Messages in this thread | | | Date | Fri, 20 Nov 2009 19:04:45 +0100 | From | Jerome Glisse <> | Subject | Re: [PATCH 3/3] drm/radeon/kms: Add range pinning to crtc/cursor bo |
| |
On Fri, Nov 20, 2009 at 11:47:22AM -0500, Alex Deucher wrote: > On Fri, Nov 20, 2009 at 8:29 AM, Jerome Glisse <jglisse@redhat.com> wrote: > > We force the crtc & cursor bo to be in the first 64M, this > > will help on legacy modesetting hw where the offset of > > scanout buffer and cursor are relative to a base address. > > This limitation only applies to pre-avivo (r1xx-r4xx) chips, there's > no need on newer hardware. Also the cursor and crtc have to be within > 128 MB of each other, so you could bump the limits to 128 MB. Another > thing to consider down the road might be to treat memory over 128 MB > as "invisible" on these pre-avivo chips (once we support that in > general) then we no longer need this limit. > > Alex >
I choosed 64M in hope to disminish VRAM fragmentation a little, this is also why i did not bother to make this conditional for legacy modesetting hw. For supporting non visible vram see my other ttm thread. I am working on adding this knowledge to ttm.
Cheers, Jerome
| |