net/bnxt: fix PF support in SR-IOV mode
authorVenkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Fri, 9 Oct 2020 11:11:21 +0000 (16:41 +0530)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 16 Oct 2020 17:47:58 +0000 (19:47 +0200)
1. Implement HWRM_FUNC_VF_RESOURCE_CFG command and use it to
   reserve resources for VFs when NEW RM is enabled.
2. Invoke PF’s FUNC_CFG before configuring VFs resources.
3. Don’t consider max_rx_em_flows in max_l2_ctx calculation
   when VFs are configured.
4. Issue HWRM_FUNC_QCFG instead of HWRM_FUNC_QCAPS to find
   out the actual allocated resources for VF.
5. Don’t add random mac to the VF.
6. Handle completion type CMPL_BASE_TYPE_HWRM_FWD_REQ instead
   of CMPL_BASE_TYPE_HWRM_FWD_RESP.
7. Don't enable HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_FWD_NONE_MODE
   when the list of HWRM commands that needs to be forwarded
   to the PF is specified in HWRM_FUNC_DRV_RGTR.
8. Update the HWRM commands list that can be forwarded to the
   PF.

Fixes: b7778e8a1c00 ("net/bnxt: refactor to properly allocate resources for PF/VF")
Cc: stable@dpdk.org
Signed-off-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
drivers/net/bnxt/bnxt.h
drivers/net/bnxt/bnxt_cpr.c
drivers/net/bnxt/bnxt_ethdev.c
drivers/net/bnxt/bnxt_hwrm.c
drivers/net/bnxt/bnxt_hwrm.h

index eca7448..a951bca 100644 (file)
 #define        BNXT_DEFAULT_VNIC_CHANGE_VF_ID_SFT              \
        HWRM_ASYNC_EVENT_CMPL_DEFAULT_VNIC_CHANGE_EVENT_DATA1_VF_ID_SFT
 
+#define BNXT_HWRM_CMD_TO_FORWARD(cmd)  \
+               (bp->pf->vf_req_fwd[(cmd) / 32] |= (1 << ((cmd) % 32)))
+
 struct bnxt_led_info {
        uint8_t      num_leds;
        uint8_t      led_id;
@@ -664,9 +667,10 @@ struct bnxt {
 #define BNXT_FW_CAP_IF_CHANGE          BIT(1)
 #define BNXT_FW_CAP_ERROR_RECOVERY     BIT(2)
 #define BNXT_FW_CAP_ERR_RECOVER_RELOAD BIT(3)
+#define BNXT_FW_CAP_HCOMM_FW_STATUS    BIT(4)
 #define BNXT_FW_CAP_ADV_FLOW_MGMT      BIT(5)
 #define BNXT_FW_CAP_ADV_FLOW_COUNTERS  BIT(6)
-#define BNXT_FW_CAP_HCOMM_FW_STATUS    BIT(7)
+#define BNXT_FW_CAP_LINK_ADMIN         BIT(7)
 
        pthread_mutex_t         flow_lock;
 
index a3a7e6a..5492394 100644 (file)
@@ -239,7 +239,7 @@ void bnxt_handle_fwd_req(struct bnxt *bp, struct cmpl_base *cmpl)
                goto reject;
        }
 
-       if (bnxt_rcv_msg_from_vf(bp, vf_id, fwd_cmd) == true) {
+       if (bnxt_rcv_msg_from_vf(bp, vf_id, fwd_cmd)) {
                /*
                 * In older firmware versions, the MAC had to be all zeros for
                 * the VF to set it's MAC via hwrm_func_vf_cfg. Set to all
@@ -254,6 +254,7 @@ void bnxt_handle_fwd_req(struct bnxt *bp, struct cmpl_base *cmpl)
                                (const uint8_t *)"\x00\x00\x00\x00\x00");
                        }
                }
+
                if (fwd_cmd->req_type == HWRM_CFA_L2_SET_RX_MASK) {
                        struct hwrm_cfa_l2_set_rx_mask_input *srm =
                                                        (void *)fwd_cmd;
@@ -265,6 +266,7 @@ void bnxt_handle_fwd_req(struct bnxt *bp, struct cmpl_base *cmpl)
                            HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_VLAN_NONVLAN |
                            HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_ANYVLAN_NONVLAN);
                }
+
                /* Forward */
                rc = bnxt_hwrm_exec_fwd_resp(bp, fw_vf_id, fwd_cmd, req_len);
                if (rc) {
@@ -306,7 +308,7 @@ int bnxt_event_hwrm_resp_handler(struct bnxt *bp, struct cmpl_base *cmp)
                bnxt_handle_async_event(bp, cmp);
                evt = 1;
                break;
-       case CMPL_BASE_TYPE_HWRM_FWD_RESP:
+       case CMPL_BASE_TYPE_HWRM_FWD_REQ:
                /* Handle HWRM forwarded responses */
                bnxt_handle_fwd_req(bp, cmp);
                evt = 1;
index 8b63134..b4654ec 100644 (file)
@@ -5208,37 +5208,14 @@ static void bnxt_config_vf_req_fwd(struct bnxt *bp)
        if (!BNXT_PF(bp))
                return;
 
-#define ALLOW_FUNC(x)  \
-       { \
-               uint32_t arg = (x); \
-               bp->pf->vf_req_fwd[((arg) >> 5)] &= \
-               ~rte_cpu_to_le_32(1 << ((arg) & 0x1f)); \
-       }
-
-       /* Forward all requests if firmware is new enough */
-       if (((bp->fw_ver >= ((20 << 24) | (6 << 16) | (100 << 8))) &&
-            (bp->fw_ver < ((20 << 24) | (7 << 16)))) ||
-           ((bp->fw_ver >= ((20 << 24) | (8 << 16))))) {
-               memset(bp->pf->vf_req_fwd, 0xff, sizeof(bp->pf->vf_req_fwd));
-       } else {
-               PMD_DRV_LOG(WARNING,
-                           "Firmware too old for VF mailbox functionality\n");
-               memset(bp->pf->vf_req_fwd, 0, sizeof(bp->pf->vf_req_fwd));
-       }
+       memset(bp->pf->vf_req_fwd, 0, sizeof(bp->pf->vf_req_fwd));
 
-       /*
-        * The following are used for driver cleanup. If we disallow these,
-        * VF drivers can't clean up cleanly.
-        */
-       ALLOW_FUNC(HWRM_FUNC_DRV_UNRGTR);
-       ALLOW_FUNC(HWRM_VNIC_FREE);
-       ALLOW_FUNC(HWRM_RING_FREE);
-       ALLOW_FUNC(HWRM_RING_GRP_FREE);
-       ALLOW_FUNC(HWRM_VNIC_RSS_COS_LB_CTX_FREE);
-       ALLOW_FUNC(HWRM_CFA_L2_FILTER_FREE);
-       ALLOW_FUNC(HWRM_STAT_CTX_FREE);
-       ALLOW_FUNC(HWRM_PORT_PHY_QCFG);
-       ALLOW_FUNC(HWRM_VNIC_TPA_CFG);
+       if (!(bp->fw_cap & BNXT_FW_CAP_LINK_ADMIN))
+               BNXT_HWRM_CMD_TO_FORWARD(HWRM_PORT_PHY_QCFG);
+       BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_CFG);
+       BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_VF_CFG);
+       BNXT_HWRM_CMD_TO_FORWARD(HWRM_CFA_L2_FILTER_ALLOC);
+       BNXT_HWRM_CMD_TO_FORWARD(HWRM_OEM_CMD);
 }
 
 uint16_t
@@ -6189,7 +6166,10 @@ bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
 
        bnxt_free_int(bp);
        bnxt_free_mem(bp, reconfig_dev);
+
        bnxt_hwrm_func_buf_unrgtr(bp);
+       rte_free(bp->pf->vf_req_buf);
+
        rc = bnxt_hwrm_func_driver_unregister(bp, 0);
        bp->flags &= ~BNXT_FLAG_REGISTERED;
        bnxt_free_ctx_mem(bp);
index faeaf4b..8133afc 100644 (file)
@@ -765,7 +765,7 @@ static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
        bp->first_vf_id = rte_le_to_cpu_16(resp->first_vf_id);
        bp->max_rx_em_flows = rte_le_to_cpu_16(resp->max_rx_em_flows);
        bp->max_l2_ctx = rte_le_to_cpu_16(resp->max_l2_ctxs);
-       if (!BNXT_CHIP_THOR(bp))
+       if (!BNXT_CHIP_THOR(bp) && !bp->pdev->max_vfs)
                bp->max_l2_ctx += bp->max_rx_em_flows;
        /* TODO: For now, do not support VMDq/RFS on VFs. */
        if (BNXT_PF(bp)) {
@@ -803,6 +803,9 @@ static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
        if (flags & HWRM_FUNC_QCAPS_OUTPUT_FLAGS_HOT_RESET_CAPABLE)
                bp->fw_cap |= BNXT_FW_CAP_HOT_RESET;
 
+       if (flags & HWRM_FUNC_QCAPS_OUTPUT_FLAGS_LINK_ADMIN_STATUS_SUPPORTED)
+               bp->fw_cap |= BNXT_FW_CAP_LINK_ADMIN;
+
        HWRM_UNLOCK();
 
        return rc;
@@ -818,16 +821,15 @@ int bnxt_hwrm_func_qcaps(struct bnxt *bp)
                if (rc)
                        return rc;
 
+               /* On older FW,
+                * bnxt_hwrm_func_resc_qcaps can fail and cause init failure.
+                * But the error can be ignored. Return success.
+                */
                rc = bnxt_hwrm_func_resc_qcaps(bp);
                if (!rc)
                        bp->flags |= BNXT_FLAG_NEW_RM;
        }
 
-       /* On older FW,
-        * bnxt_hwrm_func_resc_qcaps can fail and cause init failure.
-        * But the error can be ignored. Return success.
-        */
-
        return 0;
 }
 
@@ -916,14 +918,6 @@ int bnxt_hwrm_func_driver_register(struct bnxt *bp)
                memcpy(req.vf_req_fwd, bp->pf->vf_req_fwd,
                       RTE_MIN(sizeof(req.vf_req_fwd),
                               sizeof(bp->pf->vf_req_fwd)));
-
-               /*
-                * PF can sniff HWRM API issued by VF. This can be set up by
-                * linux driver and inherited by the DPDK PF driver. Clear
-                * this HWRM sniffer list in FW because DPDK PF driver does
-                * not support this.
-                */
-               flags |= HWRM_FUNC_DRV_RGTR_INPUT_FLAGS_FWD_NONE_MODE;
        }
 
        req.flags = rte_cpu_to_le_32(flags);
@@ -1052,21 +1046,19 @@ int bnxt_hwrm_func_resc_qcaps(struct bnxt *bp)
 
