The patch attached fixes this problem. Now SMP dom0 booted using >=8
CPUs with MMCONFIG (par of PCI Express) enabled. Basically it adds a
spurious page fault handler in the guest. See the comments in the
patch. Tested it both on none-PAE and PAE. Thanks for Nitin and Xin for
testing this patch.
Signed-off-by: Jun Nakajima <jun.nakajima@intel.com>
Ian Pratt wrote:>> I tried changing set_pte_mfs_ma to set_pte_at, and it also
>> let dom0 boot few times. I didn''t have not observed the issue
with
>> it.
>>
>> As it is a racing case kind of issue, I am not totally sure
>> that it is totally curing the issue. It can be possible we
>> are just reducing the chances of the race condition.
>> Especially with emulation, the slowness changes timing.
>
> Yes, using set_pte_at won''t cure the race (which we *definitely*
need
> to investigate), but is actually a better choice than set_pte anyway
> -- it will go x10 faster.
>
> If only we had a 16 way machine on which we could try and reproduce
> this problem :-)
>
> The race is a big worry, but the code looks correct. Please can you
> tell us more about the failure you observe. You mentioned that the
> shared info page appears to be unmapped. Please provide more details.
>
> Thanks,
> Ian
>
Jun
---
Intel Open Source Technology Center
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel