Merge branch 'mlx4-next'

Or Gerlitz says:

====================
mlx4: Flexible (asymmetric) allocation of EQs and MSI-X vectors

This series from Matan Barak is built as follows:

The 1st two patches fix small bugs w.r.t firmware spec. Next
are two patches which do more re-factoring of the init/fini flow
and a patch that adds support for the QUERY_FUNC firmware command,
these are all pre-steps for the major patch of the series. In this
patch (#6) we change the order of talking/querying the firmware
and enabling SRIOV. This allows to remote worst-case assumption
w.r.t the number of available MSI-X vectors and EQs per function.

The last patch easily enjoys this ordering change, to enable
supports > 64 VFs over a firmware that allows for that.
====================

Signed-off-by: David S. Miller <[e-mail]>

drivers/infiniband/hw/mlx4/main.c

···
 	    dev->caps.num_ports > dev->caps.comp_pool)
 		return;
 
-	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
-					dev->caps.num_ports);
+	eq_per_port = dev->caps.comp_pool / dev->caps.num_ports;
 
 	/* Init eq table */
 	added_eqs = 0;

drivers/net/ethernet/mellanox/mlx4/cmd.c

···
 int mlx4_cmd_init(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
+	int flags = 0;
 
-	mutex_init(&priv->cmd.hcr_mutex);
-	mutex_init(&priv->cmd.slave_cmd_mutex);
-	sema_init(&priv->cmd.poll_sem, 1);
-	priv->cmd.use_events = 0;
-	priv->cmd.toggle     = 1;
+	if (!priv->cmd.initialized) {
+		mutex_init(&priv->cmd.hcr_mutex);
+		mutex_init(&priv->cmd.slave_cmd_mutex);
+		sema_init(&priv->cmd.poll_sem, 1);
+		priv->cmd.use_events = 0;
+		priv->cmd.toggle     = 1;
+		priv->cmd.initialized = 1;
+		flags |= MLX4_CMD_CLEANUP_STRUCT;
+	}
 
-	priv->cmd.hcr = NULL;
-	priv->mfunc.vhcr = NULL;
-
-	if (!mlx4_is_slave(dev)) {
+	if (!mlx4_is_slave(dev) && !priv->cmd.hcr) {
 		priv->cmd.hcr = ioremap(pci_resource_start(dev->pdev, 0) +
 					MLX4_HCR_BASE, MLX4_HCR_SIZE);
 		if (!priv->cmd.hcr) {
 			mlx4_err(dev, "Couldn't map command register\n");
-			return -ENOMEM;
+			goto err;
 		}
+		flags |= MLX4_CMD_CLEANUP_HCR;
 	}
 
-	if (mlx4_is_mfunc(dev)) {
+	if (mlx4_is_mfunc(dev) && !priv->mfunc.vhcr) {
 		priv->mfunc.vhcr = dma_alloc_coherent(&(dev->pdev->dev), PAGE_SIZE,
 						      &priv->mfunc.vhcr_dma,
 						      GFP_KERNEL);
 		if (!priv->mfunc.vhcr)
-			goto err_hcr;
+			goto err;
+
+		flags |= MLX4_CMD_CLEANUP_VHCR;
 	}
 
-	priv->cmd.pool = pci_pool_create("mlx4_cmd", dev->pdev,
-					 MLX4_MAILBOX_SIZE,
-					 MLX4_MAILBOX_SIZE, 0);
-	if (!priv->cmd.pool)
-		goto err_vhcr;
+	if (!priv->cmd.pool) {
+		priv->cmd.pool = pci_pool_create("mlx4_cmd", dev->pdev,
+						 MLX4_MAILBOX_SIZE,
+						 MLX4_MAILBOX_SIZE, 0);
+		if (!priv->cmd.pool)
+			goto err;
+
+		flags |= MLX4_CMD_CLEANUP_POOL;
+	}
 
 	return 0;
 
-err_vhcr:
-	if (mlx4_is_mfunc(dev))
-		dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
-				  priv->mfunc.vhcr, priv->mfunc.vhcr_dma);
-	priv->mfunc.vhcr = NULL;
-
-err_hcr:
-	if (!mlx4_is_slave(dev))
-		iounmap(priv->cmd.hcr);
+err:
+	mlx4_cmd_cleanup(dev, flags);
 	return -ENOMEM;
 }
 
···
 	iounmap(priv->mfunc.comm);
 }
 
-void mlx4_cmd_cleanup(struct mlx4_dev *dev)
+void mlx4_cmd_cleanup(struct mlx4_dev *dev, int cleanup_mask)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
 
-	pci_pool_destroy(priv->cmd.pool);
+	if (priv->cmd.pool && (cleanup_mask & MLX4_CMD_CLEANUP_POOL)) {
+		pci_pool_destroy(priv->cmd.pool);
+		priv->cmd.pool = NULL;
+	}
 
