virtio_notify_config() needs to acquire the global mutex, which isn't
allowed from an iothread, and may lead to a deadlock like this:

 - main thead
  * Has acquired: qemu_global_mutex.
  * Is trying the acquire: iothread AioContext lock via
    AIO_WAIT_WHILE (after aio_poll).

 - iothread
  * Has acquired: AioContext lock.
  * Is trying to acquire: qemu_global_mutex (via
    virtio_notify_config->prepare_mmio_access).

If virtio_blk_resize() is called from an iothread, schedule
virtio_notify_config() to be run in the main context BH.

Signed-off-by: Sergio Lopez <slp@redhat.com>
---
Changelog

v3:
 - Unconditionally schedule the work to be done in the main context BH
   (thanks John Snow and Kevin Wolf).

v2:
 - Use aio_bh_schedule_oneshot instead of scheduling a coroutine
   (thanks Kevin Wolf).
 - Switch from RFC to v2 patch.
---
 hw/block/virtio-blk.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 18851601cb..0163285f6f 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -16,6 +16,7 @@
 #include "qemu/iov.h"
 #include "qemu/module.h"
 #include "qemu/error-report.h"
+#include "qemu/main-loop.h"
 #include "trace.h"
 #include "hw/block/block.h"
 #include "hw/qdev-properties.h"
@@ -1086,11 +1087,25 @@ static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
     return 0;
 }
 
+static void virtio_resize_cb(void *opaque)
+{
+    VirtIODevice *vdev = opaque;
+
+    assert(qemu_get_current_aio_context() == qemu_get_aio_context());
+    virtio_notify_config(vdev);
+}
+
 static void virtio_blk_resize(void *opaque)
 {
     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 
-    virtio_notify_config(vdev);
+    /*
+     * virtio_notify_config() needs to acquire the global mutex,
+     * so it can't be called from an iothread. Instead, schedule
+     * it to be run in the main context BH.
+     */
+    aio_bh_schedule_oneshot(qemu_get_aio_context(),
+                            virtio_resize_cb, vdev);
 }
 
 static const BlockDevOps virtio_block_ops = {
-- 
2.21.0

Am 16.09.2019 um 13:24 hat Sergio Lopez geschrieben:
> virtio_notify_config() needs to acquire the global mutex, which isn't
> allowed from an iothread, and may lead to a deadlock like this:
> 
>  - main thead
>   * Has acquired: qemu_global_mutex.
>   * Is trying the acquire: iothread AioContext lock via
>     AIO_WAIT_WHILE (after aio_poll).
> 
>  - iothread
>   * Has acquired: AioContext lock.
>   * Is trying to acquire: qemu_global_mutex (via
>     virtio_notify_config->prepare_mmio_access).
> 
> If virtio_blk_resize() is called from an iothread, schedule
> virtio_notify_config() to be run in the main context BH.
> 
> Signed-off-by: Sergio Lopez <slp@redhat.com>
> ---
> Changelog
> 
> v3:
>  - Unconditionally schedule the work to be done in the main context BH
>    (thanks John Snow and Kevin Wolf).
> 
> v2:
>  - Use aio_bh_schedule_oneshot instead of scheduling a coroutine
>    (thanks Kevin Wolf).
>  - Switch from RFC to v2 patch.
> ---
>  hw/block/virtio-blk.c | 17 ++++++++++++++++-
>  1 file changed, 16 insertions(+), 1 deletion(-)
> 
> diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
> index 18851601cb..0163285f6f 100644
> --- a/hw/block/virtio-blk.c
> +++ b/hw/block/virtio-blk.c
> @@ -16,6 +16,7 @@
>  #include "qemu/iov.h"
>  #include "qemu/module.h"
>  #include "qemu/error-report.h"
> +#include "qemu/main-loop.h"
>  #include "trace.h"
>  #include "hw/block/block.h"
>  #include "hw/qdev-properties.h"
> @@ -1086,11 +1087,25 @@ static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
>      return 0;
>  }
>  
> +static void virtio_resize_cb(void *opaque)
> +{
> +    VirtIODevice *vdev = opaque;
> +
> +    assert(qemu_get_current_aio_context() == qemu_get_aio_context());
> +    virtio_notify_config(vdev);
> +}
> +
>  static void virtio_blk_resize(void *opaque)
>  {
>      VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
>  
> -    virtio_notify_config(vdev);
> +    /*
> +     * virtio_notify_config() needs to acquire the global mutex,
> +     * so it can't be called from an iothread. Instead, schedule
> +     * it to be run in the main context BH.
> +     */
> +    aio_bh_schedule_oneshot(qemu_get_aio_context(),
> +                            virtio_resize_cb, vdev);

This fits on a single line. With this fixed:

Reviewed-by: Kevin Wolf <kwolf@redhat.com>

On Mon, Sep 16, 2019 at 01:24:12PM +0200, Sergio Lopez wrote:
> virtio_notify_config() needs to acquire the global mutex, which isn't
> allowed from an iothread, and may lead to a deadlock like this:
> 
>  - main thead
>   * Has acquired: qemu_global_mutex.
>   * Is trying the acquire: iothread AioContext lock via
>     AIO_WAIT_WHILE (after aio_poll).
> 
>  - iothread
>   * Has acquired: AioContext lock.
>   * Is trying to acquire: qemu_global_mutex (via
>     virtio_notify_config->prepare_mmio_access).
> 
> If virtio_blk_resize() is called from an iothread, schedule
> virtio_notify_config() to be run in the main context BH.
> 
> Signed-off-by: Sergio Lopez <slp@redhat.com>
> ---
> Changelog
> 
> v3:
>  - Unconditionally schedule the work to be done in the main context BH
>    (thanks John Snow and Kevin Wolf).
> 
> v2:
>  - Use aio_bh_schedule_oneshot instead of scheduling a coroutine
>    (thanks Kevin Wolf).
>  - Switch from RFC to v2 patch.
> ---
>  hw/block/virtio-blk.c | 17 ++++++++++++++++-
>  1 file changed, 16 insertions(+), 1 deletion(-)

Thanks, applied to my block tree:
https://github.com/stefanha/qemu/commits/block

Stefan