Re: [patch] O(1) scheduler, -H6

Re: [patch] O(1) scheduler, -H6

[Dieter Nützel]

Ingo,

you are the man!
My 1 GHz Athlon II, 640 MB, is up and running with 2.4.17-O(1)-H6.
Writing this under KDE-2.2.2 (KMail) during
time nice +19 make -j40 modules
Max load was: 41,97

Great!

dbench 32 during artsd (noatun, KDE-2.2.2) playing Ogg-Vorbis
"show" hiccup when

artsd (two sub processes) are in
wait_on_p
rt_sigsus

and/or

dbench procs are in
get_reque
wait_on_b
do_journ (ReiserFS?)

Even when I renice both artsd procs I get hiccup.
  PID USER     PRI  NI PAGEIN  SIZE  RSS SHARE WCHAN     STAT %CPU %MEM   
TIME COMMAND
 4953 nuetzel    1 -19      0  7148 7148  4324 rt_sigsus S <   3.6  1.1   
0:04 artsd
 1048 nuetzel    1 -19   1280  7148 7148  4324 schedule_ S <   2.7  1.1   
0:24 artsd

But when I put the Ogg-Vorbis file into /dev/shm I get _NO_ hiccup anymore!!!
So it is only disk IO limited with your O(1)-H6.

Wow, that is fantastic!!!
Never had that before.
Writing during both...;-)

Redid it without renice (-19):
some very few short hiccup.

-Dieter

-- 
Dieter Nützel
Graduate Student, Computer Science

University of Hamburg
Department of Computer Science
@home: Dieter.Nuetzel@hamburg.de

Re: [patch] O(1) scheduler, -H6

[Robert Love]

Ingo,

H6 seems to of fixed the boot hardlock problem on my 2.4 boxen here. 
Thanks.

	Robert Love

[patch] O(1) scheduler, -H6

[Ingo Molnar]

the -H6 patch is available:

    http://redhat.com/~mingo/O(1)-scheduler/sched-O1-2.5.2-pre11-H6.patch
    http://redhat.com/~mingo/O(1)-scheduler/sched-O1-2.4.17-H6.patch

the most important fix is from DaveM, we should not send an IPI to
ourselves. That fix was the last bug on Sparc boxes it appears.

overall stability status: all the previous bug reports about runtime
lockups (which happen in -pre11 as well) are fixed in this patchset. The
only open issue is the boot-time lockup some people are seeing with the
2.4 patch only (not the 2.5 patch), there is a chance that it's fixed in
this patch too.

Changes:

 - DaveM's the man: check for p->cpu != smp_processor_id() before doing a
   smp_send_reschedule(p->cpu). Doh!

 - task_interactive() test is done properly now - the H5 would mis-detect
   every nice +19 task as interactive. I'd like to ask everyone who had
   interactivity problems to re-test under -H6. It's all very smooth on my
   systems.

 - Rusty Russell: add comment to expire_task.

 - fix main.c in the 2.4.17 patch. (this should fix the bootup-lockups.)

Bug reports, comments, suggestions welcome. (any patch/fix that is not in
-H6 has gone lost in my mailqueue. The influx is rather high.)

	Ingo

Re: [patch] O(1) scheduler, -H6

[dan kelley]

2.4.17 w/ H6 still oopses on my dual xeon.  oops:

>>EIP; c02cc482 <stext_lock+506/8cf2>   <=====
Trace; c0114b1e <__wake_up+86/dc>
Trace; c0127292 <schedule_task+e2/e8>
Trace; c012711a <call_usermodehelper+c6/dc>
Trace; c012702c <__call_usermodehelper+0/28>
Trace; c012702c <__call_usermodehelper+0/28>
Trace; c027c905 <net_run_sbin_hotplug+79/84>
Trace; c028150f <rtmsg_ifinfo+73/78>
Trace; c0281a42 <rtnetlink_event+32/40>
Trace; c027c558 <register_netdevice+198/1a4>
Trace; c01da1e6 <register_netdev+5e/70>
Trace; c01d33b6 <vortex_probe1+ba6/c1c>
Trace; c01d2801 <vortex_init_one+35/44>
Trace; c0246d6e <pci_announce_device+36/54>
Trace; c0246dd4 <pci_register_driver+48/60>
Trace; c01050f3 <init+5b/1d8>
Trace; c01056b8 <kernel_thread+28/38>
Code;  c02cc482 <stext_lock+506/8cf2>
00000000 <_EIP>:
Code;  c02cc482 <stext_lock+506/8cf2>   <=====
   0:   80 39 00                  cmpb   $0x0,(%ecx)   <=====
Code;  c02cc485 <stext_lock+509/8cf2>
   3:   f3 90                     repz nop
Code;  c02cc487 <stext_lock+50b/8cf2>
   5:   7e f9                     jle    0 <_EIP>
Code;  c02cc489 <stext_lock+50d/8cf2>
   7:   e9 77 70 e4 ff            jmp    ffe47083 <_EIP+0xffe47083>
c0113505 <tr
y_to_wake_up+61/344>
Code;  c02cc48e <stext_lock+512/8cf2>
   c:   80 3a 00                  cmpb   $0x0,(%edx)
Code;  c02cc491 <stext_lock+515/8cf2>
   f:   f3 90                     repz nop
Code;  c02cc493 <stext_lock+517/8cf2>
  11:   7e f9                     jle    c <_EIP+0xc> c02cc48e
<stext_lock+512/8
cf2>
Code;  c02cc495 <stext_lock+519/8cf2>
  13:   e9 00 00 00 00            jmp    18 <_EIP+0x18> c02cc49a
<stext_lock+51e


> the -H6 patch is available:
>
>     http://redhat.com/~mingo/O(1)-scheduler/sched-O1-2.5.2-pre11-H6.patch
>     http://redhat.com/~mingo/O(1)-scheduler/sched-O1-2.4.17-H6.patch
>
> the most important fix is from DaveM, we should not send an IPI to
> ourselves. That fix was the last bug on Sparc boxes it appears.
>
> overall stability status: all the previous bug reports about runtime
> lockups (which happen in -pre11 as well) are fixed in this patchset. The
> only open issue is the boot-time lockup some people are seeing with the
> 2.4 patch only (not the 2.5 patch), there is a chance that it's fixed in
> this patch too.
>
> Changes:
>
>  - DaveM's the man: check for p->cpu != smp_processor_id() before doing a
>    smp_send_reschedule(p->cpu). Doh!
>
>  - task_interactive() test is done properly now - the H5 would mis-detect
>    every nice +19 task as interactive. I'd like to ask everyone who had
>    interactivity problems to re-test under -H6. It's all very smooth on my
>    systems.
>
>  - Rusty Russell: add comment to expire_task.
>
>  - fix main.c in the 2.4.17 patch. (this should fix the bootup-lockups.)
>
> Bug reports, comments, suggestions welcome. (any patch/fix that is not in
> -H6 has gone lost in my mailqueue. The influx is rather high.)
>
> 	Ingo
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>