Re: [Qemu-devel] [Qemu-stable] [PATCH] stream: fix the deadlock bug when stream finish

[Michael Roth]

Quoting Liu Yu (2014-08-22 22:43:00)
> From: Liu Yu <address@hidden>
> 
> The patch against branch stable-2.0

Hi Liu,

2.0.2 was the last planned released for stable-2.0. Are you still seeing this
issue with 2.1.0 or master? 2.1.1 is release is planned for September 9th so we
can work on getting this included there if it is.

Also, please Cc: address@hidden even for stable patches.

> 
> In case VM does IO while we run a stream job.
> When stream finishes, the stream coroutine drains all IOs before
> close the unused image, in bdrv_drain_all() it may find
> a pending request which is submitted by guest IO coroutine.
> In order to wait the pending req finish, the subsequent aio_poll()
> call poll() to wait the req. however, if the req is already done by
> threadpool and is waiting for the callback, there is no chance to switch
> back to guest IO coroutine to call the callback and so that the stream
> coroutine waits in poll() all the time.
> 
> The patch detects the deadlock case above and switch back to iothread
> coroutine to handle the callback, and work on the stream coroutine
> after the pending req get finished.
> 
> Signed-off-by: Liu Yu <address@hidden>
> ---
> the issue can be reproduced by
> 1. guest does fio test
> 2. while host runs virsh blockpull repeatedly
> 
> 
>  block.c |   27 ++++++++++++++++++++++++++-
>  1 files changed, 26 insertions(+), 1 deletions(-)
> 
> diff --git a/block.c b/block.c
> index 990a754..f8c1a8d 100644
> --- a/block.c
> +++ b/block.c
> @@ -1778,6 +1778,29 @@ static bool bdrv_requests_pending_all(void)
>      return false;
>  }
> 
> +static bool bdrv_request_coroutine_wait(void)
> +{
> +    BlockDriverState *bs;
> +    Coroutine *co;
> +
> +    if (!qemu_in_coroutine())
> +        return false;
> +
> +    co = qemu_coroutine_self();
> +    QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
> +        if (!QLIST_EMPTY(&bs->tracked_requests)) {
> +            BdrvTrackedRequest *req = QLIST_FIRST(&bs->tracked_requests);
> +
> +            if(req->co == co)
> +                continue;
> +
> +            qemu_co_queue_wait(&req->wait_queue);
> +            return true;
> +        }
> +    }
> +    return false;
> +}
> +
>  /*
>   * Wait for pending requests to complete across all BlockDriverStates
>   *
> @@ -1800,8 +1823,10 @@ void bdrv_drain_all(void)
>          QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
>              bdrv_start_throttled_reqs(bs);
>          }
> -
> +recheck:
>          busy = bdrv_requests_pending_all();
> +        if (busy && bdrv_request_coroutine_wait())
> +            goto recheck;
>          busy |= aio_poll(qemu_get_aio_context(), busy);
>      }
>  }
> -- 
> 1.7.1