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

[Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>]

On 08/19/2014 01:40 PM, David Matlack wrote:
> On Mon, Aug 18, 2014 at 10:19 PM, Xiao Guangrong
> <xiaoguangrong@linux.vnet.ibm.com> wrote:
>> On 08/19/2014 01:00 PM, David Matlack wrote:
>>> On Mon, Aug 18, 2014 at 9:41 PM, Xiao Guangrong
>>> <xiaoguangrong@linux.vnet.ibm.com> wrote:
>>>> On 08/19/2014 12:31 PM, David Matlack wrote:
>>>>> The single line patch I suggested was only intended to fix the "forever
>>>>> incorrectly exit mmio".
>>>>
>>>> My patch also fixes this case and that does not doubly increase the
>>>> number. I think this is the better one.
>>>
>>> I prefer doubly increasing the generation for this reason: the updated boolean
>>> requires extra code on the "client-side" to check if there's an update in
>>> progress. And that makes it easy to get wrong. In fact, your patch
>>> forgot to check the updated bit in mark_mmio_spte(). Doubly increasing the
>>> generation requires no "client-side" code to work.
>>
>> No, the updated patch is used to fix case 2 which i draw the scenario in
>> the last mail. I mean the original patch in this patchset which just
>> increase the number after srcu-sync.
>>
>> Then could you tell me that your approach can do but my original patch can not?
> 
> It avoids publishing new memslots with an old generation number attached to
> them (even if it only lasts for a short period of time). 

I can not see the problem if that happen, could you please draw the scenario?

> Do you have a reason
> why you don't want to doubly increase the generation?

That more easily causes the number wrap-around.