-	if (!mlx4_is_slave(dev))
+	if (!mlx4_is_slave(dev) && priv->cmd.hcr &&
+	    (cleanup_mask & MLX4_CMD_CLEANUP_HCR)) {
 		iounmap(priv->cmd.hcr);
-	if (mlx4_is_mfunc(dev))
+		priv->cmd.hcr = NULL;
+	}
+	if (mlx4_is_mfunc(dev) && priv->mfunc.vhcr &&
+	    (cleanup_mask & MLX4_CMD_CLEANUP_VHCR)) {
 		dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
 				  priv->mfunc.vhcr, priv->mfunc.vhcr_dma);
-	priv->mfunc.vhcr = NULL;
+		priv->mfunc.vhcr = NULL;
+	}
+	if (priv->cmd.initialized && (cleanup_mask & MLX4_CMD_CLEANUP_STRUCT))
+		priv->cmd.initialized = 0;
 }
 
 /*

drivers/net/ethernet/mellanox/mlx4/eq.c

···
 		goto err_out_free;
 	}
 
-	err = mlx4_bitmap_init(&priv->eq_table.bitmap, dev->caps.num_eqs,
-			       dev->caps.num_eqs - 1, dev->caps.reserved_eqs, 0);
+	err = mlx4_bitmap_init(&priv->eq_table.bitmap,
+			       roundup_pow_of_two(dev->caps.num_eqs),
+			       dev->caps.num_eqs - 1,
+			       dev->caps.reserved_eqs,
+			       roundup_pow_of_two(dev->caps.num_eqs) -
+			       dev->caps.num_eqs);
 	if (err)
 		goto err_out_free;
 

drivers/net/ethernet/mellanox/mlx4/fw.c

···
 		[13] = "Large cache line (>64B) EQE stride support",
 		[14] = "Ethernet protocol control support",
 		[15] = "Ethernet Backplane autoneg support",
-		[16] = "CONFIG DEV support"
+		[16] = "CONFIG DEV support",
+		[17] = "Asymmetric EQs support",
+		[18] = "More than 80 VFs support"
 	};
 	int i;
 
···
 	return err;
 }
 
+int mlx4_QUERY_FUNC(struct mlx4_dev *dev, struct mlx4_func *func, int slave)
+{
+	struct mlx4_cmd_mailbox *mailbox;
+	u32 *outbox;
+	u8 in_modifier;
+	u8 field;
+	u16 field16;
+	int err;
+
+#define QUERY_FUNC_BUS_OFFSET			0x00
+#define QUERY_FUNC_DEVICE_OFFSET		0x01
+#define QUERY_FUNC_FUNCTION_OFFSET		0x01
+#define QUERY_FUNC_PHYSICAL_FUNCTION_OFFSET	0x03
+#define QUERY_FUNC_RSVD_EQS_OFFSET		0x04
+#define QUERY_FUNC_MAX_EQ_OFFSET		0x06
+#define QUERY_FUNC_RSVD_UARS_OFFSET		0x0b
+
+	mailbox = mlx4_alloc_cmd_mailbox(dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+	outbox = mailbox->buf;
+
+	in_modifier = slave;
+
+	err = mlx4_cmd_box(dev, 0, mailbox->dma, in_modifier, 0,
+			   MLX4_CMD_QUERY_FUNC,
+			   MLX4_CMD_TIME_CLASS_A,
+			   MLX4_CMD_NATIVE);
+	if (err)
+		goto out;
+
+	MLX4_GET(field, outbox, QUERY_FUNC_BUS_OFFSET);
+	func->bus = field & 0xf;
+	MLX4_GET(field, outbox, QUERY_FUNC_DEVICE_OFFSET);
+	func->device = field & 0xf1;
+	MLX4_GET(field, outbox, QUERY_FUNC_FUNCTION_OFFSET);
+	func->function = field & 0x7;
+	MLX4_GET(field, outbox, QUERY_FUNC_PHYSICAL_FUNCTION_OFFSET);
+	func->physical_function = field & 0xf;
+	MLX4_GET(field16, outbox, QUERY_FUNC_RSVD_EQS_OFFSET);
+	func->rsvd_eqs = field16 & 0xffff;
+	MLX4_GET(field16, outbox, QUERY_FUNC_MAX_EQ_OFFSET);
+	func->max_eq = field16 & 0xffff;
+	MLX4_GET(field, outbox, QUERY_FUNC_RSVD_UARS_OFFSET);
+	func->rsvd_uars = field & 0x0f;
+
+	mlx4_dbg(dev, "Bus: %d, Device: %d, Function: %d, Physical function: %d, Max EQs: %d, Reserved EQs: %d, Reserved UARs: %d\n",
+		 func->bus, func->device, func->function, func->physical_function,
+		 func->max_eq, func->rsvd_eqs, func->rsvd_uars);
+
+out:
+	mlx4_free_cmd_mailbox(dev, mailbox);
+	return err;
+}
+
 int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 				struct mlx4_vhcr *vhcr,
 				struct mlx4_cmd_mailbox *inbox,
···
 	u8	field, port;
 	u32	size, proxy_qp, qkey;
 	int	err = 0;
+	struct mlx4_func func;
 
