OSDN Git Service

drm/amd/display: Refactor pageflips plane commit
authorDavid Francis <David.Francis@amd.com>
Wed, 13 Feb 2019 18:20:11 +0000 (13:20 -0500)
committerAlex Deucher <alexander.deucher@amd.com>
Tue, 19 Mar 2019 20:04:03 +0000 (15:04 -0500)
[Why]
commit_planes is indented quite far

[How]
Move the pageflip code from an if statement to after a
continue

Signed-off-by: David Francis <David.Francis@amd.com>
Reviewed-by: Nicholas Kazlauskas <Nicholas.Kazlauskas@amd.com>
Acked-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c

index 38fa2ce..3bf1157 100644 (file)
@@ -4712,74 +4712,6 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
 
                dc_plane = dm_new_plane_state->dc_state;
 
-               plane_needs_flip = old_plane_state->fb && new_plane_state->fb;
-
-               pflip_present = pflip_present || plane_needs_flip;
-
-               if (plane_needs_flip) {
-                       /*
-                        * TODO This might fail and hence better not used, wait
-                        * explicitly on fences instead
-                        * and in general should be called for
-                        * blocking commit to as per framework helpers
-                        */
-                       abo = gem_to_amdgpu_bo(fb->obj[0]);
-                       r = amdgpu_bo_reserve(abo, true);
-                       if (unlikely(r != 0))
-                               DRM_ERROR("failed to reserve buffer before flip\n");
-
-                       /*
-                        * Wait for all fences on this FB. Do limited wait to avoid
-                        * deadlock during GPU reset when this fence will not signal
-                        * but we hold reservation lock for the BO.
-                        */
-                       r = reservation_object_wait_timeout_rcu(abo->tbo.resv,
-                                                               true, false,
-                                                               msecs_to_jiffies(5000));
-                       if (unlikely(r == 0))
-                               DRM_ERROR("Waiting for fences timed out.");
-
-
-
-                       amdgpu_bo_get_tiling_flags(abo, &tiling_flags);
-
-                       amdgpu_bo_unreserve(abo);
-
-                       bundle->flip_addrs[planes_count].address.grph.addr.low_part = lower_32_bits(afb->address);
-                       bundle->flip_addrs[planes_count].address.grph.addr.high_part = upper_32_bits(afb->address);
-
-                       dcc_address = get_dcc_address(afb->address, tiling_flags);
-                       bundle->flip_addrs[planes_count].address.grph.meta_addr.low_part = lower_32_bits(dcc_address);
-                       bundle->flip_addrs[planes_count].address.grph.meta_addr.high_part = upper_32_bits(dcc_address);
-
-                       bundle->flip_addrs[planes_count].flip_immediate =
-                                       (crtc->state->pageflip_flags & DRM_MODE_PAGE_FLIP_ASYNC) != 0;
-
-                       timestamp_ns = ktime_get_ns();
-                       bundle->flip_addrs[planes_count].flip_timestamp_in_us = div_u64(timestamp_ns, 1000);
-                       bundle->surface_updates[planes_count].flip_addr = &bundle->flip_addrs[planes_count];
-                       bundle->surface_updates[planes_count].surface = dc_plane;
-
-                       if (!bundle->surface_updates[planes_count].surface) {
-                               DRM_ERROR("No surface for CRTC: id=%d\n",
-                                               acrtc_attach->crtc_id);
-                               continue;
-                       }
-
-                       if (plane == pcrtc->primary)
-                               update_freesync_state_on_stream(
-                                       dm,
-                                       acrtc_state,
-                                       acrtc_state->stream,
-                                       dc_plane,
-                                       bundle->flip_addrs[planes_count].flip_timestamp_in_us);
-
-                       DRM_DEBUG_DRIVER("%s Flipping to hi: 0x%x, low: 0x%x\n",
-                                        __func__,
-                                        bundle->flip_addrs[planes_count].address.grph.addr.high_part,
-                                        bundle->flip_addrs[planes_count].address.grph.addr.low_part);
-               }
-
                bundle->surface_updates[planes_count].surface = dc_plane;
                if (new_pcrtc_state->color_mgmt_changed) {
                        bundle->surface_updates[planes_count].gamma = dc_plane->gamma_correction;
@@ -4806,6 +4738,70 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
                bundle->plane_infos[planes_count].dcc = dc_plane->dcc;
                bundle->surface_updates[planes_count].plane_info = &bundle->plane_infos[planes_count];
 
+               plane_needs_flip = old_plane_state->fb && new_plane_state->fb;
+
+               pflip_present = pflip_present || plane_needs_flip;
+
+               if (!plane_needs_flip) {
+                       planes_count += 1;
+                       continue;
+               }
+
+               /*
+                * TODO This might fail and hence better not used, wait
+                * explicitly on fences instead
+                * and in general should be called for
+                * blocking commit to as per framework helpers
+                */
+               abo = gem_to_amdgpu_bo(fb->obj[0]);
+               r = amdgpu_bo_reserve(abo, true);
+               if (unlikely(r != 0)) {
+                       DRM_ERROR("failed to reserve buffer before flip\n");
+                       WARN_ON(1);
+               }
+
+               /* Wait for all fences on this FB */
+               WARN_ON(reservation_object_wait_timeout_rcu(abo->tbo.resv, true, false,
+                                                                           MAX_SCHEDULE_TIMEOUT) < 0);
+
+               amdgpu_bo_get_tiling_flags(abo, &tiling_flags);
+
+               amdgpu_bo_unreserve(abo);
+
+               bundle->flip_addrs[planes_count].address.grph.addr.low_part = lower_32_bits(afb->address);
+               bundle->flip_addrs[planes_count].address.grph.addr.high_part = upper_32_bits(afb->address);
+
+               dcc_address = get_dcc_address(afb->address, tiling_flags);
+               bundle->flip_addrs[planes_count].address.grph.meta_addr.low_part = lower_32_bits(dcc_address);
+               bundle->flip_addrs[planes_count].address.grph.meta_addr.high_part = upper_32_bits(dcc_address);
+
+               bundle->flip_addrs[planes_count].flip_immediate =
+                               (crtc->state->pageflip_flags & DRM_MODE_PAGE_FLIP_ASYNC) != 0;
+
+               timestamp_ns = ktime_get_ns();
+               bundle->flip_addrs[planes_count].flip_timestamp_in_us = div_u64(timestamp_ns, 1000);
+               bundle->surface_updates[planes_count].flip_addr = &bundle->flip_addrs[planes_count];
+               bundle->surface_updates[planes_count].surface = dc_plane;
+
+               if (!bundle->surface_updates[planes_count].surface) {
+                       DRM_ERROR("No surface for CRTC: id=%d\n",
+                                       acrtc_attach->crtc_id);
+                       continue;
+               }
+
+               if (plane == pcrtc->primary)
+                       update_freesync_state_on_stream(
+                               dm,
+                               acrtc_state,
+                               acrtc_state->stream,
+                               dc_plane,
+                               bundle->flip_addrs[planes_count].flip_timestamp_in_us);
+
+               DRM_DEBUG_DRIVER("%s Flipping to hi: 0x%x, low: 0x%x\n",
+                                __func__,
+                                bundle->flip_addrs[planes_count].address.grph.addr.high_part,
+                                bundle->flip_addrs[planes_count].address.grph.addr.low_part);
+
                planes_count += 1;
 
        }