Re: [PATCH v2 2/2] virtio-blk: mark disk dead on ERS permanent failure

["Michael S. Tsirkin" <mst@redhat.com>]

On Fri, Jul 31, 2026 at 11:19:00AM +0800, Xixin Liu wrote:
> After ERS reports pci_channel_io_perm_failure, virtio-pci must ask the
> virtio driver to tear down the block device, not only mark virtqueues
> broken.  Call the virtio driver shutdown hook from virtio-pci on
> perm_failure; virtio-blk implements shutdown with blk_mark_disk_dead().
> 
> Fail new requests early in virtio_queue_rq when virtqueues were removed
> during frozen reset_prepare (vqs == NULL).  A broken vq already fails in
> virtqueue_add; no separate disk_live() check is needed.
> 
> Assisted-by: DeepSeek:deepseek-v3
> Signed-off-by: Xixin Liu <liuxixin@kylinos.cn>
> ---
>  drivers/block/virtio_blk.c         | 31 ++++++++++++++++++++++++++++++
>  drivers/virtio/virtio_pci_common.c | 10 +++++++++-
>  2 files changed, 40 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 32bf3ba07a9d..fda2d4f3c7c6 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -435,6 +435,12 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
>  	blk_status_t status;
>  	int err;
>  
> +	/* VQs may be gone after frozen reset_prepare; avoid NULL deref. */
> +	if (unlikely(!vblk->vqs || !vblk->vdev)) {
> +		blk_mq_start_request(req);
> +		return BLK_STS_IOERR;
> +	}
> +
>  	status = virtblk_prep_rq(hctx, vblk, req, vbr);
>  	if (unlikely(status))
>  		return status;
> @@ -1561,6 +1567,30 @@ static int virtblk_probe(struct virtio_device *vdev)
>  	return err;
>  }
>  
> +
> +/* Stop I/O and mark the gendisk dead (ERS perm_failure or system shutdown). */
> +static void virtblk_shutdown(struct virtio_device *vdev)
> +{
> +	struct virtio_blk *vblk = vdev->priv;
> +	struct request_queue *q;
> +	unsigned int memflags;
> +
> +	if (!vblk || !vblk->disk)
> +		return;
> +
> +	virtio_break_device(vdev);
> +	flush_work(&vblk->config_work);
> +
> +	q = vblk->disk->queue;
> +	memflags = blk_mq_freeze_queue(q);
> +	blk_mq_quiesce_queue_nowait(q);
> +
> +	blk_mark_disk_dead(vblk->disk);
> +
> +	blk_mq_unquiesce_queue(q);
> +	blk_mq_unfreeze_queue(q, memflags);
> +}
> +
>  static void virtblk_remove(struct virtio_device *vdev)
>  {
>  	struct virtio_blk *vblk = vdev->priv;

what happens with requests already outstanding in the queues?
e.g. does this leak memory?

> @@ -1684,6 +1714,7 @@ static struct virtio_driver virtio_blk = {
>  	.probe				= virtblk_probe,
>  	.remove				= virtblk_remove,
>  	.config_changed			= virtblk_config_changed,
> +	.shutdown			= virtblk_shutdown,
>  #ifdef CONFIG_PM_SLEEP
>  	.freeze				= virtblk_freeze,
>  	.restore			= virtblk_restore,
> diff --git a/drivers/virtio/virtio_pci_common.c b/drivers/virtio/virtio_pci_common.c
> index fff6b6e2d0c5..957b3282865c 100644
> --- a/drivers/virtio/virtio_pci_common.c
> +++ b/drivers/virtio/virtio_pci_common.c
> @@ -849,7 +849,15 @@ static pci_ers_result_t virtio_pci_error_detected(struct pci_dev *pci_dev,
>  	case pci_channel_io_perm_failure:
>  		dev_warn(&pci_dev->dev,
>  			 "permanent failure, disconnecting device\n");
> -		virtio_break_device(&vp_dev->vdev);
> +		{
> +			struct virtio_driver *drv =
> +				drv_to_virtio(vp_dev->vdev.dev.driver);
> +
> +			if (drv && drv->shutdown)
> +				drv->shutdown(&vp_dev->vdev);
> +			else
> +				virtio_break_device(&vp_dev->vdev);
> +		}
>  		return PCI_ERS_RESULT_DISCONNECT;

So this is blk specifically, I am somewhat lost as to why
is blk treated differently from other drivers here.


>  	default:
>  		break;
> -- 
> 2.43.0