        HWRM_CHECK_RESULT_SILENT();
 
-       if (BNXT_VF(bp)) {
-               bp->max_rsscos_ctx = rte_le_to_cpu_16(resp->max_rsscos_ctx);
-               bp->max_cp_rings = rte_le_to_cpu_16(resp->max_cmpl_rings);
-               bp->max_tx_rings = rte_le_to_cpu_16(resp->max_tx_rings);
-               bp->max_rx_rings = rte_le_to_cpu_16(resp->max_rx_rings);
-               bp->max_ring_grps = rte_le_to_cpu_32(resp->max_hw_ring_grps);
-               /* func_resource_qcaps does not return max_rx_em_flows.
-                * So use the value provided by func_qcaps.
-                */
-               bp->max_l2_ctx = rte_le_to_cpu_16(resp->max_l2_ctxs);
-               if (!BNXT_CHIP_THOR(bp))
-                       bp->max_l2_ctx += bp->max_rx_em_flows;
-               bp->max_vnics = rte_le_to_cpu_16(resp->max_vnics);
-               bp->max_stat_ctx = rte_le_to_cpu_16(resp->max_stat_ctx);
-       }
+       bp->max_rsscos_ctx = rte_le_to_cpu_16(resp->max_rsscos_ctx);
+       bp->max_cp_rings = rte_le_to_cpu_16(resp->max_cmpl_rings);
+       bp->max_tx_rings = rte_le_to_cpu_16(resp->max_tx_rings);
+       bp->max_rx_rings = rte_le_to_cpu_16(resp->max_rx_rings);
+       bp->max_ring_grps = rte_le_to_cpu_32(resp->max_hw_ring_grps);
+       /* func_resource_qcaps does not return max_rx_em_flows.
+        * So use the value provided by func_qcaps.
+        */
+       bp->max_l2_ctx = rte_le_to_cpu_16(resp->max_l2_ctxs);
+       if (!BNXT_CHIP_THOR(bp) && !bp->pdev->max_vfs)
+               bp->max_l2_ctx += bp->max_rx_em_flows;
+       bp->max_vnics = rte_le_to_cpu_16(resp->max_vnics);
+       bp->max_stat_ctx = rte_le_to_cpu_16(resp->max_stat_ctx);
        bp->max_nq_rings = rte_le_to_cpu_16(resp->max_msix);
        bp->vf_resv_strategy = rte_le_to_cpu_16(resp->vf_reservation_strategy);
        if (bp->vf_resv_strategy >
@@ -3300,33 +3292,8 @@ int bnxt_hwrm_port_mac_qcfg(struct bnxt *bp)
        return 0;
 }
 
-static void copy_func_cfg_to_qcaps(struct hwrm_func_cfg_input *fcfg,
-                                  struct hwrm_func_qcaps_output *qcaps)
-{
-       qcaps->max_rsscos_ctx = fcfg->num_rsscos_ctxs;
-       memcpy(qcaps->mac_address, fcfg->dflt_mac_addr,
-              sizeof(qcaps->mac_address));
-       qcaps->max_l2_ctxs = fcfg->num_l2_ctxs;
-       qcaps->max_rx_rings = fcfg->num_rx_rings;
-       qcaps->max_tx_rings = fcfg->num_tx_rings;
-       qcaps->max_cmpl_rings = fcfg->num_cmpl_rings;
-       qcaps->max_stat_ctx = fcfg->num_stat_ctxs;
-       qcaps->max_vfs = 0;
-       qcaps->first_vf_id = 0;
-       qcaps->max_vnics = fcfg->num_vnics;
-       qcaps->max_decap_records = 0;
-       qcaps->max_encap_records = 0;
-       qcaps->max_tx_wm_flows = 0;
-       qcaps->max_tx_em_flows = 0;
-       qcaps->max_rx_wm_flows = 0;
-       qcaps->max_rx_em_flows = 0;
-       qcaps->max_flow_id = 0;
-       qcaps->max_mcast_filters = fcfg->num_mcast_filters;
-       qcaps->max_sp_tx_rings = 0;
-       qcaps->max_hw_ring_grps = fcfg->num_hw_ring_grps;
-}
-
-static int bnxt_hwrm_pf_func_cfg(struct bnxt *bp, int tx_rings)
+static int bnxt_hwrm_pf_func_cfg(struct bnxt *bp,
+                                struct bnxt_pf_resource_info *pf_resc)
 {
        struct hwrm_func_cfg_input req = {0};
        struct hwrm_func_cfg_output *resp = bp->hwrm_cmd_resp_addr;
@@ -3345,7 +3312,8 @@ static int bnxt_hwrm_pf_func_cfg(struct bnxt *bp, int tx_rings)
 
