On Wed, May 21, 2025 at 10:34:46AM +0000, Parav Pandit wrote:
From: Michael S. Tsirkin mst@redhat.com Sent: Wednesday, May 21, 2025 3:46 PM
On Wed, May 21, 2025 at 09:32:30AM +0000, Parav Pandit wrote:
From: Michael S. Tsirkin mst@redhat.com Sent: Wednesday, May 21, 2025 2:49 PM To: Parav Pandit parav@nvidia.com Cc: stefanha@redhat.com; axboe@kernel.dk; virtualization@lists.linux.dev; linux-block@vger.kernel.or; stable@vger.kernel.org; NBU-Contact-Li Rongqing (EXTERNAL) lirongqing@baidu.com; Chaitanya Kulkarni chaitanyak@nvidia.com; xuanzhuo@linux.alibaba.com; pbonzini@redhat.com; jasowang@redhat.com; Max Gurtovoy mgurtovoy@nvidia.com; Israel Rukshin israelr@nvidia.com Subject: Re: [PATCH v1] virtio_blk: Fix disk deletion hang on device surprise removal
On Wed, May 21, 2025 at 09:14:31AM +0000, Parav Pandit wrote:
From: Michael S. Tsirkin mst@redhat.com Sent: Wednesday, May 21, 2025 1:48 PM
On Wed, May 21, 2025 at 06:37:41AM +0000, Parav Pandit wrote: > When the PCI device is surprise removed, requests may not > complete the device as the VQ is marked as broken. Due to > this, the disk deletion hangs. > > Fix it by aborting the requests when the VQ is broken. > > With this fix now fio completes swiftly. > An alternative of IO timeout has been considered, however when > the driver knows about unresponsive block device, swiftly > clearing them enables users and upper layers to react quickly. > > Verified with multiple device unplug iterations with pending > requests in virtio used ring and some pending with the device. > > Fixes: 43bb40c5b926 ("virtio_pci: Support surprise removal of > virtio pci device") > Cc: stable@vger.kernel.org > Reported-by: lirongqing@baidu.com > Closes: > https://lore.kernel.org/virtualization/c45dd68698cd47238c55fb73c a9b4 74 > 1@baidu.com/ > Reviewed-by: Max Gurtovoy mgurtovoy@nvidia.com > Reviewed-by: Israel Rukshin israelr@nvidia.com > Signed-off-by: Parav Pandit parav@nvidia.com > --- > changelog: > v0->v1: > - Fixed comments from Stefan to rename a cleanup function > - Improved logic for handling any outstanding requests > in bio layer > - improved cancel callback to sync with ongoing done()
thanks for the patch! questions:
> --- > drivers/block/virtio_blk.c | 95 > ++++++++++++++++++++++++++++++++++++++ > 1 file changed, 95 insertions(+) > > diff --git a/drivers/block/virtio_blk.c > b/drivers/block/virtio_blk.c index 7cffea01d868..5212afdbd3c7 > 100644 > --- a/drivers/block/virtio_blk.c > +++ b/drivers/block/virtio_blk.c > @@ -435,6 +435,13 @@ static blk_status_t > virtio_queue_rq(struct blk_mq_hw_ctx *hctx, > blk_status_t status; > int err; > > + /* Immediately fail all incoming requests if the vq is broken. > + * Once the queue is unquiesced, upper block layer flushes > +any pending > + * queued requests; fail them right away. > + */ > + if (unlikely(virtqueue_is_broken(vblk->vqs[qid].vq))) > + return BLK_STS_IOERR; > + > status = virtblk_prep_rq(hctx, vblk, req, vbr); > if (unlikely(status)) > return status;
just below this: spin_lock_irqsave(&vblk->vqs[qid].lock, flags); err = virtblk_add_req(vblk->vqs[qid].vq, vbr); if (err) {
and virtblk_add_req calls virtqueue_add_sgs, so it will fail on a broken
vq.
Why do we need to check it one extra time here?
It may work, but for some reason if the hw queue is stopped in this flow, it
can hang the IOs flushing.
I considered it risky to rely on the error code ENOSPC returned by non virtio-
blk driver.
In other words, if lower layer changed for some reason, we may end up in
stopping the hw queue when broken, and requests would hang.
Compared to that one-time entry check seems more robust.
I don't get it. Checking twice in a row is more robust?
