1 files changed, 225 insertions, 69 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx4/catas.c b/drivers/net/ethernet/mellanox/mlx4/catas.c
index 9c656fe4983d..715de8affcc9 100644
--- a/drivers/net/ethernet/mellanox/mlx4/catas.c
+++ b/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;
+	}
 }