        if (BNXT_HAS_RING_GRPS(bp)) {
                enables |= HWRM_FUNC_CFG_INPUT_ENABLES_NUM_HW_RING_GRPS;
-               req.num_hw_ring_grps = rte_cpu_to_le_16(bp->max_ring_grps);
+               req.num_hw_ring_grps =
+                       rte_cpu_to_le_16(pf_resc->num_hw_ring_grps);
        } else if (BNXT_HAS_NQ(bp)) {
                enables |= HWRM_FUNC_CFG_INPUT_ENABLES_NUM_MSIX;
                req.num_msix = rte_cpu_to_le_16(bp->max_nq_rings);
@@ -3354,12 +3322,12 @@ static int bnxt_hwrm_pf_func_cfg(struct bnxt *bp, int tx_rings)
        req.flags = rte_cpu_to_le_32(bp->pf->func_cfg_flags);
        req.mtu = rte_cpu_to_le_16(BNXT_MAX_MTU);
        req.mru = rte_cpu_to_le_16(BNXT_VNIC_MRU(bp->eth_dev->data->mtu));
-       req.num_rsscos_ctxs = rte_cpu_to_le_16(bp->max_rsscos_ctx);
-       req.num_stat_ctxs = rte_cpu_to_le_16(bp->max_stat_ctx);
-       req.num_cmpl_rings = rte_cpu_to_le_16(bp->max_cp_rings);
-       req.num_tx_rings = rte_cpu_to_le_16(tx_rings);
-       req.num_rx_rings = rte_cpu_to_le_16(bp->max_rx_rings);
-       req.num_l2_ctxs = rte_cpu_to_le_16(bp->max_l2_ctx);
+       req.num_rsscos_ctxs = rte_cpu_to_le_16(pf_resc->num_rsscos_ctxs);
+       req.num_stat_ctxs = rte_cpu_to_le_16(pf_resc->num_stat_ctxs);
+       req.num_cmpl_rings = rte_cpu_to_le_16(pf_resc->num_cp_rings);
+       req.num_tx_rings = rte_cpu_to_le_16(pf_resc->num_tx_rings);
+       req.num_rx_rings = rte_cpu_to_le_16(pf_resc->num_rx_rings);
+       req.num_l2_ctxs = rte_cpu_to_le_16(pf_resc->num_l2_ctxs);
        req.num_vnics = rte_cpu_to_le_16(bp->max_vnics);
        req.fid = rte_cpu_to_le_16(0xffff);
        req.enables = rte_cpu_to_le_32(enables);
@@ -3374,9 +3342,43 @@ static int bnxt_hwrm_pf_func_cfg(struct bnxt *bp, int tx_rings)
        return rc;
 }
 
-static void populate_vf_func_cfg_req(struct bnxt *bp,
-                                    struct hwrm_func_cfg_input *req,
-                                    int num_vfs)
+/* min values are the guaranteed resources and max values are subject
+ * to availability. The strategy for now is to keep both min & max
+ * values the same.
+ */
+static void
+bnxt_fill_vf_func_cfg_req_new(struct bnxt *bp,
+                             struct hwrm_func_vf_resource_cfg_input *req,
+                             int num_vfs)
+{
+       req->max_rsscos_ctx = rte_cpu_to_le_16(bp->max_rsscos_ctx /
+                                              (num_vfs + 1));
+       req->min_rsscos_ctx = req->max_rsscos_ctx;
+       req->max_stat_ctx = rte_cpu_to_le_16(bp->max_stat_ctx / (num_vfs + 1));
+       req->min_stat_ctx = req->max_stat_ctx;
+       req->max_cmpl_rings = rte_cpu_to_le_16(bp->max_cp_rings /
+                                              (num_vfs + 1));
+       req->min_cmpl_rings = req->max_cmpl_rings;
+       req->max_tx_rings = rte_cpu_to_le_16(bp->max_tx_rings / (num_vfs + 1));
+       req->min_tx_rings = req->max_tx_rings;
+       req->max_rx_rings = rte_cpu_to_le_16(bp->max_rx_rings / (num_vfs + 1));
+       req->min_rx_rings = req->max_rx_rings;
+       req->max_l2_ctxs = rte_cpu_to_le_16(bp->max_l2_ctx / (num_vfs + 1));
+       req->min_l2_ctxs = req->max_l2_ctxs;
+       /* TODO: For now, do not support VMDq/RFS on VFs. */
+       req->max_vnics = rte_cpu_to_le_16(1);
+       req->min_vnics = req->max_vnics;
+       req->max_hw_ring_grps = rte_cpu_to_le_16(bp->max_ring_grps /
+                                                (num_vfs + 1));
+       req->min_hw_ring_grps = req->max_hw_ring_grps;
+       req->flags =
+        rte_cpu_to_le_16(HWRM_FUNC_VF_RESOURCE_CFG_INPUT_FLAGS_MIN_GUARANTEED);
+}
+
+static void
+bnxt_fill_vf_func_cfg_req_old(struct bnxt *bp,
+                             struct hwrm_func_cfg_input *req,
+                             int num_vfs)
 {
        req->enables = rte_cpu_to_le_32(HWRM_FUNC_CFG_INPUT_ENABLES_MTU |
                        HWRM_FUNC_CFG_INPUT_ENABLES_MRU |
@@ -3407,60 +3409,29 @@ static void populate_vf_func_cfg_req(struct bnxt *bp,
                                                 (num_vfs + 1));
 }
 
