context_drm_egl: implement n-buffering

This allows context_drm_egl to use as many buffers as libgbm or the
swapchain_depth setting allows (whichever is smaller).

On pause and on still images (cover art etc.) to make sure that output does not
lag behind user input, the swapchain is drained and reverts to working in a dual
buffered (equivalent to swapchain-depth=1) manner.

When possible (swapchain-depth>=2), the wait on the page flip event is now not
done immediately after queueing, but is deferred to the next invocation of
swap_buffers. Which should give us more CPU time between invocations.

Although, since gbm_surface_has_free_buffers() can only tell us a boolean value
and not how many buffers we have left, we are forced to do this contortionist
dance where we first overshoot until gbm_surface_has_free_buffers() reports 0,
followed by immediately waiting so we can free a buffer, to be able to get the
deferred wait on page flip rolling.

With this commit we do not rely on the default vsync fences/latency emulation of
video/out/opengl/context.c, but supply our own, since the places we create and
wait for the fences needs to be somewhat different for best performance.

Minor fixes:

 * According to GBM documentation all BO:s gotten with
   gbm_surface_lock_front_buffer must be released before gbm_surface_destroy is
   called on the surface.
 * We let the page flip handler function handle the waiting_for_flip flag.

1 files changed, 150 insertions, 59 deletions

diff --git a/video/out/opengl/context_drm_egl.c b/video/out/opengl/context_drm_egl.c
index 6aa3d95e79..7e14155d28 100644
--- a/video/out/opengl/context_drm_egl.c
+++ b/video/out/opengl/context_drm_egl.c
@@ -50,8 +50,8 @@ struct gbm
 {
     struct gbm_surface *surface;
     struct gbm_device *device;
-    struct gbm_bo *bo;
-    struct gbm_bo *next_bo;
+    struct gbm_bo **bo;
+    unsigned int num_bos;
 };
 
 struct egl
@@ -72,6 +72,9 @@ struct priv {
     struct gbm gbm;
     struct framebuffer *fb;
 
+    GLsync *vsync_fences;
+    unsigned int num_vsync_fences;
+
     uint32_t gbm_format;
 
     bool active;
@@ -80,6 +83,9 @@ struct priv {
     bool vt_switcher_active;
     struct vt_switcher vt_switcher;
 
+    bool still;
+    bool paused;
+
     struct mpv_opengl_drm_params drm_params;
     struct mpv_opengl_drm_draw_surface_size draw_surface_size;
 };
@@ -355,15 +361,6 @@ static void crtc_release(struct ra_ctx *ctx)
         return;
     p->active = false;
 
-    // wait for current page flip
-    while (p->waiting_for_flip) {
-        int ret = drmHandleEvent(p->kms->fd, &p->ev);
-        if (ret) {
-            MP_ERR(ctx->vo, "drmHandleEvent failed: %i\n", ret);
-            break;
-        }
-    }
-
     if (p->kms->atomic_context) {
         if (p->kms->atomic_context->old_state.saved) {
             if (!crtc_release_atomic(ctx))
@@ -414,44 +411,19 @@ static void acquire_vt(void *data)
     crtc_setup(ctx);
 }
 
-static bool drm_atomic_egl_start_frame(struct ra_swapchain *sw, struct ra_fbo *out_fbo)
-{
-    struct priv *p = sw->ctx->priv;
-    if (p->kms->atomic_context) {
-        if (!p->kms->atomic_context->request) {
-            p->kms->atomic_context->request = drmModeAtomicAlloc();
-            p->drm_params.atomic_request_ptr = &p->kms->atomic_context->request;
-        }
-        return ra_gl_ctx_start_frame(sw, out_fbo);
-    }
-    return false;
-}
-
-static const struct ra_swapchain_fns drm_atomic_swapchain = {
-    .start_frame   = drm_atomic_egl_start_frame,
-};
-
-static void drm_egl_swap_buffers(struct ra_ctx *ctx)
+static void queue_flip(struct ra_ctx *ctx)
 {
     struct priv *p = ctx->priv;
     struct drm_atomic_context *atomic_ctx = p->kms->atomic_context;
     int ret;
 
-    if (!p->active)
-        return;
-
-    eglSwapBuffers(p->egl.display, p->egl.surface);
-    p->gbm.next_bo = gbm_surface_lock_front_buffer(p->gbm.surface);
-    p->waiting_for_flip = true;
-    update_framebuffer_from_bo(ctx, p->gbm.next_bo);
-
     if (atomic_ctx) {
         drm_object_set_property(atomic_ctx->request, atomic_ctx->draw_plane, "FB_ID", p->fb->id);
         drm_object_set_property(atomic_ctx->request, atomic_ctx->draw_plane, "CRTC_ID", atomic_ctx->crtc->id);
         drm_object_set_property(atomic_ctx->request, atomic_ctx->draw_plane, "ZPOS", 1);
 
