>>> Ingo Molnar <mingo@elte.hu> 08.09.08 20:54 >>>
quoted text
>
>-tip testing found various kernel crashes on 32-bit testboxes and i've 
>bisected it down to:
>
>...
>
>a typical crash is like the one attached below. It's due to the ioremap 
>failing. The drivers/char/rio/rio_linux.c driver probes these addresses:
>
>   static int rio_probe_addrs[] = { 0xc0000, 0xd0000, 0xe0000 };
>
>which is questionable ...

No, they look absolutely valid, they're ISA ROM addresses.

quoted text
>for now i've reverted it from current tip/master, see commit 
>e3fdd129901. (you can reinstate the commit by doing "git revert 
>e3fdd1299"
>
>Even if we decided to fail these ioremaps it would be better to emit a 
>warning instead of crashing the box.

We shouldn't fail them, they're valid. What the crash means is that even
addresses below 1Mb are considered out of range, which I can only take
as x86_phys_bits being zero (or a bogus very small number) on
secondary (or all) CPUs. However, looking at the call tree I can't see how
that could happen (provided CPUID doesn't produce garbage output):

- smp_store_cpu_info(), as it always did, pre-initializes the new CPU's
  info with boot_cpu_data, and calls identify_secondary_cpu()
- identify_secondary_cpu() calls identify_cpu()
- identify_cpu() pre-sets x86_phys_bits to 32, and since the field didn't
  exist for 32-bits before, nothing should be able to clear or otherwise
  alter it

Jan

--
unsubscribe notice
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/