drm/radeon: Don't generate new fence for page flip.
authorMichel Dänzer <michel.daenzer@amd.com>
Wed, 13 Jul 2011 15:18:09 +0000 (15:18 +0000)
committerDave Airlie <airlied@redhat.com>
Thu, 14 Jul 2011 06:14:42 +0000 (07:14 +0100)
Use the fence of the new frontbuffer, if any.

Generating a new fence could cause us to wait for completely unrelated
rendering to finish before performing the flip.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
drivers/gpu/drm/radeon/radeon_display.c

index 292f73f0ddbd51210b8bf3494acdd79dc3d54719..ffce85a9116583ee58f6798b33350bdd99fd6e3a 100644 (file)
@@ -282,7 +282,7 @@ void radeon_crtc_handle_flip(struct radeon_device *rdev, int crtc_id)
        spin_lock_irqsave(&rdev->ddev->event_lock, flags);
        work = radeon_crtc->unpin_work;
        if (work == NULL ||
-           !radeon_fence_signaled(work->fence)) {
+           (work->fence && !radeon_fence_signaled(work->fence))) {
                spin_unlock_irqrestore(&rdev->ddev->event_lock, flags);
                return;
        }
@@ -348,7 +348,6 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
        struct radeon_framebuffer *new_radeon_fb;
        struct drm_gem_object *obj;
        struct radeon_bo *rbo;
-       struct radeon_fence *fence;
        struct radeon_unpin_work *work;
        unsigned long flags;
        u32 tiling_flags, pitch_pixels;
@@ -359,16 +358,9 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
        if (work == NULL)
                return -ENOMEM;
 
-       r = radeon_fence_create(rdev, &fence);
-       if (unlikely(r != 0)) {
-               kfree(work);
-               DRM_ERROR("flip queue: failed to create fence.\n");
-               return -ENOMEM;
-       }
        work->event = event;
        work->rdev = rdev;
        work->crtc_id = radeon_crtc->crtc_id;
-       work->fence = radeon_fence_ref(fence);
        old_radeon_fb = to_radeon_framebuffer(crtc->fb);
        new_radeon_fb = to_radeon_framebuffer(fb);
        /* schedule unpin of the old buffer */
@@ -377,6 +369,10 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
        drm_gem_object_reference(obj);
        rbo = gem_to_radeon_bo(obj);
        work->old_rbo = rbo;
+       obj = new_radeon_fb->obj;
+       rbo = gem_to_radeon_bo(obj);
+       if (rbo->tbo.sync_obj)
+               work->fence = radeon_fence_ref(rbo->tbo.sync_obj);
        INIT_WORK(&work->work, radeon_unpin_work_func);
 
        /* We borrow the event spin lock for protecting unpin_work */
@@ -391,9 +387,6 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
        spin_unlock_irqrestore(&dev->event_lock, flags);
 
        /* pin the new buffer */
-       obj = new_radeon_fb->obj;
-       rbo = gem_to_radeon_bo(obj);
-
        DRM_DEBUG_DRIVER("flip-ioctl() cur_fbo = %p, cur_bbo = %p\n",
                         work->old_rbo, rbo);
 
@@ -461,25 +454,11 @@ static int radeon_crtc_page_flip(struct drm_crtc *crtc,
                goto pflip_cleanup1;
        }
 
-       /* 32 ought to cover us */
-       r = radeon_ring_lock(rdev, 32);
-       if (r) {
-               DRM_ERROR("failed to lock the ring before flip\n");
-               goto pflip_cleanup2;
-       }
-
-       /* emit the fence */
-       radeon_fence_emit(rdev, fence);
        /* set the proper interrupt */
        radeon_pre_page_flip(rdev, radeon_crtc->crtc_id);
-       /* fire the ring */
-       radeon_ring_unlock_commit(rdev);
 
        return 0;
 
-pflip_cleanup2:
-       drm_vblank_put(dev, radeon_crtc->crtc_id);
-
 pflip_cleanup1:
        r = radeon_bo_reserve(rbo, false);
        if (unlikely(r != 0)) {
@@ -501,7 +480,7 @@ pflip_cleanup:
 unlock_free:
        drm_gem_object_unreference_unlocked(old_radeon_fb->obj);
        spin_unlock_irqrestore(&dev->event_lock, flags);
-       radeon_fence_unref(&fence);
+       radeon_fence_unref(&work->fence);
        kfree(work);
 
        return r;