On Thu, Apr 3, 2014 at 3:26 AM, LukÃÅ Czerner <lczerner@xxxxxxxxxx> wrote:
> On Thu, 3 Apr 2014, Dave Chinner wrote:
>> Date: Thu, 3 Apr 2014 12:14:11 +1100
>> From: Dave Chinner <david@xxxxxxxxxxxxx>
>> To: Theodore Ts'o <tytso@xxxxxxx>
>> Cc: Sedat Dilek <sedat.dilek@xxxxxxxxx>, lsf@xxxxxxxxxxxxxxxxxxxxxxxxxx,
>> linux-ext4@xxxxxxxxxxxxxxx, linux-fsdevel@xxxxxxxxxxxxxxx,
>> Subject: Re: [Lsf] [PATCH] xfstests-bld: Simplify determination of number of
>> CPUs in build-all
>> On Wed, Apr 02, 2014 at 10:26:20AM -0400, Theodore Ts'o wrote:
>> > On Wed, Apr 02, 2014 at 09:28:23AM +1100, Dave Chinner wrote:
>> > > > And of course, whether changes in the mainline kernel tree are
>> > > > manually propagated changes from the xfstests.git tree, or whether
>> > > > primary development happens in the kernel tree, is ultimately going to
>> > > > be up to you and the XFS developers who have stewardship of xfstests.
>> > > > I'm not sure I would be that excited about manual propagation of
>> > > > changes from one git tree to another, but that is of course, up to
>> > > > you.
>> > >
>> > > And this is exactly my point, Ted. Again, you are presuming that the
>> > > implementation is going to require syncing commits across disparate
>> > > git trees and other such games will be needed to maintain separate
>> > > packages. Nothing could be further from the truth: we already have
>> > > this problem with the shared XFS kernel/userspace code and it's a
>> > > royal PITA keeping them in sync. Hence introducing the same
>> > > maintenance problem with new code and infrastructure is highly
>> > > undesirable and something we'll try to avoid at all costs.
>> > Actually, I was presuming that the thing that makes the most sense was
>> > to move all or most of the tests in xfstests into the kernel tests
>> > tree. And then you complained that I was making a presumption that
>> > this was the only sane thing to do. That's why I said, "if you want
>> > to do something insane, be my guest".
>> So, you think there's only one "sane" solution. That's a massive
>> assumption, and that's what I'm pointing out.
>> It's perfectly sane to treat the kernel tree as just another
>> downstream consumer of xfstests, and simple add infrastructure to
>> the kernel tree to source and build xfstests from the upstream repo.
> That's what I think is the best approach as well. We'll still have
> separate tree for the xfstests which works for everyone and
> infrastructure in kernel tree which will simply use it.
> Looking at rcutorture tests in kernel tree, they are indeed using
> qemu and it looks like that they are running those tests directly
> from init script - we really need an image to boot from.
> We could use virtme
virtme will eventually be able to use a separate OS image, probably in
the form of a directory with appropriate xattrs set. I could support
images on a block device too, but that's boring :)
> which seems like a best option. Alternatively we could allow users
> passing their own image as well.
> Then we would need init script which would download and build
> xfstests, xfsprogs and possible other dependencies (such as fio)
> and finally run xfstests itself. Colnsole output from the qemu can
> be redirected to the file in the host, so that's ok, but we still
> need a way to retrieve 'retults' directory from the test.
> We'd also need a devices to test on (at least TEST_DEV and
> SCRATCH_DEV). Those could be files a of set size exported to qemu as
> block devices.
I just pushed a pair of sample scripts that invokes xfstests inside
virtme. They're a total of 42 lines, including comments and
instructions. You can now do:
PATH=.:$PATH samples/xfstests test.img scratch.img ~/apps/xfstests/
from inside a virtme checkout. You'll have to compile xfstests first, though.
They will be considerably more useful once I add read-write host
windows to virtme.
If anyone wants to use virtme for something other than
experimentation, let me know and we can talk about interface
Some xfstests thoughts:
- There's an undocumented way to write results outside the source
tree called RESULT_BASE. It would be great if it were documented and
- xfstests does not appreciate using symlinks to device nodes.
- SCRATCH_MNT needs to be in /etc/fstab. I think that this should be
changed or documented. If the latter, then SCRATCH_DEV seems
Some virtme thoughts:
- --script-sh is awkward. Suggestions for a better interface are welcome.
- It would be great if the virtconsole driver were less buggy. Even
the virtualized ttyS0 is slow enough to account for a significant
fraction of boot time. It would also be great if the connection
between virtio ports and hvc devices were exposed in sysfs and through
- earlyprintk=virtconsole would be awesome. This might need help
from qemu, though. Or maybe there's a way to get the debug port to
work for that.
- I should probably add simple way to use forcibly cached disks.