Merge branch 'mlx4-next'

Or Gerlitz says:

====================
mlx4: Fix and enhance the device reset flow

This series from Yishai Hadas fixes the device reset flow and adds SRIOV support.

Reset flows are required whenever a device experiences errors, is unresponsive,
or is not in a deterministic state. In such cases, the driver is expected to
reset the HW and continue operation. When SRIOV is enabled, these requirements
apply both to PF and VF devices.

Currently, the mlx4 reset flow doesn't work properly: when a fatal error is
detected on the FW internal buffer the chip is not reset and stays in its
bad state. There are cases that assumed to be fatal such as non-responsive FW,
errors via closing commands but are not handled today.

The AER mechanism should also be fixed:
- It should use mlx4_load_one instead of __mlx4_init_one which is done
  upon HCA probing.
- It must be aligned with concurrent catas flow, mark device to be in
  an error state, reset chip, etc.
- Port types should be restored to their original values before error occurred.

In addition, there the SRIOV use-case isn't supported.

In above cases when the device state becomes fatal we must act as follows:
1) Reset the chip and mark the HW device state as in fatal error.
2) Wake up any pending commands, preventing new ones to come in.
3) Restart the software stack.

We also address the SRIOV mode as follows: In case the PF detects a fatal error,
it lets VFs know about that, then both itself and VFs are restarted asynchronously.
However, in case only the VF encountered a fatal case or forced to be reset, they
reset the VF stuff and then restart software.

changes from V0:

No need to call pci_disable_device upon permanent PCI error. This will
be done as part of mlx4_remove_one which is called later once we
return PCI_ERS_RESULT_DISCONNECT from the pci error handler.

Initial toggle value should use only the T bit and not the whole byte value.
Not doing so sometimes broke SRIOV as of junky value seen by the VF as a
non-ready comm channel
====================

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

@@ -40,16 +40,177 @@ enum {
 	MLX4_CATAS_POLL_INTERVAL	= 5 * HZ,
 };
 
-static DEFINE_SPINLOCK(catas_lock);
 
-static LIST_HEAD(catas_list);
-static struct work_struct catas_work;
 
-static int internal_err_reset = 1;
-module_param(internal_err_reset, int, 0644);
+int mlx4_internal_err_reset = 1;
+module_param_named(internal_err_reset, mlx4_internal_err_reset,  int, 0644);
 MODULE_PARM_DESC(internal_err_reset,
-		 "Reset device on internal errors if non-zero"
-		 " (default 1, in SRIOV mode default is 0)");
+		 "Reset device on internal errors if non-zero (default 1)");
+
+static int read_vendor_id(struct mlx4_dev *dev)
+{
+	u16 vendor_id = 0;
+	int ret;
+
+	ret = pci_read_config_word(dev->persist->pdev, 0, &vendor_id);
+	if (ret) {
+		mlx4_err(dev, "Failed to read vendor ID, ret=%d\n", ret);
+		return ret;
+	}
+
+	if (vendor_id == 0xffff) {
+		mlx4_err(dev, "PCI can't be accessed to read vendor id\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int mlx4_reset_master(struct mlx4_dev *dev)
+{
+	int err = 0;
+
+	if (mlx4_is_master(dev))
+		mlx4_report_internal_err_comm_event(dev);
+
+	if (!pci_channel_offline(dev->persist->pdev)) {
+		err = read_vendor_id(dev);
+		/* If PCI can't be accessed to read vendor ID we assume that its
+		 * link was disabled and chip was already reset.
+		 */
+		if (err)
+			return 0;
+
+		err = mlx4_reset(dev);
+		if (err)
+			mlx4_err(dev, "Fail to reset HCA\n");
+	}
+
+	return err;
+}
+
+static int mlx4_reset_slave(struct mlx4_dev *dev)
+{
+#define COM_CHAN_RST_REQ_OFFSET 0x10
+#define COM_CHAN_RST_ACK_OFFSET 0x08
+
+	u32 comm_flags;
+	u32 rst_req;
+	u32 rst_ack;
+	unsigned long end;
+	struct mlx4_priv *priv = mlx4_priv(dev);
+
+	if (pci_channel_offline(dev->persist->pdev))
+		return 0;
+
+	comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
+				  MLX4_COMM_CHAN_FLAGS));
+	if (comm_flags == 0xffffffff) {
+		mlx4_err(dev, "VF reset is not needed\n");
+		return 0;
+	}
+
+	if (!(dev->caps.vf_caps & MLX4_VF_CAP_FLAG_RESET)) {
+		mlx4_err(dev, "VF reset is not supported\n");
+		return -EOPNOTSUPP;
+	}
+
+	rst_req = (comm_flags & (u32)(1 << COM_CHAN_RST_REQ_OFFSET)) >>
+		COM_CHAN_RST_REQ_OFFSET;
+	rst_ack = (comm_flags & (u32)(1 << COM_CHAN_RST_ACK_OFFSET)) >>
+		COM_CHAN_RST_ACK_OFFSET;
+	if (rst_req != rst_ack) {
+		mlx4_err(dev, "Communication channel isn't sync, fail to send reset\n");
+		return -EIO;
+	}
+
+	rst_req ^= 1;
+	mlx4_warn(dev, "VF is sending reset request to Firmware\n");
+	comm_flags = rst_req << COM_CHAN_RST_REQ_OFFSET;
+	__raw_writel((__force u32)cpu_to_be32(comm_flags),
+		     (__iomem char *)priv->mfunc.comm + MLX4_COMM_CHAN_FLAGS);
+	/* Make sure that our comm channel write doesn't
+	 * get mixed in with writes from another CPU.
+	 */
+	mmiowb();
+
+	end = msecs_to_jiffies(MLX4_COMM_TIME) + jiffies;
+	while (time_before(jiffies, end)) {
+		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
+					  MLX4_COMM_CHAN_FLAGS));
+		rst_ack = (comm_flags & (u32)(1 << COM_CHAN_RST_ACK_OFFSET)) >>
+			COM_CHAN_RST_ACK_OFFSET;
+
+		/* Reading rst_req again since the communication channel can
+		 * be reset at any time by the PF and all its bits will be
+		 * set to zero.
+		 */
+		rst_req = (comm_flags & (u32)(1 << COM_CHAN_RST_REQ_OFFSET)) >>
+			COM_CHAN_RST_REQ_OFFSET;
+
+		if (rst_ack == rst_req) {
+			mlx4_warn(dev, "VF Reset succeed\n");
+			return 0;
+		}
+		cond_resched();
+	}
+	mlx4_err(dev, "Fail to send reset over the communication channel\n");
+	return -ETIMEDOUT;
+}
+
+static int mlx4_comm_internal_err(u32 slave_read)
+{
+	return (u32)COMM_CHAN_EVENT_INTERNAL_ERR ==
+		(slave_read & (u32)COMM_CHAN_EVENT_INTERNAL_ERR) ? 1 : 0;
+}
+
+void mlx4_enter_error_state(struct mlx4_dev_persistent *persist)
+{
+	int err;
+	struct mlx4_dev *dev;
+
+	if (!mlx4_internal_err_reset)
+		return;
+
+	mutex_lock(&persist->device_state_mutex);
+	if (persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+		goto out;
+
+	dev = persist->dev;
+	mlx4_err(dev, "device is going to be reset\n");
+	if (mlx4_is_slave(dev))
+		err = mlx4_reset_slave(dev);
+	else
+		err = mlx4_reset_master(dev);
+	BUG_ON(err != 0);
+
+	dev->persist->state |= MLX4_DEVICE_STATE_INTERNAL_ERROR;
+	mlx4_err(dev, "device was reset successfully\n");
+	mutex_unlock(&persist->device_state_mutex);
+
+	/* At that step HW was already reset, now notify clients */
+	mlx4_dispatch_event(dev, MLX4_DEV_EVENT_CATASTROPHIC_ERROR, 0);
+	mlx4_cmd_wake_completions(dev);
+	return;
+
+out:
+	mutex_unlock(&persist->device_state_mutex);
+}
+
+static void mlx4_handle_error_state(struct mlx4_dev_persistent *persist)
+{
+	int err = 0;
+
+	mlx4_enter_error_state(persist);
+	mutex_lock(&persist->interface_state_mutex);
+	if (persist->interface_state & MLX4_INTERFACE_STATE_UP &&
+	    !(persist->interface_state & MLX4_INTERFACE_STATE_DELETION)) {
+		err = mlx4_restart_one(persist->pdev);
+		mlx4_info(persist->dev, "mlx4_restart_one was ended, ret=%d\n",
+			  err);
+	}
+	mutex_unlock(&persist->interface_state_mutex);
+}
 
 static void dump_err_buf(struct mlx4_dev *dev)
 {
@@ -67,58 +228,40 @@ static void poll_catas(unsigned long dev_ptr)
 {
 	struct mlx4_dev *dev = (struct mlx4_dev *) dev_ptr;
 	struct mlx4_priv *priv = mlx4_priv(dev);
+	u32 slave_read;
 