 #define QUERY_FUNC_CAP_FLAGS_OFFSET		0x0
 #define QUERY_FUNC_CAP_NUM_PORTS_OFFSET		0x1
···
 #define QUERY_FUNC_CAP_VF_ENABLE_QP0		0x08
 
 #define QUERY_FUNC_CAP_FLAGS0_FORCE_PHY_WQE_GID 0x80
+#define QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS (1 << 31)
 
 	if (vhcr->op_modifier == 1) {
 		struct mlx4_active_ports actv_ports =
···
 		size = dev->caps.num_cqs;
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET_DEP);
 
-		size = dev->caps.num_eqs;
-		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
-
-		size = dev->caps.reserved_eqs;
-		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
+		if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) ||
+		    mlx4_QUERY_FUNC(dev, &func, slave)) {
+			size = vhcr->in_modifier &
+				QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS ?
+				dev->caps.num_eqs :
+				rounddown_pow_of_two(dev->caps.num_eqs);
+			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
+			size = dev->caps.reserved_eqs;
+			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
+		} else {
+			size = vhcr->in_modifier &
+				QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS ?
+				func.max_eq :
+				rounddown_pow_of_two(func.max_eq);
+			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
+			size = func.rsvd_eqs;
+			MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
+		}
 
 		size = priv->mfunc.master.res_tracker.res_alloc[RES_MPT].quota[slave];
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
···
 	return err;
 }
 
-int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port,
+int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u8 gen_or_port,
 			struct mlx4_func_cap *func_cap)
 {
 	struct mlx4_cmd_mailbox *mailbox;
···
 	u8			field, op_modifier;
 	u32			size, qkey;
 	int			err = 0, quotas = 0;
+	u32                     in_modifier;
 
 	op_modifier = !!gen_or_port; /* 0 = general, 1 = logical port */
+	in_modifier = op_modifier ? gen_or_port :
+		QUERY_FUNC_CAP_SUPPORTS_NON_POWER_OF_2_NUM_EQS;
 
 	mailbox = mlx4_alloc_cmd_mailbox(dev);
 	if (IS_ERR(mailbox))
 		return PTR_ERR(mailbox);
 
-	err = mlx4_cmd_box(dev, 0, mailbox->dma, gen_or_port, op_modifier,
+	err = mlx4_cmd_box(dev, 0, mailbox->dma, in_modifier, op_modifier,
 			   MLX4_CMD_QUERY_FUNC_CAP,
 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
 	if (err)
···
 #define QUERY_DEV_CAP_MAX_MRW_SZ_OFFSET		0x21
 #define QUERY_DEV_CAP_RSVD_MRW_OFFSET		0x22
 #define QUERY_DEV_CAP_MAX_MTT_SEG_OFFSET	0x23
+#define QUERY_DEV_CAP_NUM_SYS_EQ_OFFSET		0x26
 #define QUERY_DEV_CAP_MAX_AV_OFFSET		0x27
 #define QUERY_DEV_CAP_MAX_REQ_QP_OFFSET		0x29
 #define QUERY_DEV_CAP_MAX_RES_QP_OFFSET		0x2b
···
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MPT_OFFSET);
 	dev_cap->max_mpts = 1 << (field & 0x3f);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_EQ_OFFSET);
-	dev_cap->reserved_eqs = field & 0xf;
+	dev_cap->reserved_eqs = 1 << (field & 0xf);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_EQ_OFFSET);
 	dev_cap->max_eqs = 1 << (field & 0xf);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_MTT_OFFSET);
···
 	dev_cap->reserved_mrws = 1 << (field & 0xf);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MTT_SEG_OFFSET);
 	dev_cap->max_mtt_seg = 1 << (field & 0x3f);
+	MLX4_GET(size, outbox, QUERY_DEV_CAP_NUM_SYS_EQ_OFFSET);
+	dev_cap->num_sys_eqs = size & 0xfff;
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_REQ_QP_OFFSET);
 	dev_cap->max_requester_per_qp = 1 << (field & 0x3f);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_RES_QP_OFFSET);
···
 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_VLAN_CONTROL;
 	if (field32 & (1 << 20))
 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FSM;
+	if (field32 & (1 << 21))
+		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_80_VFS;
 
