All of lore.kernel.org
 help / color / mirror / Atom feed
* poison_obj latency
@ 2007-06-27 10:51 Manfred Gruber
  2007-06-27 12:34 ` Ingo Molnar
  0 siblings, 1 reply; 3+ messages in thread
From: Manfred Gruber @ 2007-06-27 10:51 UTC (permalink / raw)
  To: linux-rt-users

hi !

I use a arm4vt ep93xx processor with 2.6.21.5-rt18. 
When i run cyclictest with -i 10000 and -n -p 90 -b 1000 i see some strange 
latencies in poison_obj.

the output from /proc/latency_trace where the latency is shown:
OLD:cyclicte-884   0.... 13493us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 13830us+:   336us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 13848us+:    18us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 47181us!:     4us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 47514us+:   333us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 47533us+:    18us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 90226us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 90564us+:   338us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 90584us+:    19us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 125182us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 125520us+:   338us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 125540us+:    19us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 157308us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 157644us+:   335us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 157661us+:    17us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0D..1 193339us!:     5us: 
trace_stop_sched_switched+0x14/0x374 (__schedule+0x478/0x5a4)
ACT:cyclicte-884   0.... 193662us+:   323us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 193683us+:    21us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 226083us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 226419us+:   335us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 226438us+:    19us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 258101us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 258438us+:   336us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 258456us+:    18us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 290710us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 291047us+:   336us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 291063us+:    16us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)
OLD:cyclicte-884   0.... 327645us!:     5us: check_poison_obj+0x14/0x1bc 
(cache_alloc_debugcheck_after+0x38/0x160)
ACT:cyclicte-884   0.... 327982us+:   336us: poison_obj+0x10/0x40 
(cache_alloc_debugcheck_after+0x48/0x160)
NEW:cyclicte-884   0.... 328002us+:    20us: get_unused_fd+0x10/0xf4 
(do_sys_open+0x3c/0xe8)

does someone have an idea where this comes from ?

the hole /proc/latency output is shown here:
https://dev.contec.at/micro9/tmp/l_t.txt

-- 
regards
	Manfred Gruber

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: poison_obj latency
  2007-06-27 10:51 poison_obj latency Manfred Gruber
@ 2007-06-27 12:34 ` Ingo Molnar
  2007-06-27 13:24   ` Manfred Gruber
  0 siblings, 1 reply; 3+ messages in thread
From: Ingo Molnar @ 2007-06-27 12:34 UTC (permalink / raw)
  To: Manfred Gruber; +Cc: linux-rt-users


* Manfred Gruber <m.gruber@tirol.com> wrote:

> hi !
> 
> I use a arm4vt ep93xx processor with 2.6.21.5-rt18. 
> When i run cyclictest with -i 10000 and -n -p 90 -b 1000 i see some strange 
> latencies in poison_obj.

disable SLAB_DEBUG. And look into your 'dmesg' output, the -rt kernel 
will warn about any other kernel config options as well that are known 
to cause large latencies:

VFS: Mounted root (ext3 filesystem) readonly.
*****************************************************************************
*                                                                           *
*  REMINDER, the following debugging option is turned on in your .config:   *
*                                                                           *
*        CONFIG_SLAB                                                        *
*                                                                           *
*  it may increase runtime overhead and latencies.                          *
*                                                                           *
*****************************************************************************
Freeing unused kernel memory: 324k freed

	Ingo

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: poison_obj latency
  2007-06-27 12:34 ` Ingo Molnar
@ 2007-06-27 13:24   ` Manfred Gruber
  0 siblings, 0 replies; 3+ messages in thread
From: Manfred Gruber @ 2007-06-27 13:24 UTC (permalink / raw)
  To: Ingo Molnar, linux-rt-users

Am Mittwoch, 27. Juni 2007 14:34 schrieben Sie:
> * Manfred Gruber <m.gruber@tirol.com> wrote:
> > hi !
> >
> > I use a arm4vt ep93xx processor with 2.6.21.5-rt18.
> > When i run cyclictest with -i 10000 and -n -p 90 -b 1000 i see some
> > strange latencies in poison_obj.
>
> disable SLAB_DEBUG. And look into your 'dmesg' output, the -rt kernel
> will warn about any other kernel config options as well that are known
> to cause large latencies:
>
> VFS: Mounted root (ext3 filesystem) readonly.
> ***************************************************************************
>** *                                                                        
>   * *  REMINDER, the following debugging option is turned on in your
> .config:   * *                                                             
>              * *        CONFIG_SLAB                                        
>                * *                                                         
>                  * *  it may increase runtime overhead and latencies.      
>                    * *                                                     
>                      *
> ***************************************************************************
>** Freeing unused kernel memory: 324k freed
>
> 	Ingo

ok, thanks without SLAB_DEBUG there is no poison_obj latency.

-- 
mfg
	Manfred Gruber

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2007-06-27 13:24 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-06-27 10:51 poison_obj latency Manfred Gruber
2007-06-27 12:34 ` Ingo Molnar
2007-06-27 13:24   ` Manfred Gruber

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.