On 01/04/2019 12:42 AM, Michael S. Tsirkin wrote:
On Thu, Jan 03, 2019 at 01:31:01PM +0800, Wei Wang wrote:
virtio-ccw has deadlock issues with reading config registers inside the
interrupt context, so we tweak the virtballoon_changed implementation
by moving the config read operations into the related workqueue contexts.
Signed-off-by: Wei Wang <wei.w.wang@xxxxxxxxx>
---
drivers/virtio/virtio_balloon.c | 54 ++++++++++++++++++++---------------------
1 file changed, 26 insertions(+), 28 deletions(-)
diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index 728ecd1..9a82a11 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -394,33 +394,15 @@ static void virtballoon_changed(struct virtio_device *vdev)
{
struct virtio_balloon *vb = vdev->priv;
unsigned long flags;
- s64 diff = towards_target(vb);
-
- if (diff) {
- spin_lock_irqsave(&vb->stop_update_lock, flags);
- if (!vb->stop_update)
- queue_work(system_freezable_wq,
- &vb->update_balloon_size_work);
- spin_unlock_irqrestore(&vb->stop_update_lock, flags);
- }
- if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
- virtio_cread(vdev, struct virtio_balloon_config,
- free_page_report_cmd_id, &vb->cmd_id_received);
There's one problem with this approach:
previously updating the cmd_id_received here would immediately
stop the report in send_free_pages.
With this approach we are waiting for the wq to schedule,
which might be blocked waiting for report to complete.
So host can no longer quickly stop the report in progress.
A simple work-around would be to set some kind of flag whenever there
is a change interrupt, then have send_free_pages test it
and re-read cmd_id_received.
Needs to be an atomic I guess ...
Yes, sounds better..will update the patch.
Best,
Wei
