Re: [git pull] drm patches for 2.6.27-rc1

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> [ The *non-atomic* kmap() functions are fairly high-overhead, in that they 
>   want to keep track of cached mappings and remember page addresses etc. 
>   So those are the ones we don't want to support for non-HIGHMEM setups. 
> 
>   But the atomic kmaps are pretty simple, and really only need some 
>   trivial FIXMAP support. We could easily extend it for x86-64, methinks, 
>   and do it for x86-32 even when we don't do HIGHMEM.
> 
>   Ingo? ]

agreed, and there's certainly no resistance from the x86 architecture 
side to extend our mapping APIs in such directions.

But i think the direction of the new GEM code is subtly wrong here, 
because it tries to manage memory even on 64-bit systems. IMO it should 
just map the _whole_ graphics aperture (non-cached) and be done with it. 
There's no faster method at managing pages than the CPU doing a TLB fill 
from pagetables.

The only real API need i see is on 32-bit: with a 1GB or 2GB graphics 
aperture we just cannot map that permanently, so kmap_atomic() is a 
necessity. We can certainly extend that to non-highmem as well.

But this should be an ad-hoc transitionary thing for 32-bit, and on 
64-bit we really should not be using any form of kmap.

Especially with large vertex buffers or textures, mapping a lot of pages 
via kmap is not going to be trivial overhead - even if INVLPG is faster 
than a full TLB flush, it's still on the order of 100-200 cycles - and 
with a lot of pages that mounts up quickly. And if i understood your 
workload correctly you want to do tens of thousand of map/unmap/remap 
events per frame generated - depending on the type of the 3D app/engine.

Or am i missing something subtle? Why do you want the overhead of kmap 
on 64-bit?

	Ingo