Re: [PATCH] XBox Gaming System subarchitecture.

Anders Gustafsson (andersg@0x63.nu)
Mon, 14 Jul 2003 16:54:29 +0200


On Mon, Jul 14, 2003 at 03:28:38PM +0100, Dave Jones wrote:
> > The real patch contains cleaned up HZ-ifdefs.
>
> good good..

Testing it an extra time now.

> > Or as Christoph answered: "Oh well, stupid crappy hardware..."
>
> That's a possibility, but if that were the case, I'd expect other things
> also to start randomly failing. It's unclear to me how -O2 would make
> a hardware bug more aparent. If it did so, that would also mean you'd
> have to ensure all your userspace was similarly compiled, which sounds
> very suspect. It's just a celeron based PC with nvidia nforce chipset right ?
> If that combination caused such problems, I'd expect to see the
> occasional problem report from non-Xbox regular home-built PC users too.

It only happens when paging is off. Hence no problem as soon as the kernel
is up and running. -O2 changes the memory-access patterns so it is quite
possible that it would expose hardware bugs.

> Might be one worth picking over on the gcc lists if you can identify
> which part gets miscompiled ?
>
> Tried different versions of binutils too ?

On 1.0 xboxes it works with any gcc with or without optimizations on.

On 1.1+ xboxen it does NOT work with 2.95, -O0 or -O2. But with 3.2.2-3.3 it
is known to work with -O0.

It's was not easy to localize the crashes. Cos adding more code for
debugging moves the point where it crashes around. However it does crash
reading from memory. A prefectly valid address. That it just a few
instructions earlier had written and read from.

-- 
Anders Gustafsson - andersg@0x63.nu - http://0x63.nu/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/