 	if (dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 		for (i = 1; i <= dev_cap->num_ports; ++i) {
···
 	 * we can't use any EQs whose doorbell falls on that page,
 	 * even if the EQ itself isn't reserved.
 	 */
-	dev_cap->reserved_eqs = max(dev_cap->reserved_uars * 4,
-				    dev_cap->reserved_eqs);
+	if (dev_cap->num_sys_eqs == 0)
+		dev_cap->reserved_eqs = max(dev_cap->reserved_uars * 4,
+					    dev_cap->reserved_eqs);
+	else
+		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_SYS_EQS;
 
 	mlx4_dbg(dev, "Max ICM size %lld MB\n",
 		 (unsigned long long) dev_cap->max_icm_sz >> 20);
···
 		 dev_cap->max_srqs, dev_cap->reserved_srqs, dev_cap->srq_entry_sz);
 	mlx4_dbg(dev, "Max CQs: %d, reserved CQs: %d, entry size: %d\n",
 		 dev_cap->max_cqs, dev_cap->reserved_cqs, dev_cap->cqc_entry_sz);
-	mlx4_dbg(dev, "Max EQs: %d, reserved EQs: %d, entry size: %d\n",
-		 dev_cap->max_eqs, dev_cap->reserved_eqs, dev_cap->eqc_entry_sz);
+	mlx4_dbg(dev, "Num sys EQs: %d, max EQs: %d, reserved EQs: %d, entry size: %d\n",
+		 dev_cap->num_sys_eqs, dev_cap->max_eqs, dev_cap->reserved_eqs,
+		 dev_cap->eqc_entry_sz);
 	mlx4_dbg(dev, "reserved MPTs: %d, reserved MTTs: %d\n",
 		 dev_cap->reserved_mrws, dev_cap->reserved_mtts);
 	mlx4_dbg(dev, "Max PDs: %d, reserved PDs: %d, reserved UARs: %d\n",
···
 #define	 INIT_HCA_AUXC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x50)
 #define	 INIT_HCA_EQC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x60)
 #define	 INIT_HCA_LOG_EQ_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x67)
+#define	INIT_HCA_NUM_SYS_EQS_OFFSET	(INIT_HCA_QPC_OFFSET + 0x6a)
 #define	 INIT_HCA_RDMARC_BASE_OFFSET	 (INIT_HCA_QPC_OFFSET + 0x70)
 #define	 INIT_HCA_LOG_RD_OFFSET		 (INIT_HCA_QPC_OFFSET + 0x77)
 #define INIT_HCA_MCAST_OFFSET		 0x0c0
···
 	MLX4_PUT(inbox, param->auxc_base,     INIT_HCA_AUXC_BASE_OFFSET);
 	MLX4_PUT(inbox, param->eqc_base,      INIT_HCA_EQC_BASE_OFFSET);
 	MLX4_PUT(inbox, param->log_num_eqs,   INIT_HCA_LOG_EQ_OFFSET);
+	MLX4_PUT(inbox, param->num_sys_eqs,   INIT_HCA_NUM_SYS_EQS_OFFSET);
 	MLX4_PUT(inbox, param->rdmarc_base,   INIT_HCA_RDMARC_BASE_OFFSET);
 	MLX4_PUT(inbox, param->log_rd_per_qp, INIT_HCA_LOG_RD_OFFSET);
 
···
 	MLX4_GET(param->auxc_base,     outbox, INIT_HCA_AUXC_BASE_OFFSET);
 	MLX4_GET(param->eqc_base,      outbox, INIT_HCA_EQC_BASE_OFFSET);
 	MLX4_GET(param->log_num_eqs,   outbox, INIT_HCA_LOG_EQ_OFFSET);
+	MLX4_GET(param->num_sys_eqs,   outbox, INIT_HCA_NUM_SYS_EQS_OFFSET);
 	MLX4_GET(param->rdmarc_base,   outbox, INIT_HCA_RDMARC_BASE_OFFSET);
 	MLX4_GET(param->log_rd_per_qp, outbox, INIT_HCA_LOG_RD_OFFSET);
 

drivers/net/ethernet/mellanox/mlx4/fw.h

···
 	int max_mpts;
 	int reserved_eqs;
 	int max_eqs;
+	int num_sys_eqs;
 	int reserved_mtts;
 	int max_mrw_sz;
 	int reserved_mrws;
···
 	u64	phys_port_id;
 };
 
