Re: Dual opteron various segfaults with 2.6.14.2 and earlier kernels

[Andrew Walrond <andrew@walrond.org>]

On Thursday 24 November 2005 00:03, Andrea Arcangeli wrote:
> Hello Fabio,
>
> On Thu, Nov 24, 2005 at 12:26:41AM +0100, Fabio Coatti wrote:
> > yes, uname says  2.6.14.2; on a second identical machine, I've just seen
> > this:
> >
> >
> > factorial[2352]: segfault at 0000000000020f31 rip 00000000004035ae rsp
> > 00007fffffbfaf60 error 4
> > factorial[2354]: segfault at 0000000000020f31 rip 00000000004035ae rsp
> > 00007fffffe3fc70 error 4
> > factorial[2361]: segfault at 0000000000020f31 rip 00000000004035ae rsp
> > 00007fffffb07c50 error 4
> > factorial[2358]: segfault at 0000000000020f31 rip 00000000004035ae rsp
> > 00007fffffb07c50 error 4
> > factorial[2363]: segfault at 0000000000020f31 rip 00000000004035ae rsp
> > 00007fffffe6d270 error 4
> >
> > the kernel and HW are the same.
>
> Error 4 means a read in userland on a not mapped area.
>
> The above isn't necessairly a kernel or hardware problem, it looks like
> an userland bug if it segfaults at such a low address (20f31). Nothig is
> mapped below "0x400000" exactly to catch these kind of bugs.

Which makes sense; the sed failures seen during 'make' runs were probably a 
result of the TLB flush filter errata on kernels prior to 2.6.14 , whereas 
the above is a userland bug which occurs with all kernels.

Andrew Walrond