On Wed, May 05, 2021 at 11:03:12AM -0700, Paul E. McKenney wrote:
> On Wed, May 05, 2021 at 10:36:16PM +0800, kernel test robot wrote:
> >
> >
> > Greeting,
> >
> > FYI, we noticed the following commit (built with gcc-9):
> >
> > commit: 8e614d5b58992e722f07de7c2426f2c44668092b ("clocksource: Provide kernel module to test clocksource watchdog")
> > https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git master
> >
> >
> > in testcase: boot
> >
> > on test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 16G
> >
> > caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):
> >
> >
> > +-------------------------------------------------------------------------+------------+------------+
> > | | bdbd9c673e | 8e614d5b58 |
> > +-------------------------------------------------------------------------+------------+------------+
> > | WARNING:at_kernel/time/clocksource-wdtest.c:#wdtest_func.cold | 0 | 11 |
> > | RIP:wdtest_func.cold | 0 | 11 |
> > +-------------------------------------------------------------------------+------------+------------+
>
> Might it be useful to address the lockdep issues that preceded this splat?
>
> Leaving that aside, the system appears to still be booting. There are
> RCU CPU stall warning messages later on, and then the system hangs more
> than six minutes while still booting, presumably due to the large number
> of self-tests and debug options enabled.
>
> The intent is that the clocksource-wdtest tests run after boot has
> completed. One approach would be to test it using modprobe after boot
> has completed. In addition, the clocksource-wdtest module is not designed
> to handle CPU overload conditions, and making it do so would reduce the
> effectiveness of the test.
>
> I suggest setting clocksource-wdtest.holdoff=N, where "N" is in seconds
> and is large enough that boot has completed. Alternatively, use modprobe
> to activate this module from userspace after boot has completed.
>
> What I do is just set CONFIG_TEST_CLOCKSOURCE_WATCHDOG=y in an ordinary
> rcutorture run, if that helps.
All that aside, does the patch below help in your environment?
If so, I can adjust so that my testing gets done quickly and yours
avoids false-positive failures.
Thanx, Paul
------------------------------------------------------------------------
diff --git a/kernel/time/clocksource-wdtest.c b/kernel/time/clocksource-wdtest.c
index 01df12395c0e..0d8542f8b1d2 100644
--- a/kernel/time/clocksource-wdtest.c
+++ b/kernel/time/clocksource-wdtest.c
@@ -149,7 +149,7 @@ static int wdtest_func(void *arg)
s = ", expect clock skew";
pr_info("--- Watchdog with %dx error injection, %lu retries%s.\n", i, max_cswd_read_retries, s);
WRITE_ONCE(wdtest_ktime_read_ndelays, i);
- schedule_timeout_uninterruptible(2 * HZ);
+ schedule_timeout_uninterruptible(60 * HZ);
WARN_ON_ONCE(READ_ONCE(wdtest_ktime_read_ndelays));
WARN_ON_ONCE((i <= max_cswd_read_retries) !=
!(clocksource_wdtest_ktime.flags & CLOCK_SOURCE_UNSTABLE));
