On 04/07/2017 01:54 AM, Fam Zheng wrote:
> Coroutine in block layer should always be waken up in bs->aio_context

s/waken up/awakened/

> rather than the "current" context where it is entered. They differ when
> the main loop is doing QMP tasks.
> 
> Race conditions happen without this patch, because the wrong context is
> acquired in co_schedule_bh_cb, while the entered coroutine works on a
> different one:
> 
>   main loop                                iothread
> -----------------------------------------------------------------------
>   blockdev_snapshot
>     aio_context_acquire(bs->ctx)
>     bdrv_flush(bs)
>       bdrv_co_flush(bs)
>         ...
>         qemu_coroutine_yield(co)
>       BDRV_POLL_WHILE()
>         aio_context_release(bs->ctx)
>                                             aio_context_acquire(bs->ctx)
>                                               ...
>                                                 aio_co_wake(co)
>         aio_poll(qemu_aio_context)              ...
>           co_schedule_bh_cb()                   ...
>             qemu_coroutine_enter(co)            ...
>               /* (A) bdrv_co_flush(bs)              /* (B) I/O on bs */
>                       continues... */
>                                             aio_context_release(bs->ctx)
> 
> Both (A) and (B) can access resources protected by bs->ctx, but (A) is
> not thread-safe.
> 
> Make the block layer explicitly specify a desired context for the
> entered coroutine. For the rest callers, stick to the old behavior,

s/rest/remaining/

> qemu_get_aio_context() or qemu_get_current_aio_context().
> 
> Signed-off-by: Fam Zheng <famz@redhat.com>
> ---

At this point, I'm still more comfortable waiting for Paolo's review.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org