On 3/27/24 01:46, Will Deacon wrote:
On Tue, Mar 26, 2024 at 11:43:13AM +0000, Will Deacon wrote: Ok, long shot after eyeballing the vhost code, but does the diff below help at all? It looks like vhost_vq_avail_empty() can advance the value saved in 'vq->avail_idx' but without the read barrier, possibly confusing vhost_get_vq_desc() in polling mode. diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index 045f666b4f12..87bff710331a 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -2801,6 +2801,7 @@ bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) return false; vq->avail_idx = vhost16_to_cpu(vq, avail_idx);+ smp_rmb();return vq->avail_idx == vq->last_avail_idx; } EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
Thanks, Will. I already noticed smp_rmb() has been missed in vhost_vq_avail_empty(). The issue still exists after smp_rmb() is added here. However, I'm inspired by your suggestion and recheck the code again. It seems another smp_rmb() has been missed in vhost_enable_notify(). With smp_rmb() added to vhost_vq_avail_empty() and vhost_enable_notify(), I'm unable to hit the issue. I will try for more times to make sure the issue is really resolved. After that, I will post formal patches for review. Thanks, Gavin
