MFC r330646:

Fix race between PCI error handlers and health work in mlx5core.

linux commit 05ac2c0b7438ea08c5d54b48797acf9b22cb2f6f

Submitted by:	Matthew Finlay <matt@mellanox.com>
Sponsored by:	Mellanox Technologies

3 files changed, 33 insertions, 2 deletions

diff --git a/sys/dev/mlx5/driver.h b/sys/dev/mlx5/driver.h
index 1f6520a..b5cce4a 100644
--- a/sys/dev/mlx5/driver.h
+++ b/sys/dev/mlx5/driver.h
@@ -482,7 +482,10 @@ struct mlx5_core_health {
 	u32				prev;
 	int				miss_counter;
 	bool				sick;
+	/* wq spinlock to synchronize draining */
+	spinlock_t			wq_lock;
 	struct workqueue_struct	       *wq;
+	unsigned long			flags;
 	struct work_struct		work;
 };
 
@@ -871,6 +874,7 @@ void mlx5_health_cleanup(struct mlx5_core_dev *dev);
 int mlx5_health_init(struct mlx5_core_dev *dev);
 void mlx5_start_health_poll(struct mlx5_core_dev *dev);
 void mlx5_stop_health_poll(struct mlx5_core_dev *dev);
+void mlx5_drain_health_wq(struct mlx5_core_dev *dev);
 
 #define	mlx5_buf_alloc_node(dev, size, direct, buf, node) \
 	mlx5_buf_alloc(dev, size, direct, buf)
diff --git a/sys/dev/mlx5/mlx5_core/mlx5_health.c b/sys/dev/mlx5/mlx5_core/mlx5_health.c
index eb77ea2..4e84a34 100644
--- a/sys/dev/mlx5/mlx5_core/mlx5_health.c
+++ b/sys/dev/mlx5/mlx5_core/mlx5_health.c
@@ -43,6 +43,10 @@ enum {
 	MLX5_NIC_IFC_NO_DRAM_NIC	= 2
 };
 
+enum {
+	MLX5_DROP_NEW_HEALTH_WORK,
+};
+
 static u8 get_nic_interface(struct mlx5_core_dev *dev)
 {
 	return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 3;
@@ -242,7 +246,13 @@ static void poll_health(unsigned long data)
 	if (in_fatal(dev) && !health->sick) {
 		health->sick = true;
 		print_health_info(dev);
-		queue_work(health->wq, &health->work);
+		spin_lock(&health->wq_lock);
+		if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags))
+			queue_work(health->wq, &health->work);
+		else
+			dev_err(&dev->pdev->dev,
+				"new health works are not permitted at this stage\n");
+		spin_unlock(&health->wq_lock);
 	}
 }
 
@@ -252,6 +262,7 @@ void mlx5_start_health_poll(struct mlx5_core_dev *dev)
 
 	init_timer(&health->timer);
 	health->sick = 0;
+	clear_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
 	health->health = &dev->iseg->health;
 	health->health_counter = &dev->iseg->health_counter;
 
@@ -267,6 +278,16 @@ void mlx5_stop_health_poll(struct mlx5_core_dev *dev)
 	del_timer_sync(&health->timer);
 }
 
+void mlx5_drain_health_wq(struct mlx5_core_dev *dev)
+{
+	struct mlx5_core_health *health = &dev->priv.health;
+
+	spin_lock(&health->wq_lock);
+	set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags);
+	spin_unlock(&health->wq_lock);
+	cancel_work_sync(&health->work);
+}
+
 void mlx5_health_cleanup(struct mlx5_core_dev *dev)
 {
 	struct mlx5_core_health *health = &dev->priv.health;
@@ -293,6 +314,7 @@ int mlx5_health_init(struct mlx5_core_dev *dev)
 	if (!health->wq)
 		return -ENOMEM;
 
+	spin_lock_init(&health->wq_lock);
 	INIT_WORK(&health->work, health_care);
 
 	return 0;
diff --git a/sys/dev/mlx5/mlx5_core/mlx5_main.c b/sys/dev/mlx5/mlx5_core/mlx5_main.c
index ddf70c3..a8ac5b8 100644
--- a/sys/dev/mlx5/mlx5_core/mlx5_main.c
+++ b/sys/dev/mlx5/mlx5_core/mlx5_main.c
@@ -1243,7 +1243,12 @@ static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
 	dev_info(&pdev->dev, "%s was called\n", __func__);
 	mlx5_enter_error_state(dev);
 	mlx5_unload_one(dev, priv);
-	mlx5_pci_disable_device(dev);
+	if (state) {
+		pci_save_state(pdev->dev.bsddev);
+		mlx5_drain_health_wq(dev);
+		mlx5_pci_disable_device(dev);
+	}
+
 	return state == pci_channel_io_perm_failure ?
 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
 }