On Wed, Nov 28, 2012 at 6:13 PM, Jerome Glisse j.glisse@gmail.com wrote:
On Wed, Nov 28, 2012 at 5:18 PM, Thomas Hellstrom thomas@shipmail.org wrote:
On 11/28/2012 09:09 PM, Jerome Glisse wrote:
On Wed, Nov 28, 2012 at 10:05 AM, j.glisse@gmail.com wrote:
From: Jerome Glisse jglisse@redhat.com
It make no sense to preserve caching state especialy when moving from vram to system. It burden the page allocator to match the vram caching (often WC) which just burn CPU cycle for no good reasons.
Nack. This is a driver problem.
What happens with this patch if you evict write-combined TT memory to system? That's why we want to preserve caching state in the first place.
If you need a different behavior, you can fine-tune in driver::evict_flags, or in the radeon-case in radeon_move_ram_vram / radeon_move_vram_ram.
No you can't. Because cur placement of vram will be wc, and thus it will want to preserve it for gtt. Only way to force it is to declare vram as cached too.
Cheers, Jerome
Now that i think a bit more to it, in all fairness you can do so from the driver side.
Cheers, Jerome