Virtualization - Aug 2021 - [PATCH AUTOSEL 4.4 1/2] virtio: Improve vq->broken access to avoid any compiler optimization

From: Parav Pandit <parav at nvidia.com>

[ Upstream commit 60f0779862e4ab943810187752c462e85f5fa371 ]

Currently vq->broken field is read by virtqueue_is_broken() in busy
loop in one context by virtnet_send_command().

vq->broken is set to true in other process context by
virtio_break_device(). Reader and writer are accessing it without any
synchronization. This may lead to a compiler optimization which may
result to optimize reading vq->broken only once.

Hence, force reading vq->broken on each invocation of
virtqueue_is_broken() and also force writing it so that such
update is visible to the readers.

It is a theoretical fix that isn't yet encountered in the field.

Signed-off-by: Parav Pandit <parav at nvidia.com>
Link: https://lore.kernel.org/r/20210721142648.1525924-2-parav at nvidia.com
Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
Signed-off-by: Sasha Levin <sashal at kernel.org>
---
 drivers/virtio/virtio_ring.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 6b3565feddb2..b15c24c4d91f 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -840,7 +840,7 @@ bool virtqueue_is_broken(struct virtqueue *_vq)
 {
 	struct vring_virtqueue *vq = to_vvq(_vq);
 
-	return vq->broken;
+	return READ_ONCE(vq->broken);
 }
 EXPORT_SYMBOL_GPL(virtqueue_is_broken);
 
@@ -854,7 +854,9 @@ void virtio_break_device(struct virtio_device *dev)
 
 	list_for_each_entry(_vq, &dev->vqs, list) {
 		struct vring_virtqueue *vq = to_vvq(_vq);
-		vq->broken = true;
+
+		/* Pairs with READ_ONCE() in virtqueue_is_broken(). */
+		WRITE_ONCE(vq->broken, true);
 	}
 }
 EXPORT_SYMBOL_GPL(virtio_break_device);
-- 
2.30.2

From: Neeraj Upadhyay <neeraju at codeaurora.org>

[ Upstream commit e74cfa91f42c50f7f649b0eca46aa049754ccdbd ]

As __vringh_iov() traverses a descriptor chain, it populates
each descriptor entry into either read or write vring iov
and increments that iov's ->used member. So, as we iterate
over a descriptor chain, at any point, (riov/wriov)->used
value gives the number of descriptor enteries available,
which are to be read or written by the device. As all read
iovs must precede the write iovs, wiov->used should be zero
when we are traversing a read descriptor. Current code checks
for wiov->i, to figure out whether any previous entry in the
current descriptor chain was a write descriptor. However,
iov->i is only incremented, when these vring iovs are consumed,
at a later point, and remain 0 in __vringh_iov(). So, correct
the check for read and write descriptor order, to use
wiov->used.

Acked-by: Jason Wang <jasowang at redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare at redhat.com>
Signed-off-by: Neeraj Upadhyay <neeraju at codeaurora.org>
Link: https://lore.kernel.org/r/1624591502-4827-1-git-send-email-neeraju at
codeaurora.org
Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
Signed-off-by: Sasha Levin <sashal at kernel.org>
---
 drivers/vhost/vringh.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/vhost/vringh.c b/drivers/vhost/vringh.c
index d56736655dec..da47542496cc 100644
--- a/drivers/vhost/vringh.c
+++ b/drivers/vhost/vringh.c
@@ -329,7 +329,7 @@ __vringh_iov(struct vringh *vrh, u16 i,
 			iov = wiov;
 		else {
 			iov = riov;
-			if (unlikely(wiov && wiov->i)) {
+			if (unlikely(wiov && wiov->used)) {
 				vringh_bad("Readable desc %p after writable",
 					   &descs[i]);
 				err = -EINVAL;
-- 
2.30.2