Fix oom killer braindamage
Subsystems (2)
PATCH 1/1 Fix oom killer braindamage
Hi Marcelo,
attached patch fixes the oom killer braindamage where it tries to kill processes again and again and again w/o any ending or successfull killing of the selected processes in an OOM case.
The attached, very simple but effective, patch fixes it.
All the kudos go to Rik van Riel.
Patch tested and works, and also for a long time in my tree (and maybe also others?!)
This issue is out there for several years.
Please consider it for 2.4.21-rc5, thanks.
ciao, Marc
===== mm/oom_kill.c 1.11 vs edited =====
mm/oom_kill.c +5 -0
@@ -61,11 +61,16 @@ static int badness(struct task_struct *pif (!p->mm)return 0 ; + +if (p->flags &PF_MEMDIE ) +return 0 ; +/* * Never kill init */ if (p->pid ==1 )return 0 ; +/* * The memory size of the process is the basis for the badness. */
Quoted 8 lines · Marc-Christian Petersen
Hi Marcelo,
attached patch fixes the oom killer braindamage where it tries to kill processes again and again and again w/o any ending or successfull killing of the selected processes in an OOM case.
The attached, very simple but effective, patch fixes it.
Quoted 8 lines
All the kudos go to Rik van Riel.
Patch tested and works, and also for a long time in my tree (and maybe also others?!)
This issue is out there for several years.
Please consider it for 2.4.21-rc5, thanks.
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
On Tuesday 27 May 2003 21:12, Marcelo Tosatti wrote:
Hi Marcelo,
Quoted 2 lines
Quoted 1 line
Please consider it for 2.4.21-rc5, thanks.
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
well, not my fault ;-) Rik sent this while -pre time and after I saw it wasn't applied I sent it again while -pre time.
*scnr*
Well, please apply it for 2.4.22-pre1 then. Thanks
ciao, Marc
On Tuesday 27 May 2003 21:18, Marc-Christian Petersen wrote:
Hi again,
Quoted 1 line
Quoted 1 line
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
BTW: This breaks _nothing_. It _fixes_ an annoying bug! ;)
Quoted 2 lines
well, not my fault ;-) Rik sent this while -pre time and after I saw it wasn't applied I sent it again while -pre time.
ciao, Marc
Quoted 6 lines · Marc-Christian Petersen
On Tuesday 27 May 2003 21:18, Marc-Christian Petersen wrote:
Hi again,
Quoted 1 line
Quoted 1 line
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
BTW: This breaks _nothing_. It _fixes_ an annoying bug! ;)
You're not completly sure it doesnt break nothing. MM/VM can be reallly subtle and nasty at times.
Hi!
Quoted 5 lines
Quoted 2 lines
Quoted 1 line
Quoted 1 line
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
BTW: This breaks _nothing_. It _fixes_ an annoying bug! ;)
You're not completly sure it doesnt break nothing. MM/VM can be reallly subtle and nasty at times.
OOM killer is nor subtle nor nasty. And not normally used. That patch really should be safe.
Pavel
Signature · 2 lines
When do you have a heart between your knees? [Johanka's followup: and *two* hearts?]
On Tuesday 27 May 2003 21:18, Marc-Christian Petersen wrote:
Hi again ^3 ;)
Quoted 1 line
Quoted 1 line
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
Please, if there is any chance we can fix the pause/stop bug, delay .21 final for some hours or a day (or maybe two)
I've CC'ed akpm and Axboe. I think they are the only ones knowing enough about the code to see an obvious error and even fixing the bug?!
Do you agree? Does anyone else agree? Or disagree?
ciao, Marc
Quoted 4 lines · Marc-Christian Petersen
On Tuesday 27 May 2003 21:18, Marc-Christian Petersen wrote:
Hi again ^3 ;)
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
Please, if there is any chance we can fix the pause/stop bug, delay .21 final for some hours or a day (or maybe two)
I've CC'ed akpm and Axboe. I think they are the only ones knowing enough about the code to see an obvious error and even fixing the bug?!
Do you agree? Does anyone else agree? Or disagree?
I have to agree with Marcelo, nothing in ll_rw_blk is obvious enough to just be changed at will in -rc5 time.
I need to catch up with this thread, hopefully we can get things fixed around 2.4.22-early.
Signature · 1 line
Jens Axboe
Quoted 4 lines · Marc-Christian Petersen
On Tuesday 27 May 2003 21:18, Marc-Christian Petersen wrote:
Hi again ^3 ;)
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Not suitable for -rc. Btw, -rc5 is already at bkbits.net.
Please, if there is any chance we can fix the pause/stop bug, delay .21 final for some hours or a day (or maybe two)
I've CC'ed akpm and Axboe. I think they are the only ones knowing enough about the code to see an obvious error and even fixing the bug?!
Do you agree? Does anyone else agree? Or disagree?
Few points:
- 2.4.22 is going to be a short release, meaning we will have this bug fixed soon in a final release.
- the bug is around for quite some time now, its not very critical.
- its -rc stage.
Moreover, -rc5 is already out.
I will work with Jens, Axboe and Andrea to get this properly fixed in .22 in case Andrea patch is not OK.
On Tuesday 27 May 2003 21:34, Marcelo Tosatti wrote:
Hi Marcelo,
Quoted 4 lines
- 2.4.22 is going to be a short release, meaning we will have this bug fixed soon in a final release.
- the bug is around for quite some time now, its not very critical.
- its -rc stage.
k.
Quoted 2 lines
I will work with Jens, Axboe and Andrea to get this properly fixed in .22 in case Andrea patch is not OK.
great. Just remember: fix-pausing-2 fix does _not_ fix the pauses/stops/mouse is dead/keyboard is dead/ problem. It fixes the "process stuck in D state" problem.
Please bear it in mind.
ciao, Marc
Quoted 11 lines · Marc-Christian Petersen
Hi Marcelo,
attached patch fixes the oom killer braindamage where it tries to kill processes again and again and again w/o any ending or successfull killing of the selected processes in an OOM case.
The attached, very simple but effective, patch fixes it.
All the kudos go to Rik van Riel.
Patch tested and works, and also for a long time in my tree (and maybe also others?!)
This issue is out there for several years.
Please consider it for 2.4.21-rc5, thanks.
ciao, Marc
Also in 2.5.x for some time.
-- wli