net/mlx4_core: Maintain a persistent memory for mlx4 device

Maintain a persistent memory that should survive reset flow/PCI error.
This comes as a preparation for coming series to support above flows.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/infiniband/hw/mlx4/alias_GUID.c

@@ -154,7 +154,7 @@ void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
 			continue;
 
 		slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ;
-		if (slave_id >= dev->dev->num_vfs + 1)
+		if (slave_id >= dev->dev->persist->num_vfs + 1)
 			return;
 		tmp_cur_ag = *(__be64 *)&p_data[i * GUID_REC_SIZE];
 		form_cache_ag = get_cached_alias_guid(dev, port_num,

drivers/infiniband/hw/mlx4/mad.c

@@ -1951,7 +1951,8 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,
 	ctx->ib_dev = &dev->ib_dev;
 
 	for (i = 0;
-	     i < min(dev->dev->caps.sqp_demux, (u16)(dev->dev->num_vfs + 1));
+	     i < min(dev->dev->caps.sqp_demux,
+	     (u16)(dev->dev->persist->num_vfs + 1));
 	     i++) {
 		struct mlx4_active_ports actv_ports =
 			mlx4_get_active_ports(dev->dev, i);

drivers/infiniband/hw/mlx4/main.c

@@ -198,7 +198,7 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 
 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 		0xffffff;
-	props->vendor_part_id	   = dev->dev->pdev->device;
+	props->vendor_part_id	   = dev->dev->persist->pdev->device;
 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
 
@@ -1375,7 +1375,7 @@ static ssize_t show_hca(struct device *device, struct device_attribute *attr,
 {
 	struct mlx4_ib_dev *dev =
 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
-	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
+	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
 }
 
 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
@@ -1937,7 +1937,8 @@ static void init_pkeys(struct mlx4_ib_dev *ibdev)
 	int i;
 
 	if (mlx4_is_master(ibdev->dev)) {
-		for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
+		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
+		     ++slave) {
 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
 				for (i = 0;
 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
@@ -1994,7 +1995,7 @@ static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
 		for (j = 0; j < eq_per_port; j++) {
 			snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
-				 i, j, dev->pdev->bus->name);
+				 i, j, dev->persist->pdev->bus->name);
 			/* Set IRQ for specific name (per ring) */
 			if (mlx4_assign_eq(dev, name, NULL,
 					   &ibdev->eq_table[eq])) {
@@ -2058,7 +2059,8 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 
 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
 	if (!ibdev) {
-		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
+		dev_err(&dev->persist->pdev->dev,
+			"Device struct alloc failed\n");
 		return NULL;
 	}
 
@@ -2085,7 +2087,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	ibdev->num_ports		= num_ports;
 	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
-	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
+	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
 
 	if (dev->caps.userspace_caps)
 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
@@ -2236,7 +2238,8 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 				sizeof(long),
 				GFP_KERNEL);
 		if (!ibdev->ib_uc_qpns_bitmap) {
-			dev_err(&dev->pdev->dev, "bit map alloc failed\n");
+			dev_err(&dev->persist->pdev->dev,
+				"bit map alloc failed\n");
 			goto err_steer_qp_release;
 		}
 

drivers/infiniband/hw/mlx4/mr.c

@@ -401,7 +401,8 @@ struct ib_fast_reg_page_list *mlx4_ib_alloc_fast_reg_page_list(struct ib_device
 	if (!mfrpl->ibfrpl.page_list)
 		goto err_free;
 
-	mfrpl->mapped_page_list = dma_alloc_coherent(&dev->dev->pdev->dev,
+	mfrpl->mapped_page_list = dma_alloc_coherent(&dev->dev->persist->
+						     pdev->dev,
 						     size, &mfrpl->map,
 						     GFP_KERNEL);
 	if (!mfrpl->mapped_page_list)
@@ -423,7 +424,8 @@ void mlx4_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list)
 	struct mlx4_ib_fast_reg_page_list *mfrpl = to_mfrpl(page_list);
 	int size = page_list->max_page_list_len * sizeof (u64);
 
-	dma_free_coherent(&dev->dev->pdev->dev, size, mfrpl->mapped_page_list,
+	dma_free_coherent(&dev->dev->persist->pdev->dev, size,
+			  mfrpl->mapped_page_list,
 			  mfrpl->map);
 	kfree(mfrpl->ibfrpl.page_list);
 	kfree(mfrpl);

drivers/infiniband/hw/mlx4/sysfs.c

@@ -375,7 +375,7 @@ static void get_name(struct mlx4_ib_dev *dev, char *name, int i, int max)
 	char base_name[9];
 
 	/* pci_name format is: bus:dev:func -> xxxx:yy:zz.n */
-	strlcpy(name, pci_name(dev->dev->pdev), max);
+	strlcpy(name, pci_name(dev->dev->persist->pdev), max);
 	strncpy(base_name, name, 8); /*till xxxx:yy:*/
 	base_name[8] = '\0';
 	/* with no ARI only 3 last bits are used so when the fn is higher than 8
@@ -792,7 +792,7 @@ static int register_pkey_tree(struct mlx4_ib_dev *device)
 	if (!mlx4_is_master(device->dev))
 		return 0;
 
-	for (i = 0; i <= device->dev->num_vfs; ++i)
+	for (i = 0; i <= device->dev->persist->num_vfs; ++i)
 		register_one_pkey_tree(device, i);
 
 	return 0;
@@ -807,7 +807,7 @@ static void unregister_pkey_tree(struct mlx4_ib_dev *device)
 	if (!mlx4_is_master(device->dev))
 		return;
 
-	for (slave = device->dev->num_vfs; slave >= 0; --slave) {
+	for (slave = device->dev->persist->num_vfs; slave >= 0; --slave) {
 		list_for_each_entry_safe(p, t,
 					 &device->pkeys.pkey_port_list[slave],
 					 entry) {

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

@@ -592,7 +592,7 @@ int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
 		buf->nbufs        = 1;
 		buf->npages       = 1;
 		buf->page_shift   = get_order(size) + PAGE_SHIFT;
-		buf->direct.buf   = dma_alloc_coherent(&dev->pdev->dev,
+		buf->direct.buf   = dma_alloc_coherent(&dev->persist->pdev->dev,
 						       size, &t, gfp);
 		if (!buf->direct.buf)
 			return -ENOMEM;
@@ -619,7 +619,8 @@ int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
 
 		for (i = 0; i < buf->nbufs; ++i) {
 			buf->page_list[i].buf =
-				dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
+				dma_alloc_coherent(&dev->persist->pdev->dev,
+						   PAGE_SIZE,
 						   &t, gfp);
 			if (!buf->page_list[i].buf)
 				goto err_free;
@@ -657,7 +658,8 @@ void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
 	int i;
 
 	if (buf->nbufs == 1)
-		dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
+		dma_free_coherent(&dev->persist->pdev->dev, size,
+				  buf->direct.buf,
 				  buf->direct.map);
 	else {
 		if (BITS_PER_LONG == 64 && buf->direct.buf)
@@ -665,7 +667,8 @@ void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
 
 		for (i = 0; i < buf->nbufs; ++i)
 			if (buf->page_list[i].buf)
-				dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
+				dma_free_coherent(&dev->persist->pdev->dev,
+						  PAGE_SIZE,
 						  buf->page_list[i].buf,
 						  buf->page_list[i].map);
 		kfree(buf->page_list);
@@ -738,7 +741,7 @@ int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order, gfp_t gfp
 		if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
 			goto out;
 
