On Tue, Jun 15, 2021 at 10:33:22PM +0800, Jason Wang wrote:

在 2021/6/6 下午9:29, Gautam Dawar 写道:

quoted

From: Gautam Dawar <redacted>

If some module invokes vdpa_device_unregister (usually in the module
unload function) when the userspace app (eg. QEMU) which had opened
the vhost-vdpa character device is still running, vhost_vdpa_remove()
function will block indefinitely in call to wait_for_completion().

This causes the vdpa_device_unregister caller to hang and with a
usual side-effect of rmmod command not returning when this call
is in the module_exit function.

This patch converts the wait_for_completion call to its timeout based
counterpart (wait_for_completion_timeout) and also adds a warning
message to alert the user/administrator about this hang situation.

To eventually fix this problem, a mechanism will be required to let
vhost-vdpa module inform the userspace of this situation and
userspace will close the descriptor of vhost-vdpa char device.
This will enable vhost-vdpa to continue with graceful clean-up.

Signed-off-by: Gautam Dawar <redacted>
---
  drivers/vhost/vdpa.c | 6 +++++-
  1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index bfa4c6ef554e..572b64d09b06 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c

@@ -1091,7 +1091,11 @@ static void vhost_vdpa_remove(struct vdpa_device *vdpa)
  		opened = atomic_cmpxchg(&v->opened, 0, 1);
  		if (!opened)
  			break;
-		wait_for_completion(&v->completion);
+		wait_for_completion_timeout(&v->completion,
+					    msecs_to_jiffies(1000));
+		dev_warn_ratelimited(&v->dev,
+				     "%s waiting for /dev/%s to be closed\n",
+				     __func__, dev_name(&v->dev));

Can fill up the kernel log in this case ... dev_warn_once seems more
appropriate.

quoted

  	} while (1);
  	put_device(&v->dev);

Acked-by: Jason Wang <jasowang@redhat.com>