be2net: create optimal number of queues on SR-IOV config

If SR-IOV is enabled in the adapter, the FW distributes queue resources
evenly across the PF and it's VFs. If the user is not interested in enabling
VFs, the queues set aside for VFs are wasted.
This patch adds support for the PF driver to re-configure the resource
distribution in FW based on the number of VFs enabled by the user.
This also allows for supporting RSS queues on VFs, when less number of VFs
are enabled per PF. When maximum number of VFs are enabled, each VF typically
gets only one RXQ.
Signed-off-by: Vasundhara Volam <vasundhara.volam@emulex.com>
Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/emulex/benet/be.h

@@ -501,6 +501,7 @@ struct be_adapter {
 	u32 flash_status;
 	struct completion et_cmd_compl;
 
+	struct be_resources pool_res;	/* resources available for the port */
 	struct be_resources res;	/* resources available for the func */
 	u16 num_vfs;			/* Number of VFs provisioned by PF */
 	u8 virtfn;
@@ -525,8 +526,7 @@ struct be_adapter {
 #define be_physfn(adapter)		(!adapter->virtfn)
 #define be_virtfn(adapter)		(adapter->virtfn)
 #define sriov_enabled(adapter)		(adapter->num_vfs > 0)
-#define sriov_want(adapter)             (be_physfn(adapter) &&	\
-					 (num_vfs || pci_num_vf(adapter->pdev)))
+
 #define for_all_vfs(adapter, vf_cfg, i)					\
 	for (i = 0, vf_cfg = &adapter->vf_cfg[i]; i < adapter->num_vfs;	\
 		i++, vf_cfg++)
@@ -537,7 +537,7 @@ struct be_adapter {
 #define be_max_vlans(adapter)		(adapter->res.max_vlans)
 #define be_max_uc(adapter)		(adapter->res.max_uc_mac)
 #define be_max_mc(adapter)		(adapter->res.max_mcast_mac)
-#define be_max_vfs(adapter)		(adapter->res.max_vfs)
+#define be_max_vfs(adapter)		(adapter->pool_res.max_vfs)
 #define be_max_rss(adapter)		(adapter->res.max_rss_qs)
 #define be_max_txqs(adapter)		(adapter->res.max_tx_qs)
 #define be_max_prio_txqs(adapter)	(adapter->res.max_prio_tx_qs)

drivers/net/ethernet/emulex/benet/be_cmds.c

@@ -3523,38 +3523,39 @@ int be_cmd_get_profile_config(struct be_adapter *adapter,
 	return status;
 }
 
-int be_cmd_set_profile_config(struct be_adapter *adapter, void *desc,
-			      int size, u8 version, u8 domain)
+/* Will use MBOX only if MCCQ has not been created */
+static int be_cmd_set_profile_config(struct be_adapter *adapter, void *desc,
+				     int size, int count, u8 version, u8 domain)
 {
 	struct be_cmd_req_set_profile_config *req;
-	struct be_mcc_wrb *wrb;
+	struct be_mcc_wrb wrb = {0};
+	struct be_dma_mem cmd;
 	int status;
 
-	spin_lock_bh(&adapter->mcc_lock);
-
-	wrb = wrb_from_mccq(adapter);
-	if (!wrb) {
-		status = -EBUSY;
-		goto err;
-	}
+	memset(&cmd, 0, sizeof(struct be_dma_mem));
+	cmd.size = sizeof(struct be_cmd_req_set_profile_config);
+	cmd.va = pci_alloc_consistent(adapter->pdev, cmd.size, &cmd.dma);
+	if (!cmd.va)
+		return -ENOMEM;
 
-	req = embedded_payload(wrb);
+	req = cmd.va;
 	be_wrb_cmd_hdr_prepare(&req->hdr, CMD_SUBSYSTEM_COMMON,
-			       OPCODE_COMMON_SET_PROFILE_CONFIG, sizeof(*req),
-			       wrb, NULL);
+			       OPCODE_COMMON_SET_PROFILE_CONFIG, cmd.size,
+			       &wrb, &cmd);
 	req->hdr.version = version;
 	req->hdr.domain = domain;
-	req->desc_count = cpu_to_le32(1);
+	req->desc_count = cpu_to_le32(count);
 	memcpy(req->desc, desc, size);
 
-	status = be_mcc_notify_wait(adapter);
-err:
-	spin_unlock_bh(&adapter->mcc_lock);
+	status = be_cmd_notify_wait(adapter, &wrb);
+
+	if (cmd.va)
+		pci_free_consistent(adapter->pdev, cmd.size, cmd.va, cmd.dma);
 	return status;
 }
 