         ret = drmModeAtomicCommit(p->kms->fd, atomic_ctx->request,
-                                  DRM_MODE_ATOMIC_NONBLOCK | DRM_MODE_PAGE_FLIP_EVENT, NULL);
+                                  DRM_MODE_ATOMIC_NONBLOCK | DRM_MODE_PAGE_FLIP_EVENT, p);
         if (ret)
             MP_WARN(ctx->vo, "Failed to commit atomic request (%d)\n", ret);
     } else {
@@ -461,30 +433,129 @@ static void drm_egl_swap_buffers(struct ra_ctx *ctx)
             MP_WARN(ctx->vo, "Failed to queue page flip: %s\n", mp_strerror(errno));
         }
     }
+    p->waiting_for_flip = true;
+
+    if (atomic_ctx) {
+        drmModeAtomicFree(atomic_ctx->request);
+        atomic_ctx->request = drmModeAtomicAlloc();
+    }
+}
+
+static void wait_on_flip(struct ra_ctx *ctx)
+{
+    struct priv *p = ctx->priv;
 
     // poll page flip finish event
-    const int timeout_ms = 3000;
-    struct pollfd fds[1] = { { .events = POLLIN, .fd = p->kms->fd } };
-    poll(fds, 1, timeout_ms);
-    if (fds[0].revents & POLLIN) {
-        ret = drmHandleEvent(p->kms->fd, &p->ev);
-        if (ret != 0) {
-            MP_ERR(ctx->vo, "drmHandleEvent failed: %i\n", ret);
-            p->waiting_for_flip = false;
-            return;
+    while (p->waiting_for_flip) {
+        const int timeout_ms = 3000;
+        struct pollfd fds[1] = { { .events = POLLIN, .fd = p->kms->fd } };
+        poll(fds, 1, timeout_ms);
+        if (fds[0].revents & POLLIN) {
+            const int ret = drmHandleEvent(p->kms->fd, &p->ev);
+            if (ret != 0)
+                MP_ERR(ctx->vo, "drmHandleEvent failed: %i\n", ret);
         }
     }
-    p->waiting_for_flip = false;
+}
 
-    if (atomic_ctx) {
-        drmModeAtomicFree(atomic_ctx->request);
-        atomic_ctx->request = drmModeAtomicAlloc();
+static void swapchain_step(struct ra_ctx *ctx)
+{
+    struct priv *p = ctx->priv;
+
+    if (p->gbm.num_bos > 0 && p->gbm.bo[0])
+        gbm_surface_release_buffer(p->gbm.surface, p->gbm.bo[0]);
+    MP_TARRAY_REMOVE_AT(p->gbm.bo, p->gbm.num_bos, 0);
+}
+
+static void new_fence(struct ra_ctx *ctx)
+{
+    struct priv *p = ctx->priv;
+
+    if (p->gl.FenceSync) {
+        GLsync fence = p->gl.FenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
+        if (fence)
+            MP_TARRAY_APPEND(p, p->vsync_fences, p->num_vsync_fences, fence);
     }
+}
+
+static void wait_fence(struct ra_ctx *ctx)
+{
+    struct priv *p = ctx->priv;
+
+    while (p->num_vsync_fences && (p->num_vsync_fences >= p->gbm.num_bos)) {
+        p->gl.ClientWaitSync(p->vsync_fences[0], GL_SYNC_FLUSH_COMMANDS_BIT, 1e9);
+        p->gl.DeleteSync(p->vsync_fences[0]);
+        MP_TARRAY_REMOVE_AT(p->vsync_fences, p->num_vsync_fences, 0);
+    }
+}
+
+static bool drm_egl_start_frame(struct ra_swapchain *sw, struct ra_fbo *out_fbo)
+{
+    struct ra_ctx *ctx = sw->ctx;
+    struct priv *p = ctx->priv;
+
+    if (p->kms->atomic_context && !p->kms->atomic_context->request) {
+        p->kms->atomic_context->request = drmModeAtomicAlloc();
+        p->drm_params.atomic_request_ptr = &p->kms->atomic_context->request;
+    }
+
+    return ra_gl_ctx_start_frame(sw, out_fbo);
+}
+
+static bool drm_egl_submit_frame(struct ra_swapchain *sw, const struct vo_frame *frame)
+{
+    struct ra_ctx *ctx = sw->ctx;
+    struct priv *p = ctx->priv;
 
-    gbm_surface_release_buffer(p->gbm.surface, p->gbm.bo);
-    p->gbm.bo = p->gbm.next_bo;
+    p->still = frame->still;
+
+    return ra_gl_ctx_submit_frame(sw, frame);
 }
 