+struct mlx4_func {
+	int	bus;
+	int	device;
+	int	function;
+	int	physical_function;
+	int	rsvd_eqs;
+	int	max_eq;
+	int	rsvd_uars;
+};
+
 struct mlx4_adapter {
 	char board_id[MLX4_BOARD_ID_LEN];
 	u8   inta_pin;
···
 	u8  log_num_srqs;
 	u8  log_num_cqs;
 	u8  log_num_eqs;
+	u16 num_sys_eqs;
 	u8  log_rd_per_qp;
 	u8  log_mc_table_sz;
 	u8  log_mpt_sz;
···
 };
 
 int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap);
-int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port,
+int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u8 gen_or_port,
 			struct mlx4_func_cap *func_cap);
 int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 				struct mlx4_vhcr *vhcr,
 				struct mlx4_cmd_mailbox *inbox,
 				struct mlx4_cmd_mailbox *outbox,
 				struct mlx4_cmd_info *cmd);
+int mlx4_QUERY_FUNC(struct mlx4_dev *dev, struct mlx4_func *func, int slave);
 int mlx4_MAP_FA(struct mlx4_dev *dev, struct mlx4_icm *icm);
 int mlx4_UNMAP_FA(struct mlx4_dev *dev);
 int mlx4_RUN_FW(struct mlx4_dev *dev);

drivers/net/ethernet/mellanox/mlx4/main.c

···
 		dev->caps.port_mask[i] = dev->caps.port_type[i];
 }
 
+enum {
+	MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
+};
+
+static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
+{
+	int err = 0;
+	struct mlx4_func func;
+
+	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
+		err = mlx4_QUERY_FUNC(dev, &func, 0);
+		if (err) {
+			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
+			return err;
+		}
+		dev_cap->max_eqs = func.max_eq;
+		dev_cap->reserved_eqs = func.rsvd_eqs;
+		dev_cap->reserved_uars = func.rsvd_uars;
+		err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
+	}
+	return err;
+}
+
 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
 {
 	struct mlx4_caps *dev_cap = &dev->caps;
···
 	}
 
 	dev->caps.num_ports	     = dev_cap->num_ports;
-	dev->phys_caps.num_phys_eqs  = MLX4_MAX_EQ_NUM;
+	dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
+	dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
+				      dev->caps.num_sys_eqs :
+				      MLX4_MAX_EQ_NUM;
 	for (i = 1; i <= dev->caps.num_ports; ++i) {
 		dev->caps.vl_cap[i]	    = dev_cap->max_vl[i];
 		dev->caps.ib_mtu_cap[i]	    = dev_cap->ib_mtu[i];
···
 	struct mlx4_dev_cap	   dev_cap;
 	struct mlx4_func_cap	   func_cap;
 	struct mlx4_init_hca_param hca_param;
-	int			   i;
+	u8			   i;
 
 	memset(&hca_param, 0, sizeof(hca_param));
 	err = mlx4_QUERY_HCA(dev, &hca_param);
···
 	}
 
 	for (i = 1; i <= dev->caps.num_ports; ++i) {
-		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
+		err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
 		if (err) {
 			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
 				 i, err);
···
 	if (err)
 		goto err_srq;
 
-	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
-		  dev->caps.num_eqs;
+	num_eqs = dev->phys_caps.num_phys_eqs;
 	err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
 				  cmpt_base +
 				  ((u64) (MLX4_CMPT_TYPE_EQ *
···
 	}
 
 
-	num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
-		   dev->caps.num_eqs;
+	num_eqs = dev->phys_caps.num_phys_eqs;
 	err = mlx4_init_icm_table(dev, &priv->eq_table.table,
 				  init_hca->eqc_base, dev_cap->eqc_entry_sz,
 				  num_eqs, num_eqs, 0, 0);
···
 	else {
 		mlx4_CLOSE_HCA(dev, 0);
 		mlx4_free_icms(dev);
+	}
+}
+
+static void mlx4_close_fw(struct mlx4_dev *dev)
+{
+	if (!mlx4_is_slave(dev)) {
 		mlx4_UNMAP_FA(dev);
 		mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
 	}
···
 		 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
 }
 
-static int mlx4_init_hca(struct mlx4_dev *dev)
+static int mlx4_init_fw(struct mlx4_dev *dev)
 {
-	struct mlx4_priv	  *priv = mlx4_priv(dev);
-	struct mlx4_adapter	   adapter;
-	struct mlx4_dev_cap	   dev_cap;
 	struct mlx4_mod_stat_cfg   mlx4_cfg;
-	struct mlx4_profile	   profile;
-	struct mlx4_init_hca_param init_hca;
-	u64 icm_size;
-	int err;
-	struct mlx4_config_dev_params params;
+	int err = 0;
 
 	if (!mlx4_is_slave(dev)) {
 		err = mlx4_QUERY_FW(dev);
···
 		err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
 		if (err)
 			mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
+	}
 