 /* Mark all fields invalid */
-void be_reset_nic_desc(struct be_nic_res_desc *nic)
+static void be_reset_nic_desc(struct be_nic_res_desc *nic)
 {
 	memset(nic, 0, sizeof(*nic));
 	nic->unicast_mac_count = 0xFFFF;
@@ -3575,9 +3576,20 @@ void be_reset_nic_desc(struct be_nic_res_desc *nic)
 	nic->wol_param = 0x0F;
 	nic->tunnel_iface_count = 0xFFFF;
 	nic->direct_tenant_iface_count = 0xFFFF;
+	nic->bw_min = 0xFFFFFFFF;
 	nic->bw_max = 0xFFFFFFFF;
 }
 
+/* Mark all fields invalid */
+static void be_reset_pcie_desc(struct be_pcie_res_desc *pcie)
+{
+	memset(pcie, 0, sizeof(*pcie));
+	pcie->sriov_state = 0xFF;
+	pcie->pf_state = 0xFF;
+	pcie->pf_type = 0xFF;
+	pcie->num_vfs = 0xFFFF;
+}
+
 int be_cmd_config_qos(struct be_adapter *adapter, u32 max_rate, u16 link_speed,
 		      u8 domain)
 {
@@ -3608,7 +3620,63 @@ int be_cmd_config_qos(struct be_adapter *adapter, u32 max_rate, u16 link_speed,
 
 	return be_cmd_set_profile_config(adapter, &nic_desc,
 					 nic_desc.hdr.desc_len,
-					 version, domain);
+					 1, version, domain);
+}
+
+int be_cmd_set_sriov_config(struct be_adapter *adapter,
+			    struct be_resources res, u16 num_vfs)
+{
+	struct {
+		struct be_pcie_res_desc pcie;
+		struct be_nic_res_desc nic_vft;
+	} __packed desc;
+	u16 vf_q_count;
+
+	if (BEx_chip(adapter) || lancer_chip(adapter))
+		return 0;
+
+	/* PF PCIE descriptor */
+	be_reset_pcie_desc(&desc.pcie);
+	desc.pcie.hdr.desc_type = PCIE_RESOURCE_DESC_TYPE_V1;
+	desc.pcie.hdr.desc_len = RESOURCE_DESC_SIZE_V1;
+	desc.pcie.flags = (1 << IMM_SHIFT) | (1 << NOSV_SHIFT);
+	desc.pcie.pf_num = adapter->pdev->devfn;
+	desc.pcie.sriov_state = num_vfs ? 1 : 0;
+	desc.pcie.num_vfs = cpu_to_le16(num_vfs);
+
+	/* VF NIC Template descriptor */
+	be_reset_nic_desc(&desc.nic_vft);
+	desc.nic_vft.hdr.desc_type = NIC_RESOURCE_DESC_TYPE_V1;
+	desc.nic_vft.hdr.desc_len = RESOURCE_DESC_SIZE_V1;
+	desc.nic_vft.flags = (1 << VFT_SHIFT) | (1 << IMM_SHIFT) |
+				(1 << NOSV_SHIFT);
+	desc.nic_vft.pf_num = adapter->pdev->devfn;
+	desc.nic_vft.vf_num = 0;
+
+	if (num_vfs && res.vf_if_cap_flags & BE_IF_FLAGS_RSS) {
+		/* If number of VFs requested is 8 less than max supported,
+		 * assign 8 queue pairs to the PF and divide the remaining
+		 * resources evenly among the VFs
+		 */
+		if (num_vfs < (be_max_vfs(adapter) - 8))
+			vf_q_count = (res.max_rss_qs - 8) / num_vfs;
+		else
+			vf_q_count = res.max_rss_qs / num_vfs;
+
+		desc.nic_vft.rq_count = cpu_to_le16(vf_q_count);
+		desc.nic_vft.txq_count = cpu_to_le16(vf_q_count);
+		desc.nic_vft.rssq_count = cpu_to_le16(vf_q_count - 1);
+		desc.nic_vft.cq_count = cpu_to_le16(3 * vf_q_count);
+	} else {
+		desc.nic_vft.txq_count = cpu_to_le16(1);
+		desc.nic_vft.rq_count = cpu_to_le16(1);
+		desc.nic_vft.rssq_count = cpu_to_le16(0);
+		/* One CQ for each TX, RX and MCCQ */
+		desc.nic_vft.cq_count = cpu_to_le16(3);
+	}
+
+	return be_cmd_set_profile_config(adapter, &desc,
+					 2 * RESOURCE_DESC_SIZE_V1, 2, 1, 0);
 }
 
 int be_cmd_manage_iface(struct be_adapter *adapter, u32 iface, u8 op)
@@ -3660,7 +3728,7 @@ int be_cmd_set_vxlan_port(struct be_adapter *adapter, __be16 port)
 	}
 
