Re: Fix 80d20d35af1e ("nohz: Fix local_timer_softirq_pending()") may have revealed another problem

From: Heiner Kallweit
Date: Thu Jan 24 2019 - 14:54:30 EST


On 16.01.2019 07:24, Frederic Weisbecker wrote:
> On Fri, Dec 28, 2018 at 12:11:12AM +0100, Heiner Kallweit wrote:
>>
>> # tracer: nop
>> #
>> # _-----=> irqs-off
>> # / _----=> need-resched
>> # | / _---=> hardirq/softirq
>> # || / _--=> preempt-depth
>> # ||| / delay
>> # TASK-PID CPU# |||| TIMESTAMP FUNCTION
>> # | | | |||| | |
>> <idle>-0 [001] d.h2 1479.099092: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.099098: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.099106: softirq_raise: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.099114: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.099120: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.099121: softirq_entry: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.099134: softirq_exit: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.099135: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.099141: softirq_exit: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.100094: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.100109: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.100116: softirq_exit: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.101091: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.101113: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.101118: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.102094: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.102114: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.102121: softirq_exit: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.103091: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.103097: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.103105: softirq_raise: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.103114: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.103118: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.103119: softirq_entry: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.103131: softirq_exit: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.103132: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.103138: softirq_exit: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.105092: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.105115: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.105119: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.106092: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.106112: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] .Ns2 1479.106144: softirq_exit: vec=9 [action=RCU]
>> cpuhp/1-13 [001] d..2 1479.106279: timer_cancel: timer=0000000009a25653
>> <idle>-0 [001] d.h2 1479.106965: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.106969: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.106974: softirq_raise: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.106981: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.106984: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.106985: softirq_entry: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.106994: softirq_exit: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.106995: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.106999: softirq_exit: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.107996: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.108010: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.108014: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.109009: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.109013: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.109024: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.109028: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.109028: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.109033: softirq_exit: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.110013: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.110033: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] ..s2 1479.110040: softirq_exit: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.111011: softirq_raise: vec=1 [action=TIMER]
>> <idle>-0 [001] d.h2 1479.111017: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] d.h2 1479.111026: softirq_raise: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.111035: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.111040: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] ..s2 1479.111040: softirq_entry: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.111052: softirq_exit: vec=7 [action=SCHED]
>> <idle>-0 [001] ..s2 1479.111052: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] .Ns2 1479.111079: softirq_exit: vec=9 [action=RCU]
>> cpuhp/1-13 [001] dNh2 1479.112930: softirq_raise: vec=1 [action=TIMER]
>> cpuhp/1-13 [001] dNh2 1479.112935: softirq_raise: vec=9 [action=RCU]
>> <idle>-0 [001] d..1 1479.113077: <stack trace>
>> => can_stop_idle_tick.isra.14
>> => tick_nohz_get_sleep_length
>> => menu_select
>> => cpuidle_select
>> => do_idle
>> => cpu_startup_entry
>> => start_secondary
>> => secondary_startup_64
>> <idle>-0 [001] .Ns2 1479.113110: softirq_entry: vec=1 [action=TIMER]
>> <idle>-0 [001] .Ns2 1479.113114: softirq_exit: vec=1 [action=TIMER]
>> <idle>-0 [001] .Ns2 1479.113115: softirq_entry: vec=9 [action=RCU]
>> <idle>-0 [001] .Ns2 1479.113139: softirq_exit: vec=9 [action=RCU]
>
>
> So, I'm being very doubtful now actually.
>
> First of all, I realize my patch only fixes things halfway. Only softirq raised from
> hardirqs are going to be handled after ksoftirqd is parked. If it's raised outside
> interrupts, the softirq is going to be ignored. If other smpboot kthreads, following
> ksoftirqd parking, raise softirqs from their park() callback we should at least warn
> that we are ignoring that.
>
> Also, I'm still not sure if you're running threaded IRQs. Very likely not as
> the softirqs in your traces are executed in idle, which means from an IRQ interrupting
> idle. You haven't passed "threadirqs" in your kernel parameters, right?
> In any case we need to figure out why your softirqs raised at 1479.112930 and
> 1479.112935 don't get handled.
>
> Finally I don't understand why you have so many timer softirqs and yet no timer traces
> at all, I turned them on in the script I gave you. So something weird is happening with
> the timers code. At best we are wasting lots of CPU cycles with those empty softirqs, at
> worst we are miscalculating some expiration somewhere, possibly ignoring a timer.
>
> So I'll need to cook up an updated debug patch for you to trigger if you don't
> mind.
>
Hi Frederic,
is the update debug patch still to come?

> Thanks.
>
Heiner