+	return err;
+}
+
+static int mlx4_init_hca(struct mlx4_dev *dev)
+{
+	struct mlx4_priv	  *priv = mlx4_priv(dev);
+	struct mlx4_adapter	   adapter;
+	struct mlx4_dev_cap	   dev_cap;
+	struct mlx4_profile	   profile;
+	struct mlx4_init_hca_param init_hca;
+	u64 icm_size;
+	struct mlx4_config_dev_params params;
+	int err;
+
+	if (!mlx4_is_slave(dev)) {
 		err = mlx4_dev_cap(dev, &dev_cap);
 		if (err) {
 			mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
···
 			mlx4_err(dev, "INIT_HCA command failed, aborting\n");
 			goto err_free_icm;
 		}
+
+		if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
+			err = mlx4_query_func(dev, &dev_cap);
+			if (err < 0) {
+				mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
+				goto err_stop_fw;
+			} else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
+				dev->caps.num_eqs = dev_cap.max_eqs;
+				dev->caps.reserved_eqs = dev_cap.reserved_eqs;
+				dev->caps.reserved_uars = dev_cap.reserved_uars;
+			}
+		}
+
 		/*
 		 * If TS is supported by FW
 		 * read HCA frequency by QUERY_HCA command
···
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
 	struct msix_entry *entries;
-	int nreq = min_t(int, dev->caps.num_ports *
-			 min_t(int, num_online_cpus() + 1,
-			       MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
 	int i;
 
 	if (msi_x) {
+		int nreq = dev->caps.num_ports * num_online_cpus() + MSIX_LEGACY_SZ;
+
 		nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
 			     nreq);
 
···
 	iounmap(owner);
 }
 
+#define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV)	==\
+				  !!((flags) & MLX4_FLAG_MASTER))
+
+static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
+			     u8 total_vfs, int existing_vfs)
+{
+	u64 dev_flags = dev->flags;
+
+	dev->dev_vfs = kzalloc(
+			total_vfs * sizeof(*dev->dev_vfs),
+			GFP_KERNEL);
+	if (NULL == dev->dev_vfs) {
+		mlx4_err(dev, "Failed to allocate memory for VFs\n");
+		goto disable_sriov;
+	} else if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
+		int err = 0;
+
+		atomic_inc(&pf_loading);
+		if (existing_vfs) {
+			if (existing_vfs != total_vfs)
+				mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
+					 existing_vfs, total_vfs);
+		} else {
+			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
+			err = pci_enable_sriov(pdev, total_vfs);
+		}
+		if (err) {
+			mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
+				 err);
+			atomic_dec(&pf_loading);
+			goto disable_sriov;
+		} else {
+			mlx4_warn(dev, "Running in master mode\n");
+			dev_flags |= MLX4_FLAG_SRIOV |
+				MLX4_FLAG_MASTER;
+			dev_flags &= ~MLX4_FLAG_SLAVE;
+			dev->num_vfs = total_vfs;
+		}
+	}
+	return dev_flags;
+
+disable_sriov:
+	dev->num_vfs = 0;
+	kfree(dev->dev_vfs);
+	return dev_flags & ~MLX4_FLAG_MASTER;
+}
+
+enum {
+	MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
+};
+
+static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
+			      int *nvfs)
+{
+	int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
+	/* Checking for 64 VFs as a limitation of CX2 */
+	if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
+	    requested_vfs >= 64) {
+		mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
+			 requested_vfs);
+		return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
+	}
+	return 0;
+}
+
 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
 			 int total_vfs, int *nvfs, struct mlx4_priv *priv)
 {
···
 	int err;
 	int port;
 	int i;
+	struct mlx4_dev_cap *dev_cap = NULL;
 	int existing_vfs = 0;
 
 	dev = &priv->dev;
···
 			}
 		}
 
-		if (total_vfs) {
-			mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n",
-				  total_vfs);
-			dev->dev_vfs = kzalloc(
-				total_vfs * sizeof(*dev->dev_vfs),
-				GFP_KERNEL);
-			if (NULL == dev->dev_vfs) {
-				mlx4_err(dev, "Failed to allocate memory for VFs\n");
-				err = -ENOMEM;
-				goto err_free_own;
-			} else {
-				atomic_inc(&pf_loading);
-				existing_vfs = pci_num_vf(pdev);
-				if (existing_vfs) {
-					err = 0;
-					if (existing_vfs != total_vfs)
-						mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
-							 existing_vfs, total_vfs);
-				} else {
-					err = pci_enable_sriov(pdev, total_vfs);
-				}
-				if (err) {
-					mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
-						 err);
-					atomic_dec(&pf_loading);
-				} else {
-					mlx4_warn(dev, "Running in master mode\n");
-					dev->flags |= MLX4_FLAG_SRIOV |
-						MLX4_FLAG_MASTER;
-					dev->num_vfs = total_vfs;
-				}
-			}
-		}
-
 		atomic_set(&priv->opreq_count, 0);
 		INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
 
