2004-03-27 23:03:45

by Shawn Starr

[permalink] [raw]
Subject: 2.6.x strangeness with large buffer usage via network transfer/disk and SEGV processes

I don't get something maybe someone can explain why this is happening:

1) When using a large amount of buffers via sending say a 800MB file from
one PC to another, the Linux system will segfault processes but not preform
an OOM. Even though the system itself has not touched swap memory. Why is
the kernel killing/or why are the processes dying with Segfault?

I see this happening when I extract a Linux source tarball and have certain
processes running, while tar extracts the process will just receive a
segmentation fault w/o core.

When using a virtual OS emulator, the emulator will just die.

I don't remember this behaviour in 2.4 at all and I don't think this is
correct. I have PREEMPT enabled as well.

Is this a problem or is this correct behavour?

Thanks

Shawn S.


2004-03-28 02:58:50

by Eric Bambach

[permalink] [raw]
Subject: Re: 2.6.x strangeness with large buffer usage via network transfer/disk and SEGV processes

On Saturday 27 March 2004 5:07 pm, Shawn Starr wrote:
> I don't get something maybe someone can explain why this is happening:
>
> 1) When using a large amount of buffers via sending say a 800MB file from
> one PC to another, the Linux system will segfault processes but not preform
> an OOM. Even though the system itself has not touched swap memory. Why is
> the kernel killing/or why are the processes dying with Segfault?
>
> I see this happening when I extract a Linux source tarball and have certain
> processes running, while tar extracts the process will just receive a
> segmentation fault w/o core.
>
> When using a virtual OS emulator, the emulator will just die.
>
> I don't remember this behaviour in 2.4 at all and I don't think this is
> correct. I have PREEMPT enabled as well.
>
> Is this a problem or is this correct behavour?
>
Crashes a correct behavior?
You must be used to windows where crashes are sometimes the correct behavior
hehe.

Sorry, I had to say it.
Otherwise I don't have a serious answer to your problem ;)

> Thanks
>
> Shawn S.

-------------------------
Eric Bambach
Eric at cisu dot net
-------------------------