On 12/24/25 08:45, Michael S. Tsirkin wrote:
On Tue, Dec 23, 2025 at 10:25:32PM +0700, Bui Quang Minh wrote:
Calling napi_disable() on an already disabled napi can cause the
deadlock. Because the delayed refill work will call napi_disable(), we
must ensure that refill work is only enabled and scheduled after we have
enabled the rx queue's NAPI.
Signed-off-by: Bui Quang Minh <[email protected]>
---
drivers/net/virtio_net.c | 31 ++++++++++++++++++++++++-------
1 file changed, 24 insertions(+), 7 deletions(-)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 63126e490bda..8016d2b378cf 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -3208,16 +3208,31 @@ static int virtnet_open(struct net_device *dev)
int i, err;
for (i = 0; i < vi->max_queue_pairs; i++) {
+ bool schedule_refill = false;
+
+ /* - We must call try_fill_recv before enabling napi of the same
+ * receive queue so that it doesn't race with the call in
+ * virtnet_receive.
+ * - We must enable and schedule delayed refill work only when
+ * we have enabled all the receive queue's napi. Otherwise, in
+ * refill_work, we have a deadlock when calling napi_disable on
+ * an already disabled napi.
+ */
I would do:
bool refill = i < vi->curr_queue_pairs;
in fact this is almost the same as resume with
one small difference. pass a flag so we do not duplicate code?
I'll fix it in next version.
if (i < vi->curr_queue_pairs) {
- enable_delayed_refill(&vi->rq[i]);
/* Make sure we have some buffers: if oom use wq. */
if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
- schedule_delayed_work(&vi->rq[i].refill, 0);
+ schedule_refill = true;
}
err = virtnet_enable_queue_pair(vi, i);
if (err < 0)
goto err_enable_qp;
+
+ if (i < vi->curr_queue_pairs) {
+ enable_delayed_refill(&vi->rq[i]);
+ if (schedule_refill)
+ schedule_delayed_work(&vi->rq[i].refill, 0);
hmm. should not schedule be under the lock?
I see that schedule is safe to be called concurrently.
struct work_struct {
atomic_long_t data;
struct list_head entry;
work_func_t func;
#ifdef CONFIG_LOCKDEP
struct lockdep_map lockdep_map;
#endif
};
The atomic_long_t field to set pending bit and worker pool's lock help
with the synchronization.
+ }
}
if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
@@ -3456,11 +3471,16 @@ static void __virtnet_rx_resume(struct virtnet_info *vi,
bool running = netif_running(vi->dev);
bool schedule_refill = false;
+ /* See the comment in virtnet_open for the ordering rule
+ * of try_fill_recv, receive queue napi_enable and delayed
+ * refill enable/schedule.
+ */
so maybe common code?
if (refill && !try_fill_recv(vi, rq, GFP_KERNEL))
schedule_refill = true;
if (running)
virtnet_napi_enable(rq);
+ enable_delayed_refill(rq);
if (schedule_refill)
schedule_delayed_work(&rq->refill, 0);
hmm. should not schedule be under the lock?
}
@@ -3470,18 +3490,15 @@ static void virtnet_rx_resume_all(struct virtnet_info
*vi)
int i;
for (i = 0; i < vi->max_queue_pairs; i++) {
- if (i < vi->curr_queue_pairs) {
- enable_delayed_refill(&vi->rq[i]);
+ if (i < vi->curr_queue_pairs)
__virtnet_rx_resume(vi, &vi->rq[i], true);
- } else {
+ else
__virtnet_rx_resume(vi, &vi->rq[i], false);
- }
}
}
static void virtnet_rx_resume(struct virtnet_info *vi, struct receive_queue *rq)
{
- enable_delayed_refill(rq);
__virtnet_rx_resume(vi, rq, true);
}
so I would add bool to virtnet_rx_resume and call it everywhere,
removing __virtnet_rx_resume. can be a patch on top.
I'll create another patch after this patch to clean up the
__virtnet_rx_resume in the next version.
Thanks,
Quang Minh.
--
2.43.0