···
 		if (err) {
 			mlx4_err(dev, "Failed to reset HCA, aborting\n");
 			goto err_sriov;
+		}
+
+		if (total_vfs) {
+			existing_vfs = pci_num_vf(pdev);
+			dev->flags = MLX4_FLAG_MASTER;
+			dev->num_vfs = total_vfs;
 		}
 	}
 
···
 	 * before posting commands. Also, init num_slaves before calling
 	 * mlx4_init_hca */
 	if (mlx4_is_mfunc(dev)) {
-		if (mlx4_is_master(dev))
+		if (mlx4_is_master(dev)) {
 			dev->num_slaves = MLX4_MAX_NUM_SLAVES;
-		else {
+
+		} else {
 			dev->num_slaves = 0;
 			err = mlx4_multi_func_init(dev);
 			if (err) {
···
 		}
 	}
 
+	err = mlx4_init_fw(dev);
+	if (err) {
+		mlx4_err(dev, "Failed to init fw, aborting.\n");
+		goto err_mfunc;
+	}
+
+	if (mlx4_is_master(dev)) {
+		if (!dev_cap) {
+			dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
+
+			if (!dev_cap) {
+				err = -ENOMEM;
+				goto err_fw;
+			}
+
+			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
+			if (err) {
+				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
+				goto err_fw;
+			}
+
+			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
+				goto err_fw;
+
+			if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
+				u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
+								  existing_vfs);
+
+				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
+				dev->flags = dev_flags;
+				if (!SRIOV_VALID_STATE(dev->flags)) {
+					mlx4_err(dev, "Invalid SRIOV state\n");
+					goto err_sriov;
+				}
+				err = mlx4_reset(dev);
+				if (err) {
+					mlx4_err(dev, "Failed to reset HCA, aborting.\n");
+					goto err_sriov;
+				}
+				goto slave_start;
+			}
+		} else {
+			/* Legacy mode FW requires SRIOV to be enabled before
+			 * doing QUERY_DEV_CAP, since max_eq's value is different if
+			 * SRIOV is enabled.
+			 */
+			memset(dev_cap, 0, sizeof(*dev_cap));
+			err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
+			if (err) {
+				mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
+				goto err_fw;
+			}
+
+			if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
+				goto err_fw;
+		}
+	}
+
 	err = mlx4_init_hca(dev);
 	if (err) {
 		if (err == -EACCES) {
 			/* Not primary Physical function
 			 * Running in slave mode */
-			mlx4_cmd_cleanup(dev);
+			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
+			/* We're not a PF */
+			if (dev->flags & MLX4_FLAG_SRIOV) {
+				if (!existing_vfs)
+					pci_disable_sriov(pdev);
+				if (mlx4_is_master(dev))
+					atomic_dec(&pf_loading);
+				dev->flags &= ~MLX4_FLAG_SRIOV;
+			}
+			if (!mlx4_is_slave(dev))
+				mlx4_free_ownership(dev);
 			dev->flags |= MLX4_FLAG_SLAVE;
 			dev->flags &= ~MLX4_FLAG_MASTER;
 			goto slave_start;
 		} else
-			goto err_mfunc;
+			goto err_fw;
+	}
+
+	if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
+		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs, existing_vfs);
+
+		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
+			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
+			dev->flags = dev_flags;
+			err = mlx4_cmd_init(dev);
+			if (err) {
+				/* Only VHCR is cleaned up, so could still
+				 * send FW commands
+				 */
+				mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
+				goto err_close;
+			}
+		} else {
+			dev->flags = dev_flags;
+		}
+
+		if (!SRIOV_VALID_STATE(dev->flags)) {
+			mlx4_err(dev, "Invalid SRIOV state\n");
+			goto err_close;
+		}
 	}
 
 	/* check if the device is functioning at its maximum possible speed.
···
 err_close:
 	mlx4_close_hca(dev);
 
+err_fw:
+	mlx4_close_fw(dev);
+
 err_mfunc:
 	if (mlx4_is_slave(dev))
 		mlx4_multi_func_cleanup(dev);
 
 err_cmd:
-	mlx4_cmd_cleanup(dev);
+	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
 
 err_sriov:
 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs)
···
 
 	kfree(priv->dev.dev_vfs);
 
-err_free_own:
 	if (!mlx4_is_slave(dev))
 		mlx4_free_ownership(dev);
 
+	kfree(dev_cap);
 	return err;
 }
 
···
 	if (mlx4_is_master(dev))
 		mlx4_multi_func_cleanup(dev);
 	mlx4_close_hca(dev);
+	mlx4_close_fw(dev);
 	if (mlx4_is_slave(dev))
 		mlx4_multi_func_cleanup(dev);
-	mlx4_cmd_cleanup(dev);
+	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
 
 	if (dev->flags & MLX4_FLAG_MSI_X)
 		pci_disable_msix(pdev);
 	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
 		mlx4_warn(dev, "Disabling SR-IOV\n");
 		pci_disable_sriov(pdev);
+		dev->flags &= ~MLX4_FLAG_SRIOV;
 		dev->num_vfs = 0;
 	}
 

drivers/net/ethernet/mellanox/mlx4/mlx4.h

···
 	u8			use_events;
 	u8			toggle;
 	u8			comm_toggle;
+	u8			initialized;
 };
 
 enum {
···
 
 int mlx4_GEN_EQE(struct mlx4_dev *dev, int slave, struct mlx4_eqe *eqe);
 
