Re: [Qemu-devel] dataplane: latest rc with selected patches hangs very often

[Andrey Korolyov]

On Fri, Jul 11, 2014 at 11:01 PM, Andrey Korolyov <address@hidden> wrote:
> Hello,
>
> is it worthy to re-test dp until next tag point? I am checking aio
> with -rc1 and following picked commits (sorry for patchwork
> references):
>
> stefanha:
> 368191 New          [v3,1/4] virtio-blk: avoid dataplane
> VirtIOBlockReq early free
> 368787 New          [v3,2/4] dataplane: do not free VirtQueueElement
> in vring_push()
> 368360 New          [v3,3/4] virtio-blk: avoid g_slice_new0() for
> VirtIOBlockReq and VirtQueueElement
> 368863 New          [v3,4/4] virtio-blk: embed VirtQueueElement in
> VirtIOBlockReq
>
> pbonzini:
> 367551 New          [1/4] block: prefer aio_poll to qemu_aio_wait
> 367553 New          [2/4] block: drop aio functions that operate on
> the main AioContext
> 367552 New          [3/4] test-aio: fix GSource-based timer test
> 367554 New          [4/4] AioContext: speed up aio_notify
>
> Emulator itself stop responding in approximately ninety percent of
> cases, e.g. it does not react to anything except SIGKILL. Windows VM
> is able to read some bits before lockup happens because it draws an
> initial logo.
>
> hangs on vl.c:
> (gdb) list
> 2890        g_free(dummy);
> 2891        if (err) {
> 2892            qerror_report_err(err);
> 2893            return -1;
> 2894        }
> 2895        return 0;
> 2896    }
> 2897
> 2898    int main(int argc, char **argv, char **envp)
> 2899    {
> (gdb) next
> Single stepping until exit from function __lll_lock_wait,
> which has no line number information.
>
> cmdline follows:
> if=none,id=drive-virtio-disk0,format=raw,cache=writeback,aio=native
> ... -object iothread,id=blk0 -set device.virtio-disk0.config-wce=off
> -set device.virtio-disk0.scsi=off -set
> device.virtio-disk0.iothread=blk0

Forgot to attach backtrace, though the issue and bt itself can be
easily reproduced.

aio-dp-bt.txt
Description: Text document