File _service:download_url:1441.patch of Package mutter

Overview Repositories Revisions Requests Users Attributes Meta

File _service:download_url:1441.patch of Package mutter (Revision e09bea664856ffb3db92802976cbb216)

Currently displaying revision e09bea664856ffb3db92802976cbb216 , Show latest

From b3b1fa7d2a9fd4459c2bf36c75e031f6bd04f536 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Fri, 17 Sep 2021 17:48:20 +0800
Subject: [PATCH 01/32] cogl/onscreen: Add function
 cogl_onscreen_count_pending_frames

---
 cogl/cogl/cogl-onscreen-private.h | 3 +++
 cogl/cogl/cogl-onscreen.c         | 8 ++++++++
 2 files changed, 11 insertions(+)

diff --git a/cogl/cogl/cogl-onscreen-private.h b/cogl/cogl/cogl-onscreen-private.h
index 959a60533b3..785f67aca16 100644
--- a/cogl/cogl/cogl-onscreen-private.h
+++ b/cogl/cogl/cogl-onscreen-private.h
@@ -79,3 +79,6 @@ cogl_onscreen_peek_tail_frame_info (CoglOnscreen *onscreen);
 
 COGL_EXPORT CoglFrameInfo *
 cogl_onscreen_pop_head_frame_info (CoglOnscreen *onscreen);
+
+COGL_EXPORT unsigned int
+cogl_onscreen_count_pending_frames (CoglOnscreen *onscreen);
diff --git a/cogl/cogl/cogl-onscreen.c b/cogl/cogl/cogl-onscreen.c
index f4b460a83f1..3e3f73a69fa 100644
--- a/cogl/cogl/cogl-onscreen.c
+++ b/cogl/cogl/cogl-onscreen.c
@@ -511,6 +511,14 @@ cogl_onscreen_pop_head_frame_info (CoglOnscreen *onscreen)
   return g_queue_pop_head (&priv->pending_frame_infos);
 }
 
+unsigned int
+cogl_onscreen_count_pending_frames (CoglOnscreen *onscreen)
+{
+  CoglOnscreenPrivate *priv = cogl_onscreen_get_instance_private (onscreen);
+
+  return g_queue_get_length (&priv->pending_frame_infos);
+}
+
 CoglFrameClosure *
 cogl_onscreen_add_frame_callback (CoglOnscreen *onscreen,
                                   CoglFrameCallback callback,
-- 
GitLab

From c9d4950a91530adfd732c58a03eb60798515b793 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Mon, 12 Feb 2024 17:16:03 +0800
Subject: [PATCH 02/32] cogl/onscreen: Indent declaration parameters to align
 with above/below

This fixes warnings from check-code-style.
---
 cogl/cogl/cogl-onscreen-private.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/cogl/cogl/cogl-onscreen-private.h b/cogl/cogl/cogl-onscreen-private.h
index 785f67aca16..86d8ea2d5ff 100644
--- a/cogl/cogl/cogl-onscreen-private.h
+++ b/cogl/cogl/cogl-onscreen-private.h
@@ -78,7 +78,7 @@ COGL_EXPORT CoglFrameInfo *
 cogl_onscreen_peek_tail_frame_info (CoglOnscreen *onscreen);
 
 COGL_EXPORT CoglFrameInfo *
-cogl_onscreen_pop_head_frame_info (CoglOnscreen *onscreen);
+cogl_onscreen_pop_head_frame_info  (CoglOnscreen *onscreen);
 
 COGL_EXPORT unsigned int
 cogl_onscreen_count_pending_frames (CoglOnscreen *onscreen);
-- 
GitLab

From d88b84a647aa4d1d7e67bffd5d175703b7867fb3 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Wed, 20 Apr 2022 18:33:43 +0800
Subject: [PATCH 03/32] kms: Keep a shutting_down flag

