On Wed, 12 Feb 2014 23:07:15 +0000
Chris Wilson <ch...@chris-wilson.co.uk> wrote:
> On Wed, Feb 12, 2014 at 12:26:29PM -0800, Jesse Barnes wrote:
> > + /* Find the largest fb */
> > + list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
> > + intel_crtc = to_intel_crtc(crtc);
> > +
> > + if (!intel_crtc->active || !intel_crtc->plane_config.fb) {
> > + DRM_DEBUG_KMS("pipe %c not active or no fb, skipping\n",
> > + pipe_name(intel_crtc->pipe));
> > + continue;
> > + }
> > +
> > + if (intel_crtc->plane_config.size > max_size) {
> > + DRM_DEBUG_KMS("found possible fb from plane %c\n",
> > + pipe_name(intel_crtc->pipe));
> > + plane_config = &intel_crtc->plane_config;
> > + fb = plane_config->fb;
> > + max_size = plane_config->size;
> > + }
> > + }
>
> I had another idea... Can you select the fb with offset 0 first and see
> if that fits every pipe? That way we reduce the potential fragmentation
> from the permanently pinned in-aperture object.
So if they happen to be the same size we'll prefer the one at 0? I
guess we can tweak this all sorts of ways... but we'll generally
probably only have a single buffer from the BIOS, so I'm not sure it'll
make much difference in practice.
> > +
> > + if (!fb) {
> > + DRM_DEBUG_KMS("no active fbs found, not using BIOS config\n");
> > + goto out;
> > + }
> > +
> > + /* Now make sure all the pipes will fit into it */
> > + list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
> > + unsigned int cur_size;
> > +
> > + intel_crtc = to_intel_crtc(crtc);
> > +
> > + if (!intel_crtc->active) {
> > + DRM_DEBUG_KMS("pipe %c not active, skipping\n",
> > + pipe_name(intel_crtc->pipe));
> > + continue;
> > + }
> > +
> > + DRM_DEBUG_KMS("checking plane %c for BIOS fb\n",
> > + pipe_name(intel_crtc->pipe));
> > +
> > + /*
> > + * See if the plane fb we found above will fit on this
> > + * pipe. Note we need to use the selected fb's bpp rather
> > + * than the current pipe's, since they could be different.
> > + */
> > + cur_size = intel_crtc->config.adjusted_mode.crtc_hdisplay *
> > + intel_crtc->config.adjusted_mode.crtc_vdisplay;
> > + DRM_DEBUG_KMS("pipe %c area: %d\n", pipe_name(intel_crtc->pipe),
> > + cur_size);
> > + cur_size *= fb->base.bits_per_pixel / 8;
> > + DRM_DEBUG_KMS("total size %d (bpp %d)\n", cur_size,
> > + fb->base.bits_per_pixel / 8);
>
> But the display hardware will be using the stride from the fb, so the
> end of the last row will still be vdisplay*fb->stride, right? Unless the
> fb pitch itself is recomputed to match?
I think there are two cases:
1) pipe A re-uses plane A's fb - no changes needed, should match
nicely
2) pipe B switches to plane A's fb - we'll end up doing a set_base and
switching the stride bits and depth over to the plane A fb params,
so it should work
right?
--
Jesse Barnes, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