-	if (readl(priv->catas_err.map)) {
-		/* If the device is off-line, we cannot try to recover it */
-		if (pci_channel_offline(dev->pdev))
-			mod_timer(&priv->catas_err.timer,
-				  round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL));
-		else {
-			dump_err_buf(dev);
-			mlx4_dispatch_event(dev, MLX4_DEV_EVENT_CATASTROPHIC_ERROR, 0);
-
-			if (internal_err_reset) {
-				spin_lock(&catas_lock);
-				list_add(&priv->catas_err.list, &catas_list);
-				spin_unlock(&catas_lock);
-
-				queue_work(mlx4_wq, &catas_work);
-			}
+	if (mlx4_is_slave(dev)) {
+		slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
+		if (mlx4_comm_internal_err(slave_read)) {
+			mlx4_warn(dev, "Internal error detected on the communication channel\n");
+			goto internal_err;
 		}
-	} else
-		mod_timer(&priv->catas_err.timer,
-			  round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL));
+	} else if (readl(priv->catas_err.map)) {
+		dump_err_buf(dev);
+		goto internal_err;
+	}
+
+	if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+		mlx4_warn(dev, "Internal error mark was detected on device\n");
+		goto internal_err;
+	}
+
+	mod_timer(&priv->catas_err.timer,
+		  round_jiffies(jiffies + MLX4_CATAS_POLL_INTERVAL));
+	return;
+
+internal_err:
+	if (mlx4_internal_err_reset)
+		queue_work(dev->persist->catas_wq, &dev->persist->catas_work);
 }
 
 static void catas_reset(struct work_struct *work)
 {
-	struct mlx4_priv *priv, *tmppriv;
-	struct mlx4_dev *dev;
+	struct mlx4_dev_persistent *persist =
+		container_of(work, struct mlx4_dev_persistent,
+			     catas_work);
 
-	LIST_HEAD(tlist);
-	int ret;
-
-	spin_lock_irq(&catas_lock);
-	list_splice_init(&catas_list, &tlist);
-	spin_unlock_irq(&catas_lock);
-
-	list_for_each_entry_safe(priv, tmppriv, &tlist, catas_err.list) {
-		struct pci_dev *pdev = priv->dev.pdev;
-
-		/* If the device is off-line, we cannot reset it */
-		if (pci_channel_offline(pdev))
-			continue;
-
-		ret = mlx4_restart_one(priv->dev.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");
-		}
-	}
+	mlx4_handle_error_state(persist);
 }
 
 void mlx4_start_catas_poll(struct mlx4_dev *dev)
@@ -126,22 +269,21 @@ void mlx4_start_catas_poll(struct mlx4_dev *dev)
 	struct mlx4_priv *priv = mlx4_priv(dev);
 	phys_addr_t addr;
 
-	/*If we are in SRIOV the default of the module param must be 0*/
-	if (mlx4_is_mfunc(dev))
-		internal_err_reset = 0;
-
 	INIT_LIST_HEAD(&priv->catas_err.list);
 	init_timer(&priv->catas_err.timer);
 	priv->catas_err.map = NULL;
 
-	addr = pci_resource_start(dev->pdev, priv->fw.catas_bar) +
-		priv->fw.catas_offset;
+	if (!mlx4_is_slave(dev)) {
+		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);
-	if (!priv->catas_err.map) {
-		mlx4_warn(dev, "Failed to map internal error buffer at 0x%llx\n",
-			  (unsigned long long) addr);
-		return;
+		priv->catas_err.map = ioremap(addr, priv->fw.catas_size * 4);
+		if (!priv->catas_err.map) {
+			mlx4_warn(dev, "Failed to map internal error buffer at 0x%llx\n",
+				  (unsigned long long)addr);
+			return;
+		}
 	}
 
 	priv->catas_err.timer.data     = (unsigned long) dev;
@@ -157,15 +299,29 @@ void mlx4_stop_catas_poll(struct mlx4_dev *dev)
 
 	del_timer_sync(&priv->catas_err.timer);
 
-	if (priv->catas_err.map)
+	if (priv->catas_err.map) {
 		iounmap(priv->catas_err.map);
+		priv->catas_err.map = NULL;
+	}
 
-	spin_lock_irq(&catas_lock);
-	list_del(&priv->catas_err.list);
-	spin_unlock_irq(&catas_lock);
+	if (dev->persist->interface_state & MLX4_INTERFACE_STATE_DELETION)
+		flush_workqueue(dev->persist->catas_wq);
 }
 
-void  __init mlx4_catas_init(void)
+int  mlx4_catas_init(struct mlx4_dev *dev)
 {
-	INIT_WORK(&catas_work, catas_reset);
+	INIT_WORK(&dev->persist->catas_work, catas_reset);
+	dev->persist->catas_wq = create_singlethread_workqueue("mlx4_health");
+	if (!dev->persist->catas_wq)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void mlx4_catas_end(struct mlx4_dev *dev)
+{
+	if (dev->persist->catas_wq) {
+		destroy_workqueue(dev->persist->catas_wq);
+		dev->persist->catas_wq = NULL;
+	}
 }

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

@@ -42,6 +42,7 @@
 #include <linux/mlx4/device.h>
 #include <linux/semaphore.h>
 #include <rdma/ib_smi.h>
+#include <linux/delay.h>
 
 #include <asm/io.h>
 
@@ -182,6 +183,72 @@ static u8 mlx4_errno_to_status(int errno)
 	}
 }
 
+static int mlx4_internal_err_ret_value(struct mlx4_dev *dev, u16 op,
+				       u8 op_modifier)
+{
+	switch (op) {
+	case MLX4_CMD_UNMAP_ICM:
+	case MLX4_CMD_UNMAP_ICM_AUX:
+	case MLX4_CMD_UNMAP_FA:
+	case MLX4_CMD_2RST_QP:
+	case MLX4_CMD_HW2SW_EQ:
+	case MLX4_CMD_HW2SW_CQ:
+	case MLX4_CMD_HW2SW_SRQ:
+	case MLX4_CMD_HW2SW_MPT:
+	case MLX4_CMD_CLOSE_HCA:
+	case MLX4_QP_FLOW_STEERING_DETACH:
+	case MLX4_CMD_FREE_RES:
+	case MLX4_CMD_CLOSE_PORT:
+		return CMD_STAT_OK;
+
+	case MLX4_CMD_QP_ATTACH:
+		/* On Detach case return success */
+		if (op_modifier == 0)
+			return CMD_STAT_OK;
+		return mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
+
+	default:
+		return mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
+	}
+}
+
+static int mlx4_closing_cmd_fatal_error(u16 op, u8 fw_status)
+{
+	/* Any error during the closing commands below is considered fatal */
+	if (op == MLX4_CMD_CLOSE_HCA ||
+	    op == MLX4_CMD_HW2SW_EQ ||
+	    op == MLX4_CMD_HW2SW_CQ ||
+	    op == MLX4_CMD_2RST_QP ||
+	    op == MLX4_CMD_HW2SW_SRQ ||
+	    op == MLX4_CMD_SYNC_TPT ||
+	    op == MLX4_CMD_UNMAP_ICM ||
+	    op == MLX4_CMD_UNMAP_ICM_AUX ||
+	    op == MLX4_CMD_UNMAP_FA)
+		return 1;
+	/* Error on MLX4_CMD_HW2SW_MPT is fatal except when fw status equals
+	  * CMD_STAT_REG_BOUND.
+	  * This status indicates that memory region has memory windows bound to it
+	  * which may result from invalid user space usage and is not fatal.
+	  */
+	if (op == MLX4_CMD_HW2SW_MPT && fw_status != CMD_STAT_REG_BOUND)
+		return 1;
+	return 0;
+}
+
+static int mlx4_cmd_reset_flow(struct mlx4_dev *dev, u16 op, u8 op_modifier,
+			       int err)
+{
+	/* Only if reset flow is really active return code is based on
+	  * command, otherwise current error code is returned.
+	  */
+	if (mlx4_internal_err_reset) {
+		mlx4_enter_error_state(dev->persist);
+		err = mlx4_internal_err_ret_value(dev, op, op_modifier);
+	}
+
+	return err;
+}
+
 static int comm_pending(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
@@ -190,16 +257,30 @@ static int comm_pending(struct mlx4_dev *dev)
 	return (swab32(status) >> 31) != priv->cmd.comm_toggle;
 }
 
