Re: [Qemu-devel] qemu/qemu-kvm floppy regression brought by 212ec7baa28cc9d819234fed1541fc1423cfe3d8

[Gleb Natapov <gleb@redhat.com>]

On Fri, Oct 28, 2011 at 09:46:33AM -0200, Lucas Meneghel Rodrigues wrote:
> On 10/27/2011 11:17 PM, Zhi Yong Wu wrote:
> >On Fri, Oct 28, 2011 at 2:57 AM, Lucas Meneghel Rodrigues
> ><lmr@redhat.com>  wrote:
> >>On 10/27/2011 05:17 AM, Stefan Hajnoczi wrote:
> >>>
> >>>On Wed, Oct 26, 2011 at 03:19:17PM -0200, Lucas Meneghel Rodrigues wrote:
> >>>>
> >>>>On 10/26/2011 01:47 PM, Kevin Wolf wrote:
> >>>>>
> >>>>>Am 26.10.2011 16:41, schrieb Lucas Meneghel Rodrigues:
> >>>>>>
> >>>>>>Hi folks:
> >>>>>>
> >>>>>>We've captured a regression with floppy disk on recent qemu (and
> >>>>>>qemu-kvm, after a code merge). We bisected it to be caused by:
> >>>>>>
> >>>>>>commit 212ec7baa28cc9d819234fed1541fc1423cfe3d8
> >>>>>>Author: Richard Henderson<rth@twiddle.net>
> >>>>>>Date:   Mon Aug 15 15:08:45 2011 -0700
> >>>>>>
> >>>>>>      fdc: Convert to isa_register_portio_list
> >>>>>>
> >>>>>>      Signed-off-by: Richard Henderson<rth@twiddle.net>
> >>>>>>      Signed-off-by: Avi Kivity<avi@redhat.com>
> >>>>>>
> >>>>>>Since this commit, the guest doesn't see a floppy disk attached to it
> >>>>>>anymore, blocking kvm autotest ability to install windows guests
> >>>>>>automatically. This is a big deal for kvm autotest (ruins our automated
> >>>>>>regression jobs), so please take a look at it.
> >>>>>
> >>>>>Can you please try again with the latest block branch? I think there is
> >>>>>a patch queued that will fix it.
> >>>>
> >>>>Kevin, I did try with HEAD of your repo:
> >>>>
> >>>>git://repo.or.cz/qemu/kevin.git
> >>>>
> >>>>[lmr@freedom qemu-kwolf]$ git branch -r
> >>>>   origin/HEAD ->    origin/master
> >>>>   origin/blkqueue
> >>>>   origin/blkqueue-v1
> >>>>   origin/block
> >>>>   origin/coroutine
> >>>>   origin/coroutine-block
> >>>>   origin/coroutine-devel
> >>>>   origin/devel
> >>>>   origin/ehci
> >>>>   origin/for-anthony
> >>>>   origin/for-stable-0.14
> >>>>   origin/inplace-conversion
> >>>>   origin/master
> >>>>
> >>>>With this repo, master branch, the problem persists. With the block
> >>>>branch, the problem persists.
> >>>>
> >>>>Now, with the blkqueue branch the problem is resolved. Cleber had
> >>>>the same results booting a FreeDOS floppy. So the fix is indeed in
> >>>>blkqueue.
> >>>>
> >>>>Oh, you might want to check the blkqueue branch, it does have quite
> >>>>a bunch of set but unused variables, which will cause compilation
> >>>>errors unless --disable-werror is passed to the configure script.
> >>>
> >>>I think blkqueue is an older development branch of the "block queue"
> >>>feature that Kevin was working on.  It is not Kevin's block tree (see
> >>>his "block" branch).
> >>
> >>So no, the block branch does not resolve the floppy access problem.
> >>
> >>Well, considering the tests of the stable set I'm running against qemu right
> >>now, this is not the biggest of our problems... I'm verifying qemu is
> >>segfaulting on nearly every prolonged attempt of doing migration... I'm
> >>about to write an email about it.
> >What is the OS of your guest? fedora16 or RHEL6? I would like recently
> >to use floppy device in guest.
> 
> Ok, correcting my answer, it was Windows 7 SP1. For some reason, I
> read 'host' instead of guest. I had a long day when I wrote the 1st
> reply.
> 
> It's been a while that we've verified that the linux floppy driver
> is quite unstable. In general linux + floppy devices on a guest
> tends to crash the guest kernel, and that's why in kvm autotest we
> moved from floppy devices to cdroms to hold the kickstart file. The
> fact you managed to get things working under F14 means you are
> lucky, and that particular floppy driver bug does not happen under
> F14's kernel. We never had such a kernel crash problem with any of
> the Windows's kernels.
> 
> So if I were you, I would *not* use a floppy with a linux guest.
> Well, unless you want to debug the floppy driver bug and fix this on
> upstream linux once for all, so future versions of linux won't have
> this problem.
AFAIK the problem exists only on smp guest.

--
			Gleb.