 	return be_cmd_set_profile_config(adapter, &port_desc,
-					 RESOURCE_DESC_SIZE_V1, 1, 0);
+					 RESOURCE_DESC_SIZE_V1, 1, 1, 0);
 }
 
 int be_cmd_get_if_id(struct be_adapter *adapter, struct be_vf_cfg *vf_cfg,

drivers/net/ethernet/emulex/benet/be_cmds.h

@@ -1963,8 +1963,8 @@ struct be_cmd_req_set_profile_config {
 	struct be_cmd_req_hdr hdr;
 	u32 rsvd;
 	u32 desc_count;
-	u8 desc[RESOURCE_DESC_SIZE_V1];
-};
+	u8 desc[2 * RESOURCE_DESC_SIZE_V1];
+} __packed;
 
 struct be_cmd_resp_set_profile_config {
 	struct be_cmd_resp_hdr hdr;
@@ -2158,8 +2158,6 @@ int be_cmd_get_func_config(struct be_adapter *adapter,
 			   struct be_resources *res);
 int be_cmd_get_profile_config(struct be_adapter *adapter,
 			      struct be_resources *res, u8 domain);
-int be_cmd_set_profile_config(struct be_adapter *adapter, void *desc,
-			      int size, u8 version, u8 domain);
 int be_cmd_get_active_profile(struct be_adapter *adapter, u16 *profile);
 int be_cmd_get_if_id(struct be_adapter *adapter, struct be_vf_cfg *vf_cfg,
 		     int vf_num);
@@ -2169,3 +2167,5 @@ int be_cmd_set_logical_link_config(struct be_adapter *adapter,
 					  int link_state, u8 domain);
 int be_cmd_set_vxlan_port(struct be_adapter *adapter, __be16 port);
 int be_cmd_manage_iface(struct be_adapter *adapter, u32 iface, u8 op);
+int be_cmd_set_sriov_config(struct be_adapter *adapter,
+			    struct be_resources res, u16 num_vfs);

drivers/net/ethernet/emulex/benet/be_main.c

@@ -3098,6 +3098,13 @@ static int be_clear(struct be_adapter *adapter)
 	if (sriov_enabled(adapter))
 		be_vf_clear(adapter);
 
+	/* Re-configure FW to distribute resources evenly across max-supported
+	 * number of VFs, only when VFs are not already enabled.
+	 */
+	if (be_physfn(adapter) && !pci_vfs_assigned(adapter->pdev))
+		be_cmd_set_sriov_config(adapter, adapter->pool_res,
+					pci_sriov_get_totalvfs(adapter->pdev));
+
 #ifdef CONFIG_BE2NET_VXLAN
 	be_disable_vxlan_offloads(adapter);
 #endif
@@ -3170,19 +3177,6 @@ static int be_vf_setup(struct be_adapter *adapter)
 	u32 privileges;
 
 	old_vfs = pci_num_vf(adapter->pdev);
-	if (old_vfs) {
-		dev_info(dev, "%d VFs are already enabled\n", old_vfs);
-		if (old_vfs != num_vfs)
-			dev_warn(dev, "Ignoring num_vfs=%d setting\n", num_vfs);
-		adapter->num_vfs = old_vfs;
-	} else {
-		if (num_vfs > be_max_vfs(adapter))
-			dev_info(dev, "Device supports %d VFs and not %d\n",
-				 be_max_vfs(adapter), num_vfs);
-		adapter->num_vfs = min_t(u16, num_vfs, be_max_vfs(adapter));
-		if (!adapter->num_vfs)
-			return 0;
-	}
 
 	status = be_vf_setup_init(adapter);
 	if (status)
@@ -3194,17 +3188,15 @@ static int be_vf_setup(struct be_adapter *adapter)
 			if (status)
 				goto err;
 		}
-	} else {
-		status = be_vfs_if_create(adapter);
-		if (status)
-			goto err;
-	}
 
-	if (old_vfs) {
 		status = be_vfs_mac_query(adapter);
 		if (status)
 			goto err;
 	} else {
+		status = be_vfs_if_create(adapter);
+		if (status)
+			goto err;
+
 		status = be_vf_eth_addr_config(adapter);
 		if (status)
 			goto err;
@@ -3270,19 +3262,7 @@ static u8 be_convert_mc_type(u32 function_mode)
 static void BEx_get_resources(struct be_adapter *adapter,
 			      struct be_resources *res)
 {
-	struct pci_dev *pdev = adapter->pdev;
-	bool use_sriov = false;
-	int max_vfs = 0;
-
-	if (be_physfn(adapter) && BE3_chip(adapter)) {
-		be_cmd_get_profile_config(adapter, res, 0);
-		/* Some old versions of BE3 FW don't report max_vfs value */
-		if (res->max_vfs == 0) {
-			max_vfs = pci_sriov_get_totalvfs(pdev);
-			res->max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
-		}
-		use_sriov = res->max_vfs && sriov_want(adapter);
-	}
+	bool use_sriov = adapter->num_vfs ? 1 : 0;
 
 	if (be_physfn(adapter))
 		res->max_uc_mac = BE_UC_PMAC_COUNT;
@@ -3349,6 +3329,54 @@ static void be_setup_init(struct be_adapter *adapter)
 		adapter->cmd_privileges = MIN_PRIVILEGES;
 }
 