-	pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev), gfp);
+	pgdir = mlx4_alloc_db_pgdir(&dev->persist->pdev->dev, gfp);
 	if (!pgdir) {
 		ret = -ENOMEM;
 		goto out;
@@ -775,7 +778,7 @@ void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
 	set_bit(i, db->u.pgdir->bits[o]);
 
 	if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) {
-		dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
+		dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
 				  db->u.pgdir->db_page, db->u.pgdir->db_dma);
 		list_del(&db->u.pgdir->list);
 		kfree(db->u.pgdir);

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

@@ -70,7 +70,7 @@ static void poll_catas(unsigned long dev_ptr)
 
 	if (readl(priv->catas_err.map)) {
 		/* If the device is off-line, we cannot try to recover it */
-		if (pci_channel_offline(dev->pdev))
+		if (pci_channel_offline(dev->persist->pdev))
 			mod_timer(&priv->catas_err.timer,
 				  round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL));
 		else {
@@ -94,6 +94,7 @@ static void catas_reset(struct work_struct *work)
 {
 	struct mlx4_priv *priv, *tmppriv;
 	struct mlx4_dev *dev;
+	struct mlx4_dev_persistent *persist;
 
 	LIST_HEAD(tlist);
 	int ret;
@@ -103,20 +104,20 @@ static void catas_reset(struct work_struct *work)
 	spin_unlock_irq(&catas_lock);
 
 	list_for_each_entry_safe(priv, tmppriv, &tlist, catas_err.list) {
-		struct pci_dev *pdev = priv->dev.pdev;
+		struct pci_dev *pdev = priv->dev.persist->pdev;
 
 		/* If the device is off-line, we cannot reset it */
 		if (pci_channel_offline(pdev))
 			continue;
 
-		ret = mlx4_restart_one(priv->dev.pdev);
+		ret = mlx4_restart_one(priv->dev.persist->pdev);
 		/* 'priv' now is not valid */
 		if (ret)
 			pr_err("mlx4 %s: Reset failed (%d)\n",
 			       pci_name(pdev), ret);
 		else {
-			dev  = pci_get_drvdata(pdev);
-			mlx4_dbg(dev, "Reset succeeded\n");
+			persist  = pci_get_drvdata(pdev);
+			mlx4_dbg(persist->dev, "Reset succeeded\n");
 		}
 	}
 }
@@ -134,7 +135,7 @@ void mlx4_start_catas_poll(struct mlx4_dev *dev)
 	init_timer(&priv->catas_err.timer);
 	priv->catas_err.map = NULL;
 
-	addr = pci_resource_start(dev->pdev, priv->fw.catas_bar) +
+	addr = pci_resource_start(dev->persist->pdev, priv->fw.catas_bar) +
 		priv->fw.catas_offset;
 
 	priv->catas_err.map = ioremap(addr, priv->fw.catas_size * 4);

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

@@ -307,7 +307,7 @@ static int cmd_pending(struct mlx4_dev *dev)
 {
 	u32 status;
 
-	if (pci_channel_offline(dev->pdev))
+	if (pci_channel_offline(dev->persist->pdev))
 		return -EIO;
 
 	status = readl(mlx4_priv(dev)->cmd.hcr + HCR_STATUS_OFFSET);
@@ -328,7 +328,7 @@ static int mlx4_cmd_post(struct mlx4_dev *dev, u64 in_param, u64 out_param,
 
 	mutex_lock(&cmd->hcr_mutex);
 
-	if (pci_channel_offline(dev->pdev)) {
+	if (pci_channel_offline(dev->persist->pdev)) {
 		/*
 		 * Device is going through error recovery
 		 * and cannot accept commands.
@@ -342,7 +342,7 @@ static int mlx4_cmd_post(struct mlx4_dev *dev, u64 in_param, u64 out_param,
 		end += msecs_to_jiffies(GO_BIT_TIMEOUT_MSECS);
 
 	while (cmd_pending(dev)) {
-		if (pci_channel_offline(dev->pdev)) {
+		if (pci_channel_offline(dev->persist->pdev)) {
 			/*
 			 * Device is going through error recovery
 			 * and cannot accept commands.
@@ -464,7 +464,7 @@ static int mlx4_cmd_poll(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 
 	down(&priv->cmd.poll_sem);
 
-	if (pci_channel_offline(dev->pdev)) {
+	if (pci_channel_offline(dev->persist->pdev)) {
 		/*
 		 * Device is going through error recovery
 		 * and cannot accept commands.
@@ -487,7 +487,7 @@ static int mlx4_cmd_poll(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 
 	end = msecs_to_jiffies(timeout) + jiffies;
 	while (cmd_pending(dev) && time_before(jiffies, end)) {
-		if (pci_channel_offline(dev->pdev)) {
+		if (pci_channel_offline(dev->persist->pdev)) {
 			/*
 			 * Device is going through error recovery
 			 * and cannot accept commands.
@@ -612,7 +612,7 @@ int __mlx4_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 	       int out_is_imm, u32 in_modifier, u8 op_modifier,
 	       u16 op, unsigned long timeout, int native)
 {
-	if (pci_channel_offline(dev->pdev))
+	if (pci_channel_offline(dev->persist->pdev))
 		return -EIO;
 
 	if (!mlx4_is_mfunc(dev) || (native && mlx4_is_master(dev))) {
@@ -1997,11 +1997,12 @@ int mlx4_multi_func_init(struct mlx4_dev *dev)
 
 	if (mlx4_is_master(dev))
 		priv->mfunc.comm =
-		ioremap(pci_resource_start(dev->pdev, priv->fw.comm_bar) +
+		ioremap(pci_resource_start(dev->persist->pdev,
+					   priv->fw.comm_bar) +
 			priv->fw.comm_base, MLX4_COMM_PAGESIZE);
 	else
 		priv->mfunc.comm =
-		ioremap(pci_resource_start(dev->pdev, 2) +
+		ioremap(pci_resource_start(dev->persist->pdev, 2) +
 			MLX4_SLAVE_COMM_BASE, MLX4_COMM_PAGESIZE);
 	if (!priv->mfunc.comm) {
 		mlx4_err(dev, "Couldn't map communication vector\n");
@@ -2107,9 +2108,9 @@ err_comm_admin:
 err_comm:
 	iounmap(priv->mfunc.comm);
 err_vhcr:
-	dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
-					     priv->mfunc.vhcr,
-					     priv->mfunc.vhcr_dma);
+	dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
+			  priv->mfunc.vhcr,
+			  priv->mfunc.vhcr_dma);
 	priv->mfunc.vhcr = NULL;
 	return -ENOMEM;
 }
@@ -2130,8 +2131,8 @@ int mlx4_cmd_init(struct mlx4_dev *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);
+		priv->cmd.hcr = ioremap(pci_resource_start(dev->persist->pdev,
+					0) + MLX4_HCR_BASE, MLX4_HCR_SIZE);
 		if (!priv->cmd.hcr) {
 			mlx4_err(dev, "Couldn't map command register\n");
 			goto err;
@@ -2140,7 +2141,8 @@ int mlx4_cmd_init(struct mlx4_dev *dev)
 	}
 
