Re: [PATCH 1/2] KVM: fix cache stale memslot info with correct mmio generation number

[Paolo Bonzini <pbonzini@redhat.com>]

Il 20/08/2014 03:03, David Matlack ha scritto:
> On Tue, Aug 19, 2014 at 5:29 PM, Xiao Guangrong
> <xiaoguangrong@linux.vnet.ibm.com> wrote:
>> On 08/19/2014 05:03 PM, Paolo Bonzini wrote:
>>> Il 19/08/2014 10:50, Xiao Guangrong ha scritto:
>>>> Okay, what confused me it that it seems that the single line patch
>>>> is ok to you. :)
>>>
>>> No, it was late and I was confused. :)
>>>
>>>> Now, do we really need to care the case 2? like David said:
>>>> "Sorry I didn't explain myself very well: Since we can get a single wrong
>>>> mmio exit no matter what, it has to be handled in userspace. So my point
>>>> was, it doesn't really help to fix that one very specific way that it can
>>>> happen, because it can just happen in other ways. (E.g. update memslots
>>>> occurs after is_noslot_pfn() and before mmio exit)."
>>>>
>>>> What's your idea?
>>>>
>>>>> I think if you always treat the low bit as zero in mmio sptes, you can
>>>>> do that without losing a bit of the generation.
>>>>
>>>> What's you did is avoiding cache a invalid generation number into spte, but
>>>> actually if we can figure it out when we check mmio access, it's ok. Like the
>>>> updated patch i posted should fix it, that way avoids doubly increase the number.
>>>
>>> Yes.
>>>
>>>> Okay, if you're interested increasing the number doubly, there is the simpler
>>>> one:
>>>
>>> This wastes a bit in the mmio spte though.  My idea is to increase the
>>> memslots generation twice, but drop the low bit in the mmio spte.
>>
>> Yeah, really smart idea. :)
>>
>> Paolo/David, would you mind making a patch for this (+ the comments in David's
>> patch)?
> 
> Paolo, since it was your idea would you like to write it? I don't mind either
> way.

Sure, I'll post the patch for review.

Paolo