-static void mlx4_comm_cmd_post(struct mlx4_dev *dev, u8 cmd, u16 param)
+static int mlx4_comm_cmd_post(struct mlx4_dev *dev, u8 cmd, u16 param)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
 	u32 val;
 
+	/* To avoid writing to unknown addresses after the device state was
+	 * changed to internal error and the function was rest,
+	 * check the INTERNAL_ERROR flag which is updated under
+	 * device_state_mutex lock.
+	 */
+	mutex_lock(&dev->persist->device_state_mutex);
+
+	if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+		mutex_unlock(&dev->persist->device_state_mutex);
+		return -EIO;
+	}
+
 	priv->cmd.comm_toggle ^= 1;
 	val = param | (cmd << 16) | (priv->cmd.comm_toggle << 31);
 	__raw_writel((__force u32) cpu_to_be32(val),
 		     &priv->mfunc.comm->slave_write);
 	mmiowb();
+	mutex_unlock(&dev->persist->device_state_mutex);
+	return 0;
 }
 
 static int mlx4_comm_cmd_poll(struct mlx4_dev *dev, u8 cmd, u16 param,
@@ -219,7 +300,13 @@ static int mlx4_comm_cmd_poll(struct mlx4_dev *dev, u8 cmd, u16 param,
 
 	/* Write command */
 	down(&priv->cmd.poll_sem);
-	mlx4_comm_cmd_post(dev, cmd, param);
+	if (mlx4_comm_cmd_post(dev, cmd, param)) {
+		/* Only in case the device state is INTERNAL_ERROR,
+		 * mlx4_comm_cmd_post returns with an error
+		 */
+		err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
+		goto out;
+	}
 
 	end = msecs_to_jiffies(timeout) + jiffies;
 	while (comm_pending(dev) && time_before(jiffies, end))
@@ -231,18 +318,23 @@ static int mlx4_comm_cmd_poll(struct mlx4_dev *dev, u8 cmd, u16 param,
 		 * is MLX4_DELAY_RESET_SLAVE*/
 		if ((MLX4_COMM_CMD_RESET == cmd)) {
 			err = MLX4_DELAY_RESET_SLAVE;
+			goto out;
 		} else {
-			mlx4_warn(dev, "Communication channel timed out\n");
-			err = -ETIMEDOUT;
+			mlx4_warn(dev, "Communication channel command 0x%x timed out\n",
+				  cmd);
+			err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
 		}
 	}
 
+	if (err)
+		mlx4_enter_error_state(dev->persist);
+out:
 	up(&priv->cmd.poll_sem);
 	return err;
 }
 
-static int mlx4_comm_cmd_wait(struct mlx4_dev *dev, u8 op,
-			      u16 param, unsigned long timeout)
+static int mlx4_comm_cmd_wait(struct mlx4_dev *dev, u8 vhcr_cmd,
+			      u16 param, u16 op, unsigned long timeout)
 {
 	struct mlx4_cmd *cmd = &mlx4_priv(dev)->cmd;
 	struct mlx4_cmd_context *context;
@@ -258,34 +350,49 @@ static int mlx4_comm_cmd_wait(struct mlx4_dev *dev, u8 op,
 	cmd->free_head = context->next;
 	spin_unlock(&cmd->context_lock);
 
-	init_completion(&context->done);
+	reinit_completion(&context->done);
 
-	mlx4_comm_cmd_post(dev, op, param);
+	if (mlx4_comm_cmd_post(dev, vhcr_cmd, param)) {
+		/* Only in case the device state is INTERNAL_ERROR,
+		 * mlx4_comm_cmd_post returns with an error
+		 */
+		err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
+		goto out;
+	}
 
 	if (!wait_for_completion_timeout(&context->done,
 					 msecs_to_jiffies(timeout))) {
-		mlx4_warn(dev, "communication channel command 0x%x timed out\n",
-			  op);
-		err = -EBUSY;
-		goto out;
+		mlx4_warn(dev, "communication channel command 0x%x (op=0x%x) timed out\n",
+			  vhcr_cmd, op);
+		goto out_reset;
 	}
 
 	err = context->result;
 	if (err && context->fw_status != CMD_STAT_MULTI_FUNC_REQ) {
 		mlx4_err(dev, "command 0x%x failed: fw status = 0x%x\n",
-			 op, context->fw_status);
-		goto out;
+			 vhcr_cmd, context->fw_status);
+		if (mlx4_closing_cmd_fatal_error(op, context->fw_status))
+			goto out_reset;
 	}
 
-out:
 	/* wait for comm channel ready
 	 * this is necessary for prevention the race
 	 * when switching between event to polling mode
+	 * Skipping this section in case the device is in FATAL_ERROR state,
+	 * In this state, no commands are sent via the comm channel until
+	 * the device has returned from reset.
 	 */
-	end = msecs_to_jiffies(timeout) + jiffies;
-	while (comm_pending(dev) && time_before(jiffies, end))
-		cond_resched();
+	if (!(dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)) {
+		end = msecs_to_jiffies(timeout) + jiffies;
+		while (comm_pending(dev) && time_before(jiffies, end))
+			cond_resched();
+	}
+	goto out;
 
+out_reset:
+	err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
+	mlx4_enter_error_state(dev->persist);
+out:
 	spin_lock(&cmd->context_lock);
 	context->next = cmd->free_head;
 	cmd->free_head = context - cmd->context;
@@ -296,10 +403,13 @@ out:
 }
 
 int mlx4_comm_cmd(struct mlx4_dev *dev, u8 cmd, u16 param,
-		  unsigned long timeout)
+		  u16 op, unsigned long timeout)
 {
+	if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+		return mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
+
 	if (mlx4_priv(dev)->cmd.use_events)
-		return mlx4_comm_cmd_wait(dev, cmd, param, timeout);
+		return mlx4_comm_cmd_wait(dev, cmd, param, op, timeout);
 	return mlx4_comm_cmd_poll(dev, cmd, param, timeout);
 }
 
@@ -307,7 +417,7 @@ 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);
@@ -323,17 +433,21 @@ static int mlx4_cmd_post(struct mlx4_dev *dev, u64 in_param, u64 out_param,
 {
 	struct mlx4_cmd *cmd = &mlx4_priv(dev)->cmd;
 	u32 __iomem *hcr = cmd->hcr;
-	int ret = -EAGAIN;
+	int ret = -EIO;
 	unsigned long end;
 
-	mutex_lock(&cmd->hcr_mutex);
-
-	if (pci_channel_offline(dev->pdev)) {
+	mutex_lock(&dev->persist->device_state_mutex);
+	/* To avoid writing to unknown addresses after the device state was
+	  * changed to internal error and the chip was reset,
+	  * check the INTERNAL_ERROR flag which is updated under
+	  * device_state_mutex lock.
+	  */
+	if (pci_channel_offline(dev->persist->pdev) ||
+	    (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)) {
 		/*
 		 * Device is going through error recovery
 		 * and cannot accept commands.
 		 */
-		ret = -EIO;
 		goto out;
 	}
 
@@ -342,12 +456,11 @@ 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.
 			 */
-			ret = -EIO;
 			goto out;
 		}
 
@@ -391,7 +504,11 @@ static int mlx4_cmd_post(struct mlx4_dev *dev, u64 in_param, u64 out_param,
 	ret = 0;
 
 out:
-	mutex_unlock(&cmd->hcr_mutex);
+	if (ret)
+		mlx4_warn(dev, "Could not post command 0x%x: ret=%d, in_param=0x%llx, in_mod=0x%x, op_mod=0x%x\n",
+			  op, ret, in_param, in_modifier, op_modifier);
+	mutex_unlock(&dev->persist->device_state_mutex);
+
 	return ret;
 }
 
