Re: [CRASH] kunit failures in platform-device-devm

[Frederic Weisbecker]

Le Tue, Mar 03, 2026 at 08:45:42AM +0800, David Gow a écrit :
> Le 02/03/2026 à 11:32 PM, Alice Ryhl a écrit :
> > On Sun, Mar 01, 2026 at 02:06:49PM +0000, Alice Ryhl wrote:
> > > Hi DRIVER CORE, KOBJECTS, DEBUGFS AND SYSFS,
> > > 
> > > I've experienced a fair number of different crashes when running kunit
> > > locally. I just ran this command on v7.0-rc1 or upstream/master:
> > > 
> > > ./tools/testing/kunit/kunit.py run --make_options LLVM=1 --arch x86_64 --kconfig_add CONFIG_RUST=y --kconfig_add CONFIG_PCI=y
> > 
> > Bisection has been carried out!
> > 
> > The culprit is commit 5564c12385b7 ("kthread: Include unbound kthreads
> > in the managed affinity list"). Cc'ing everyone from this commit.
> > 
> > On this commit, kunit (executed using above command) crashes for me 70%
> > of the time (14 out of 20 runs) with kernel crashes (Oops) and
> > refcount_t use-after-free warnings in pm_runtime_test_cases and
> > platform-device-devm. On the parent commit, I was able to run kunit 24
> > times in a row with zero kunit failures.
> > 
> > The same crashes occur on tag v7.0-rc2, so this issue is not fixed.
> > 
> 
> Hmm... I've not been able to reproduce any crashes with stacktraces which
> look similar to yours, but given the bisection, could this be the kthread
> exit use-after-free?
> 
> Could you try the patch here, and see if that fixes it:
> https://lore.kernel.org/all/20260226-ungeziefer-erzfeind-13425179c7b2@brauner/

Urgh, I missed that one. Thanks a lot! Can someone bounce it to me for review as
it's not on LKML? Namely it would be nice to remove kthread_exit() on the
callsites (there aren't that many) and call do_exit() directly?

Thanks.

-- 
Frederic Weisbecker
SUSE Labs