[Qemu-devel] Re: [PATCH 3/3] disk: don't read from disk until the guest starts

[Juan Quintela]

Anthony Liguori <address@hidden> wrote:
> On 09/12/2010 05:42 AM, Avi Kivity wrote:
>>  On 09/11/2010 05:04 PM, Anthony Liguori wrote:
>>> This fixes a couple nasty problems relating to live migration.
>>>
>>> 1) When dealing with shared storage with weak coherence (i.e. NFS),
>>> even if
>>>     we re-read, we may end up with undesired caching.  By delaying
>>> any reads
>>>     until we absolutely have to, we decrease the likelihood of any
>>> undesirable
>>>     caching.
>>>
>>> 2) When dealing with copy-on-read, the local storage acts as a
>>> cache.  We need
>>>     to make sure to avoid any reads to avoid polluting the local cache.
>>>
>>> +
>>>   static void ide_identify(IDEState *s)
>>>   {
>>>       uint16_t *p;
>>> @@ -105,6 +132,8 @@ static void ide_identify(IDEState *s)
>>>       return;
>>>       }
>>>
>>> +    guess_geometry(s);
>>> +
>>
>> This can cause a disk read, no?  Shouldn't it be made asynchronous?
>
> Yes, it should.  I'm not sure there's an easy way to make it
> asynchronous though not because of the block layer but because of how
> these functions are called.
>
>>
>> Or just move it to just before the guest starts?
>
> We don't really have a notion of "guest starts" today although maybe
> we should.

I agree that we should have more states.  Expecting Migration/Migration
done are quite special casses that are handled adhoc now.  Having
some state like: NeverRun will make trivial to have vm_start() do the
right thing on the 1st run.

Later, Juan.

> Regards,
>
> Anthony Liguori