net/mlx4_core: Activate reset flow upon fatal command cases

We activate reset flow upon command fatal errors, when the device enters an
erroneous state, and must be reset.

The cases below are assumed to be fatal: FW command timed-out, an error from FW
on closing commands, pci is offline when posting/pending a command.

In those cases we place the device into an error state: chip is reset, pending
commands are awakened and completed immediately. Subsequent commands will
return immediately.

The return code in the above cases will depend on the command. Commands which
free and close resources will return success (because the chip was reset, so
callers may safely free their kernel resources). Other commands will return -EIO.

Since the device's state was marked as error, the catas poller will
detect this and restart the device's software stack (as is done when a FW
internal error is directly detected). The device state is protected by a
persistent mutex lives on its mlx4_dev, as such no need any more for the
hcr_mutex which is removed.
Signed-off-by: NYishai Hadas <yishaih@mellanox.com>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

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

@@ -42,8 +42,8 @@ enum {
 
 
 
-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)");
@@ -92,7 +92,7 @@ void mlx4_enter_error_state(struct mlx4_dev_persistent *persist)
 	int err;
 	struct mlx4_dev *dev;
 
-	if (!internal_err_reset)
+	if (!mlx4_internal_err_reset)
 		return;
 
 	mutex_lock(&persist->device_state_mutex);
@@ -110,6 +110,7 @@ void mlx4_enter_error_state(struct mlx4_dev_persistent *persist)
 
 	/* 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:
@@ -157,7 +158,7 @@ static void poll_catas(unsigned long dev_ptr)
 	return;
 
 internal_err:
-	if (internal_err_reset)
+	if (mlx4_internal_err_reset)
 		queue_work(dev->persist->catas_wq, &dev->persist->catas_work);
 }
 
@@ -177,7 +178,7 @@ void mlx4_start_catas_poll(struct mlx4_dev *dev)
 
 	/*If we are in SRIOV the default of the module param must be 0*/
 	if (mlx4_is_mfunc(dev))
-		internal_err_reset = 0;
+		mlx4_internal_err_reset = 0;
 
 	INIT_LIST_HEAD(&priv->catas_err.list);
 	init_timer(&priv->catas_err.timer);

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

@@ -182,6 +182,72 @@ 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);
@@ -258,7 +324,7 @@ 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);
 
@@ -323,17 +389,21 @@ 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->persist->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;
 	}
 
@@ -347,7 +417,6 @@ static int mlx4_cmd_post(struct mlx4_dev *dev, u64 in_param, u64 out_param,
 			 * Device is going through error recovery
 			 * and cannot accept commands.
 			 */
-			ret = -EIO;
 			goto out;
 		}
 
@@ -391,7 +460,11 @@ 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;
 }
 
@@ -464,12 +537,12 @@ 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->persist->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;
 	}
 
@@ -483,7 +556,7 @@ 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)) {
@@ -493,6 +566,11 @@ static int mlx4_cmd_poll(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 			 * 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 +580,8 @@ 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 +593,17 @@ 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 +650,19 @@ 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 +679,20 @@ 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;
@@ -613,9 +708,12 @@ int __mlx4_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
 	       u16 op, unsigned long timeout, int native)
 {
 	if (pci_channel_offline(dev->persist->pdev))
-		return -EIO;
+		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,
@@ -2121,7 +2219,6 @@ 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;
@@ -2232,6 +2329,11 @@ 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;
@@ -2329,6 +2431,25 @@ 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;

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

@@ -1318,6 +1318,9 @@ int mlx4_qp_detach_common(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
 	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;
 }
 

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

@@ -235,6 +235,7 @@ do {									\
 
 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
@@ -607,7 +608,6 @@ 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;

include/linux/mlx4/cmd.h

@@ -279,6 +279,7 @@ 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);
 /*
  * mlx4_get_slave_default_vlan -
  * return true if VST ( default vlan)