a dual board (meant for pentium) with one k6 200 and
a 2.4.0-test11 kernel with APIC support enabled does
oops here. removed the APIC support, and now everything is fine.
i read here it?s a known problem ? at least someone
else reported this, and it?s the same problem here.
regards, andreas
>>>>> "Andreas" == Andreas Jellinghaus <[email protected]> writes:
Andreas> a dual board (meant for pentium) with one k6 200 and a
Andreas> 2.4.0-test11 kernel with APIC support enabled does oops
Andreas> here. removed the APIC support, and now everything is fine. i
Andreas> read here it?s a known problem ? at least someone else reported
Andreas> this, and it?s the same problem here.
Yes, a colleague of mine has a similar beast, hence Caldera ships for quite a
while now with an appropriate patch. Now Christoph Hellwig has identified a
simpler solution (updated for -test11 by me):
--- linux/arch/i386/kernel/setup.c~ Fri Jul 7 04:42:06 2000
+++ linux/arch/i386/kernel/setup.c Tue Jul 18 19:22:48 2000
@@ -785,7 +785,8 @@
/*
* get boot-time SMP configuration:
*/
- if (smp_found_config)
+ if (smp_found_config && /* try only if the cpu has a local apic */
+ test_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability))
get_smp_config();
#endif
#ifdef CONFIG_X86_LOCAL_APIC
I think Alan has a similar thing in his test11-ac* series.
Hope that helps,
Torsten
> while now with an appropriate patch. Now Christoph Hellwig has identifi=
> ed a
> simpler solution (updated for -test11 by me):
>
> --- linux/arch/i386/kernel/setup.c~ Fri Jul 7 04:42:06 2000
> +++ linux/arch/i386/kernel/setup.c Tue Jul 18 19:22:48 2000
> @@ -785,7 +785,8 @@
> /*
> * get boot-time SMP configuration:
> */
> - if (smp_found_config)
> + if (smp_found_config && /* try only if the cpu has a local apic */
> + test_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability))
> get_smp_config();
This patch unfortunately does break some real setups, since it is legal to
have non APIC on the CPU but an APIC as an addon chip.
> I think Alan has a similar thing in his test11-ac* series.
Maciej Rozycki provided a much better test, that from reports so far seems
to be working