+enum {
+	MLX4_CMD_CLEANUP_STRUCT = 1UL << 0,
+	MLX4_CMD_CLEANUP_POOL	= 1UL << 1,
+	MLX4_CMD_CLEANUP_HCR	= 1UL << 2,
+	MLX4_CMD_CLEANUP_VHCR	= 1UL << 3,
+	MLX4_CMD_CLEANUP_ALL	= (MLX4_CMD_CLEANUP_VHCR << 1) - 1
+};
+
 int mlx4_cmd_init(struct mlx4_dev *dev);
-void mlx4_cmd_cleanup(struct mlx4_dev *dev);
+void mlx4_cmd_cleanup(struct mlx4_dev *dev, int cleanup_mask);
 int mlx4_multi_func_init(struct mlx4_dev *dev);
 void mlx4_multi_func_cleanup(struct mlx4_dev *dev);
 void mlx4_cmd_event(struct mlx4_dev *dev, u16 token, u8 status, u64 out_param);

drivers/net/ethernet/mellanox/mlx4/profile.c

···
 	profile[MLX4_RES_AUXC].num    = request->num_qp;
 	profile[MLX4_RES_SRQ].num     = request->num_srq;
 	profile[MLX4_RES_CQ].num      = request->num_cq;
-	profile[MLX4_RES_EQ].num      = mlx4_is_mfunc(dev) ?
-					dev->phys_caps.num_phys_eqs :
+	profile[MLX4_RES_EQ].num = mlx4_is_mfunc(dev) ? dev->phys_caps.num_phys_eqs :
 					min_t(unsigned, dev_cap->max_eqs, MAX_MSIX);
 	profile[MLX4_RES_DMPT].num    = request->num_mpt;
 	profile[MLX4_RES_CMPT].num    = MLX4_NUM_CMPTS;
···
 			init_hca->log_num_cqs = profile[i].log_num;
 			break;
 		case MLX4_RES_EQ:
-			dev->caps.num_eqs     = roundup_pow_of_two(min_t(unsigned, dev_cap->max_eqs,
-									 MAX_MSIX));
-			init_hca->eqc_base    = profile[i].start;
-			init_hca->log_num_eqs = ilog2(dev->caps.num_eqs);
+			if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
+				init_hca->log_num_eqs = 0x1f;
+				init_hca->eqc_base    = profile[i].start;
+				init_hca->num_sys_eqs = dev_cap->num_sys_eqs;
+			} else {
+				dev->caps.num_eqs     = roundup_pow_of_two(
+								min_t(unsigned,
+								      dev_cap->max_eqs,
+								      MAX_MSIX));
+				init_hca->eqc_base    = profile[i].start;
+				init_hca->log_num_eqs = ilog2(dev->caps.num_eqs);
+			}
 			break;
 		case MLX4_RES_DMPT:
 			dev->caps.num_mpts	= profile[i].num;

include/linux/mlx4/device.h

···
 
 enum {
 	MLX4_MAX_NUM_PF		= 16,
-	MLX4_MAX_NUM_VF		= 64,
+	MLX4_MAX_NUM_VF		= 126,
 	MLX4_MAX_NUM_VF_P_PORT  = 64,
 	MLX4_MFUNC_MAX		= 80,
 	MLX4_MAX_EQ_NUM		= 1024,
···
 	MLX4_DEV_CAP_FLAG2_EQE_STRIDE		= 1LL <<  13,
 	MLX4_DEV_CAP_FLAG2_ETH_PROT_CTRL        = 1LL <<  14,
 	MLX4_DEV_CAP_FLAG2_ETH_BACKPL_AN_REP	= 1LL <<  15,
-	MLX4_DEV_CAP_FLAG2_CONFIG_DEV		= 1LL <<  16
+	MLX4_DEV_CAP_FLAG2_CONFIG_DEV		= 1LL <<  16,
+	MLX4_DEV_CAP_FLAG2_SYS_EQS		= 1LL <<  17,
+	MLX4_DEV_CAP_FLAG2_80_VFS		= 1LL <<  18
 };
 
 enum {
···
 	int			num_cqs;
 	int			max_cqes;
 	int			reserved_cqs;
+	int			num_sys_eqs;
 	int			num_eqs;
 	int			reserved_eqs;
 	int			num_comp_vectors;