-static void add_random_mac_if_needed(struct bnxt *bp,
-                                    struct hwrm_func_cfg_input *cfg_req,
-                                    int vf)
-{
-       struct rte_ether_addr mac;
-
-       if (bnxt_hwrm_func_qcfg_vf_default_mac(bp, vf, &mac))
-               return;
-
-       if (memcmp(mac.addr_bytes, "\x00\x00\x00\x00\x00", 6) == 0) {
-               cfg_req->enables |=
-               rte_cpu_to_le_32(HWRM_FUNC_CFG_INPUT_ENABLES_DFLT_MAC_ADDR);
-               rte_eth_random_addr(cfg_req->dflt_mac_addr);
-               bp->pf->vf_info[vf].random_mac = true;
-       } else {
-               memcpy(cfg_req->dflt_mac_addr, mac.addr_bytes,
-                       RTE_ETHER_ADDR_LEN);
-       }
-}
-
-static int reserve_resources_from_vf(struct bnxt *bp,
-                                    struct hwrm_func_cfg_input *cfg_req,
+/* Update the port wide resource values based on how many resources
+ * got allocated to the VF.
+ */
+static int bnxt_update_max_resources(struct bnxt *bp,
                                     int vf)
 {
-       struct hwrm_func_qcaps_input req = {0};
-       struct hwrm_func_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
+       struct hwrm_func_qcfg_input req = {0};
+       struct hwrm_func_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
        int rc;
 
        /* Get the actual allocated values now */
-       HWRM_PREP(&req, HWRM_FUNC_QCAPS, BNXT_USE_CHIMP_MB);
+       HWRM_PREP(&req, HWRM_FUNC_QCFG, BNXT_USE_CHIMP_MB);
        req.fid = rte_cpu_to_le_16(bp->pf->vf_info[vf].fid);
        rc = bnxt_hwrm_send_message(bp, &req, sizeof(req), BNXT_USE_CHIMP_MB);
+       HWRM_CHECK_RESULT();
 
-       if (rc) {
-               PMD_DRV_LOG(ERR, "hwrm_func_qcaps failed rc:%d\n", rc);
-               copy_func_cfg_to_qcaps(cfg_req, resp);
-       } else if (resp->error_code) {
-               rc = rte_le_to_cpu_16(resp->error_code);
-               PMD_DRV_LOG(ERR, "hwrm_func_qcaps error %d\n", rc);
-               copy_func_cfg_to_qcaps(cfg_req, resp);
-       }
-
-       bp->max_rsscos_ctx -= rte_le_to_cpu_16(resp->max_rsscos_ctx);
-       bp->max_stat_ctx -= rte_le_to_cpu_16(resp->max_stat_ctx);
-       bp->max_cp_rings -= rte_le_to_cpu_16(resp->max_cmpl_rings);
-       bp->max_tx_rings -= rte_le_to_cpu_16(resp->max_tx_rings);
-       bp->max_rx_rings -= rte_le_to_cpu_16(resp->max_rx_rings);
-       bp->max_l2_ctx -= rte_le_to_cpu_16(resp->max_l2_ctxs);
-       /*
-        * TODO: While not supporting VMDq with VFs, max_vnics is always
-        * forced to 1 in this case
-        */
-       //bp->max_vnics -= rte_le_to_cpu_16(esp->max_vnics);
-       bp->max_ring_grps -= rte_le_to_cpu_16(resp->max_hw_ring_grps);
+       bp->max_rsscos_ctx -= rte_le_to_cpu_16(resp->alloc_rsscos_ctx);
+       bp->max_stat_ctx -= rte_le_to_cpu_16(resp->alloc_stat_ctx);
+       bp->max_cp_rings -= rte_le_to_cpu_16(resp->alloc_cmpl_rings);
+       bp->max_tx_rings -= rte_le_to_cpu_16(resp->alloc_tx_rings);
+       bp->max_rx_rings -= rte_le_to_cpu_16(resp->alloc_rx_rings);
+       bp->max_l2_ctx -= rte_le_to_cpu_16(resp->alloc_l2_ctx);
+       bp->max_ring_grps -= rte_le_to_cpu_16(resp->alloc_hw_ring_grps);
 
        HWRM_UNLOCK();
 
@@ -3485,7 +3456,8 @@ int bnxt_hwrm_func_qcfg_current_vf_vlan(struct bnxt *bp, int vf)
        return rc;
 }
 
-static int update_pf_resource_max(struct bnxt *bp)
+static int bnxt_query_pf_resources(struct bnxt *bp,
+                                  struct bnxt_pf_resource_info *pf_resc)
 {
        struct hwrm_func_qcfg_input req = {0};
        struct hwrm_func_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
@@ -3497,8 +3469,13 @@ static int update_pf_resource_max(struct bnxt *bp)
        rc = bnxt_hwrm_send_message(bp, &req, sizeof(req), BNXT_USE_CHIMP_MB);
        HWRM_CHECK_RESULT();
 
-       /* Only TX ring value reflects actual allocation? TODO */
-       bp->max_tx_rings = rte_le_to_cpu_16(resp->alloc_tx_rings);
+       pf_resc->num_tx_rings = rte_le_to_cpu_16(resp->alloc_tx_rings);
+       pf_resc->num_rsscos_ctxs = rte_le_to_cpu_16(resp->alloc_rsscos_ctx);
+       pf_resc->num_stat_ctxs = rte_le_to_cpu_16(resp->alloc_stat_ctx);
+       pf_resc->num_cp_rings = rte_le_to_cpu_16(resp->alloc_cmpl_rings);
+       pf_resc->num_rx_rings = rte_le_to_cpu_16(resp->alloc_rx_rings);
+       pf_resc->num_l2_ctxs = rte_le_to_cpu_16(resp->alloc_l2_ctx);
+       pf_resc->num_hw_ring_grps = rte_le_to_cpu_32(resp->alloc_hw_ring_grps);
        bp->pf->evb_mode = resp->evb_mode;
 