---
 src/backends/native/meta-kms.c | 9 +++++++++
 src/backends/native/meta-kms.h | 2 ++
 2 files changed, 11 insertions(+)

diff --git a/src/backends/native/meta-kms.c b/src/backends/native/meta-kms.c
index 795008b210f..70d1e792c45 100644
--- a/src/backends/native/meta-kms.c
+++ b/src/backends/native/meta-kms.c
@@ -63,6 +63,8 @@ struct _MetaKms
   int kernel_thread_inhibit_count;
 
   MetaKmsCursorManager *cursor_manager;
+
+  gboolean shutting_down;
 };
 
 G_DEFINE_TYPE (MetaKms, meta_kms, META_TYPE_THREAD)
@@ -354,6 +356,7 @@ static void
 on_prepare_shutdown (MetaBackend *backend,
                      MetaKms     *kms)
 {
+  kms->shutting_down = TRUE;
   meta_kms_run_impl_task_sync (kms, prepare_shutdown_in_impl, NULL, NULL);
   meta_thread_flush_callbacks (META_THREAD (kms));
 
@@ -408,6 +411,12 @@ meta_kms_new (MetaBackend   *backend,
   return kms;
 }
 
+gboolean
+meta_kms_is_shutting_down (MetaKms *kms)
+{
+  return kms->shutting_down;
+}
+
 static void
 meta_kms_finalize (GObject *object)
 {
diff --git a/src/backends/native/meta-kms.h b/src/backends/native/meta-kms.h
index 74340140639..f6b19520bec 100644
--- a/src/backends/native/meta-kms.h
+++ b/src/backends/native/meta-kms.h
@@ -60,6 +60,8 @@ MetaKmsDevice * meta_kms_create_device (MetaKms            *kms,
                                         MetaKmsDeviceFlag   flags,
                                         GError            **error);
 
+gboolean meta_kms_is_shutting_down (MetaKms *kms);
+
 MetaKms * meta_kms_new (MetaBackend   *backend,
                         MetaKmsFlags   flags,
                         GError       **error);
-- 
GitLab

From f1ea6eb7e2111bdf8044b1e6f132ab46cde0b95a Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 26 Oct 2021 18:50:50 +0800
Subject: [PATCH 04/32] renderer/native: Avoid requeuing the same onscreen for
 a power save flip

This is a case that triple buffering will encounter. We don't want it
to queue the same onscreen multiple times because that would represent
multiple flips occurring simultaneously.

It's a linear search but the list length is typically only 1 or 2 so
no need for anything fancier yet.
---
 src/backends/native/meta-renderer-native.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/src/backends/native/meta-renderer-native.c b/src/backends/native/meta-renderer-native.c
index 7b64ff3f464..158e816d437 100644
--- a/src/backends/native/meta-renderer-native.c
+++ b/src/backends/native/meta-renderer-native.c
@@ -743,6 +743,9 @@ meta_renderer_native_queue_power_save_page_flip (MetaRendererNative *renderer_na
 {
   const unsigned int timeout_ms = 100;
 
+  if (g_list_find (renderer_native->power_save_page_flip_onscreens, onscreen))
+    return;
+
   if (!renderer_native->power_save_page_flip_source_id)
     {
       renderer_native->power_save_page_flip_source_id =
-- 
GitLab

From c742115ce4db5c6c3da16de945ce64926b6ca00d Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Mon, 1 Nov 2021 19:35:34 +0800
Subject: [PATCH 05/32] renderer/native: Steal the power save flip list before
 iterating over it

Because a single iteration might also grow the list again.
---
 src/backends/native/meta-renderer-native.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/src/backends/native/meta-renderer-native.c b/src/backends/native/meta-renderer-native.c
index 158e816d437..25ee1ba35c7 100644
--- a/src/backends/native/meta-renderer-native.c
+++ b/src/backends/native/meta-renderer-native.c
@@ -726,12 +726,18 @@ static gboolean
 dummy_power_save_page_flip_cb (gpointer user_data)
 {
   MetaRendererNative *renderer_native = user_data;
+  GList *old_list =
+    g_steal_pointer (&renderer_native->power_save_page_flip_onscreens);
 
