Re: [Qemu-devel] Re: [PATCH 3/3] Add KVM support to QEMU

[Glauber Costa <glommer@redhat.com>]

On Wed, Oct 29, 2008 at 02:39:57PM +0200, Avi Kivity wrote:
> Glauber Costa wrote:
>>>> Another place "hook" is updating a slot's dirty bitmap.  Right now, 
>>>>  with my patchset we don't have live migration or the VGA RAM   
>>>> optimization.  There's nothing about the VGA RAM optimization that  
>>>> wouldn't work for QEMU.  I'm not sure that it really is an   
>>>> optimization in the context of TCG, but I certainly don't think 
>>>> it's  any worse.  The only thing you really need is to query the 
>>>> KVM dirty  bitmap when it comes time to enable start over querying 
>>>> the VGA dirty  bits.
>>>>       
>>> I don't understand this.  The VGA optimization really is qemu's, the 
>>> kvm  modifications only cater to the different way of getting the 
>>> dirty bits.
>>>     
>>
>> As it seems to me, the real difference is that qemu has to explicitly set
>> certain regions as dirty, while kvm get dirty bit "automatically" from the kernel.
>>
>>   
>
> I'm completely lost.  I don't see how one or the other is more or less  
> automatic, or how qemu has to explicitly set regions as dirty (except  
> when emulating bitblt).
Or maybe I am. But I don't see any way in which qemu sets dirty bits but
explicitly with cpu_physical_memory_set_dirty(). This is pretty explicit.

>
>> So I believe we can have markers on the code to refresh dirty bitmap for certain
>> area ranges (for kvm use), and also enable a manual override (for qemu). After that,
>> the cpu_physical_memory_get_dirty() will simply return whether or not the page is
>> dirty.
>>   
>
> Does not cpu_p_m_g_dirty() simply return whether or not the page is  
> dirty now?

If you look at the vga code, you see something like:

    cpu_physical_memory_get_dirty(page0, VGA_DIRTY_FLAG) |
    cpu_physical_memory_get_dirty(page1, VGA_DIRTY_FLAG);
    if (kvm_enabled()) {
        update |= bitmap_get_dirty(bitmap, (page0 - s->vram_offset) >> TARGET_PAGE_BITS);
        update |= bitmap_get_dirty(bitmap, (page1 - s->vram_offset) >> TARGET_PAGE_BITS);
    }

so if the page is not dirty to cpu_p_m_g_dirty() (I liked that abb), it can still be dirty
for kvm. Ideally, it would not be necessary.

>
>> Also, kvm only tracks "dirty" bits, whereas qemu has at least three kinds of them.
>> But I think for now we can assume that kvm's dirty mean "all dirty
>
> kvm's dirty bits mean that kvm has seen the page written to since the  
> last query.  A zero doesn't mean the page is clean though -- it could  
> have been written to by qemu.

Right.
The point here is more like kvm has 1 type of dirty whereas qemu has many.