@@ -428,8 +545,11 @@ static int mlx4_slave_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 			}
 			ret = mlx4_status_to_errno(vhcr->status);
 		}
+		if (ret &&
+		    dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+			ret = mlx4_internal_err_ret_value(dev, op, op_modifier);
 	} else {
-		ret = mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_POST, 0,
+		ret = mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_POST, 0, op,
 				    MLX4_COMM_TIME + timeout);
 		if (!ret) {
 			if (out_is_imm) {
@@ -443,9 +563,14 @@ static int mlx4_slave_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 				}
 			}
 			ret = mlx4_status_to_errno(vhcr->status);
-		} else
-			mlx4_err(dev, "failed execution of VHCR_POST command opcode 0x%x\n",
-				 op);
+		} else {
+			if (dev->persist->state &
+			    MLX4_DEVICE_STATE_INTERNAL_ERROR)
+				ret = mlx4_internal_err_ret_value(dev, op,
+								  op_modifier);
+			else
+				mlx4_err(dev, "failed execution of VHCR_POST command opcode 0x%x\n", op);
+		}
 	}
 
 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
@@ -464,12 +589,12 @@ 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 (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
 		/*
 		 * Device is going through error recovery
 		 * and cannot accept commands.
 		 */
-		err = -EIO;
+		err = mlx4_internal_err_ret_value(dev, op, op_modifier);
 		goto out;
 	}
 
@@ -483,16 +608,21 @@ static int mlx4_cmd_poll(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 	err = mlx4_cmd_post(dev, in_param, out_param ? *out_param : 0,
 			    in_modifier, op_modifier, op, CMD_POLL_TOKEN, 0);
 	if (err)
-		goto out;
+		goto out_reset;
 
 	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.
 			 */
 			err = -EIO;
+			goto out_reset;
+		}
+
+		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
+			err = mlx4_internal_err_ret_value(dev, op, op_modifier);
 			goto out;
 		}
 
@@ -502,8 +632,8 @@ static int mlx4_cmd_poll(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 	if (cmd_pending(dev)) {
 		mlx4_warn(dev, "command 0x%x timed out (go bit not cleared)\n",
 			  op);
-		err = -ETIMEDOUT;
-		goto out;
+		err = -EIO;
+		goto out_reset;
 	}
 
 	if (out_is_imm)
@@ -515,10 +645,17 @@ static int mlx4_cmd_poll(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 	stat = be32_to_cpu((__force __be32)
 			   __raw_readl(hcr + HCR_STATUS_OFFSET)) >> 24;
 	err = mlx4_status_to_errno(stat);
-	if (err)
+	if (err) {
 		mlx4_err(dev, "command 0x%x failed: fw status = 0x%x\n",
 			 op, stat);
+		if (mlx4_closing_cmd_fatal_error(op, stat))
+			goto out_reset;
+		goto out;
+	}
 
+out_reset:
+	if (err)
+		err = mlx4_cmd_reset_flow(dev, op, op_modifier, err);
 out:
 	up(&priv->cmd.poll_sem);
 	return err;
@@ -565,17 +702,19 @@ static int mlx4_cmd_wait(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 		goto out;
 	}
 
-	init_completion(&context->done);
+	reinit_completion(&context->done);
 
-	mlx4_cmd_post(dev, in_param, out_param ? *out_param : 0,
-		      in_modifier, op_modifier, op, context->token, 1);
+	err = mlx4_cmd_post(dev, in_param, out_param ? *out_param : 0,
+			    in_modifier, op_modifier, op, context->token, 1);
+	if (err)
+		goto out_reset;
 
 	if (!wait_for_completion_timeout(&context->done,
 					 msecs_to_jiffies(timeout))) {
 		mlx4_warn(dev, "command 0x%x timed out (go bit not cleared)\n",
 			  op);
-		err = -EBUSY;
-		goto out;
+		err = -EIO;
+		goto out_reset;
 	}
 
 	err = context->result;
@@ -592,12 +731,20 @@ static int mlx4_cmd_wait(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 		else
 			mlx4_err(dev, "command 0x%x failed: fw status = 0x%x\n",
 				 op, context->fw_status);
+		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+			err = mlx4_internal_err_ret_value(dev, op, op_modifier);
+		else if (mlx4_closing_cmd_fatal_error(op, context->fw_status))
+			goto out_reset;
+
 		goto out;
 	}
 
 	if (out_is_imm)
 		*out_param = context->out_param;
 
+out_reset:
+	if (err)
+		err = mlx4_cmd_reset_flow(dev, op, op_modifier, err);
 out:
 	spin_lock(&cmd->context_lock);
 	context->next = cmd->free_head;
@@ -612,10 +759,13 @@ 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))
-		return -EIO;
+	if (pci_channel_offline(dev->persist->pdev))
+		return mlx4_cmd_reset_flow(dev, op, op_modifier, -EIO);
 
 	if (!mlx4_is_mfunc(dev) || (native && mlx4_is_master(dev))) {
+		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+			return mlx4_internal_err_ret_value(dev, op,
+							  op_modifier);
 		if (mlx4_priv(dev)->cmd.use_events)
 			return mlx4_cmd_wait(dev, in_param, out_param,
 					     out_is_imm, in_modifier,
@@ -631,7 +781,7 @@ int __mlx4_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 EXPORT_SYMBOL_GPL(__mlx4_cmd);
 
 
-static int mlx4_ARM_COMM_CHANNEL(struct mlx4_dev *dev)
+int mlx4_ARM_COMM_CHANNEL(struct mlx4_dev *dev)
 {
 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_ARM_COMM_CHANNEL,
 			MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
@@ -1460,8 +1610,10 @@ static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
 				      ALIGN(sizeof(struct mlx4_vhcr_cmd),
 					    MLX4_ACCESS_MEM_ALIGN), 1);
 		if (ret) {
-			mlx4_err(dev, "%s: Failed reading vhcr ret: 0x%x\n",
-				 __func__, ret);
+			if (!(dev->persist->state &
+			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
+				mlx4_err(dev, "%s: Failed reading vhcr ret: 0x%x\n",
+					 __func__, ret);
 			kfree(vhcr);
 			return ret;
 		}
@@ -1500,11 +1652,14 @@ static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
 			goto out_status;
 		}
 
-		if (mlx4_ACCESS_MEM(dev, inbox->dma, slave,
-				    vhcr->in_param,
-				    MLX4_MAILBOX_SIZE, 1)) {
-			mlx4_err(dev, "%s: Failed reading inbox (cmd:0x%x)\n",
-				 __func__, cmd->opcode);
+		ret = mlx4_ACCESS_MEM(dev, inbox->dma, slave,
+				      vhcr->in_param,
+				      MLX4_MAILBOX_SIZE, 1);
+		if (ret) {
+			if (!(dev->persist->state &
+			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
+				mlx4_err(dev, "%s: Failed reading inbox (cmd:0x%x)\n",
+					 __func__, cmd->opcode);
 			vhcr_cmd->status = CMD_STAT_INTERNAL_ERR;
 			goto out_status;
 		}
@@ -1552,8 +1707,9 @@ static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
 	}
 
 	if (err) {
-		mlx4_warn(dev, "vhcr command:0x%x slave:%d failed with error:%d, status %d\n",
-			  vhcr->op, slave, vhcr->errno, err);
+		if (!(dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR))
+			mlx4_warn(dev, "vhcr command:0x%x slave:%d failed with error:%d, status %d\n",
+				  vhcr->op, slave, vhcr->errno, err);
 		vhcr_cmd->status = mlx4_errno_to_status(err);
 		goto out_status;
 	}
@@ -1568,7 +1724,9 @@ static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
 			/* If we failed to write back the outbox after the
 			 *command was successfully executed, we must fail this
 			 * slave, as it is now in undefined state */
-			mlx4_err(dev, "%s:Failed writing outbox\n", __func__);
+			if (!(dev->persist->state &
+			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
+				mlx4_err(dev, "%s:Failed writing outbox\n", __func__);
 			goto out;
 		}
 	}
@@ -1847,8 +2005,11 @@ static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd,
 		break;
 	case MLX4_COMM_CMD_VHCR_POST:
 		if ((slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_EN) &&
-		    (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_POST))
+		    (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_POST)) {
+			mlx4_warn(dev, "slave:%d is out of sync, cmd=0x%x, last command=0x%x, reset is needed\n",
+				  slave, cmd, slave_state[slave].last_cmd);
 			goto reset_slave;
+		}
 
