Re: 2.1.63 - testing Pentium bug workaround..

Thomas Wouters (thomas@xs4all.nl)
Thu, 13 Nov 1997 05:31:01 +0100 (CET)


(Can't CC to the writer of the F0 0F bug workaround since i don't know who
it is :-P Hope he/she reads it.)

On Wed, 12 Nov 1997, Linus Torvalds wrote:

> Please give it a good testing, especially the Pentium bug workaround.
> Throw all the tests you have at it, to see that it really works.

The pentium bug workaround doesn't seem to be water proof -- at least, not in
2.1.63-VGER-CVS971112. That is, i just ran 20 of these:

while 1
/root/crashme
sleep 1
end

where crashme is a binary that works fine (crashes fine) under the other
linux versions. After a few minutes my harddisk began sputtering, first
slowly but gradually increasing. I was just trying to find out what was
going on (could be just the logfile, though, or the (not limited) corefiles)
when my machine hung, F00F-style (i checked, finger method ;-) I was running
X at the time, as well as two 'screen's (for the 20 scripts) but I haven't
had *any* trouble with hungup X sessions yet, so i hardly think so.

It might be something load related, though, my X was beginning to show some
slugishness.

I'm running on a P166 MMX, Asus motherboard with a triton chipset and a
couple of IDE disks, 48Mb RAM and 100+ Mb swap (of which about 10 Mb was in
use when my pentium did *ick*.) My syslog is quite full of

Nov 13 05:07:02 centurion kernel: <c3800fc8/c3800ff8>
Nov 13 05:07:02 centurion kernel: <handler c01094b8... ... done>

but they all nicely match up. Still, it'd taken the processor about 600 or
700 tries before it locked up (rough guess) so the bug workaround is a *lot*
better than nothing. If i was depending on pentiums and linux for
user-account-machines i'd be very glad with it, especially with the enormous
speedy development ;-)

I will do some tests on plain 2.1.63 in the morning, just for the fun of it
(haven't compiled it yet.) Time for bed right now *rubeye*.

G'night,
Thomas.