 	if (mlx4_is_mfunc(dev) && !priv->mfunc.vhcr) {
-		priv->mfunc.vhcr = dma_alloc_coherent(&(dev->pdev->dev), PAGE_SIZE,
+		priv->mfunc.vhcr = dma_alloc_coherent(&dev->persist->pdev->dev,
+						      PAGE_SIZE,
 						      &priv->mfunc.vhcr_dma,
 						      GFP_KERNEL);
 		if (!priv->mfunc.vhcr)
@@ -2150,7 +2152,8 @@ int mlx4_cmd_init(struct mlx4_dev *dev)
 	}
 
 	if (!priv->cmd.pool) {
-		priv->cmd.pool = pci_pool_create("mlx4_cmd", dev->pdev,
+		priv->cmd.pool = pci_pool_create("mlx4_cmd",
+						 dev->persist->pdev,
 						 MLX4_MAILBOX_SIZE,
 						 MLX4_MAILBOX_SIZE, 0);
 		if (!priv->cmd.pool)
@@ -2202,7 +2205,7 @@ void mlx4_cmd_cleanup(struct mlx4_dev *dev, int cleanup_mask)
 	}
 	if (mlx4_is_mfunc(dev) && priv->mfunc.vhcr &&
 	    (cleanup_mask & MLX4_CMD_CLEANUP_VHCR)) {
-		dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
+		dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
 				  priv->mfunc.vhcr, priv->mfunc.vhcr_dma);
 		priv->mfunc.vhcr = NULL;
 	}
@@ -2306,8 +2309,9 @@ u32 mlx4_comm_get_version(void)
 
 static int mlx4_get_slave_indx(struct mlx4_dev *dev, int vf)
 {
-	if ((vf < 0) || (vf >= dev->num_vfs)) {
-		mlx4_err(dev, "Bad vf number:%d (number of activated vf: %d)\n", vf, dev->num_vfs);
+	if ((vf < 0) || (vf >= dev->persist->num_vfs)) {
+		mlx4_err(dev, "Bad vf number:%d (number of activated vf: %d)\n",
+			 vf, dev->persist->num_vfs);
 		return -EINVAL;
 	}
 
@@ -2316,7 +2320,7 @@ static int mlx4_get_slave_indx(struct mlx4_dev *dev, int vf)
 
 int mlx4_get_vf_indx(struct mlx4_dev *dev, int slave)
 {
-	if (slave < 1 || slave > dev->num_vfs) {
+	if (slave < 1 || slave > dev->persist->num_vfs) {
 		mlx4_err(dev,
 			 "Bad slave number:%d (number of activated slaves: %lu)\n",
 			 slave, dev->num_slaves);
@@ -2388,7 +2392,7 @@ struct mlx4_slaves_pport mlx4_phys_to_slaves_pport(struct mlx4_dev *dev,
 	if (port <= 0 || port > dev->caps.num_ports)
 		return slaves_pport;
 
-	for (i = 0; i < dev->num_vfs + 1; i++) {
+	for (i = 0; i < dev->persist->num_vfs + 1; i++) {
 		struct mlx4_active_ports actv_ports =
 			mlx4_get_active_ports(dev, i);
 		if (test_bit(port - 1, actv_ports.ports))
@@ -2408,7 +2412,7 @@ struct mlx4_slaves_pport mlx4_phys_to_slaves_pport_actv(
 
 	bitmap_zero(slaves_pport.slaves, MLX4_MFUNC_MAX);
 
-	for (i = 0; i < dev->num_vfs + 1; i++) {
+	for (i = 0; i < dev->persist->num_vfs + 1; i++) {
 		struct mlx4_active_ports actv_ports =
 			mlx4_get_active_ports(dev, i);
 		if (bitmap_equal(crit_ports->ports, actv_ports.ports,

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

@@ -70,10 +70,10 @@ int mlx4_en_create_cq(struct mlx4_en_priv *priv,
 	/* Allocate HW buffers on provided NUMA node.
 	 * dev->numa_node is used in mtt range allocation flow.
 	 */
-	set_dev_node(&mdev->dev->pdev->dev, node);
+	set_dev_node(&mdev->dev->persist->pdev->dev, node);
 	err = mlx4_alloc_hwq_res(mdev->dev, &cq->wqres,
 				cq->buf_size, 2 * PAGE_SIZE);
-	set_dev_node(&mdev->dev->pdev->dev, mdev->dev->numa_node);
+	set_dev_node(&mdev->dev->persist->pdev->dev, mdev->dev->numa_node);
 	if (err)
 		goto err_cq;
 

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

@@ -92,7 +92,7 @@ mlx4_en_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *drvinfo)
 		(u16) (mdev->dev->caps.fw_ver >> 32),
 		(u16) ((mdev->dev->caps.fw_ver >> 16) & 0xffff),
 		(u16) (mdev->dev->caps.fw_ver & 0xffff));
-	strlcpy(drvinfo->bus_info, pci_name(mdev->dev->pdev),
+	strlcpy(drvinfo->bus_info, pci_name(mdev->dev->persist->pdev),
 		sizeof(drvinfo->bus_info));
 	drvinfo->n_stats = 0;
 	drvinfo->regdump_len = 0;

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

@@ -241,8 +241,8 @@ static void *mlx4_en_add(struct mlx4_dev *dev)
 	spin_lock_init(&mdev->uar_lock);
 
 	mdev->dev = dev;
-	mdev->dma_device = &(dev->pdev->dev);
-	mdev->pdev = dev->pdev;
+	mdev->dma_device = &dev->persist->pdev->dev;
+	mdev->pdev = dev->persist->pdev;
 	mdev->device_up = false;
 
 	mdev->LSO_support = !!(dev->caps.flags & (1 << 15));

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

@@ -2457,7 +2457,7 @@ int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
 
-	SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
+	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
 	dev->dev_port = port - 1;
 
 	/*

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

@@ -387,10 +387,10 @@ int mlx4_en_create_rx_ring(struct mlx4_en_priv *priv,
 		 ring->rx_info, tmp);
 
 	/* Allocate HW buffers on provided NUMA node */
-	set_dev_node(&mdev->dev->pdev->dev, node);
+	set_dev_node(&mdev->dev->persist->pdev->dev, node);
 	err = mlx4_alloc_hwq_res(mdev->dev, &ring->wqres,
 				 ring->buf_size, 2 * PAGE_SIZE);
-	set_dev_node(&mdev->dev->pdev->dev, mdev->dev->numa_node);
+	set_dev_node(&mdev->dev->persist->pdev->dev, mdev->dev->numa_node);
 	if (err)
 		goto err_info;
 

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

@@ -91,10 +91,10 @@ int mlx4_en_create_tx_ring(struct mlx4_en_priv *priv,
 	ring->buf_size = ALIGN(size * ring->stride, MLX4_EN_PAGE_SIZE);
 
 	/* Allocate HW buffers on provided NUMA node */
-	set_dev_node(&mdev->dev->pdev->dev, node);
+	set_dev_node(&mdev->dev->persist->pdev->dev, node);
 	err = mlx4_alloc_hwq_res(mdev->dev, &ring->wqres, ring->buf_size,
 				 2 * PAGE_SIZE);
-	set_dev_node(&mdev->dev->pdev->dev, mdev->dev->numa_node);
+	set_dev_node(&mdev->dev->persist->pdev->dev, mdev->dev->numa_node);
 	if (err) {
 		en_err(priv, "Failed allocating hwq resources\n");
 		goto err_bounce;

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

@@ -237,7 +237,7 @@ int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
 	struct mlx4_eqe eqe;
 
