On Tue, Feb 17, 2009 at 23:18, Maarten Maathuis <[email protected]> wrote:
> On Tue, Feb 17, 2009 at 11:01 PM, Keith Packard <[email protected]> wrote:
>> On Tue, 2009-02-17 at 16:04 -0500, Adam Jackson wrote:
>>
>>> So, your kernel memory manager is done then?
>>
>> My point was that we've demonstrated for several years that the legacy
>> user-mode memory management stuff is pretty much unworkable. Instead of
>> investing any time attempting to make it go faster, we should simply
>> make it work correctly and then work on switching drivers to kernel-mode
>> memory management.
>>
>> --
>> [email protected]
>>
>> _______________________________________________
>> xorg-devel mailing list
>> [email protected]
>> http://lists.x.org/mailman/listinfo/xorg-devel
>>
>>
>
> I personally think that exa's memory manager isn't stupid, it's just
> unsuited to non-linear memory. Improving existing code has the nice
> side effect of knowing what bottlenecks exist, since you'll most
> likely encounter them again when switching to a more sophisticated
> memory manager. It's quite possible that intel uma's perform poorly in
> combination with exa_offscreen/migration, but that doesn't make it
> horrible for other hardware. I wouldn't be surprised if quite a few of
> the optimisations in classic exa also make it into future memory
> managers (obviously on the userspace side of things).
>

Yeah, things like initial pixmap placement and migration hints still
make sense in a system with kernel memory management. And you can't
guess the right placement or migration decisions in the kernel, user
space has a better view of the situation. The current system lets us
figure those out nicely.

Stéphane
_______________________________________________
xorg-devel mailing list
[email protected]
http://lists.x.org/mailman/listinfo/xorg-devel

Reply via email to