Re: Negative values in /proc/latency_stats

[Andrew Morton <akpm@linux-foundation.org>]

On Sat, 31 Jan 2009 14:42:20 +0100 Corrado Zoccolo <czoccolo@gmail.com> wrote:

> Hello,
> I found negative numbers sometimes appear in /proc/latency_stats
> (vanilla kernel 2.6.29-rc3, on x86 cpu, configuration attached)
> 
> [corrado@localhost ~]$ while sleep 1; do grep -- - /proc/latency_stats
> >> neg_stats; done
> ^Z
> [1]+  Stopped                 sleep 1
> [corrado@localhost ~]$ cat neg_stats
> 1 -486373534 -486373534 sys_rt_sigsuspend sysenter_do_call
> 1 -486373534 -486373534 sys_rt_sigsuspend sysenter_do_call
> 1 -486373534 -486373534 sys_rt_sigsuspend sysenter_do_call
> 
> 
> I suspect this can be the cause for
> https://bugs.launchpad.net/ubuntu/+source/latencytop/+bug/297776 , as
> I saw it happening on my machine during a kernel recompilation.
> 

<discovers kernel/latencytop.c>

- It implements a kernel/userspace interface yet it has zero documentation.

- It was committed with a 2-line changelog which tells us practically
  nothing.

- It implements an up-to-1536-loops loop followed by an
  up-to-384-loops loop on a scheduler hotpath.

  All under spin_lock_irqsave()!

- store_stacktrace() unnecessarily initalises trace.skip.

- account_scheduler_latency() should be an inline:

	if (unlikely(latencytop_enabled))
		__account_scheduler_latency(...);

- ditto clear_all_latency_tracing()

- it's schizophrenic in its placement of spaces around semicolons in
  `for' statements.

- it seems to only be implemented if CONFIG_FAIR_GROUP_SCHED=y.

- lstats_fops should be const.


And it emits negative numbers too ;)