 	/*don't send if we don't have the that slave */
-	if (dev->num_vfs < slave)
+	if (dev->persist->num_vfs < slave)
 		return 0;
 	memset(&eqe, 0, sizeof eqe);
 
@@ -255,7 +255,7 @@ int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
 	struct mlx4_eqe eqe;
 
 	/*don't send if we don't have the that slave */
-	if (dev->num_vfs < slave)
+	if (dev->persist->num_vfs < slave)
 		return 0;
 	memset(&eqe, 0, sizeof eqe);
 
@@ -310,7 +310,7 @@ static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
 	struct mlx4_slaves_pport slaves_pport = mlx4_phys_to_slaves_pport(dev,
 									  port);
 
-	for (i = 0; i < dev->num_vfs + 1; i++)
+	for (i = 0; i < dev->persist->num_vfs + 1; i++)
 		if (test_bit(i, slaves_pport.slaves))
 			set_and_calc_slave_port_state(dev, i, port,
 						      event, &gen_event);
@@ -560,7 +560,8 @@ static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
 				mlx4_priv(dev)->sense.do_sense_port[port] = 1;
 				if (!mlx4_is_master(dev))
 					break;
-				for (i = 0; i < dev->num_vfs + 1; i++) {
+				for (i = 0; i < dev->persist->num_vfs + 1;
+				     i++) {
 					if (!test_bit(i, slaves_port.slaves))
 						continue;
 					if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
@@ -596,7 +597,9 @@ static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
 				if (!mlx4_is_master(dev))
 					break;
 				if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
-					for (i = 0; i < dev->num_vfs + 1; i++) {
+					for (i = 0;
+					     i < dev->persist->num_vfs + 1;
+					     i++) {
 						if (!test_bit(i, slaves_port.slaves))
 							continue;
 						if (i == mlx4_master_func_num(dev))
@@ -865,7 +868,7 @@ static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
 
 	if (!priv->eq_table.uar_map[index]) {
 		priv->eq_table.uar_map[index] =
-			ioremap(pci_resource_start(dev->pdev, 2) +
+			ioremap(pci_resource_start(dev->persist->pdev, 2) +
 				((eq->eqn / 4) << PAGE_SHIFT),
 				PAGE_SIZE);
 		if (!priv->eq_table.uar_map[index]) {
@@ -928,8 +931,10 @@ static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
 	eq_context = mailbox->buf;
 
 	for (i = 0; i < npages; ++i) {
-		eq->page_list[i].buf = dma_alloc_coherent(&dev->pdev->dev,
-							  PAGE_SIZE, &t, GFP_KERNEL);
+		eq->page_list[i].buf = dma_alloc_coherent(&dev->persist->
+							  pdev->dev,
+							  PAGE_SIZE, &t,
+							  GFP_KERNEL);
 		if (!eq->page_list[i].buf)
 			goto err_out_free_pages;
 
@@ -995,7 +1000,7 @@ err_out_free_eq:
 err_out_free_pages:
 	for (i = 0; i < npages; ++i)
 		if (eq->page_list[i].buf)
-			dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
+			dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
 					  eq->page_list[i].buf,
 					  eq->page_list[i].map);
 
@@ -1044,9 +1049,9 @@ static void mlx4_free_eq(struct mlx4_dev *dev,
 
 	mlx4_mtt_cleanup(dev, &eq->mtt);
 	for (i = 0; i < npages; ++i)
-		dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
-				    eq->page_list[i].buf,
-				    eq->page_list[i].map);
+		dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
+				  eq->page_list[i].buf,
+				  eq->page_list[i].map);
 
 	kfree(eq->page_list);
 	mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn, MLX4_USE_RR);
@@ -1060,7 +1065,7 @@ static void mlx4_free_irqs(struct mlx4_dev *dev)
 	int	i, vec;
 
 	if (eq_table->have_irq)
-		free_irq(dev->pdev->irq, dev);
+		free_irq(dev->persist->pdev->irq, dev);
 
 	for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
 		if (eq_table->eq[i].have_irq) {
@@ -1089,7 +1094,8 @@ static int mlx4_map_clr_int(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
 
-	priv->clr_base = ioremap(pci_resource_start(dev->pdev, priv->fw.clr_int_bar) +
+	priv->clr_base = ioremap(pci_resource_start(dev->persist->pdev,
+				 priv->fw.clr_int_bar) +
 				 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
 	if (!priv->clr_base) {
 		mlx4_err(dev, "Couldn't map interrupt clear register, aborting\n");
@@ -1212,13 +1218,13 @@ int mlx4_init_eq_table(struct mlx4_dev *dev)
 					 i * MLX4_IRQNAME_SIZE,
 					 MLX4_IRQNAME_SIZE,
 					 "mlx4-comp-%d@pci:%s", i,
-					 pci_name(dev->pdev));
+					 pci_name(dev->persist->pdev));
 			} else {
 				snprintf(priv->eq_table.irq_names +
 					 i * MLX4_IRQNAME_SIZE,
 					 MLX4_IRQNAME_SIZE,
 					 "mlx4-async@pci:%s",
-					 pci_name(dev->pdev));
+					 pci_name(dev->persist->pdev));
 			}
 
 			eq_name = priv->eq_table.irq_names +
@@ -1235,8 +1241,8 @@ int mlx4_init_eq_table(struct mlx4_dev *dev)
 		snprintf(priv->eq_table.irq_names,
 			 MLX4_IRQNAME_SIZE,
 			 DRV_NAME "@pci:%s",
-			 pci_name(dev->pdev));
-		err = request_irq(dev->pdev->irq, mlx4_interrupt,
+			 pci_name(dev->persist->pdev));
+		err = request_irq(dev->persist->pdev->irq, mlx4_interrupt,
 				  IRQF_SHARED, priv->eq_table.irq_names, dev);
 		if (err)
 			goto err_out_async;

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

@@ -56,7 +56,7 @@ static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct mlx4_icm_chunk *chu
 	int i;
 
 	if (chunk->nsg > 0)
-		pci_unmap_sg(dev->pdev, chunk->mem, chunk->npages,
+		pci_unmap_sg(dev->persist->pdev, chunk->mem, chunk->npages,
 			     PCI_DMA_BIDIRECTIONAL);
 
 	for (i = 0; i < chunk->npages; ++i)
@@ -69,7 +69,8 @@ static void mlx4_free_icm_coherent(struct mlx4_dev *dev, struct mlx4_icm_chunk *
 	int i;
 
 	for (i = 0; i < chunk->npages; ++i)
-		dma_free_coherent(&dev->pdev->dev, chunk->mem[i].length,
+		dma_free_coherent(&dev->persist->pdev->dev,
+				  chunk->mem[i].length,
 				  lowmem_page_address(sg_page(&chunk->mem[i])),
 				  sg_dma_address(&chunk->mem[i]));
 }
@@ -173,7 +174,7 @@ struct mlx4_icm *mlx4_alloc_icm(struct mlx4_dev *dev, int npages,
 			--cur_order;
 
 		if (coherent)
-			ret = mlx4_alloc_icm_coherent(&dev->pdev->dev,
+			ret = mlx4_alloc_icm_coherent(&dev->persist->pdev->dev,
 						      &chunk->mem[chunk->npages],
 						      cur_order, gfp_mask);
 		else
@@ -193,7 +194,7 @@ struct mlx4_icm *mlx4_alloc_icm(struct mlx4_dev *dev, int npages,
 		if (coherent)
 			++chunk->nsg;
 		else if (chunk->npages == MLX4_ICM_CHUNK_LEN) {
-			chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
+			chunk->nsg = pci_map_sg(dev->persist->pdev, chunk->mem,
 						chunk->npages,
 						PCI_DMA_BIDIRECTIONAL);
 
