Illegal seek (reprise)

bill davidsen davidsen at tmr.com
Tue Jan 27 18:59:05 UTC 2004


In article <Pine.LNX.4.55.0401270901330.3540 at td.office.support.nl>,
Patrick Schreurs  <patricks at support.nl> wrote:
| On Mon, 26 Jan 2004, bill davidsen wrote:
| 
| > By any chance is this an SMP machine? There are some memory hardware
| > problems which seem to only show in shared memory with SMP. I certainly
| > don't have a memory checker which would play with the same memory with
| > multiple processors, so all I can do is ask.
| 
| What kind of memory hardware problems are you refering to?
| We are seeing some "filemap.c:2170: bad pmd 044001e3" on SMP
| servers the last few days.

Marginal memory which works fine except when you have it shared between
multiple processors, each of which is heavily accessing the memory. I
can't give a gate level explanation of the issue, although I suspect
that the refresh is being held off just a hair too long in some cases.
Take that as a guess.

In any case I've seen memory which was unreliable in SMP and fine in
uni, and mmap certainly will lead to shared memory. Just a thought. The
error message doesn't mean much to me, is is a Linux kernel message, but
it is only in the arm and m68k hardware sections in 2.6, 2.4 had it in
many places. Could be a kernel or hardware issue then.

Sorry I can't pin it better than that, I'll let you know if I find more.
-- 
bill davidsen <davidsen at tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.


More information about the inn-workers mailing list