threads & /proc

threads & /proc

[Anton Petrusevich]

Hi Guys,

Well, I'm a bit surprised that nobody asked it yet. Do we have sound
thread support? I am able to put my linux-2.4.15-pre{1,7} --
definitely, and if I remember right, 2.4.14-pre{7,8} too in some strange
state, when any program like top, killall or ps that wanna get some
information from /proc (even midnight commander if you are trying to
look at state of any process) blocks indefinitely. It goes to unclean
shutdown, for example. kill doesn block, but do nothing. (I tried to
kill processes from ls /proc list). And I see it only after several
[unsuccessful] runs of my multithreaded program. Well, I can't say 
it's a correctly written program, I am still looking for bugs there. 
I don't have 100% way to get into this state, but I suspect some locking
issues with /proc. 
-- 
Anton

Re: threads & /proc

[Andrew Morton]

Anton Petrusevich wrote:
> 
> Hi Guys,
> 
> Well, I'm a bit surprised that nobody asked it yet. Do we have sound
> thread support? I am able to put my linux-2.4.15-pre{1,7} --
> definitely, and if I remember right, 2.4.14-pre{7,8} too in some strange
> state, when any program like top, killall or ps that wanna get some
> information from /proc (even midnight commander if you are trying to
> look at state of any process) blocks indefinitely. It goes to unclean
> shutdown, for example. kill doesn block, but do nothing. (I tried to
> kill processes from ls /proc list). And I see it only after several
> [unsuccessful] runs of my multithreaded program. Well, I can't say
> it's a correctly written program, I am still looking for bugs there.
> I don't have 100% way to get into this state, but I suspect some locking
> issues with /proc.

I've seen this once, on a semi-production machine which, unfortunately,
wasn't set up for diagnostic work.  Uniprocessor.

If you can send out some code which demonstrates the bug then
that will be invaluable - it will be fixed quickly.

-

Re: threads & /proc

[Peter Zaitsev]

Hello Anton,

Saturday, November 24, 2001, 8:38:57 AM, you wrote:


I would confirm the problem exists from the first 2.4.x kernels. Also
programs "w","vmstat","snmpd" hangs.  It seems to happen both for
UP/SMP kernels. I have seen this only on High Memory (1-2G), therefore
I do not have much low memory machines loaded.  The workload seems to
be the thing to affect this - I do not remember any hang on low VM
load system (for example Web Server) therefore it's offen happens
then there are a lot of I/O  and swapping goes (i.e Mysql Server).
Often this happens after appearing some of new __order allocation
failed errors.




Other issue which looks like connected to this - a thread hangs in
"D" state forever and it's unable to be killed in this case even
reboot -f does not work.

AP> Hi Guys,

AP> Well, I'm a bit surprised that nobody asked it yet. Do we have sound
AP> thread support? I am able to put my linux-2.4.15-pre{1,7} --
AP> definitely, and if I remember right, 2.4.14-pre{7,8} too in some strange
AP> state, when any program like top, killall or ps that wanna get some
AP> information from /proc (even midnight commander if you are trying to
AP> look at state of any process) blocks indefinitely. It goes to unclean
AP> shutdown, for example. kill doesn block, but do nothing. (I tried to
AP> kill processes from ls /proc list). And I see it only after several
AP> [unsuccessful] runs of my multithreaded program. Well, I can't say 
AP> it's a correctly written program, I am still looking for bugs there. 
AP> I don't have 100% way to get into this state, but I suspect some locking
AP> issues with /proc. 



-- 
Best regards,
 Peter                            mailto:pz@spylog.ru

Re: threads & /proc

[FORT David]

Peter Zaitsev wrote:

>Hello Anton,
>
>Saturday, November 24, 2001, 8:38:57 AM, you wrote:
>
>
>I would confirm the problem exists from the first 2.4.x kernels. Also
>programs "w","vmstat","snmpd" hangs.  It seems to happen both for
>UP/SMP kernels. I have seen this only on High Memory (1-2G), therefore
>I do not have much low memory machines loaded.  The workload seems to
>be the thing to affect this - I do not remember any hang on low VM
>load system (for example Web Server) therefore it's offen happens
>then there are a lot of I/O  and swapping goes (i.e Mysql Server).
>Often this happens after appearing some of new __order allocation
>failed errors.
>
>
>
>
>Other issue which looks like connected to this - a thread hangs in
>"D" state forever and it's unable to be killed in this case even
>reboot -f does not work.
>
>AP> Hi Guys,
>
>AP> Well, I'm a bit surprised that nobody asked it yet. Do we have sound
>AP> thread support? I am able to put my linux-2.4.15-pre{1,7} --
>AP> definitely, and if I remember right, 2.4.14-pre{7,8} too in some strange
>AP> state, when any program like top, killall or ps that wanna get some
>AP> information from /proc (even midnight commander if you are trying to
>AP> look at state of any process) blocks indefinitely. It goes to unclean
>AP> shutdown, for example. kill doesn block, but do nothing. (I tried to
>AP> kill processes from ls /proc list). And I see it only after several
>AP> [unsuccessful] runs of my multithreaded program. Well, I can't say 
>AP> it's a correctly written program, I am still looking for bugs there. 
>AP> I don't have 100% way to get into this state, but I suspect some locking
>AP> issues with /proc. 
>
>
>
I've also seen the case on a UP, ps was locking. And i had launched a
lot of multi-threaded processes(but i don't know if it's related).
I can had that i had also launched a lot of gdb on the multi-threaded
program. The program i was debugging is forking a process in a thread
that dies after the fork.

-- 
%--LINUX-HTTPD-PIOGENE----------------------------------------------------%
%  -datamining <-/                        |   .~.                         %
%  -networking/PHP/java/JSPs              |   /V\        L  I  N  U  X    %
%  -opensource                            |  // \\     >Fear the Penguin< %
%  -GNOME/enlightenment/GIMP              | /(   )\                       %
%           feel enlightened....          |  ^^-^^                        %
%                              HomePage: http://www.enlightened-popo.net  %
%---------- -- This was sent by Djinn running Linux 2.4.13 -- ------------%

Re: threads & /proc

[Luke]

Here's some more data about the problem reported by Anton where "ps",
"top", "killall", etc block and "kill" doesn't work.

I have encountered this behaviour with kernels 2.4.14 and 2.4.16 but not
2.4.13 nor 2.4.7 (although 2.4.13 hung this box in a different way).

This was observed on an SMP Pentium III, which does some multithreaded
computation.

I can't give precise instructions on how to replicate this bug, but
perhaps it can be repeated simply by exercising kernel threading?

Luke.

Re: threads & /proc

[Luigi Genoni]

on 2.4.14 I could not do kill -9 -1 as root with success to kill
everything was running (very bad idea, but it was for tests :) ).


On Wed, 28 Nov 2001, Luke wrote:

> Here's some more data about the problem reported by Anton where "ps",
> "top", "killall", etc block and "kill" doesn't work.
>
> I have encountered this behaviour with kernels 2.4.14 and 2.4.16 but not
> 2.4.13 nor 2.4.7 (although 2.4.13 hung this box in a different way).
>
> This was observed on an SMP Pentium III, which does some multithreaded
> computation.
>
> I can't give precise instructions on how to replicate this bug, but
> perhaps it can be repeated simply by exercising kernel threading?
>
> Luke.
>
>
> -
> 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/
>