@@ -208,7 +209,7 @@ struct mlx4_icm *mlx4_alloc_icm(struct mlx4_dev *dev, int npages,
 	}
 
 	if (!coherent && chunk) {
-		chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
+		chunk->nsg = pci_map_sg(dev->persist->pdev, chunk->mem,
 					chunk->npages,
 					PCI_DMA_BIDIRECTIONAL);
 

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

@@ -318,10 +318,11 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 		return -ENODEV;
 	}
 
-	if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
+	if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
 		mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
 			 dev_cap->uar_size,
-			 (unsigned long long) pci_resource_len(dev->pdev, 2));
+			 (unsigned long long)
+			 pci_resource_len(dev->persist->pdev, 2));
 		return -ENODEV;
 	}
 
@@ -541,8 +542,10 @@ static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
 	*speed = PCI_SPEED_UNKNOWN;
 	*width = PCIE_LNK_WIDTH_UNKNOWN;
 
-	err1 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP, &lnkcap1);
-	err2 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP2, &lnkcap2);
+	err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
+					  &lnkcap1);
+	err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
+					  &lnkcap2);
 	if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
 		if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
 			*speed = PCIE_SPEED_8_0GT;
@@ -587,7 +590,7 @@ static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
 		return;
 	}
 
-	err = pcie_get_minimum_link(dev->pdev, &speed, &width);
+	err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
 	if (err || speed == PCI_SPEED_UNKNOWN ||
 	    width == PCIE_LNK_WIDTH_UNKNOWN) {
 		mlx4_warn(dev,
@@ -837,10 +840,12 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
 
 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
 				       dev->caps.reserved_uars) >
-				       pci_resource_len(dev->pdev, 2)) {
+				       pci_resource_len(dev->persist->pdev,
+							2)) {
 		mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
 			 dev->caps.uar_page_size * dev->caps.num_uars,
-			 (unsigned long long) pci_resource_len(dev->pdev, 2));
+			 (unsigned long long)
+			 pci_resource_len(dev->persist->pdev, 2));
 		goto err_mem;
 	}
 
@@ -1492,9 +1497,9 @@ static int map_bf_area(struct mlx4_dev *dev)
 	if (!dev->caps.bf_reg_size)
 		return -ENXIO;
 
-	bf_start = pci_resource_start(dev->pdev, 2) +
+	bf_start = pci_resource_start(dev->persist->pdev, 2) +
 			(dev->caps.num_uars << PAGE_SHIFT);
-	bf_len = pci_resource_len(dev->pdev, 2) -
+	bf_len = pci_resource_len(dev->persist->pdev, 2) -
 			(dev->caps.num_uars << PAGE_SHIFT);
 	priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
 	if (!priv->bf_mapping)
@@ -1536,7 +1541,8 @@ static int map_internal_clock(struct mlx4_dev *dev)
 	struct mlx4_priv *priv = mlx4_priv(dev);
 
 	priv->clock_mapping =
-		ioremap(pci_resource_start(dev->pdev, priv->fw.clock_bar) +
+		ioremap(pci_resource_start(dev->persist->pdev,
+					   priv->fw.clock_bar) +
 			priv->fw.clock_offset, MLX4_CLOCK_SIZE);
 
 	if (!priv->clock_mapping)
@@ -1705,7 +1711,8 @@ static void choose_steering_mode(struct mlx4_dev *dev,
 	if (mlx4_log_num_mgm_entry_size <= 0 &&
 	    dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
 	    (!mlx4_is_mfunc(dev) ||
-	     (dev_cap->fs_max_num_qp_per_entry >= (dev->num_vfs + 1))) &&
+	     (dev_cap->fs_max_num_qp_per_entry >=
+	     (dev->persist->num_vfs + 1))) &&
 	    choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
 		MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
 		dev->oper_log_mgm_entry_size =
@@ -2288,7 +2295,8 @@ static void mlx4_enable_msi_x(struct mlx4_dev *dev)
 		for (i = 0; i < nreq; ++i)
 			entries[i].entry = i;
 
-		nreq = pci_enable_msix_range(dev->pdev, entries, 2, nreq);
+		nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
+					     nreq);
 
 		if (nreq < 0) {
 			kfree(entries);
@@ -2316,7 +2324,7 @@ no_msi:
 	dev->caps.comp_pool	   = 0;
 
 	for (i = 0; i < 2; ++i)
-		priv->eq_table.eq[i].irq = dev->pdev->irq;
+		priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
 }
 
 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
@@ -2344,7 +2352,7 @@ static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
 	info->port_attr.show      = show_port_type;
 	sysfs_attr_init(&info->port_attr.attr);
 
-	err = device_create_file(&dev->pdev->dev, &info->port_attr);
+	err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
 	if (err) {
 		mlx4_err(dev, "Failed to create file for port %d\n", port);
 		info->port = -1;
@@ -2361,10 +2369,12 @@ static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
 	info->port_mtu_attr.show      = show_port_ib_mtu;
 	sysfs_attr_init(&info->port_mtu_attr.attr);
 
-	err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
+	err = device_create_file(&dev->persist->pdev->dev,
+				 &info->port_mtu_attr);
 	if (err) {
 		mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
-		device_remove_file(&info->dev->pdev->dev, &info->port_attr);
+		device_remove_file(&info->dev->persist->pdev->dev,
+				   &info->port_attr);
 		info->port = -1;
 	}
 
@@ -2376,8 +2386,9 @@ static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
 	if (info->port < 0)
 		return;
 
-	device_remove_file(&info->dev->pdev->dev, &info->port_attr);
-	device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
+	device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
+	device_remove_file(&info->dev->persist->pdev->dev,
+			   &info->port_mtu_attr);
 }
 
 static int mlx4_init_steering(struct mlx4_dev *dev)
@@ -2444,10 +2455,11 @@ static int mlx4_get_ownership(struct mlx4_dev *dev)
 	void __iomem *owner;
 	u32 ret;
 
-	if (pci_channel_offline(dev->pdev))
+	if (pci_channel_offline(dev->persist->pdev))
 		return -EIO;
 
-	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
+	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
+			MLX4_OWNER_BASE,
 			MLX4_OWNER_SIZE);
 	if (!owner) {
 		mlx4_err(dev, "Failed to obtain ownership bit\n");
@@ -2463,10 +2475,11 @@ static void mlx4_free_ownership(struct mlx4_dev *dev)
 {
 	void __iomem *owner;
 
-	if (pci_channel_offline(dev->pdev))
+	if (pci_channel_offline(dev->persist->pdev))
 		return;
 
