Re: 2.3.44 bug

[David Wragg]

Tigran Aivazian <[e-mail]> writes:
> I was using 2.3.43 (and all other recent 2.3.x) without observing this
> specific problem so it is something new to 2.3.44. I have not narrowed
> it down yet but the symptom is that I can't start any large programs
> (e.g. X) and can't compile almost anything, gcc randomly dies with
> SIGSEGV.

Me too.

Different programs though; X is fine, gcc is fine, xemacs is fine. The
things that triggered it for me are nnrpd (innd itself is fine) and
/usr/bin/netscape -- not netscape itself but the wrapper shell script
(from the RedHat 6.1 netscape-common-4.61-12 RPM). But other shell
scripts run without problems. Very odd.

> (No, please do not tell me that my 256M DIMM is broken - it was
> running absolutely every 2.3.x just fine since the day I bought it).
> 
> The machine is a plain dual PIII (Katmai) 550 on a GA BXD motherboard,
> nothing too exotic - old good slot 1 design.

Dual PPro machine. 

> I will let you know if I find more - I intend to narrow it down and
> then look closer at the changes. The first look at 2.3.44 patch
> reveals only numa changes (irrelevant) and the innocent change to
> page_alloc.c (not counting my change to comment in vmscan.c,
> of course :), so there is something non-trivial going on here.

strace showed that the SIGSEGV seems to follow an sbrk. If it is
related to memory allocation behaviour, that could explain why one
shell script can trigger it but not another.