+static int be_get_sriov_config(struct be_adapter *adapter)
+{
+	struct device *dev = &adapter->pdev->dev;
+	struct be_resources res = {0};
+	int status, max_vfs, old_vfs;
+
+	status = be_cmd_get_profile_config(adapter, &res, 0);
+	if (status)
+		return status;
+
+	adapter->pool_res = res;
+
+	/* Some old versions of BE3 FW don't report max_vfs value */
+	if (BE3_chip(adapter) && !res.max_vfs) {
+		max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
+		res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
+	}
+
+	adapter->pool_res.max_vfs = res.max_vfs;
+	pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
+
+	if (!be_max_vfs(adapter)) {
+		if (num_vfs)
+			dev_warn(dev, "device doesn't support SRIOV\n");
+		adapter->num_vfs = 0;
+		return 0;
+	}
+
+	/* validate num_vfs module param */
+	old_vfs = pci_num_vf(adapter->pdev);
+	if (old_vfs) {
+		dev_info(dev, "%d VFs are already enabled\n", old_vfs);
+		if (old_vfs != num_vfs)
+			dev_warn(dev, "Ignoring num_vfs=%d setting\n", num_vfs);
+		adapter->num_vfs = old_vfs;
+	} else {
+		if (num_vfs > be_max_vfs(adapter)) {
+			dev_info(dev, "Resources unavailable to init %d VFs\n",
+				 num_vfs);
+			dev_info(dev, "Limiting to %d VFs\n",
+				 be_max_vfs(adapter));
+		}
+		adapter->num_vfs = min_t(u16, num_vfs, be_max_vfs(adapter));
+	}
+
+	return 0;
+}
+
 static int be_get_resources(struct be_adapter *adapter)
 {
 	struct device *dev = &adapter->pdev->dev;
@@ -3374,14 +3402,6 @@ static int be_get_resources(struct be_adapter *adapter)
 			res.max_evt_qs /= 2;
 		adapter->res = res;
 
-		if (be_physfn(adapter)) {
-			status = be_cmd_get_profile_config(adapter, &res, 0);
-			if (status)
-				return status;
-			adapter->res.max_vfs = res.max_vfs;
-			adapter->res.vf_if_cap_flags = res.vf_if_cap_flags;
-		}
-
 		dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
 			 be_max_txqs(adapter), be_max_rxqs(adapter),
 			 be_max_rss(adapter), be_max_eqs(adapter),
@@ -3394,7 +3414,6 @@ static int be_get_resources(struct be_adapter *adapter)
 	return 0;
 }
 
-/* Routine to query per function resource limits */
 static int be_get_config(struct be_adapter *adapter)
 {
 	u16 profile_id;
@@ -3412,6 +3431,26 @@ static int be_get_config(struct be_adapter *adapter)
 		if (!status)
 			dev_info(&adapter->pdev->dev,
 				 "Using profile 0x%x\n", profile_id);
+
+		status = be_get_sriov_config(adapter);
+		if (status)
+			return status;
+
+		/* When the HW is in SRIOV capable configuration, the PF-pool
+		 * resources are equally distributed across the max-number of
+		 * VFs. The user may request only a subset of the max-vfs to be
+		 * enabled. Based on num_vfs, redistribute the resources across
+		 * num_vfs so that each VF will have access to more number of
+		 * resources. This facility is not available in BE3 FW.
+		 * Also, this is done by FW in Lancer chip.
+		 */
+		if (!pci_num_vf(adapter->pdev)) {
+			status = be_cmd_set_sriov_config(adapter,
+							 adapter->pool_res,
+							 adapter->num_vfs);
+			if (status)
+				return status;
+		}
 	}
 
 	status = be_get_resources(adapter);
@@ -3597,12 +3636,8 @@ static int be_setup(struct be_adapter *adapter)
 		be_cmd_set_logical_link_config(adapter,
 					       IFLA_VF_LINK_STATE_AUTO, 0);
 
-	if (sriov_want(adapter)) {
-		if (be_max_vfs(adapter))
+	if (adapter->num_vfs)
 		be_vf_setup(adapter);
-		else
-			dev_warn(dev, "device doesn't support SRIOV\n");
-	}
 
 	status = be_cmd_get_phy_info(adapter);
 	if (!status && be_pause_supported(adapter))