 		mutex_lock(&priv->cmd.slave_cmd_mutex);
 		if (mlx4_master_process_vhcr(dev, slave, NULL)) {
@@ -1882,7 +2043,18 @@ static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd,
 
 reset_slave:
 	/* cleanup any slave resources */
-	mlx4_delete_all_resources_for_slave(dev, slave);
+	if (dev->persist->interface_state & MLX4_INTERFACE_STATE_UP)
+		mlx4_delete_all_resources_for_slave(dev, slave);
+
+	if (cmd != MLX4_COMM_CMD_RESET) {
+		mlx4_warn(dev, "Turn on internal error to force reset, slave=%d, cmd=0x%x\n",
+			  slave, cmd);
+		/* Turn on internal error letting slave reset itself immeditaly,
+		 * otherwise it might take till timeout on command is passed
+		 */
+		reply |= ((u32)COMM_CHAN_EVENT_INTERNAL_ERR);
+	}
+
 	spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
 	if (!slave_state[slave].is_slave_going_down)
 		slave_state[slave].last_cmd = MLX4_COMM_CMD_RESET;
@@ -1958,17 +2130,28 @@ void mlx4_master_comm_channel(struct work_struct *work)
 static int sync_toggles(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
-	int wr_toggle;
-	int rd_toggle;
+	u32 wr_toggle;
+	u32 rd_toggle;
 	unsigned long end;
 
-	wr_toggle = swab32(readl(&priv->mfunc.comm->slave_write)) >> 31;
-	end = jiffies + msecs_to_jiffies(5000);
+	wr_toggle = swab32(readl(&priv->mfunc.comm->slave_write));
+	if (wr_toggle == 0xffffffff)
+		end = jiffies + msecs_to_jiffies(30000);
+	else
+		end = jiffies + msecs_to_jiffies(5000);
 
 	while (time_before(jiffies, end)) {
-		rd_toggle = swab32(readl(&priv->mfunc.comm->slave_read)) >> 31;
-		if (rd_toggle == wr_toggle) {
-			priv->cmd.comm_toggle = rd_toggle;
+		rd_toggle = swab32(readl(&priv->mfunc.comm->slave_read));
+		if (wr_toggle == 0xffffffff || rd_toggle == 0xffffffff) {
+			/* PCI might be offline */
+			msleep(100);
+			wr_toggle = swab32(readl(&priv->mfunc.comm->
+					   slave_write));
+			continue;
+		}
+
+		if (rd_toggle >> 31 == wr_toggle >> 31) {
+			priv->cmd.comm_toggle = rd_toggle >> 31;
 			return 0;
 		}
 
@@ -1997,11 +2180,12 @@ 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");
@@ -2073,13 +2257,6 @@ int mlx4_multi_func_init(struct mlx4_dev *dev)
 		if (mlx4_init_resource_tracker(dev))
 			goto err_thread;
 
-		err = mlx4_ARM_COMM_CHANNEL(dev);
-		if (err) {
-			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
-				 err);
-			goto err_resource;
-		}
-
 	} else {
 		err = sync_toggles(dev);
 		if (err) {
@@ -2089,8 +2266,6 @@ int mlx4_multi_func_init(struct mlx4_dev *dev)
 	}
 	return 0;
 
-err_resource:
-	mlx4_free_resource_tracker(dev, RES_TR_FREE_ALL);
 err_thread:
 	flush_workqueue(priv->mfunc.master.comm_wq);
 	destroy_workqueue(priv->mfunc.master.comm_wq);
@@ -2107,9 +2282,9 @@ 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;
 }
@@ -2120,7 +2295,6 @@ int mlx4_cmd_init(struct mlx4_dev *dev)
 	int flags = 0;
 
 	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;
@@ -2130,8 +2304,8 @@ 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 +2314,8 @@ 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 +2325,8 @@ 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)
@@ -2166,6 +2342,27 @@ err:
 	return -ENOMEM;
 }
 
+void mlx4_report_internal_err_comm_event(struct mlx4_dev *dev)
+{
+	struct mlx4_priv *priv = mlx4_priv(dev);
+	int slave;
+	u32 slave_read;
+
+	/* Report an internal error event to all
+	 * communication channels.
+	 */
+	for (slave = 0; slave < dev->num_slaves; slave++) {
+		slave_read = swab32(readl(&priv->mfunc.comm[slave].slave_read));
+		slave_read |= (u32)COMM_CHAN_EVENT_INTERNAL_ERR;
+		__raw_writel((__force u32)cpu_to_be32(slave_read),
+			     &priv->mfunc.comm[slave].slave_read);
+		/* Make sure that our comm channel write doesn't
+		 * get mixed in with writes from another CPU.
+		 */
+		mmiowb();
+	}
+}
+
 void mlx4_multi_func_cleanup(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
@@ -2181,6 +2378,7 @@ void mlx4_multi_func_cleanup(struct mlx4_dev *dev)
 		kfree(priv->mfunc.master.slave_state);
 		kfree(priv->mfunc.master.vf_admin);
 		kfree(priv->mfunc.master.vf_oper);
+		dev->num_slaves = 0;
 	}
 
 	iounmap(priv->mfunc.comm);
@@ -2202,7 +2400,7 @@ 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;
 	}
@@ -2229,6 +2427,11 @@ int mlx4_cmd_use_events(struct mlx4_dev *dev)
 	for (i = 0; i < priv->cmd.max_cmds; ++i) {
 		priv->cmd.context[i].token = i;
 		priv->cmd.context[i].next  = i + 1;
+		/* To support fatal error flow, initialize all
+		 * cmd contexts to allow simulating completions
+		 * with complete() at any time.
+		 */
+		init_completion(&priv->cmd.context[i].done);
 	}
 
 	priv->cmd.context[priv->cmd.max_cmds - 1].next = -1;
@@ -2306,8 +2509,9 @@ 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 +2520,7 @@ 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);
@@ -2325,6 +2529,25 @@ int mlx4_get_vf_indx(struct mlx4_dev *dev, int slave)
 	return slave - 1;
 }
 
+void mlx4_cmd_wake_completions(struct mlx4_dev *dev)
+{
+	struct mlx4_priv *priv = mlx4_priv(dev);
+	struct mlx4_cmd_context *context;
+	int i;
+
+	spin_lock(&priv->cmd.context_lock);
+	if (priv->cmd.context) {
+		for (i = 0; i < priv->cmd.max_cmds; ++i) {
+			context = &priv->cmd.context[i];
+			context->fw_status = CMD_STAT_INTERNAL_ERR;
+			context->result    =
+				mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
+			complete(&context->done);
+		}
+	}
+	spin_unlock(&priv->cmd.context_lock);
+}
+
 struct mlx4_active_ports mlx4_get_active_ports(struct mlx4_dev *dev, int slave)
 {
 	struct mlx4_active_ports actv_ports;
@@ -2388,7 +2611,7 @@ 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 +2631,7 @@ 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);
@@ -429,8 +429,14 @@ void mlx4_master_handle_slave_flr(struct work_struct *work)
 		if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
 			mlx4_dbg(dev, "mlx4_handle_slave_flr: clean slave: %d\n",
 				 i);
-
-			mlx4_delete_all_resources_for_slave(dev, i);
+			/* In case of 'Reset flow' FLR can be generated for
+			 * a slave before mlx4_load_one is done.
+			 * make sure interface is up before trying to delete
+			 * slave resources which weren't allocated yet.
+			 */
+			if (dev->persist->interface_state &
+			    MLX4_INTERFACE_STATE_UP)
+				mlx4_delete_all_resources_for_slave(dev, i);
 			/*return the slave to running mode*/
 			spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
 			slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
@@ -560,7 +566,8 @@ 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 +603,9 @@ 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 +874,7 @@ 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 +937,10 @@ 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 +1006,7 @@ 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 +1055,9 @@ 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 +1071,7 @@ 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 +1100,8 @@ 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 +1224,13 @@ 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 +1247,8 @@ 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/intf.c

