On Thu, Feb 5, 2015 at 7:45 PM, Paul E. McKenney
<paulmck@linux.vnet.ibm.com> wrote:
> On Thu, Feb 05, 2015 at 10:35:33AM -0800, Dave Hansen wrote:
>> On 02/05/2015 10:34 AM, Paul E. McKenney wrote:
>> >> >                               Did I actually need to be
>> >> > onlining/offlining CPUs to hit the splat that Sedat was reporting?
>> > Yep, you do need to offline at least one CPU to hit that splat.
>>
>> Heh, do we need a debugging mode that will randomly offline/online CPUs? :)
>
> For that, kernel/rcu/rcutorture.c and kernel/locking/locktorture.c
> are your friends.  ;-)
>
> The problem is that I only run RCU-relevant combinations of Kconfigs,
> which means that I missed the ones that Sedat used to find this problem.
> So I guess it is a good thing that others run -next testing.
>

[ Revived by a voltaren resinat pill... ]

I reverted "x86/mm: Omit switch_mm() tracing for offline CPUs"
...and...
applied "tlb: Don't do trace_tlb_flush() on offline CPUs"
...in my build-dir.
( I did not build from scratch but re-invoking make "updated" the
files touched by Steven's patch, see attached build-log. )

Unfortunately, the call-trace remains when doing an offlining of cpu1.
( It's good to see it's reproducible. )

root# echo 0 > /sys/devices/system/cpu/cpu1/online

[  121.652796] intel_pstate CPU 1 exiting
[  121.666272]
[  121.666274] ===============================
[  121.666274] [ INFO: suspicious RCU usage. ]
[  121.666277] 3.19.0-rc7-next-20150204.7-iniza-small #4 Not tainted
[  121.666278] -------------------------------
[  121.666280] include/trace/events/tlb.h:37 suspicious
rcu_dereference_check() usage!
[  121.666281]
[  121.666281] other info that might help us debug this:
[  121.666281]
[  121.666282]
[  121.666282] RCU used illegally from offline CPU!
[  121.666282] rcu_scheduler_active = 1, debug_locks = 0
[  121.666283] no locks held by swapper/1/0.
[  121.666284]
[  121.666284] stack backtrace:
[  121.666287] CPU: 1 PID: 0 Comm: swapper/1 Not tainted
3.19.0-rc7-next-20150204.7-iniza-small #4
[  121.666288] Hardware name: SAMSUNG ELECTRONICS CO., LTD.
530U3BI/530U4BI/530U4BH/530U3BI/530U4BI/530U4BH, BIOS 13XK 03/28/2013
[  121.666293]  0000000000000001 ffff88011a44fe18 ffffffff817e39cd
0000000000000011
[  121.666296]  ffff88011a448290 ffff88011a44fe48 ffffffff810d6af7
ffff8800d3dfaac0
[  121.666299]  0000000000000001 ffffffff81d32ce0 0000000000000005
ffff88011a44fe78
[  121.666300] Call Trace:
[  121.666308]  [<ffffffff817e39cd>] dump_stack+0x4c/0x65
[  121.666313]  [<ffffffff810d6af7>] lockdep_rcu_suspicious+0xe7/0x120
[  121.666318]  [<ffffffff810b73f9>] idle_task_exit+0x1c9/0x260
[  121.666322]  [<ffffffff81054c4e>] play_dead_common+0xe/0x50
[  121.666325]  [<ffffffff81054ca5>] native_play_dead+0x15/0x140
[  121.666330]  [<ffffffff8102963f>] arch_cpu_idle_dead+0xf/0x20
[  121.666333]  [<ffffffff810cdb4e>] cpu_startup_entry+0x37e/0x580
[  121.666336]  [<ffffffff81053e20>] start_secondary+0x140/0x150
[  121.666744] smpboot: CPU 1 is now offline

>From rcu point this is now safe?
But another area (linux-pm?) is still affected?
I will try to test "vanilla" pm-next if the problem exists with
intel_pstate as suggested by Rafael.
Hmmm, not sure how I can get the pm-next code which went into
next-20150204 as linux-pm.git#linux-next was feeded with new stuff.


- Sedat -