        HWRM_UNLOCK();
@@ -3506,8 +3483,42 @@ static int update_pf_resource_max(struct bnxt *bp)
        return rc;
 }
 
+static void
+bnxt_calculate_pf_resources(struct bnxt *bp,
+                           struct bnxt_pf_resource_info *pf_resc,
+                           int num_vfs)
+{
+       if (!num_vfs) {
+               pf_resc->num_rsscos_ctxs = bp->max_rsscos_ctx;
+               pf_resc->num_stat_ctxs = bp->max_stat_ctx;
+               pf_resc->num_cp_rings = bp->max_cp_rings;
+               pf_resc->num_tx_rings = bp->max_tx_rings;
+               pf_resc->num_rx_rings = bp->max_rx_rings;
+               pf_resc->num_l2_ctxs = bp->max_l2_ctx;
+               pf_resc->num_hw_ring_grps = bp->max_ring_grps;
+
+               return;
+       }
+
+       pf_resc->num_rsscos_ctxs = bp->max_rsscos_ctx / (num_vfs + 1) +
+                                  bp->max_rsscos_ctx % (num_vfs + 1);
+       pf_resc->num_stat_ctxs = bp->max_stat_ctx / (num_vfs + 1) +
+                                bp->max_stat_ctx % (num_vfs + 1);
+       pf_resc->num_cp_rings = bp->max_cp_rings / (num_vfs + 1) +
+                               bp->max_cp_rings % (num_vfs + 1);
+       pf_resc->num_tx_rings = bp->max_tx_rings / (num_vfs + 1) +
+                               bp->max_tx_rings % (num_vfs + 1);
+       pf_resc->num_rx_rings = bp->max_rx_rings / (num_vfs + 1) +
+                               bp->max_rx_rings % (num_vfs + 1);
+       pf_resc->num_l2_ctxs = bp->max_l2_ctx / (num_vfs + 1) +
+                              bp->max_l2_ctx % (num_vfs + 1);
+       pf_resc->num_hw_ring_grps = bp->max_ring_grps / (num_vfs + 1) +
+                                   bp->max_ring_grps % (num_vfs + 1);
+}
+
 int bnxt_hwrm_allocate_pf_only(struct bnxt *bp)
 {
+       struct bnxt_pf_resource_info pf_resc = { 0 };
        int rc;
 
        if (!BNXT_PF(bp)) {
@@ -3519,82 +3530,100 @@ int bnxt_hwrm_allocate_pf_only(struct bnxt *bp)
        if (rc)
                return rc;
 
+       bnxt_calculate_pf_resources(bp, &pf_resc, 0);
+
        bp->pf->func_cfg_flags &=
                ~(HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_ENABLE |
                  HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_DISABLE);
        bp->pf->func_cfg_flags |=
                HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_DISABLE;
-       rc = bnxt_hwrm_pf_func_cfg(bp, bp->max_tx_rings);
+       rc = bnxt_hwrm_pf_func_cfg(bp, &pf_resc);
        rc = __bnxt_hwrm_func_qcaps(bp);
        return rc;
 }
 
-int bnxt_hwrm_allocate_vfs(struct bnxt *bp, int num_vfs)
+static int
+bnxt_configure_vf_req_buf(struct bnxt *bp, int num_vfs)
 {
-       struct hwrm_func_cfg_input req = {0};
-       struct hwrm_func_cfg_output *resp = bp->hwrm_cmd_resp_addr;
-       int i;
-       size_t sz;
-       int rc = 0;
-       size_t req_buf_sz;
-
-       if (!BNXT_PF(bp)) {
-               PMD_DRV_LOG(ERR, "Attempt to allcoate VFs on a VF!\n");
-               return -EINVAL;
-       }
-
-       rc = bnxt_hwrm_func_qcaps(bp);
-
-       if (rc)
-               return rc;
-
-       bp->pf->active_vfs = num_vfs;
-
-       /*
-        * First, configure the PF to only use one TX ring.  This ensures that
-        * there are enough rings for all VFs.
-        *
-        * If we don't do this, when we call func_alloc() later, we will lock
-        * extra rings to the PF that won't be available during func_cfg() of
-        * the VFs.
-        *
-        * This has been fixed with firmware versions above 20.6.54
-        */
-       bp->pf->func_cfg_flags &=
-               ~(HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_ENABLE |
-                 HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_DISABLE);
-       bp->pf->func_cfg_flags |=
-               HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_ENABLE;
-       rc = bnxt_hwrm_pf_func_cfg(bp, 1);
-       if (rc)
-               return rc;
+       size_t req_buf_sz, sz;
+       int i, rc;
 
-       /*
-        * Now, create and register a buffer to hold forwarded VF requests
-        */
        req_buf_sz = num_vfs * HWRM_MAX_REQ_LEN;
        bp->pf->vf_req_buf = rte_malloc("bnxt_vf_fwd", req_buf_sz,
                page_roundup(num_vfs * HWRM_MAX_REQ_LEN));
        if (bp->pf->vf_req_buf == NULL) {
-               rc = -ENOMEM;
-               goto error_free;
+               return -ENOMEM;
        }
+
        for (sz = 0; sz < req_buf_sz; sz += getpagesize())
                rte_mem_lock_page(((char *)bp->pf->vf_req_buf) + sz);
+
        for (i = 0; i < num_vfs; i++)
                bp->pf->vf_info[i].req_buf = ((char *)bp->pf->vf_req_buf) +
-                                       (i * HWRM_MAX_REQ_LEN);
+                                            (i * HWRM_MAX_REQ_LEN);
 