@@ -138,13 +138,13 @@ int mlx4_register_device(struct mlx4_dev *dev)
 
 	mutex_lock(&intf_mutex);
 
+	dev->persist->interface_state |= MLX4_INTERFACE_STATE_UP;
 	list_add_tail(&priv->dev_list, &dev_list);
 	list_for_each_entry(intf, &intf_list, list)
 		mlx4_add_device(intf, priv);
 
 	mutex_unlock(&intf_mutex);
-	if (!mlx4_is_slave(dev))
-		mlx4_start_catas_poll(dev);
+	mlx4_start_catas_poll(dev);
 
 	return 0;
 }
@@ -154,14 +154,14 @@ void mlx4_unregister_device(struct mlx4_dev *dev)
 	struct mlx4_priv *priv = mlx4_priv(dev);
 	struct mlx4_interface *intf;
 
-	if (!mlx4_is_slave(dev))
-		mlx4_stop_catas_poll(dev);
+	mlx4_stop_catas_poll(dev);
 	mutex_lock(&intf_mutex);
 
 	list_for_each_entry(intf, &intf_list, list)
 		mlx4_remove_device(intf, priv);
 
 	list_del(&priv->dev_list);
+	dev->persist->interface_state &= ~MLX4_INTERFACE_STATE_UP;
 
 	mutex_unlock(&intf_mutex);
 }

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

@@ -108,6 +108,8 @@ MODULE_PARM_DESC(enable_64b_cqe_eqe,
 					 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
 					 MLX4_FUNC_CAP_DMFS_A0_STATIC)
 
+#define RESET_PERSIST_MASK_FLAGS	(MLX4_FLAG_SRIOV)
+
 static char mlx4_version[] =
 	DRV_NAME ": Mellanox ConnectX core driver v"
 	DRV_VERSION " (" DRV_RELDATE ")\n";
@@ -318,10 +320,11 @@ 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 +544,10 @@ 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 +592,7 @@ 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 +842,12 @@ 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;
 	}
 
@@ -1477,7 +1484,8 @@ static void mlx4_slave_exit(struct mlx4_dev *dev)
 	struct mlx4_priv *priv = mlx4_priv(dev);
 
 	mutex_lock(&priv->cmd.slave_cmd_mutex);
-	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
+	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
+			  MLX4_COMM_TIME))
 		mlx4_warn(dev, "Failed to close slave function\n");
 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
 }
@@ -1492,9 +1500,9 @@ 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 +1544,8 @@ 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)
@@ -1573,6 +1582,50 @@ static void mlx4_close_fw(struct mlx4_dev *dev)
 	}
 }
 
+static int mlx4_comm_check_offline(struct mlx4_dev *dev)
+{
+#define COMM_CHAN_OFFLINE_OFFSET 0x09
+
+	u32 comm_flags;
+	u32 offline_bit;
+	unsigned long end;
+	struct mlx4_priv *priv = mlx4_priv(dev);
+
+	end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
+	while (time_before(jiffies, end)) {
+		comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
+					  MLX4_COMM_CHAN_FLAGS));
+		offline_bit = (comm_flags &
+			       (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
+		if (!offline_bit)
+			return 0;
+		/* There are cases as part of AER/Reset flow that PF needs
+		 * around 100 msec to load. We therefore sleep for 100 msec
+		 * to allow other tasks to make use of that CPU during this
+		 * time interval.
+		 */
+		msleep(100);
+	}
+	mlx4_err(dev, "Communication channel is offline.\n");
+	return -EIO;
+}
+
+static void mlx4_reset_vf_support(struct mlx4_dev *dev)
+{
+#define COMM_CHAN_RST_OFFSET 0x1e
+
+	struct mlx4_priv *priv = mlx4_priv(dev);
+	u32 comm_rst;
+	u32 comm_caps;
+
+	comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
+				 MLX4_COMM_CHAN_CAPS));
+	comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
+
+	if (comm_rst)
+		dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
+}
+
 static int mlx4_init_slave(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
@@ -1588,9 +1641,15 @@ static int mlx4_init_slave(struct mlx4_dev *dev)
 
 	mutex_lock(&priv->cmd.slave_cmd_mutex);
 	priv->cmd.max_cmds = 1;
+	if (mlx4_comm_check_offline(dev)) {
+		mlx4_err(dev, "PF is not responsive, skipping initialization\n");
+		goto err_offline;
+	}
+
+	mlx4_reset_vf_support(dev);
 	mlx4_warn(dev, "Sending reset\n");
 	ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
-				       MLX4_COMM_TIME);
+				       MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
 	/* if we are in the middle of flr the slave will try
 	 * NUM_OF_RESET_RETRIES times before leaving.*/
 	if (ret_from_reset) {
@@ -1615,22 +1674,24 @@ static int mlx4_init_slave(struct mlx4_dev *dev)
 
 	mlx4_warn(dev, "Sending vhcr0\n");
 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
-						    MLX4_COMM_TIME))
+			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
 		goto err;
 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
-						    MLX4_COMM_TIME))
+			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
 		goto err;
 	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
-						    MLX4_COMM_TIME))
+			     MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
 		goto err;
-	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
+	if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
+			  MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
 		goto err;
 
 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
 	return 0;
 
 err:
-	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
+	mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
+err_offline:
 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
 	return -EIO;
 }
@@ -1705,7 +1766,8 @@ 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 +2350,8 @@ 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 +2379,7 @@ 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 +2407,7 @@ 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 +2424,12 @@ 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 +2441,9 @@ 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 +2510,11 @@ 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 +2530,11 @@ 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");
@@ -2481,11 +2549,19 @@ static void mlx4_free_ownership(struct mlx4_dev *dev)
 				  !!((flags) & MLX4_FLAG_MASTER))
 
 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
-			     u8 total_vfs, int existing_vfs)
+			     u8 total_vfs, int existing_vfs, int reset_flow)
 {
 	u64 dev_flags = dev->flags;
 	int err = 0;
 
+	if (reset_flow) {
+		dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
+				       GFP_KERNEL);
+		if (!dev->dev_vfs)
+			goto free_mem;
+		return dev_flags;
+	}
+
 	atomic_inc(&pf_loading);
 	if (dev->flags &  MLX4_FLAG_SRIOV) {
 		if (existing_vfs != total_vfs) {
@@ -2514,13 +2590,14 @@ 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;
+free_mem:
+	dev->persist->num_vfs = 0;
 	kfree(dev->dev_vfs);
 	return dev_flags & ~MLX4_FLAG_MASTER;
 }
@@ -2544,7 +2621,8 @@ static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap
 }
 
 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
-			 int total_vfs, int *nvfs, struct mlx4_priv *priv)
+			 int total_vfs, int *nvfs, struct mlx4_priv *priv,
+			 int reset_flow)
 {
 	struct mlx4_dev *dev;
 	unsigned sum = 0;
@@ -2607,10 +2685,15 @@ 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;
 		}
 	}
 
+	/* on load remove any previous indication of internal error,
+	 * device is up.
+	 */
+	dev->persist->state = MLX4_DEVICE_STATE_UP;
+
 slave_start:
 	err = mlx4_cmd_init(dev);
 	if (err) {
@@ -2661,8 +2744,10 @@ slave_start:
 				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);
+				u64 dev_flags = mlx4_enable_sriov(dev, pdev,
+								  total_vfs,
+								  existing_vfs,
+								  reset_flow);
 
 				mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
 				dev->flags = dev_flags;
@@ -2704,7 +2789,7 @@ slave_start:
 			if (dev->flags & MLX4_FLAG_SRIOV) {
 				if (!existing_vfs)
 					pci_disable_sriov(pdev);
-				if (mlx4_is_master(dev))
+				if (mlx4_is_master(dev) && !reset_flow)
 					atomic_dec(&pf_loading);
 				dev->flags &= ~MLX4_FLAG_SRIOV;
 			}
@@ -2718,7 +2803,8 @@ slave_start:
 	}
 
 	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);
+		u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
+						  existing_vfs, reset_flow);
 
 		if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
 			mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
@@ -2771,12 +2857,14 @@ 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;
@@ -2828,6 +2916,17 @@ slave_start:
 		goto err_steer;
 
 	mlx4_init_quotas(dev);
+	/* When PF resources are ready arm its comm channel to enable
+	 * getting commands
+	 */
+	if (mlx4_is_master(dev)) {
+		err = mlx4_ARM_COMM_CHANNEL(dev);
+		if (err) {
+			mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
+				 err);
+			goto err_steer;
+		}
+	}
 
 	for (port = 1; port <= dev->caps.num_ports; port++) {
 		err = mlx4_init_port_info(dev, port);
@@ -2846,7 +2945,7 @@ slave_start:
 
 	priv->removed = 0;
 