No. I am not confident on the relying on the error code -ENOSPC from layers
outside of virtio-blk driver.
You can rely on virtio core to return an error on a broken vq. The error won't be -ENOSPC though, why would it?
Presently that is not the API contract between virtio core and driver. When the VQ is broken the error code is EIO. This is from the code inspection.
yes
If you prefer to rely on the code inspection of lower layer to define the virtio-blk, I am fine and remove the two checks. I just find it fragile, but if you prefer this way, I am fine.
I think it's better, yes.
If for a broken VQ, ENOSPC arrives, then hw queue is stopped and requests
could be stuck.
Can you describe the scenario in more detail pls? where does ENOSPC arrive from? when is the vq get broken ...
ENOSPC arrives when it fails to enqueue the request in present form. EIO arrives when VQ is broken.
If in the future, ENOSPC arrives for broken VQ, following flow can trigger a hang.
cpu_0: virtblk_broken_device_cleanup() ... blk_mq_unquiesce_queue(); ... stage_1: blk_mq_freeze_queue_wait().
Cpu_1: Queue_rq() virtio_queue_rq() virtblk_add_req() -ENOSPC Stop_hw_queue() At this point, new requests in block layer may get stuck and may not be enqueued to queue_rq().
What am I missing? Can you describe the scenario in more detail?
> @@ -508,6 +515,11 @@ static void virtio_queue_rqs(struct > rq_list
*rqlist)
> while ((req = rq_list_pop(rqlist))) { > struct virtio_blk_vq *this_vq = get_virtio_blk_vq(req- >mq_hctx); > > + if (unlikely(virtqueue_is_broken(this_vq->vq))) { > + rq_list_add_tail(&requeue_list, req); > + continue; > + } > + > if (vq && vq != this_vq) > virtblk_add_req_batch(vq, &submit_list); > vq = this_vq;
similarly
The error code is not surfacing up here from virtblk_add_req().
but wait a sec:
static void virtblk_add_req_batch(struct virtio_blk_vq *vq, struct rq_list *rqlist) { struct request *req; unsigned long flags; bool kick;
spin_lock_irqsave(&vq->lock, flags); while ((req = rq_list_pop(rqlist))) { struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); int err; err = virtblk_add_req(vq->vq, vbr); if (err) { virtblk_unmap_data(req, vbr); virtblk_cleanup_cmd(req); blk_mq_requeue_request(req, true); } } kick = virtqueue_kick_prepare(vq->vq); spin_unlock_irqrestore(&vq->lock, flags); if (kick) virtqueue_notify(vq->vq); }it actually handles the error internally?
For all the errors it requeues the request here.
ok and they will not prevent removal will they?
It should not prevent removal. One must be careful every single time changing it to make sure that hw queues are not stopped in lower layer, but may be this is ok.
It would end up adding checking for special error code here as well to abort
by translating broken VQ -> EIO to break the loop in
virtblk_add_req_batch().
Weighing on specific error code-based data path that may require audit from
lower layers now and future, an explicit check of broken in this layer could be better.
[..]
Checking add was successful is preferred because it has to be done *anyway* - device can get broken after you check before add.
So I would like to understand why are we also checking explicitly and I do not get it so far.
checking explicitly to not depend on specific error code-based logic.
I do not understand. You must handle vq add errors anyway.
I believe removal of the two vq broken checks should also be fine. I would probably add the comment in the code indicating virtio block driver assumes that ENOSPC is not returned for broken VQ.
You can include this in the series if you like. Tweak to taste:
-->
virtio: document ENOSPC
drivers handle ENOSPC specially since it's an error one can get from a working VQ. Document the semantics.
Signed-off-by: Michael S. Tsirkin mst@redhat.com
---
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c index b784aab66867..97ab0cce527d 100644 --- a/drivers/virtio/virtio_ring.c +++ b/drivers/virtio/virtio_ring.c @@ -2296,6 +2296,10 @@ static inline int virtqueue_add(struct virtqueue *_vq, * at the same time (except where noted). * * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). + * + * NB: ENOSPC is a special code that is only returned on an attempt to add a + * buffer to a full VQ. It indicates that some buffers are outstanding and that + * the operation can be retried after some buffers have been used. */ int virtqueue_add_sgs(struct virtqueue *_vq, struct scatterlist *sgs[],