On Thu, Nov 23, 2017 at 08:47:47AM +0800, Fam Zheng wrote: > On Wed, 11/22 04:55, Jeff Cody wrote: > > On Wed, Nov 22, 2017 at 09:09:02AM +0100, Christian Borntraeger wrote: > > > > > > > > > On 11/22/2017 04:23 AM, Michael Roth wrote: > > > > Quoting Christian Borntraeger (2017-11-21 15:38:32) > > > >> forgot to cc qemu-devel.... > > > >> > > > >> On 11/21/2017 10:37 PM, Christian Borntraeger wrote: > > > >>> a quick heads up . Rc2 now triggers > > > >>> +qemu-img: block/block-backend.c:2088: blk_root_drained_end: > > > >>> Assertion `blk->quiesce_counter' failed. > > > >>> for several qemu iotests. > > > >>> > > > >>> I have not looked into any details. > > > > > > > > It looks to be due to: > > > > > > > > 4afeffc8572f40d8844b946a30c00b10da4442b1 > > > > blockjob: do not allow coroutine double entry or entry-after-completion > > > > > > Yes, I can confirm that reverting this patch gets rid of this assertion, > > > but > > > I see things like > > > > > > --- /home/cborntra/REPOS/qemu/tests/qemu-iotests/020.out 2017-11-21 > > > 20:19:34.785519323 +0100 > > > +++ /home/cborntra/REPOS/qemu/build/tests/qemu-iotests/020.out.bad > > > 2017-11-22 09:04:50.127612500 +0100 > > > @@ -537,7 +537,8 @@ > > > wrote 65536/65536 bytes at offset 4295098368 > > > 64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) > > > No errors were found on the image. > > > -Image committed. > > > +qemu_aio_coroutine_enter: Co-routine was already scheduled in > > > 'co_aio_sleep_ns' > > > +./common.rc: line 61: 88002 Aborted (core dumped) ( exec > > > "$QEMU_IMG_PROG" $QEMU_IMG_OPTIONS "$@" ) > > > > > > > That is from the subsequent patches in the series - you will want to revert > > the whole series to test, as the introduced aborts catch the illegal > > entries that the reverted patch sidestepped. > > > > The series patches are: > > > > 4afeffc > > 6133b39 > > a233969 > > d975301 > > > > Of course, these new aborts prevent improper behavior, so we may want to > > figure out why this is getting hit. > > > > Unfortunately, I am traveling at the moment (waiting to board my flight), so > > will have limited connectivity. > > I'll take a look at this today and the bottom line is we revert the series > until > a proper fix is found. >
My hunch is the series is a proper fix, but uncovered other latent bugs that were relying on dangerous behavior.