-	if (mlx4_is_master(dev) && dev->num_vfs)
+	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
 		atomic_dec(&pf_loading);
 
 	kfree(dev_cap);
@@ -2905,10 +3004,12 @@ err_cmd:
 	mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
 
 err_sriov:
-	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs)
+	if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
 		pci_disable_sriov(pdev);
+		dev->flags &= ~MLX4_FLAG_SRIOV;
+	}
 
-	if (mlx4_is_master(dev) && dev->num_vfs)
+	if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
 		atomic_dec(&pf_loading);
 
 	kfree(priv->dev.dev_vfs);
@@ -3049,11 +3150,19 @@ static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
 		}
 	}
 
-	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv);
+	err = mlx4_catas_init(&priv->dev);
 	if (err)
 		goto err_release_regions;
+
+	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
+	if (err)
+		goto err_catas;
+
 	return 0;
 
+err_catas:
+	mlx4_catas_end(&priv->dev);
+
 err_release_regions:
 	pci_release_regions(pdev);
 
@@ -3076,38 +3185,60 @@ 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;
+	mutex_init(&dev->persist->device_state_mutex);
+	mutex_init(&dev->persist->interface_state_mutex);
 
 	ret =  __mlx4_init_one(pdev, id->driver_data, priv);
-	if (ret)
+	if (ret) {
+		kfree(dev->persist);
 		kfree(priv);
+	} else {
+		pci_save_state(pdev);
+	}
 
 	return ret;
 }
 
+static void mlx4_clean_dev(struct mlx4_dev *dev)
+{
+	struct mlx4_dev_persistent *persist = dev->persist;
+	struct mlx4_priv *priv = mlx4_priv(dev);
+	unsigned long	flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
+
+	memset(priv, 0, sizeof(*priv));
+	priv->dev.persist = persist;
+	priv->dev.flags = flags;
+}
+
 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;
-	int active_vfs = 0;
+	int p, i;
 
 	if (priv->removed)
 		return;
 
+	/* saving current ports type for further use */
+	for (i = 0; i < dev->caps.num_ports; i++) {
+		dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
+		dev->persist->curr_port_poss_type[i] = dev->caps.
+						       possible_type[i + 1];
+	}
+
 	pci_dev_data = priv->pci_dev_data;
 
-	/* Disabling SR-IOV is not allowed while there are active vf's */
-	if (mlx4_is_master(dev)) {
-		active_vfs = mlx4_how_many_lives_vf(dev);
-		if (active_vfs) {
-			pr_warn("Removing PF when there are active VF's !!\n");
-			pr_warn("Will not disable SR-IOV.\n");
-		}
-	}
 	mlx4_stop_sense(dev);
 	mlx4_unregister_device(dev);
 
@@ -3151,12 +3282,6 @@ static void mlx4_unload_one(struct pci_dev *pdev)
 
 	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;
-	}
 
 	if (!mlx4_is_slave(dev))
 		mlx4_free_ownership(dev);
@@ -3168,42 +3293,96 @@ static void mlx4_unload_one(struct pci_dev *pdev)
 	kfree(dev->caps.qp1_proxy);
 	kfree(dev->dev_vfs);
 
-	memset(priv, 0, sizeof(*priv));
+	mlx4_clean_dev(dev);
 	priv->pci_dev_data = pci_dev_data;
 	priv->removed = 1;
 }
 
 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);
+	int active_vfs = 0;
+
+	mutex_lock(&persist->interface_state_mutex);
+	persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
+	mutex_unlock(&persist->interface_state_mutex);
+
+	/* Disabling SR-IOV is not allowed while there are active vf's */
+	if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
+		active_vfs = mlx4_how_many_lives_vf(dev);
+		if (active_vfs) {
+			pr_warn("Removing PF when there are active VF's !!\n");
+			pr_warn("Will not disable SR-IOV.\n");
+		}
+	}
+
+	/* device marked to be under deletion running now without the lock
+	 * letting other tasks to be terminated
+	 */
+	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
+		mlx4_unload_one(pdev);
+	else
+		mlx4_info(dev, "%s: interface is down\n", __func__);
+	mlx4_catas_end(dev);
+	if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
+		mlx4_warn(dev, "Disabling SR-IOV\n");
+		pci_disable_sriov(pdev);
+	}
 
-	mlx4_unload_one(pdev);
 	pci_release_regions(pdev);
 	pci_disable_device(pdev);
+	kfree(dev->persist);
 	kfree(priv);
 	pci_set_drvdata(pdev, NULL);
 }
 
+static int restore_current_port_types(struct mlx4_dev *dev,
+				      enum mlx4_port_type *types,
+				      enum mlx4_port_type *poss_types)
+{
+	struct mlx4_priv *priv = mlx4_priv(dev);
+	int err, i;
+
+	mlx4_stop_sense(dev);
+
+	mutex_lock(&priv->port_mutex);
+	for (i = 0; i < dev->caps.num_ports; i++)
+		dev->caps.possible_type[i + 1] = poss_types[i];
+	err = mlx4_change_port_types(dev, types);
+	mlx4_start_sense(dev);
+	mutex_unlock(&priv->port_mutex);
+
+	return err;
+}
+
 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);
+	err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
 	if (err) {
 		mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
 			 __func__, pci_name(pdev), err);
 		return err;
 	}
 
+	err = restore_current_port_types(dev, dev->persist->curr_port_type,
+					 dev->persist->curr_port_poss_type);
+	if (err)
+		mlx4_err(dev, "could not restore original port types (%d)\n",
+			 err);
+
 	return err;
 }
 
@@ -3258,23 +3437,79 @@ MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
 					      pci_channel_state_t state)
 {
-	mlx4_unload_one(pdev);
+	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
+
+	mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
+	mlx4_enter_error_state(persist);
 
-	return state == pci_channel_io_perm_failure ?
-		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
+	mutex_lock(&persist->interface_state_mutex);
+	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
+		mlx4_unload_one(pdev);
+
+	mutex_unlock(&persist->interface_state_mutex);
+	if (state == pci_channel_io_perm_failure)
+		return PCI_ERS_RESULT_DISCONNECT;
+
+	pci_disable_device(pdev);
+	return PCI_ERS_RESULT_NEED_RESET;
 }
 
 static pci_ers_result_t mlx4_pci_slot_reset(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               ret;
+	int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
+	int total_vfs;
+
+	mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
+	ret = pci_enable_device(pdev);
+	if (ret) {
+		mlx4_err(dev, "Can not re-enable device, ret=%d\n", ret);
+		return PCI_ERS_RESULT_DISCONNECT;
+	}
+
+	pci_set_master(pdev);
+	pci_restore_state(pdev);
+	pci_save_state(pdev);
+
+	total_vfs = dev->persist->num_vfs;
+	memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
+
+	mutex_lock(&persist->interface_state_mutex);
+	if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
+		ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
+				    priv, 1);
+		if (ret) {
+			mlx4_err(dev, "%s: mlx4_load_one failed, ret=%d\n",
+				 __func__,  ret);
+			goto end;
+		}
 
-	ret = __mlx4_init_one(pdev, priv->pci_dev_data, priv);
+		ret = restore_current_port_types(dev, dev->persist->
+						 curr_port_type, dev->persist->
+						 curr_port_poss_type);
+		if (ret)
+			mlx4_err(dev, "could not restore original port types (%d)\n", ret);
+	}
+end:
+	mutex_unlock(&persist->interface_state_mutex);
 
 	return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
 }
 
+static void mlx4_shutdown(struct pci_dev *pdev)
+{
+	struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
+
+	mlx4_info(persist->dev, "mlx4_shutdown was called\n");
+	mutex_lock(&persist->interface_state_mutex);
+	if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
+		mlx4_unload_one(pdev);
+	mutex_unlock(&persist->interface_state_mutex);
+}
+
 static const struct pci_error_handlers mlx4_err_handler = {
 	.error_detected = mlx4_pci_err_detected,
 	.slot_reset     = mlx4_pci_slot_reset,
@@ -3284,7 +3519,7 @@ static struct pci_driver mlx4_driver = {
 	.name		= DRV_NAME,
 	.id_table	= mlx4_pci_table,
 	.probe		= mlx4_init_one,
-	.shutdown	= mlx4_unload_one,
+	.shutdown	= mlx4_shutdown,
 	.remove		= mlx4_remove_one,
 	.err_handler    = &mlx4_err_handler,
 };
@@ -3336,7 +3571,6 @@ static int __init mlx4_init(void)
 	if (mlx4_verify_params())
 		return -EINVAL;
 