-	owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
+	owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
+			MLX4_OWNER_BASE,
 			MLX4_OWNER_SIZE);
 	if (!owner) {
 		mlx4_err(dev, "Failed to obtain ownership bit\n");
@@ -2514,13 +2527,13 @@ static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
 		dev_flags |= MLX4_FLAG_SRIOV |
 			MLX4_FLAG_MASTER;
 		dev_flags &= ~MLX4_FLAG_SLAVE;
-		dev->num_vfs = total_vfs;
+		dev->persist->num_vfs = total_vfs;
 	}
 	return dev_flags;
 
 disable_sriov:
 	atomic_dec(&pf_loading);
-	dev->num_vfs = 0;
+	dev->persist->num_vfs = 0;
 	kfree(dev->dev_vfs);
 	return dev_flags & ~MLX4_FLAG_MASTER;
 }
@@ -2607,7 +2620,7 @@ static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
 			existing_vfs = pci_num_vf(pdev);
 			if (existing_vfs)
 				dev->flags |= MLX4_FLAG_SRIOV;
-			dev->num_vfs = total_vfs;
+			dev->persist->num_vfs = total_vfs;
 		}
 	}
 
@@ -2771,12 +2784,14 @@ slave_start:
 				 dev->caps.num_ports);
 			goto err_close;
 		}
-		memcpy(dev->nvfs, nvfs, sizeof(dev->nvfs));
+		memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
 
-		for (i = 0; i < sizeof(dev->nvfs)/sizeof(dev->nvfs[0]); i++) {
+		for (i = 0;
+		     i < sizeof(dev->persist->nvfs)/
+		     sizeof(dev->persist->nvfs[0]); i++) {
 			unsigned j;
 
-			for (j = 0; j < dev->nvfs[i]; ++sum, ++j) {
+			for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
 				dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
 				dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
 					dev->caps.num_ports;
@@ -2846,7 +2861,7 @@ slave_start:
 
 	priv->removed = 0;
 
-	if (mlx4_is_master(dev) && dev->num_vfs)
+	if (mlx4_is_master(dev) && dev->persist->num_vfs)
 		atomic_dec(&pf_loading);
 
 	kfree(dev_cap);
@@ -2908,7 +2923,7 @@ err_sriov:
 	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs)
 		pci_disable_sriov(pdev);
 
-	if (mlx4_is_master(dev) && dev->num_vfs)
+	if (mlx4_is_master(dev) && dev->persist->num_vfs)
 		atomic_dec(&pf_loading);
 
 	kfree(priv->dev.dev_vfs);
@@ -3076,20 +3091,28 @@ static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
 		return -ENOMEM;
 
 	dev       = &priv->dev;
-	dev->pdev = pdev;
-	pci_set_drvdata(pdev, dev);
+	dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
+	if (!dev->persist) {
+		kfree(priv);
+		return -ENOMEM;
+	}
+	dev->persist->pdev = pdev;
+	dev->persist->dev = dev;
+	pci_set_drvdata(pdev, dev->persist);
 	priv->pci_dev_data = id->driver_data;
 
 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
-	if (ret)
+	if (ret) {
+		kfree(dev->persist);
 		kfree(priv);
-
+	}
 	return ret;
 }
 
 static void mlx4_unload_one(struct pci_dev *pdev)
 {
-	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
+	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
+	struct mlx4_dev  *dev  = persist->dev;
 	struct mlx4_priv *priv = mlx4_priv(dev);
 	int               pci_dev_data;
 	int p;
@@ -3155,7 +3178,7 @@ static void mlx4_unload_one(struct pci_dev *pdev)
 		mlx4_warn(dev, "Disabling SR-IOV\n");
 		pci_disable_sriov(pdev);
 		dev->flags &= ~MLX4_FLAG_SRIOV;
-		dev->num_vfs = 0;
+		dev->persist->num_vfs = 0;
 	}
 
 	if (!mlx4_is_slave(dev))
@@ -3175,26 +3198,29 @@ static void mlx4_unload_one(struct pci_dev *pdev)
 
 static void mlx4_remove_one(struct pci_dev *pdev)
 {
-	struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
+	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
+	struct mlx4_dev  *dev  = persist->dev;
 	struct mlx4_priv *priv = mlx4_priv(dev);
 
 	mlx4_unload_one(pdev);
 	pci_release_regions(pdev);
 	pci_disable_device(pdev);
+	kfree(dev->persist);
 	kfree(priv);
 	pci_set_drvdata(pdev, NULL);
 }
 
 int mlx4_restart_one(struct pci_dev *pdev)
 {
-	struct mlx4_dev	 *dev  = pci_get_drvdata(pdev);
+	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
+	struct mlx4_dev	 *dev  = persist->dev;
 	struct mlx4_priv *priv = mlx4_priv(dev);
 	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
 	int pci_dev_data, err, total_vfs;
 
 	pci_dev_data = priv->pci_dev_data;
-	total_vfs = dev->num_vfs;
-	memcpy(nvfs, dev->nvfs, sizeof(dev->nvfs));
+	total_vfs = dev->persist->num_vfs;
+	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
 
 	mlx4_unload_one(pdev);
 	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv);

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

@@ -221,16 +221,17 @@ extern int mlx4_debug_level;
 #define mlx4_dbg(mdev, format, ...)					\
 do {									\
 	if (mlx4_debug_level)						\
-		dev_printk(KERN_DEBUG, &(mdev)->pdev->dev, format,	\
+		dev_printk(KERN_DEBUG,					\
+			   &(mdev)->persist->pdev->dev, format,		\
 			   ##__VA_ARGS__);				\
 } while (0)
 
 #define mlx4_err(mdev, format, ...)					\
-	dev_err(&(mdev)->pdev->dev, format, ##__VA_ARGS__)
+	dev_err(&(mdev)->persist->pdev->dev, format, ##__VA_ARGS__)
 #define mlx4_info(mdev, format, ...)					\
-	dev_info(&(mdev)->pdev->dev, format, ##__VA_ARGS__)
+	dev_info(&(mdev)->persist->pdev->dev, format, ##__VA_ARGS__)
 #define mlx4_warn(mdev, format, ...)					\
-	dev_warn(&(mdev)->pdev->dev, format, ##__VA_ARGS__)
+	dev_warn(&(mdev)->persist->pdev->dev, format, ##__VA_ARGS__)
 
 extern int mlx4_log_num_mgm_entry_size;
 extern int log_mtts_per_seg;

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

@@ -708,13 +708,13 @@ static int mlx4_write_mtt_chunk(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
 	if (!mtts)
 		return -ENOMEM;
 
-	dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle,
+	dma_sync_single_for_cpu(&dev->persist->pdev->dev, dma_handle,
 				npages * sizeof (u64), DMA_TO_DEVICE);
 
 	for (i = 0; i < npages; ++i)
 		mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
 
-	dma_sync_single_for_device(&dev->pdev->dev, dma_handle,
+	dma_sync_single_for_device(&dev->persist->pdev->dev, dma_handle,
 				   npages * sizeof (u64), DMA_TO_DEVICE);
 
 	return 0;
@@ -1020,13 +1020,13 @@ int mlx4_map_phys_fmr(struct mlx4_dev *dev, struct mlx4_fmr *fmr, u64 *page_list
 	/* Make sure MPT status is visible before writing MTT entries */
 	wmb();
 
-	dma_sync_single_for_cpu(&dev->pdev->dev, fmr->dma_handle,
+	dma_sync_single_for_cpu(&dev->persist->pdev->dev, fmr->dma_handle,
 				npages * sizeof(u64), DMA_TO_DEVICE);
 
 	for (i = 0; i < npages; ++i)
 		fmr->mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
 
-	dma_sync_single_for_device(&dev->pdev->dev, fmr->dma_handle,
+	dma_sync_single_for_device(&dev->persist->pdev->dev, fmr->dma_handle,
 				   npages * sizeof(u64), DMA_TO_DEVICE);
 
 	fmr->mpt->key    = cpu_to_be32(key);

