Re: [Syzkaller & bisect] There is "task hung in synchronize_rcu bisect" in v6.3-rc2 kernel

[Pengfei Xu <pengfei.xu@intel.com>]

On 2023-04-13 at 16:30:17 +0800, Hillf Danton wrote:
> On Thu, 13 Apr 2023 16:03:51 +0800 Pengfei Xu <pengfei.xu@intel.com>
> > 
> >   I newly updated previous alder lake S reproduced syzkaller repro.prog and
> >   machineInfo0(vm info) info in log link:
> > https://github.com/xupengfe/syzkaller_logs/tree/main/230315_023443_synchronize_rcu
> > 
> > And tried debug kernel on alder lake S, there were 2 cpu in guest, and
> > CPU1 reported "interrupt took 331596 ns", then kernel should not trigger
> > "rcu_tasks_wait_gp" for this issue.
> > [  104.171591] memfd_create() without MFD_EXEC nor MFD_NOEXEC_SEAL, pid=334 'systemd'
> > [  126.770981] hrtimer CPU1: interrupt took 331596 ns      //  "rcu_tasks_wait_gp" after 15s
> > [  141.127282] rcu_tasks_wait_gp: rcu_tasks grace period number 9 (since boot) is 2571 jiffies old.
> > [  171.272100] rcu_tasks_wait_gp: rcu_tasks grace period number 9 (since boot) is 10107 jiffies old.
> > [  261.767189] rcu_tasks_wait_gp: rcu_tasks grace period number 9 (since boot) is 32731 jiffies old.
> > [  300.102911] INFO: task repro_rcu:398 blocked for more than 147 seconds.
> > [  300.103408]       Not tainted 6.3.0-rc6-dbg-dirty #1
> 
> Similar issue, and hrtimer trace helps in both cases.
I enabled "hrtimer_*" and tried again.
I didn't find some clue for this issue due to lack of experience for hrtimer
logs.
Anyway I put the issue dmesg and hrtimer in link:
https://github.com/xupengfe/syzkaller_logs/tree/main/230412_031722_synchronize_rcu/0413_debug_on_alder_lake-s

Thanks!
BR.