-       rc = bnxt_hwrm_func_buf_rgtr(bp);
+       rc = bnxt_hwrm_func_buf_rgtr(bp, num_vfs);
        if (rc)
-               goto error_free;
+               rte_free(bp->pf->vf_req_buf);
+
+       return rc;
+}
 
-       populate_vf_func_cfg_req(bp, &req, num_vfs);
+static int
+bnxt_process_vf_resc_config_new(struct bnxt *bp, int num_vfs)
+{
+       struct hwrm_func_vf_resource_cfg_output *resp = bp->hwrm_cmd_resp_addr;
+       struct hwrm_func_vf_resource_cfg_input req = {0};
+       int i, rc = 0;
 
+       bnxt_fill_vf_func_cfg_req_new(bp, &req, num_vfs);
        bp->pf->active_vfs = 0;
        for (i = 0; i < num_vfs; i++) {
-               add_random_mac_if_needed(bp, &req, i);
+               HWRM_PREP(&req, HWRM_FUNC_VF_RESOURCE_CFG, BNXT_USE_CHIMP_MB);
+               req.vf_id = rte_cpu_to_le_16(bp->pf->vf_info[i].fid);
+               rc = bnxt_hwrm_send_message(bp,
+                                           &req,
+                                           sizeof(req),
+                                           BNXT_USE_CHIMP_MB);
+               if (rc || resp->error_code) {
+                       PMD_DRV_LOG(ERR,
+                               "Failed to initialize VF %d\n", i);
+                       PMD_DRV_LOG(ERR,
+                               "Not all VFs available. (%d, %d)\n",
+                               rc, resp->error_code);
+                       HWRM_UNLOCK();
+
+                       /* If the first VF configuration itself fails,
+                        * unregister the vf_fwd_request buffer.
+                        */
+                       if (i == 0)
+                               bnxt_hwrm_func_buf_unrgtr(bp);
+                       break;
+               }
+               HWRM_UNLOCK();
+
+               /* Update the max resource values based on the resource values
+                * allocated to the VF.
+                */
+               bnxt_update_max_resources(bp, i);
+               bp->pf->active_vfs++;
+               bnxt_hwrm_func_clr_stats(bp, bp->pf->vf_info[i].fid);
+       }
+
+       return 0;
+}
+
+static int
+bnxt_process_vf_resc_config_old(struct bnxt *bp, int num_vfs)
+{
+       struct hwrm_func_cfg_output *resp = bp->hwrm_cmd_resp_addr;
+       struct hwrm_func_cfg_input req = {0};
+       int i, rc;
 
+       bnxt_fill_vf_func_cfg_req_old(bp, &req, num_vfs);
+
+       bp->pf->active_vfs = 0;
+       for (i = 0; i < num_vfs; i++) {
                HWRM_PREP(&req, HWRM_FUNC_CFG, BNXT_USE_CHIMP_MB);
                req.flags = rte_cpu_to_le_32(bp->pf->vf_info[i].func_cfg_flags);
                req.fid = rte_cpu_to_le_16(bp->pf->vf_info[i].fid);
@@ -3609,40 +3638,107 @@ int bnxt_hwrm_allocate_vfs(struct bnxt *bp, int num_vfs)
 
                if (rc || resp->error_code) {
                        PMD_DRV_LOG(ERR,
-                               "Failed to initizlie VF %d\n", i);
+                               "Failed to initialize VF %d\n", i);
                        PMD_DRV_LOG(ERR,
                                "Not all VFs available. (%d, %d)\n",
                                rc, resp->error_code);
                        HWRM_UNLOCK();
+
+                       /* If the first VF configuration itself fails,
+                        * unregister the vf_fwd_request buffer.
+                        */
+                       if (i == 0)
+                               bnxt_hwrm_func_buf_unrgtr(bp);
                        break;
                }
 
                HWRM_UNLOCK();
 
-               reserve_resources_from_vf(bp, &req, i);
+               /* Update the max resource values based on the resource values
+                * allocated to the VF.
+                */
+               bnxt_update_max_resources(bp, i);
                bp->pf->active_vfs++;
                bnxt_hwrm_func_clr_stats(bp, bp->pf->vf_info[i].fid);
        }
 