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

@@ -151,11 +151,13 @@ int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar)
 		return -ENOMEM;
 
 	if (mlx4_is_slave(dev))
-		offset = uar->index % ((int) pci_resource_len(dev->pdev, 2) /
+		offset = uar->index % ((int)pci_resource_len(dev->persist->pdev,
+							     2) /
 				       dev->caps.uar_page_size);
 	else
 		offset = uar->index;
-	uar->pfn = (pci_resource_start(dev->pdev, 2) >> PAGE_SHIFT) + offset;
+	uar->pfn = (pci_resource_start(dev->persist->pdev, 2) >> PAGE_SHIFT)
+		    + offset;
 	uar->map = NULL;
 	return 0;
 }

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

@@ -553,9 +553,9 @@ int mlx4_get_slave_num_gids(struct mlx4_dev *dev, int slave, int port)
 		slaves_pport_actv = mlx4_phys_to_slaves_pport_actv(
 				    dev, &exclusive_ports);
 		slave_gid -= bitmap_weight(slaves_pport_actv.slaves,
-					   dev->num_vfs + 1);
+					   dev->persist->num_vfs + 1);
 	}
-	vfs = bitmap_weight(slaves_pport.slaves, dev->num_vfs + 1) - 1;
+	vfs = bitmap_weight(slaves_pport.slaves, dev->persist->num_vfs + 1) - 1;
 	if (slave_gid <= ((MLX4_ROCE_MAX_GIDS - MLX4_ROCE_PF_GIDS) % vfs))
 		return ((MLX4_ROCE_MAX_GIDS - MLX4_ROCE_PF_GIDS) / vfs) + 1;
 	return (MLX4_ROCE_MAX_GIDS - MLX4_ROCE_PF_GIDS) / vfs;
@@ -590,10 +590,10 @@ int mlx4_get_base_gid_ix(struct mlx4_dev *dev, int slave, int port)
 		slaves_pport_actv = mlx4_phys_to_slaves_pport_actv(
 				    dev, &exclusive_ports);
 		slave_gid -= bitmap_weight(slaves_pport_actv.slaves,
-					   dev->num_vfs + 1);
+					   dev->persist->num_vfs + 1);
 	}
 	gids = MLX4_ROCE_MAX_GIDS - MLX4_ROCE_PF_GIDS;
-	vfs = bitmap_weight(slaves_pport.slaves, dev->num_vfs + 1) - 1;
+	vfs = bitmap_weight(slaves_pport.slaves, dev->persist->num_vfs + 1) - 1;
 	if (slave_gid <= gids % vfs)
 		return MLX4_ROCE_PF_GIDS + ((gids / vfs) + 1) * (slave_gid - 1);
 
@@ -644,7 +644,7 @@ void mlx4_reset_roce_gids(struct mlx4_dev *dev, int slave)
 	int num_eth_ports, err;
 	int i;
 
-	if (slave < 0 || slave > dev->num_vfs)
+	if (slave < 0 || slave > dev->persist->num_vfs)
 		return;
 
 	actv_ports = mlx4_get_active_ports(dev, slave);
@@ -1214,7 +1214,8 @@ int mlx4_get_slave_from_roce_gid(struct mlx4_dev *dev, int port, u8 *gid,
 		return -EINVAL;
 
 	slaves_pport = mlx4_phys_to_slaves_pport(dev, port);
-	num_vfs = bitmap_weight(slaves_pport.slaves, dev->num_vfs + 1) - 1;
+	num_vfs = bitmap_weight(slaves_pport.slaves,
+				dev->persist->num_vfs + 1) - 1;
 
 	for (i = 0; i < MLX4_ROCE_MAX_GIDS; i++) {
 		if (!memcmp(priv->port[port].gid_table.roce_gids[i].raw, gid,
@@ -1258,7 +1259,7 @@ int mlx4_get_slave_from_roce_gid(struct mlx4_dev *dev, int port, u8 *gid,
 							dev, &exclusive_ports);
 				num_vfs_before += bitmap_weight(
 						slaves_pport_actv.slaves,
-						dev->num_vfs + 1);
+						dev->persist->num_vfs + 1);
 			}
 
 			/* candidate_slave_gid isn't necessarily the correct slave, but
@@ -1288,7 +1289,7 @@ int mlx4_get_slave_from_roce_gid(struct mlx4_dev *dev, int port, u8 *gid,
 						dev, &exclusive_ports);
 				slave_gid += bitmap_weight(
 						slaves_pport_actv.slaves,
-						dev->num_vfs + 1);
+						dev->persist->num_vfs + 1);
 			}
 		}
 		*slave_id = slave_gid;

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

@@ -76,19 +76,21 @@ int mlx4_reset(struct mlx4_dev *dev)
 		goto out;
 	}
 
-	pcie_cap = pci_pcie_cap(dev->pdev);
+	pcie_cap = pci_pcie_cap(dev->persist->pdev);
 
 	for (i = 0; i < 64; ++i) {
 		if (i == 22 || i == 23)
 			continue;
-		if (pci_read_config_dword(dev->pdev, i * 4, hca_header + i)) {
+		if (pci_read_config_dword(dev->persist->pdev, i * 4,
+					  hca_header + i)) {
 			err = -ENODEV;
 			mlx4_err(dev, "Couldn't save HCA PCI header, aborting\n");
 			goto out;
 		}
 	}
 
-	reset = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_RESET_BASE,
+	reset = ioremap(pci_resource_start(dev->persist->pdev, 0) +
+			MLX4_RESET_BASE,
 			MLX4_RESET_SIZE);
 	if (!reset) {
 		err = -ENOMEM;
@@ -122,8 +124,8 @@ int mlx4_reset(struct mlx4_dev *dev)
 
 	end = jiffies + MLX4_RESET_TIMEOUT_JIFFIES;
 	do {
-		if (!pci_read_config_word(dev->pdev, PCI_VENDOR_ID, &vendor) &&
-		    vendor != 0xffff)
+		if (!pci_read_config_word(dev->persist->pdev, PCI_VENDOR_ID,
+					  &vendor) && vendor != 0xffff)
 			break;
 
 		msleep(1);
@@ -138,14 +140,16 @@ int mlx4_reset(struct mlx4_dev *dev)
 	/* Now restore the PCI headers */
 	if (pcie_cap) {
 		devctl = hca_header[(pcie_cap + PCI_EXP_DEVCTL) / 4];
-		if (pcie_capability_write_word(dev->pdev, PCI_EXP_DEVCTL,
+		if (pcie_capability_write_word(dev->persist->pdev,
+					       PCI_EXP_DEVCTL,
 					       devctl)) {
 			err = -ENODEV;
 			mlx4_err(dev, "Couldn't restore HCA PCI Express Device Control register, aborting\n");
 			goto out;
 		}
 		linkctl = hca_header[(pcie_cap + PCI_EXP_LNKCTL) / 4];
-		if (pcie_capability_write_word(dev->pdev, PCI_EXP_LNKCTL,
+		if (pcie_capability_write_word(dev->persist->pdev,
+					       PCI_EXP_LNKCTL,
 					       linkctl)) {
 			err = -ENODEV;
 			mlx4_err(dev, "Couldn't restore HCA PCI Express Link control register, aborting\n");
@@ -157,7 +161,8 @@ int mlx4_reset(struct mlx4_dev *dev)
 		if (i * 4 == PCI_COMMAND)
 			continue;
 
