Re: Bus error while building Erlang

Re: Bus error while building Erlang

[BERTRAND Joël]

Sergei Golovan wrote:
> Hi!

	Hello,

> Recent build of erlang package failed on sparc architecture
> (http://buildd.debian.org/build.php?&pkg=erlang&arch=sparc&file=log).
> I cannot find a reason looking at the log file, and I don't have an
> access to any sparc machine to debug the failure.

	I have seen the same bus error with gcc 4.x (internal error that 
returns a bus error on Sbus based sparc _and_ sparc64 linux boxes). I'm 
trying to find this bug for a long time without any success. I never 
have seen this bug on PCI based sparc64 boxes. Please note that I'm not 
sure that this trouble comes from Sbus subsystem ;-) I suspect a bug in 
sparc/sparc64 kernel.

Tested boxes :
- SS5 (MS/85), 256 MB -> bug (but system not stable)
- SS20 (SM71), 512 MB -> bug (but system not stable)
- SS20 (RT626) 512 MB -> system crashes before I can launch a gcc ;-)
- U1E (U1/170), 640 MB -> bug
- U2E (2*U2/300), 2048 MB -> bug
- U5 (U2/270), 512 MB -> works fine
- U60 (2*U2/450), 1024 MB -> works fine
- U80 (4*U2/450), 2048 MB -> works fine
- U420 (4*U2/450), 4096 MB -> works fine

	Today, I don't try to debug on sparc32 due to a bad support of SMP 
boxes (even in UP, I have some trouble with sparc32). I use a U2/SMP, 
but this bug randomly appears and is very difficult to reproduce. I have 
seen if only with iceape (all versions) and gcc 4.x.
In a first time, I though that my U2 was broken, or that fans were 
halted for some reasons. I have used SUNvts on Solaris9 to test hardware 
and it don't return error.

	I have seen this bug on 2.6.23.1 kernel. Now, my U2 runs with a 
2.6.23.8 (built by gcc-4.1) and I don't see any bus error when I write 
these lines.

	Regards,

	JKB

CC: sparc kernel mailing list

Re: Bus error while building Erlang

[Josip Rodin]

On Sun, Nov 25, 2007 at 03:53:37PM +0100, BERTRAND Joël wrote:
> >Recent build of erlang package failed on sparc architecture
> >(http://buildd.debian.org/build.php?&pkg=erlang&arch=sparc&file=log).
> >I cannot find a reason looking at the log file, and I don't have an
> >access to any sparc machine to debug the failure.
> 
> 	I have seen the same bus error with gcc 4.x (internal error that 
> returns a bus error on Sbus based sparc _and_ sparc64 linux boxes). I'm 
> trying to find this bug for a long time without any success. I never 
> have seen this bug on PCI based sparc64 boxes. Please note that I'm not 
> sure that this trouble comes from Sbus subsystem ;-) I suspect a bug in 
> sparc/sparc64 kernel.

FWIW, the box that produced that log linked above was running 2.6.23.8, and
it's a sparc64 CPU with PCI.

-- 
     2. That which causes joy or happiness.

Re: Bus error while building Erlang

[BERTRAND Joël]

Josip Rodin wrote:
> On Sun, Nov 25, 2007 at 03:53:37PM +0100, BERTRAND Joël wrote:
>>> Recent build of erlang package failed on sparc architecture
>>> (http://buildd.debian.org/build.php?&pkg=erlang&arch=sparc&file=log).
>>> I cannot find a reason looking at the log file, and I don't have an
>>> access to any sparc machine to debug the failure.
>> 	I have seen the same bus error with gcc 4.x (internal error that 
>> returns a bus error on Sbus based sparc _and_ sparc64 linux boxes). I'm 
>> trying to find this bug for a long time without any success. I never 
>> have seen this bug on PCI based sparc64 boxes. Please note that I'm not 
>> sure that this trouble comes from Sbus subsystem ;-) I suspect a bug in 
>> sparc/sparc64 kernel.
> 
> FWIW, the box that produced that log linked above was running 2.6.23.8, and
> it's a sparc64 CPU with PCI.

	For me, it's the same bug... I've never seen it on PCI boxes, but 
result is the same :-(

	Regards,

	JKB

Re: Bus error while building Erlang

[Julian Calaby]

On Nov 26, 2007 1:53 AM, BERTRAND Joël <joel.bertrand@systella.fr> wrote:
> Sergei Golovan wrote:
> > Hi!
>
>         Hello,
>
> > Recent build of erlang package failed on sparc architecture
> > (http://buildd.debian.org/build.php?&pkg=erlang&arch=sparc&file=log).
> > I cannot find a reason looking at the log file, and I don't have an
> > access to any sparc machine to debug the failure.

I've been plagued by bus errors on my SS10 and U1E in the past whilst
using Debian Etch. It appeared to be a bug in the domain name => IP
code - which caused all the stuff I wanted to run on it to immediately
fail.

What notes I could find on bus errors on Sparc led me to believe that
this error was caused by unaligned memory accesses. (please correct me
if I'm talking out of my arse) I guessed that the problem was
somewhere in libc. (and hence inaccessible for me) The problems all
went away when I upgraded the boxes to the new version of libc in
Lenny. (before Sparc32 support was dropped)

I'm guessing that something in erlc is doing a cast when it should do
a memcopy. However I've never even tried to get to the bottom of this
on my boxes, and given that the U1E now works fine, I've never had any
reason to.

Hope this is useful,

-- 

Julian Calaby

Email: julian.calaby@gmail.com