+       return 0;
+}
+
+static void
+bnxt_configure_vf_resources(struct bnxt *bp, int num_vfs)
+{
+       if (bp->flags & BNXT_FLAG_NEW_RM)
+               bnxt_process_vf_resc_config_new(bp, num_vfs);
+       else
+               bnxt_process_vf_resc_config_old(bp, num_vfs);
+}
+
+static void
+bnxt_update_pf_resources(struct bnxt *bp,
+                        struct bnxt_pf_resource_info *pf_resc)
+{
+       bp->max_rsscos_ctx = pf_resc->num_rsscos_ctxs;
+       bp->max_stat_ctx = pf_resc->num_stat_ctxs;
+       bp->max_cp_rings = pf_resc->num_cp_rings;
+       bp->max_tx_rings = pf_resc->num_tx_rings;
+       bp->max_rx_rings = pf_resc->num_rx_rings;
+       bp->max_ring_grps = pf_resc->num_hw_ring_grps;
+}
+
+static int32_t
+bnxt_configure_pf_resources(struct bnxt *bp,
+                           struct bnxt_pf_resource_info *pf_resc)
+{
        /*
-        * Now configure the PF to use "the rest" of the resources
-        * We're using STD_TX_RING_MODE here though which will limit the TX
-        * rings.  This will allow QoS to function properly.  Not setting this
+        * We're using STD_TX_RING_MODE here which will limit the TX
+        * rings. This will allow QoS to function properly. Not setting this
         * will cause PF rings to break bandwidth settings.
         */
-       rc = bnxt_hwrm_pf_func_cfg(bp, bp->max_tx_rings);
+       bp->pf->func_cfg_flags &=
+               ~(HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_ENABLE |
+                 HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_DISABLE);
+       bp->pf->func_cfg_flags |=
+               HWRM_FUNC_CFG_INPUT_FLAGS_STD_TX_RING_MODE_ENABLE;
+       return bnxt_hwrm_pf_func_cfg(bp, pf_resc);
+}
+
+int bnxt_hwrm_allocate_vfs(struct bnxt *bp, int num_vfs)
+{
+       struct bnxt_pf_resource_info pf_resc = { 0 };
+       int rc;
+
+       if (!BNXT_PF(bp)) {
+               PMD_DRV_LOG(ERR, "Attempt to allocate VFs on a VF!\n");
+               return -EINVAL;
+       }
+
+       rc = bnxt_hwrm_func_qcaps(bp);
        if (rc)
-               goto error_free;
+               return rc;
+
+       bnxt_calculate_pf_resources(bp, &pf_resc, num_vfs);
 
-       rc = update_pf_resource_max(bp);
+       rc = bnxt_configure_pf_resources(bp, &pf_resc);
        if (rc)
-               goto error_free;
+               return rc;
 
-       return rc;
+       rc = bnxt_query_pf_resources(bp, &pf_resc);
+       if (rc)
+               return rc;
 
-error_free:
-       bnxt_hwrm_func_buf_unrgtr(bp);
-       return rc;
+       /*
+        * Now, create and register a buffer to hold forwarded VF requests
+        */
+       rc = bnxt_configure_vf_req_buf(bp, num_vfs);
+       if (rc)
+               return rc;
+
+       bnxt_configure_vf_resources(bp, num_vfs);
+
+       bnxt_update_pf_resources(bp, &pf_resc);
+
+       return 0;
 }
 
 int bnxt_hwrm_pf_evb_mode(struct bnxt *bp)
@@ -3747,23 +3843,24 @@ int bnxt_set_rx_mask_no_vlan(struct bnxt *bp, struct bnxt_vnic_info *vnic)
        return bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
 }
 
-int bnxt_hwrm_func_buf_rgtr(struct bnxt *bp)
+int bnxt_hwrm_func_buf_rgtr(struct bnxt *bp, int num_vfs)
 {
-       int rc = 0;
-       struct hwrm_func_buf_rgtr_input req = {.req_type = 0 };
        struct hwrm_func_buf_rgtr_output *resp = bp->hwrm_cmd_resp_addr;
+       struct hwrm_func_buf_rgtr_input req = {.req_type = 0 };
+       int rc;
 
        HWRM_PREP(&req, HWRM_FUNC_BUF_RGTR, BNXT_USE_CHIMP_MB);
 
        req.req_buf_num_pages = rte_cpu_to_le_16(1);
-       req.req_buf_page_size = rte_cpu_to_le_16(
-                        page_getenum(bp->pf->active_vfs * HWRM_MAX_REQ_LEN));
+       req.req_buf_page_size =
+               rte_cpu_to_le_16(page_getenum(num_vfs * HWRM_MAX_REQ_LEN));
        req.req_buf_len = rte_cpu_to_le_16(HWRM_MAX_REQ_LEN);
        req.req_buf_page_addr0 =
                rte_cpu_to_le_64(rte_malloc_virt2iova(bp->pf->vf_req_buf));
        if (req.req_buf_page_addr0 == RTE_BAD_IOVA) {
                PMD_DRV_LOG(ERR,
                        "unable to map buffer address to physical memory\n");
+               HWRM_UNLOCK();
                return -ENOMEM;
        }
 
index e98b1fe..a7fa7f6 100644 (file)
@@ -107,6 +107,16 @@ enum bnxt_flow_dir {
        BNXT_DIR_MAX
 };
 
+struct bnxt_pf_resource_info {
+       uint16_t num_rsscos_ctxs;
+       uint16_t num_stat_ctxs;
+       uint16_t num_tx_rings;
+       uint16_t num_rx_rings;
+       uint16_t num_cp_rings;
+       uint16_t num_l2_ctxs;
+       uint32_t num_hw_ring_grps;
+};
+
 #define BNXT_CTX_VAL_INVAL     0xFFFF
 
 int bnxt_hwrm_cfa_l2_clear_rx_mask(struct bnxt *bp,
@@ -127,7 +137,7 @@ int bnxt_hwrm_exec_fwd_resp(struct bnxt *bp, uint16_t target_id,
 int bnxt_hwrm_reject_fwd_resp(struct bnxt *bp, uint16_t target_id,
                              void *encaped, size_t ec_size);
 
-int bnxt_hwrm_func_buf_rgtr(struct bnxt *bp);
+int bnxt_hwrm_func_buf_rgtr(struct bnxt *bp, int num_vfs);
 int bnxt_hwrm_func_buf_unrgtr(struct bnxt *bp);
 int bnxt_hwrm_func_driver_register(struct bnxt *bp);
 int bnxt_hwrm_func_qcaps(struct bnxt *bp);