-		if (pci_write_config_dword(dev->pdev, i * 4, hca_header[i])) {
+		if (pci_write_config_dword(dev->persist->pdev, i * 4,
+					   hca_header[i])) {
 			err = -ENODEV;
 			mlx4_err(dev, "Couldn't restore HCA reg %x, aborting\n",
 				 i);
@@ -165,7 +170,7 @@ int mlx4_reset(struct mlx4_dev *dev)
 		}
 	}
 
-	if (pci_write_config_dword(dev->pdev, PCI_COMMAND,
+	if (pci_write_config_dword(dev->persist->pdev, PCI_COMMAND,
 				   hca_header[PCI_COMMAND / 4])) {
 		err = -ENODEV;
 		mlx4_err(dev, "Couldn't restore HCA COMMAND, aborting\n");

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

@@ -309,12 +309,13 @@ static inline int mlx4_grant_resource(struct mlx4_dev *dev, int slave,
 	int allocated, free, reserved, guaranteed, from_free;
 	int from_rsvd;
 
-	if (slave > dev->num_vfs)
+	if (slave > dev->persist->num_vfs)
 		return -EINVAL;
 
 	spin_lock(&res_alloc->alloc_lock);
 	allocated = (port > 0) ?
-		res_alloc->allocated[(port - 1) * (dev->num_vfs + 1) + slave] :
+		res_alloc->allocated[(port - 1) *
+		(dev->persist->num_vfs + 1) + slave] :
 		res_alloc->allocated[slave];
 	free = (port > 0) ? res_alloc->res_port_free[port - 1] :
 		res_alloc->res_free;
@@ -352,7 +353,8 @@ static inline int mlx4_grant_resource(struct mlx4_dev *dev, int slave,
 	if (!err) {
 		/* grant the request */
 		if (port > 0) {
-			res_alloc->allocated[(port - 1) * (dev->num_vfs + 1) + slave] += count;
+			res_alloc->allocated[(port - 1) *
+			(dev->persist->num_vfs + 1) + slave] += count;
 			res_alloc->res_port_free[port - 1] -= count;
 			res_alloc->res_port_rsvd[port - 1] -= from_rsvd;
 		} else {
@@ -376,13 +378,14 @@ static inline void mlx4_release_resource(struct mlx4_dev *dev, int slave,
 		&priv->mfunc.master.res_tracker.res_alloc[res_type];
 	int allocated, guaranteed, from_rsvd;
 
-	if (slave > dev->num_vfs)
+	if (slave > dev->persist->num_vfs)
 		return;
 
 	spin_lock(&res_alloc->alloc_lock);
 
 	allocated = (port > 0) ?
-		res_alloc->allocated[(port - 1) * (dev->num_vfs + 1) + slave] :
+		res_alloc->allocated[(port - 1) *
+		(dev->persist->num_vfs + 1) + slave] :
 		res_alloc->allocated[slave];
 	guaranteed = res_alloc->guaranteed[slave];
 
@@ -397,7 +400,8 @@ static inline void mlx4_release_resource(struct mlx4_dev *dev, int slave,
 	}
 
 	if (port > 0) {
-		res_alloc->allocated[(port - 1) * (dev->num_vfs + 1) + slave] -= count;
+		res_alloc->allocated[(port - 1) *
+		(dev->persist->num_vfs + 1) + slave] -= count;
 		res_alloc->res_port_free[port - 1] += count;
 		res_alloc->res_port_rsvd[port - 1] += from_rsvd;
 	} else {
@@ -415,7 +419,8 @@ static inline void initialize_res_quotas(struct mlx4_dev *dev,
 					 enum mlx4_resource res_type,
 					 int vf, int num_instances)
 {
-	res_alloc->guaranteed[vf] = num_instances / (2 * (dev->num_vfs + 1));
+	res_alloc->guaranteed[vf] = num_instances /
+				    (2 * (dev->persist->num_vfs + 1));
 	res_alloc->quota[vf] = (num_instances / 2) + res_alloc->guaranteed[vf];
 	if (vf == mlx4_master_func_num(dev)) {
 		res_alloc->res_free = num_instances;
@@ -486,21 +491,26 @@ int mlx4_init_resource_tracker(struct mlx4_dev *dev)
 	for (i = 0; i < MLX4_NUM_OF_RESOURCE_TYPE; i++) {
 		struct resource_allocator *res_alloc =
 			&priv->mfunc.master.res_tracker.res_alloc[i];
-		res_alloc->quota = kmalloc((dev->num_vfs + 1) * sizeof(int), GFP_KERNEL);
-		res_alloc->guaranteed = kmalloc((dev->num_vfs + 1) * sizeof(int), GFP_KERNEL);
+		res_alloc->quota = kmalloc((dev->persist->num_vfs + 1) *
+					   sizeof(int), GFP_KERNEL);
+		res_alloc->guaranteed = kmalloc((dev->persist->num_vfs + 1) *
+						sizeof(int), GFP_KERNEL);
 		if (i == RES_MAC || i == RES_VLAN)
 			res_alloc->allocated = kzalloc(MLX4_MAX_PORTS *
-						       (dev->num_vfs + 1) * sizeof(int),
-							GFP_KERNEL);
+						       (dev->persist->num_vfs
+						       + 1) *
+						       sizeof(int), GFP_KERNEL);
 		else
-			res_alloc->allocated = kzalloc((dev->num_vfs + 1) * sizeof(int), GFP_KERNEL);
+			res_alloc->allocated = kzalloc((dev->persist->
+							num_vfs + 1) *
+						       sizeof(int), GFP_KERNEL);
 
 		if (!res_alloc->quota || !res_alloc->guaranteed ||
 		    !res_alloc->allocated)
 			goto no_mem_err;
 
 		spin_lock_init(&res_alloc->alloc_lock);
-		for (t = 0; t < dev->num_vfs + 1; t++) {
+		for (t = 0; t < dev->persist->num_vfs + 1; t++) {
 			struct mlx4_active_ports actv_ports =
 				mlx4_get_active_ports(dev, t);
 			switch (i) {

include/linux/mlx4/device.h

@@ -744,8 +744,15 @@ struct mlx4_vf_dev {
 	u8			n_ports;
 };
 
-struct mlx4_dev {
+struct mlx4_dev_persistent {
 	struct pci_dev	       *pdev;
+	struct mlx4_dev	       *dev;
+	int                     nvfs[MLX4_MAX_PORTS + 1];
+	int			num_vfs;
+};
+
+struct mlx4_dev {
+	struct mlx4_dev_persistent *persist;
 	unsigned long		flags;
 	unsigned long		num_slaves;
 	struct mlx4_caps	caps;
@@ -754,13 +761,11 @@ struct mlx4_dev {
 	struct radix_tree_root	qp_table_tree;
 	u8			rev_id;
 	char			board_id[MLX4_BOARD_ID_LEN];
-	int			num_vfs;
 	int			numa_node;
 	int			oper_log_mgm_entry_size;
 	u64			regid_promisc_array[MLX4_MAX_PORTS + 1];
 	u64			regid_allmulti_array[MLX4_MAX_PORTS + 1];
 	struct mlx4_vf_dev     *dev_vfs;
-	int                     nvfs[MLX4_MAX_PORTS + 1];
 };
 
 struct mlx4_eqe {