Re: [lkp-robot] [locking/ww_mutex] 857811a371: INFO:task_blocked_for_more_than#seconds

[Fengguang Wu]

On Wed, Mar 08, 2017 at 12:13:12PM +0000, Chris Wilson wrote:
> On Wed, Mar 08, 2017 at 09:08:54AM +0800, kernel test robot wrote:
> > FYI, we noticed the following commit:
> >
> > commit: 857811a37129f5d2ba162d7be3986eff44724014 ("locking/ww_mutex: Adjust the lock number for stress test")
> > https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
> >
> > in testcase: boot
> >
> > on test machine: qemu-system-i386 -enable-kvm -m 320M
> >
> > caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):
>
> Now the test is running, it takes too long. :)

Sorry that's right. Up to now the 0day robot still cannot guarantee
the timely reporting of a runtime regression, nor can it guarantee
bisecting of a new regression even when some test actually triggered
the bug.

One fundamental challenge is, there are ~50,000 runtime "regressions"
queued for bisect. Obviously there is no way to bisect them all. So a
large portion of real regressions never get a chance to be bisected.
Not to mention the problem of bisect reliability and efficiency.

Most of the test "regressions" may be duplicates to each other (eg. a
bug in mainline kernel will also show up in various developer trees).
A great portion of them may also be random noises (eg. performance
fluctuations). We've tried various approaches to improve the
de-duplicate, filtering, prioritize etc. algorithms. Together with
increased test coverage, they have been reflected in our slowly
increasing report numbers. However there is still a long way to go.

Thanks,
Fengguang