+static void drm_egl_swap_buffers(struct ra_swapchain *sw)
+{
+    struct ra_ctx *ctx = sw->ctx;
+    struct priv *p = ctx->priv;
+    const bool drain = p->paused || p->still;  // True when we need to drain the swapchain
+
+    if (!p->active)
+        return;
+
+    wait_fence(ctx);
+
+    eglSwapBuffers(p->egl.display, p->egl.surface);
+
+    struct gbm_bo *new_bo = gbm_surface_lock_front_buffer(p->gbm.surface);
+    if (!new_bo) {
+        MP_ERR(ctx->vo, "Couldn't lock front buffer\n");
+        return;
+    }
+    MP_TARRAY_APPEND(p, p->gbm.bo, p->gbm.num_bos, new_bo);
+    new_fence(ctx);
+
+    while (drain || p->gbm.num_bos > ctx->opts.swapchain_depth || !gbm_surface_has_free_buffers(p->gbm.surface)) {
+        if (p->waiting_for_flip) {
+            wait_on_flip(ctx);
+            swapchain_step(ctx);
+        }
+        if (p->gbm.num_bos <= 1)
+            break;
+        if (!p->gbm.bo[1]) {
+            MP_ERR(ctx->vo, "Hole in swapchain?\n");
+            swapchain_step(ctx);
+            continue;
+        }
+        update_framebuffer_from_bo(ctx, p->gbm.bo[1]);
+        queue_flip(ctx);
+    }
+}
+
+static const struct ra_swapchain_fns drm_egl_swapchain = {
+    .start_frame   = drm_egl_start_frame,
+    .submit_frame  = drm_egl_submit_frame,
+    .swap_buffers  = drm_egl_swap_buffers,
+};
+
 static void drm_egl_uninit(struct ra_ctx *ctx)
 {
     struct priv *p = ctx->priv;
@@ -503,6 +574,12 @@ static void drm_egl_uninit(struct ra_ctx *ctx)
     if (p->vt_switcher_active)
         vt_switcher_destroy(&p->vt_switcher);
 
+    // According to GBM documentation all BO:s must be released before
+    // gbm_surface_destroy can be called on the surface.
+    while (p->gbm.num_bos) {
+        swapchain_step(ctx);
+    }
+
     eglMakeCurrent(p->egl.display, EGL_NO_SURFACE, EGL_NO_SURFACE,
                    EGL_NO_CONTEXT);
     eglDestroyContext(p->egl.display, p->egl.context);
@@ -565,6 +642,13 @@ static bool probe_gbm_format(struct ra_ctx *ctx, uint32_t argb_format, uint32_t
     return result;
 }
 
+static void page_flipped(int fd, unsigned int frame, unsigned int sec,
+                         unsigned int usec, void *data)
+{
+    struct priv *p = data;
+    p->waiting_for_flip = false;
+}
+
 static bool drm_egl_init(struct ra_ctx *ctx)
 {
     if (ctx->opts.probing) {
@@ -574,6 +658,7 @@ static bool drm_egl_init(struct ra_ctx *ctx)
 
     struct priv *p = ctx->priv = talloc_zero(ctx, struct priv);
     p->ev.version = DRM_EVENT_CONTEXT_VERSION;
+    p->ev.page_flip_handler = page_flipped;
 
     p->vt_switcher_active = vt_switcher_init(&p->vt_switcher, ctx->vo->log);
     if (p->vt_switcher_active) {
@@ -644,12 +729,13 @@ static bool drm_egl_init(struct ra_ctx *ctx)
     eglSwapBuffers(p->egl.display, p->egl.surface);
 
     MP_VERBOSE(ctx, "Preparing framebuffer\n");
-    p->gbm.bo = gbm_surface_lock_front_buffer(p->gbm.surface);
-    if (!p->gbm.bo) {
+    struct gbm_bo *new_bo = gbm_surface_lock_front_buffer(p->gbm.surface);
+    if (!new_bo) {
         MP_ERR(ctx, "Failed to lock GBM surface.\n");
         return false;
     }
-    update_framebuffer_from_bo(ctx, p->gbm.bo);
+    MP_TARRAY_APPEND(p, p->gbm.bo, p->gbm.num_bos, new_bo);
+    update_framebuffer_from_bo(ctx, new_bo);
     if (!p->fb || !p->fb->id) {
         MP_ERR(ctx, "Failed to create framebuffer.\n");
         return false;
@@ -681,9 +767,7 @@ static bool drm_egl_init(struct ra_ctx *ctx)
     }
 
     struct ra_gl_ctx_params params = {
-        .swap_buffers = drm_egl_swap_buffers,
-        .external_swapchain = p->kms->atomic_context ? &drm_atomic_swapchain :
-                                                       NULL,
+        .external_swapchain = &drm_egl_swapchain,
     };
     if (!ra_gl_ctx_init(ctx, &p->gl, params))
         return false;
@@ -715,6 +799,13 @@ static int drm_egl_control(struct ra_ctx *ctx, int *events, int request,
         *(double*)arg = fps;
         return VO_TRUE;
     }
+    case VOCTRL_PAUSE:
+        ctx->vo->want_redraw = true;
+        p->paused = true;
+        return VO_TRUE;
+    case VOCTRL_RESUME:
+        p->paused = false;
+        return VO_TRUE;
     }
     return VO_NOTIMPL;
 }