-	mlx4_catas_init();
 
 	mlx4_wq = create_singlethread_workqueue("mlx4");
 	if (!mlx4_wq)

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

@@ -1318,6 +1318,9 @@ out:
 	mutex_unlock(&priv->mcg_table.mutex);
 
 	mlx4_free_cmd_mailbox(dev, mailbox);
+	if (err && dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+		/* In case device is under an error, return success as a closing command */
+		err = 0;
 	return err;
 }
 
@@ -1347,6 +1350,9 @@ static int mlx4_QP_ATTACH(struct mlx4_dev *dev, struct mlx4_qp *qp,
 		       MLX4_CMD_WRAPPED);
 
 	mlx4_free_cmd_mailbox(dev, mailbox);
+	if (err && !attach &&
+	    dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
+		err = 0;
 	return err;
 }
 

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

@@ -85,7 +85,9 @@ enum {
 	MLX4_CLR_INT_SIZE	= 0x00008,
 	MLX4_SLAVE_COMM_BASE	= 0x0,
 	MLX4_COMM_PAGESIZE	= 0x1000,
-	MLX4_CLOCK_SIZE		= 0x00008
+	MLX4_CLOCK_SIZE		= 0x00008,
+	MLX4_COMM_CHAN_CAPS	= 0x8,
+	MLX4_COMM_CHAN_FLAGS	= 0xc
 };
 
 enum {
@@ -120,6 +122,10 @@ enum mlx4_mpt_state {
 };
 
 #define MLX4_COMM_TIME		10000
+#define MLX4_COMM_OFFLINE_TIME_OUT 30000
+#define MLX4_COMM_CMD_NA_OP    0x0
+
+
 enum {
 	MLX4_COMM_CMD_RESET,
 	MLX4_COMM_CMD_VHCR0,
@@ -221,19 +227,21 @@ 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;
+extern int mlx4_internal_err_reset;
 
 #define MLX4_MAX_NUM_SLAVES	(MLX4_MAX_NUM_PF + MLX4_MAX_NUM_VF)
 #define ALL_SLAVES 0xff
@@ -606,7 +614,6 @@ struct mlx4_mgm {
 struct mlx4_cmd {
 	struct pci_pool	       *pool;
 	void __iomem	       *hcr;
-	struct mutex		hcr_mutex;
 	struct mutex		slave_cmd_mutex;
 	struct semaphore	poll_sem;
 	struct semaphore	event_sem;
@@ -994,7 +1001,8 @@ void __mlx4_xrcd_free(struct mlx4_dev *dev, u32 xrcdn);
 
 void mlx4_start_catas_poll(struct mlx4_dev *dev);
 void mlx4_stop_catas_poll(struct mlx4_dev *dev);
-void mlx4_catas_init(void);
+int mlx4_catas_init(struct mlx4_dev *dev);
+void mlx4_catas_end(struct mlx4_dev *dev);
 int mlx4_restart_one(struct pci_dev *pdev);
 int mlx4_register_device(struct mlx4_dev *dev);
 void mlx4_unregister_device(struct mlx4_dev *dev);
@@ -1160,13 +1168,14 @@ enum {
 int mlx4_cmd_init(struct mlx4_dev *dev);
 void mlx4_cmd_cleanup(struct mlx4_dev *dev, int cleanup_mask);
 int mlx4_multi_func_init(struct mlx4_dev *dev);
+int mlx4_ARM_COMM_CHANNEL(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);
 int mlx4_cmd_use_events(struct mlx4_dev *dev);
 void mlx4_cmd_use_polling(struct mlx4_dev *dev);
 
 int mlx4_comm_cmd(struct mlx4_dev *dev, u8 cmd, u16 param,
-		  unsigned long timeout);
+		  u16 op, unsigned long timeout);
 
 void mlx4_cq_tasklet_cb(unsigned long data);
 void mlx4_cq_completion(struct mlx4_dev *dev, u32 cqn);
@@ -1176,7 +1185,7 @@ void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type);
 
 void mlx4_srq_event(struct mlx4_dev *dev, u32 srqn, int event_type);
 
-void mlx4_handle_catas_err(struct mlx4_dev *dev);
+void mlx4_enter_error_state(struct mlx4_dev_persistent *persist);
 
 int mlx4_SENSE_PORT(struct mlx4_dev *dev, int port,
 		    enum mlx4_port_type *type);

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/cmd.h

@@ -279,6 +279,8 @@ int mlx4_get_vf_config(struct mlx4_dev *dev, int port, int vf, struct ifla_vf_in
 int mlx4_set_vf_link_state(struct mlx4_dev *dev, int port, int vf, int link_state);
 int mlx4_config_dev_retrieval(struct mlx4_dev *dev,
 			      struct mlx4_config_dev_params *params);
+void mlx4_cmd_wake_completions(struct mlx4_dev *dev);
+void mlx4_report_internal_err_comm_event(struct mlx4_dev *dev);
 /*
  * mlx4_get_slave_default_vlan -
  * return true if VST ( default vlan)
@@ -288,5 +290,6 @@ bool mlx4_get_slave_default_vlan(struct mlx4_dev *dev, int port, int slave,
 				 u16 *vlan, u8 *qos);
 
 #define MLX4_COMM_GET_IF_REV(cmd_chan_ver) (u8)((cmd_chan_ver) >> 8)
+#define COMM_CHAN_EVENT_INTERNAL_ERR (1 << 17)
 
 #endif /* MLX4_CMD_H */

include/linux/mlx4/device.h

@@ -208,6 +208,10 @@ enum {
 	MLX4_QUERY_FUNC_FLAGS_A0_RES_QP		= 1LL << 1
 };
 
+enum {
+	MLX4_VF_CAP_FLAG_RESET			= 1 << 0
+};
+
 /* bit enums for an 8-bit flags field indicating special use
  * QPs which require special handling in qp_reserve_range.
  * Currently, this only includes QPs used by the ETH interface,
@@ -411,6 +415,16 @@ enum {
 	MLX4_EQ_PORT_INFO_MSTR_SM_SL_CHANGE_MASK	= 1 << 4,
 };
 
+enum {
+	MLX4_DEVICE_STATE_UP			= 1 << 0,
+	MLX4_DEVICE_STATE_INTERNAL_ERROR	= 1 << 1,
+};
+
+enum {
+	MLX4_INTERFACE_STATE_UP		= 1 << 0,
+	MLX4_INTERFACE_STATE_DELETION	= 1 << 1,
+};
+
 #define MSTR_SM_CHANGE_MASK (MLX4_EQ_PORT_INFO_MSTR_SM_SL_CHANGE_MASK | \
 			     MLX4_EQ_PORT_INFO_MSTR_SM_LID_CHANGE_MASK)
 
@@ -535,6 +549,7 @@ struct mlx4_caps {
 	u8			alloc_res_qp_mask;
 	u32			dmfs_high_rate_qpn_base;
 	u32			dmfs_high_rate_qpn_range;
+	u32			vf_caps;
 };
 
 struct mlx4_buf_list {
@@ -744,8 +759,23 @@ 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;
+	enum mlx4_port_type curr_port_type[MLX4_MAX_PORTS + 1];
+	enum mlx4_port_type curr_port_poss_type[MLX4_MAX_PORTS + 1];
+	struct work_struct      catas_work;
+	struct workqueue_struct *catas_wq;
+	struct mutex	device_state_mutex; /* protect HW state */
+	u8		state;
+	struct mutex	interface_state_mutex; /* protect SW state */
+	u8	interface_state;
+};
+
+struct mlx4_dev {
+	struct mlx4_dev_persistent *persist;
 	unsigned long		flags;
 	unsigned long		num_slaves;
 	struct mlx4_caps	caps;
@@ -754,13 +784,11 @@ 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 {