Bug 16802

Summary:	An user math program have the system on K6-2/III freeze
Product:	Base System	Reporter:	Takayuki HOSODA <lyuka>
Component:	i386	Assignee:	freebsd-bugs (Nobody) <bugs>
Status:	Closed FIXED
Severity:	Affects Only Me
Priority:	Normal
Version:	3.4-RELEASE
Hardware:	Any
OS:	Any

Description Takayuki HOSODA 2000-02-18 09:00:02 UTC

An user math (FFT) program compliled by gcc-2.95 with 
'-O2 -mpreferred-stack-boundary=2' options often have the
system freeze, which is running on AMD's K6-2 or K6-III processor.
 It won't occur on the system running on the intel pentium processor.
 Also, another user program that making use of a public FFT library
(ftp://ftp.fftw.org/pub/fftw/fftw-2.1.3.tar.gz)
configured with '--enable-i386-hacks' shows the same phenomenon.
 No core dumped, no signal recieved, though the user program compiled,
in this case, by gcc-2.7.1 with -g options was runed within the gdb.

Fix: 

I have no idea for the permanent fix, however, the followings
maight be help for some case at the user hand.

1. You'd better use gcc-2.7.1 instead of gcc-2.95.
2. Don't use -O3 compile option.
3. Never use -mpreferrd-stack-boundary=2 compile option.
How-To-Repeat: Generally, tight and fast floating math program compiled by
gcc-2.95 with -mpreferred-stack-boundary=2 options runned on
K6-2 or K6-III processor often cause the system freeze, however,
the same program won't cause the problem runned on the pentium
processor.
 I can send you the program souces in trouble (4KB).
(It can run on pentium processor without problem)

Comment 1 borjamar 2000-02-22 09:50:18 UTC

	I can test it in a K7 computer.




	Borja.
-- 
***********************************************************************
Borja Marcos                    * Internet: borjam@we.lc.ehu.es
Alangoeta, 11 1 izq             *           borjamar@sarenet.es
48990 - Algorta (Vizcaya)       *           borjam@uninet.edu
SPAIN                           *           borjam@well.com
***********************************************************************
--- FreeBSD, turning PCs into workstations

Comment 2 iedowse freebsd_committer

2001-08-27 21:40:46 UTC

State Changed
From-To: open->feedback


Does this problem still exist? It sounds more like a hardware 
problem than a software bug.

Comment 3 iedowse 2001-09-26 20:39:07 UTC

Adding this to the audit trail.

------- Forwarded Message

From: Takayuki HOSODA <lyuka@finetune.co.jp>
Date: Tue, 28 Aug 2001 18:33:26 +0900

Hi.
Thank you for your reply and consideration on the problem.

The problem still exist. It seems to be K6-2/III specific, and
won't happen on the pentium based machines.
For intensive use on FPU of the K6-2/III seemingly often cause
the problem, I might think, it might be a critical path or any
other timing problem of the proccessor than a software bug.
I think we have nothing to do for the problem by a software.
Slowing the proccessor clock down much may be a help which I do
not want to do. Forget the old K6-2/III and use latest proccessor
may be a best way. 
Thank you.

------- End of Forwarded Message

Comment 4 iedowse freebsd_committer

2001-09-26 20:44:35 UTC

State Changed
From-To: feedback->open


Apparently this is still a problem.

Comment 5 Oscar Bonilla 2002-06-19 18:15:34 UTC

My wild guess would be that this is a heating problem. AMD processors 
heat up a lot
and can usually freeze when overheated. Pentiums continue to work albeit 
slower when
overheated. If other people have not been able to replicate the problem, 
or if slowing
down the clock (or cooling down the machine) makes the problem go away 
then it's
the heat.

Regards,

-Oscar

Comment 6 Doug Barton freebsd_committer

2003-01-23 10:14:54 UTC

State Changed
From-To: open->feedback


Is this still a problem on more recent FreeBSD systems?

Comment 7 Doug Barton freebsd_committer

2003-01-23 11:08:59 UTC

State Changed
From-To: feedback->closed


Originator reports that this is no longer a problem.