Re: [PATCH 06/11] KVM: MMU: let page fault handler be aware tracked page

[Kai Huang <kai.huang@linux.intel.com>]

On 12/15/2015 05:03 PM, Xiao Guangrong wrote:
>
>
> On 12/15/2015 04:11 PM, Kai Huang wrote:
>>
>>
>> On 12/01/2015 02:26 AM, Xiao Guangrong wrote:
>>> The page fault caused by write access on the write tracked page can not
>>> be fixed, it always need to be emulated. page_fault_handle_page_track()
>>> is the fast path we introduce here to skip holding mmu-lock and shadow
>>> page table walking
>> Why can it be out side of mmu-lock? Is it OK that some other thread 
>> removes tracking of this page
>> simultaneously? Shall we assuming the emulation code should handle 
>> this case?
>>
>
> What your mentioned is the worst case, if that happen the vcpu will spend
> longer time to emulate the access rather then retry it. It is bad but 
> it is
> the rare contention. It is worth speeding up the common / most case.
My concern is when this case happen, whether emulating the access is 
still the right behavior, you know, after other thread removed the GFN 
from tracking..
And as the notifier's  track_write call back will be called in the 
emulating code, won't there be problem if the GFN has been removed from 
tracking by other thread?

Thanks,
-Kai
>
>> Even it works for write protection, is it OK for other purpose 
>> tracking (as your tracking framework
>> can be extended for other purpose)?
>
> We only need to make sure that no track event is lost, i.e, we can not
> skip the case that the index is changed from 0 to 1.
>
> If we see index is 0, the vcpu can hold the mmu-lock and go to slow path
> anyway so no track event will be lost.
>
>>>
>>> However, if the page table is not present, it is worth making the page
>>> table entry present and readonly to make the read access happy
>> It's  fine for tracking write from guest. But what if I want to track 
>> every read from guest?
>> Probably I am exaggerating :)
>>
>
> Then we do not go to the real page fault handler, just keep the shadow
> page entry non-present.
> -- 
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>