When trying to talk to a device that has gone out to lunch, the virtnet_send_command() will sit and spin forever, causing a soft lockup and eventually crashing the kernel. Add a limit to the spin and return false if we hit the timeout. The 2 second time limit seems a bit arbitrary, but a reasonable place to start.
This is a little more brute force than Jason's suggestions in [1], but at least prevents the soft lockups and eventual kernel crash that we were seeing in testing. [1]: https://lore.kernel.org/netdev/20230524081842.3060-1-jasow...@redhat.com/ Fixes: 2a41f71d3bd9 ("virtio_net: Add a virtqueue for outbound control commands") Signed-off-by: Shannon Nelson <shannon.nel...@amd.com> --- drivers/net/virtio_net.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 0db14f6b87d3..c3bf1c9f3244 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -2264,6 +2264,8 @@ static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, { struct scatterlist *sgs[4], hdr, stat; unsigned out_num = 0, tmp; + unsigned long deadline; + bool timeout; int ret; /* Caller should know better */ @@ -2297,11 +2299,16 @@ static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, /* Spin for a response, the kick causes an ioport write, trapping * into the hypervisor, so the request should be handled immediately. */ + deadline = jiffies + 2 * HZ; + timeout = false; while (!virtqueue_get_buf(vi->cvq, &tmp) && - !virtqueue_is_broken(vi->cvq)) + !virtqueue_is_broken(vi->cvq) && + !timeout) { cpu_relax(); + timeout = time_after(jiffies, deadline); + } - return vi->ctrl->status == VIRTIO_NET_OK; + return vi->ctrl->status == VIRTIO_NET_OK && !timeout; } static int virtnet_set_mac_address(struct net_device *dev, void *p) -- 2.17.1 _______________________________________________ Virtualization mailing list Virtualization@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/virtualization
