RDMA/bnxt_re: Optimize bnxt_qplib_get_dev_attr function

Optimize bnxt_qplib_get_dev_attr() by separating out query_version which
uses creq notification method to host. Due to serialization of cmdq by
firmware, expected latency in response to heavy multi-threaded rdma
applications might be observed.

This patch separates the version_query logic out of device attribute
query and called only during rdma driver init.

Signed-off-by: Damodharam Ammepalli <damodharam.ammepalli@broadcom.com>
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Link: https://patch.msgid.link/20250814112555.221665-6-kalesh-anakkur.purayil@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 059cc1a151650d61ac29d062729dde8af0f537c3..10dd498bd20658aa868293796c209f0a5770a0b8 100644 (file)
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -2178,6 +2178,7 @@ static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 op_type)
        if (rc)
                goto disable_rcfw;
 
+       bnxt_qplib_query_version(&rdev->rcfw);
        bnxt_re_set_resource_limits(rdev);
 
        rc = bnxt_qplib_alloc_ctx(&rdev->qplib_res, &rdev->qplib_ctx, 0,

diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
index 68981399598d8d6b9732043467b5d89e2110fa19..b602a1da19cd8f5c750e5f50087309a0a2e67685 100644 (file)
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -66,14 +66,15 @@ static bool bnxt_qplib_is_atomic_cap(struct bnxt_qplib_rcfw *rcfw)
        return (pcie_ctl2 & PCI_EXP_DEVCTL2_ATOMIC_REQ);
 }
 
-static void bnxt_qplib_query_version(struct bnxt_qplib_rcfw *rcfw,
-                                    char *fw_ver)
+void bnxt_qplib_query_version(struct bnxt_qplib_rcfw *rcfw)
 {
        struct creq_query_version_resp resp = {};
        struct bnxt_qplib_cmdqmsg msg = {};
        struct cmdq_query_version req = {};
+       struct bnxt_qplib_dev_attr *attr;
        int rc;
 
+       attr = rcfw->res->dattr;
        bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
                                 CMDQ_BASE_OPCODE_QUERY_VERSION,
                                 sizeof(req));
@@ -82,10 +83,10 @@ static void bnxt_qplib_query_version(struct bnxt_qplib_rcfw *rcfw,
        rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
        if (rc)
                return;
-       fw_ver[0] = resp.fw_maj;
-       fw_ver[1] = resp.fw_minor;
-       fw_ver[2] = resp.fw_bld;
-       fw_ver[3] = resp.fw_rsvd;
+       attr->fw_ver[0] = resp.fw_maj;
+       attr->fw_ver[1] = resp.fw_minor;
+       attr->fw_ver[2] = resp.fw_bld;
+       attr->fw_ver[3] = resp.fw_rsvd;
 }
 
 int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw)
@@ -179,8 +180,6 @@ int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw)
        if (_is_max_srq_ext_supported(attr->dev_cap_flags2))
                attr->max_srq += le16_to_cpu(sb->max_srq_ext);
 
-       bnxt_qplib_query_version(rcfw, attr->fw_ver);
-
        for (i = 0; i < MAX_TQM_ALLOC_REQ / 4; i++) {
                temp = le32_to_cpu(sb->tqm_alloc_reqs[i]);
                tqm_alloc = (u8 *)&temp;

diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.h b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
index 09faf4a1e849cd0ef403164384e446468c4819e5..e9834e7fc38363ede9dd4b8495151b4393fd7f07 100644 (file)
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
@@ -358,6 +358,7 @@ int bnxt_qplib_read_context(struct bnxt_qplib_rcfw *rcfw, u8 type, u32 xid,
                            u32 resp_size, void *resp_va);
 int bnxt_qplib_query_cc_param(struct bnxt_qplib_res *res,
                              struct bnxt_qplib_cc_param *cc_param);
+void bnxt_qplib_query_version(struct bnxt_qplib_rcfw *rcfw);
 
 #define BNXT_VAR_MAX_WQE       4352
 #define BNXT_VAR_MAX_SLOT_ALIGN 256