Re: [Qemu-devel] vga optmization

[Stefano Stabellini <stefano.stabellini@eu.citrix.com>]

Glauber Costa wrote:

> On Wed, Nov 05, 2008 at 02:42:22PM +0000, Stefano Stabellini wrote:
>> Glauber Costa wrote:
>>
>>>> That's also how currently qemu-xen works.
>>>> I am glad that we agree :)
>>> I'm attaching a new version. Let me know if it's better this way.
>>>
>>
>>
>> Yes, this is much better thanks.
>>
>> I still have comments about possible improvements, so I wrote patch
>> (against your patch).
>> This patch is not meant to be applied, is only meant to be read (I believe
>> that C code is more meaningful than English :).
>>
>> Some of the changes include:
>>
>> - instead of adding cirrus_lfb_addr, add a more generic lfb_addr to
>> VGAState, so that can be reused in the future for possible stdvga only
>> mappings;
>>
>> - instead of using cirrus_lfb_mapped as a boolean, use it as the
>> mapping address, it is more useful that way;
>>
>> - instead of keeping the kvm dirty map always enabled, enable it only
>> when the framebuffer is linear and in graphical mode;
>>
>> - look at the changes to vga.c, there is a simple check to reduce the
>> dirty area to sync.
>>
>> It would be nice to check if the last two changes are actually a
>> performance improvement.
> I'm not opposed to the changes. I'll redo a patch and resend it today.
> However, I'm not planning to include the last optimization you sent.
> It is better that we do it in a separate commit. If you can get numbers,
> even better. I'll try it myself too.

Fine with me.