Re: [PATCH 11/11] KVM: MMU: improve write flooding detected

[Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>]

On 07/27/2011 07:08 PM, Avi Kivity wrote:
> On 07/27/2011 01:20 PM, Xiao Guangrong wrote:
>> >>    }
>> >
>> >  I think this is a little dangerous.  A guest kernel may be instantiating multiple gptes on a page fault, but guest userspace hits only one of them (the one which caused the page fault) - I think Windows does this, but I'm not sure.
>> >
>>
>> I think this case is not bad: if the guest kernel need to write multiple gptes (>=3),
>> it will cause many page fault, we do better zap the shadow page and let it become writable as
>> soon as possible.
>> (And, we have pte-fetch, it can quickly establish the mapping for a new shadow page)
> 
> Actually, what should save us is unsync pages.  Why are we hitting this path at all?
> 

Avi,

The shadow page can not became unsync if it has other sp that sp.gfn = gfn && sp.role.level != 1,
for example:
- if the gfn is not only used for the last page structure(PTE page)
or
- gfn was used for upper page structure before but we do not zap the old shadow pages

So, if this gfn is written, #PF is generated, we hope that these sp can be zapped earlier,
the later #PF can detect this gfn is not have shadow pages, and the mapping can became writable.