Re: [PATCH 3/4] KVM: Switch to srcu-less get_dirty_log()

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

On 03/16/2012 03:55 PM, Takuya Yoshikawa wrote:

> On Fri, 16 Mar 2012 15:30:45 +0800
> Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> wrote:
> 
>>>> There is a example:
>>>>
>>>>              CPU A                                   CPU  B
>>>> guest page is written by write-emulation
>>>>
>>>>                                                   hold mmu-lock and see dirty-bitmap
>>>>                                                   is not be changed, then migration is
>>>>                                                   completed.
>>>
>>> We do not allow this break.
>>>
>>
>>
>> Hmm? what can avoid this? Could you please point it out?
> 
> Stopping the guest before actualy migrating the guest means VCPU threads
> must be back in the userspace at the moment, no?
> 
> So when the final GET_DIRTY_LOG is being executed, thread A cannot be
> in KVM.
> 
>> The problem is the guest page is written before dirty-bitmap is set,
>> we may log the dirty page in this window like above case...
> 
> Exactly, but the next GET_DIRTY_LOG call can take that because, as I
> wrote above, at this time the GET_DIRTY_LOG must not be the final one.
> 


Thanks for your explanation, maybe you are right, i do not know migration
much.

What i worried about is, you have changed the behaviour of GET_DIRTY_LOG,
in the current one, it can get all the dirty pages when it is called; after
your change, GET_DIRTY_LOG can get a empty dirty bitmap but dirty page exists.

Migration may work correctly depends on the final GET_DIRTY_LOG, in that time,
guest is stopped. But i am not sure whether other components using GET_DIRTY_LOG
are happy, e.g. frame-buffer.