error;
+       }
        if (cdev->config.dbnc != config->dbnc) {
                DRV_LOG(ERR, "\"" MLX5_SQ_DB_NC "\" "
                        "configuration mismatch for device %s.",
diff --git a/drivers/common/mlx5/mlx5_common.h 
b/drivers/common/mlx5/mlx5_common.h
index 6cb40f54dd..f1b59d6f07 100644
--- a/drivers/common/mlx5/mlx5_common.h
+++ b/drivers/common/mlx5/mlx5_common.h
@@ -183,6 +183,7 @@ struct mlx5_dev_info {
        uint32_t port_num;
        uint32_t ibindex;
        char ibname[MLX5_FS_NAME_MAX];
+       uint8_t probe_opt;
        struct mlx5_port_nl_info *port_info;
 };
 
@@ -525,6 +526,7 @@ struct mlx5_common_dev_config {
        int pd_handle; /* Protection Domain handle for importation.  */
        unsigned int devx:1; /* Whether devx interface is available or not. */
        unsigned int sys_mem_en:1; /* The default memory allocator. */
+       unsigned int probe_opt:1; /* Optimize probing . */
        unsigned int mr_mempool_reg_en:1;
        /* Allow/prevent implicit mempool memory registration. */
        unsigned int mr_ext_memseg_en:1;
diff --git a/drivers/net/mlx5/linux/mlx5_ethdev_os.c 
b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
index 08ac6dd939..88d3c57c6e 100644
--- a/drivers/net/mlx5/linux/mlx5_ethdev_os.c
+++ b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
@@ -691,6 +691,8 @@ mlx5_handle_port_info_update(struct mlx5_dev_info 
*dev_info, uint32_t if_index,
        if (dev_info->port_num <= 1 || dev_info->port_info == NULL)
                return;
 
+       DRV_LOG(DEBUG, "IB device %s ifindex %u received netlink event %u",
+                       dev_info->ibname, if_index, msg_type);
        for (i = 1; i <= dev_info->port_num; i++) {
                if (!dev_info->port_info[i].valid)
                        continue;
@@ -734,7 +736,8 @@ mlx5_dev_interrupt_nl_cb(struct nlmsghdr *hdr, void *cb_arg)
 
        if (mlx5_nl_parse_link_status_update(hdr, &if_index) < 0)
                return;
-       mlx5_handle_port_info_update(&sh->cdev->dev_info, if_index, 
hdr->nlmsg_type);
+       if (sh->cdev->config.probe_opt && sh->cdev->dev_info.port_num > 1)
+               mlx5_handle_port_info_update(&sh->cdev->dev_info, if_index, 
hdr->nlmsg_type);
 
        for (i = 0; i < sh->max_port; i++) {
                struct mlx5_dev_shared_port *port = &sh->port[i];
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index dcf1ff917b..a408790d1e 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -2335,7 +2335,7 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
        while (ret-- > 0) {
                struct rte_pci_addr pci_addr;
 
-               if (cdev->dev_info.port_num) {
+               if (cdev->config.probe_opt && cdev->dev_info.port_num) {
                        if (strcmp(ibv_list[ret]->name, cdev->dev_info.ibname)) 
{
                                DRV_LOG(INFO, "Unmatched caching device \"%s\" 
\"%s\"",
                                        cdev->dev_info.ibname, 
ibv_list[ret]->name);
-- 
2.34.1

Reply via email to