Re: 4.2-rc5 rcu stalls.

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Aug 03, 2015 at 05:55:35PM -0400, Dave Jones wrote:
>  Content analysis details:   (-2.9 points, 5.0 required)
> 
>   pts rule name              description
>  ---- ---------------------- --------------------------------------------------
>  -1.0 ALL_TRUSTED            Passed through trusted hosts only via SMTP
>  -1.9 BAYES_00               BODY: Bayes spam probability is 0 to 1%
>                              [score: 0.0000]
> X-Authenticated-User: davej@codemonkey.org.uk
> X-ZLA-Header: unknown; 0
> X-ZLA-DetailInfo: BA=6.00003574; NDR=6.00000001; ZLA=6.00000002; ZF=6.00000004; ZB=6.00041153; ZH=6.00102541; ZP=6.00079473; ZU=6.00000001; UDB=6.00244665; UTC=2015-08-03 21:55:44
> x-cbid: 15080321-0049-0000-0000-0000023893EC
> X-IBM-ISS-SpamDetectors: Score=0.399202; BY=0; FL=0; FP=0; FZ=0; HX=0; KW=0;
>  PH=0; RB=0; SC=0.399202; ST=0; TS=0; UL=0; ISC=
> X-IBM-ISS-DetailInfo:  BY=3.00004248; HX=3.00000236; KW=3.00000007;
>  PH=3.00000003; SC=3.00000115; SDB=6.00568916; UDB=6.00244665; UTC=2015-08-03
>  21:55:45
> X-TM-AS-MML: disable
> 
> On Mon, Aug 03, 2015 at 02:37:23PM -0700, Paul E. McKenney wrote:
>  > On Mon, Aug 03, 2015 at 05:08:35PM -0400, Dave Jones wrote:
>  > > [ 2120.855128] 	Tasks blocked on level-0 rcu_node (CPUs 0-3): P1497
>  > > [ 2120.855263] 	(detected by 2, t=65002 jiffies, g=78835, c=78834, q=0)
>  > > [ 2120.855403] trinity-watchdo R  running task    14336  1497   1496 0x00080000
>  > > [ 2120.855563]  ffff8804b94e3c88 ffffffffa17fa0b0 ffff8805010a1b40 ffff8804f58b51c0
>  > > [ 2120.855728]  ffff8805010a1b40 ffff8804b94e3c78 0000000000000000 ffff8804b94e4000
>  > > [ 2120.855893]  0000000000000001 0000000000000001 0000000000000002 ffff8804b94e3ca8
>  > > [ 2120.856062] Call Trace:
>  > > [ 2120.856116]  [<ffffffffa17fa0b0>] ? preempt_schedule_irq+0x40/0xa0
>  > > [ 2120.856252]  [<ffffffffa17fa0b6>] preempt_schedule_irq+0x46/0xa0
>  > 
>  > Sasha Levin (CCed) was having roughly similar stalls, and found that
>  > reverting b30f0e3ffedf (sched/preempt: Optimize preemption operations
>  > on __schedule() callers) made the stalls go away.  Does that help in
>  > your case?
> 
> Ugh, that doesn't revert cleanly.  Got something handy ?

I do not, but perhaps either Sasha or Frederic do.

> The curious thing is why I've only just started seeing these.
> I've had that machine for a month now, so it's been fuzzing since ~4.1,
> and until yesterday I'd not seen this at all.

It could well be a new problem.  I just saw the preempt_schedule_irq()
and was reminded of the problem that Sasha found.  ;-)

							Thanx, Paul