lkml.org 
[lkml]   [2011]   [Mar]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: excessive kworker activity when idle. (was Re: vma corruption in today's -git)
On Wed, Mar 30, 2011 at 08:37:21PM -0700, Linus Torvalds wrote:
> You don't see some nice thread description
> in 'top' any more (like you used to when everybody created their own
> threads and didn't do the common worker thread thing), and the best
> approach literally seems to be something like
>
> perf record -ag sleep 10
> perf report
>
> which does tend to show what's going on, but it's still a ridiculous
> way to this.

I tried that, and wasn't particularly enlightened.

+ 6.53% kworker/1:2 [kernel.kallsyms] [k] read_hpet
+ 4.83% kworker/0:0 [kernel.kallsyms] [k] read_hpet
+ 4.28% kworker/0:0 [kernel.kallsyms] [k] arch_local_irq_restore
+ 4.03% kworker/1:2 [kernel.kallsyms] [k] arch_local_irq_restore
+ 3.10% kworker/0:0 [kernel.kallsyms] [k] do_raw_spin_trylock
+ 2.88% kworker/1:2 [kernel.kallsyms] [k] do_raw_spin_trylock
+ 2.85% kworker/1:2 [kernel.kallsyms] [k] debug_locks_off
+ 2.69% kworker/0:0 [kernel.kallsyms] [k] debug_locks_off
+ 2.48% kworker/0:0 [kernel.kallsyms] [k] lock_release
+ 2.26% kworker/1:2 [kernel.kallsyms] [k] lock_release
+ 2.03% kworker/0:0 [kernel.kallsyms] [k] lock_acquire
+ 1.88% kworker/0:0 [kernel.kallsyms] [k] arch_local_save_flags
+ 1.87% kworker/1:2 [kernel.kallsyms] [k] lock_acquire
+ 1.82% kworker/1:2 [kernel.kallsyms] [k] arch_local_save_flags
+ 1.81% kworker/1:2 [kernel.kallsyms] [k] arch_local_irq_save
+ 1.78% kworker/0:0 [kernel.kallsyms] [k] arch_local_irq_save
+ 1.56% kworker/0:0 [kernel.kallsyms] [k] lock_acquired
+ 1.53% kworker/1:2 [kernel.kallsyms] [k] __lock_acquire
+ 1.51% kworker/0:0 [kernel.kallsyms] [k] __lock_acquire
+ 1.29% kworker/0:0 [kernel.kallsyms] [k] native_write_msr_safe
+ 1.23% kworker/1:2 [kernel.kallsyms] [k] cpu_relax
+ 1.17% kworker/1:2 [kernel.kallsyms] [k] lock_acquired
+ 1.17% kworker/0:0 [kernel.kallsyms] [k] trace_hardirqs_off_caller
+ 1.11% kworker/1:2 [kernel.kallsyms] [k] trace_hardirqs_off_caller
+ 1.08% kworker/1:2 [kernel.kallsyms] [k] native_write_msr_safe
+ 1.02% kworker/0:0 [kernel.kallsyms] [k] _raw_spin_lock_irqsave
+ 0.92% kworker/0:0 [kernel.kallsyms] [k] process_one_work
+ 0.87% kworker/1:2 [kernel.kallsyms] [k] _raw_spin_lock_irqsave
+ 0.80% kworker/0:0 [kernel.kallsyms] [k] flush_to_ldisc
+ 0.76% kworker/1:2 [kernel.kallsyms] [k] process_one_work
+ 0.76% kworker/1:2 [kernel.kallsyms] [k] flush_to_ldisc
+ 0.72% kworker/0:0 [kernel.kallsyms] [k] arch_local_irq_restore
+ 0.71% kworker/1:2 [kernel.kallsyms] [k] arch_local_irq_restore
+ 0.64% kworker/1:2 [kernel.kallsyms] [k] do_raw_spin_unlock
+ 0.63% kworker/0:0 [kernel.kallsyms] [k] perf_event_task_tick
+ 0.61% kworker/1:2 [kernel.kallsyms] [k] ktime_get
+ 0.59% kworker/0:0 [kernel.kallsyms] [k] _raw_spin_unlock_irqrestore

this is what led me to try the other perf methods. the kmem traces were
the only things that really jumped out.

> (Powertop can also do it, and is probably a better thing to use, I'm
> just used to "perf record" for other reasons, so..)

Tried that too, here's what it said..

Summary: 0.0 wakeups/second, 0.0 GPU ops/second and 0.0 VFS ops/sec

Usage Events/s Category Description
-2147483648 ms/s 0.0 Timer
-2147483648 ms/s 0.0 kWork
35151589 ms/s 0.0 Timer
35151588 ms/s 0.0 Timer
35151587 ms/s 0.0 Timer
35151586 ms/s 0.0 Timer
35151585 ms/s 0.0 Timer
35151585 ms/s 0.0 Timer
35151584 ms/s 0.0 Timer
35151583 ms/s 0.0 Timer
35151582 ms/s 0.0 Timer
35151581 ms/s 0.0 Timer
35151581 ms/s 0.0 Timer
35151580 ms/s 0.0 Timer

Not exactly helpful.

Dave



\
 
 \ /
  Last update: 2011-03-31 05:59    [W:0.103 / U:0.308 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site