-  g_list_foreach (renderer_native->power_save_page_flip_onscreens,
+  g_list_foreach (old_list,
                   (GFunc) meta_onscreen_native_dummy_power_save_page_flip,
                   NULL);
-  g_clear_list (&renderer_native->power_save_page_flip_onscreens,
+  g_clear_list (&old_list,
                 g_object_unref);
+
+  if (renderer_native->power_save_page_flip_onscreens != NULL)
+    return G_SOURCE_CONTINUE;
+
   renderer_native->power_save_page_flip_source_id = 0;
 
   return G_SOURCE_REMOVE;
-- 
GitLab

From 1bb4eab9980e1de3b13164283de71dc30520599f Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Fri, 10 Dec 2021 16:40:58 +0800
Subject: [PATCH 06/32] onscreen/native: Log swapbuffers and N-buffering when
 MUTTER_DEBUG=kms

---
 src/backends/native/meta-onscreen-native.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 85d86f6f40b..06ceae0ff79 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -1300,6 +1300,19 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
   COGL_TRACE_BEGIN_SCOPED (MetaRendererNativeSwapBuffers,
                            "Meta::OnscreenNative::swap_buffers_with_damage()");
 
+  if (meta_is_topic_enabled (META_DEBUG_KMS))
+    {
+      unsigned int frames_pending =
+        cogl_onscreen_count_pending_frames (onscreen);
+
+      meta_topic (META_DEBUG_KMS,
+                  "Swap buffers: %u frames pending (%s-buffering)",
+                  frames_pending,
+                  frames_pending == 1 ? "double" :
+                  frames_pending == 2 ? "triple" :
+                  "?");
+    }
+
   secondary_gpu_fb =
     update_secondary_gpu_state_pre_swap_buffers (onscreen,
                                                  rectangles,
-- 
GitLab

From 46e899cec8095e74bcb78d5e64f8692438e5e9cc Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Wed, 28 Jul 2021 16:35:56 +0800
Subject: [PATCH 07/32] onscreen/native: Replace an assertion that double
 buffering is the maximum

Because it soon won't be the maximum. But we do want to verify that the
frame info queue is not empty, to avoid NULL dereferencing and catch logic
errors.
---
 src/backends/native/meta-onscreen-native.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 06ceae0ff79..60fab0634a3 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -206,7 +206,7 @@ meta_onscreen_native_notify_frame_complete (CoglOnscreen *onscreen)
 
   info = cogl_onscreen_pop_head_frame_info (onscreen);
 
-  g_assert (!cogl_onscreen_peek_head_frame_info (onscreen));
+  g_return_if_fail (info);
 
   _cogl_onscreen_notify_frame_sync (onscreen, info);
   _cogl_onscreen_notify_complete (onscreen, info);
-- 
GitLab

From 5519229beaea87e3714a9d88294264bf9ad55ec0 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 16 Sep 2021 16:26:25 +0800
Subject: [PATCH 08/32] onscreen/native: Deduplicate calls to
 clutter_frame_set_result

All paths out of `meta_onscreen_native_swap_buffers_with_damage` from
here onward would set the same `CLUTTER_FRAME_RESULT_PENDING_PRESENTED`
(or terminate with `g_assert_not_reached`).

Even failed posts set this result because they will do a
`meta_onscreen_native_notify_frame_complete` in
`page_flip_feedback_discarded`.
---
 src/backends/native/meta-onscreen-native.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 60fab0634a3..032b4b050db 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -1396,6 +1396,9 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
       return;
     }
 
+  clutter_frame_set_result (frame,
+                            CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
+
   /*
    * If we changed EGL context, cogl will have the wrong idea about what is
    * current, making it fail to set it when it needs to. Avoid that by making
@@ -1432,8 +1435,6 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
     {
       meta_renderer_native_queue_power_save_page_flip (renderer_native,
                                                        onscreen);
-      clutter_frame_set_result (frame,
-                                CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
       return;
     }
 
@@ -1453,8 +1454,6 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
           kms_update = meta_frame_native_steal_kms_update (frame_native);
           meta_renderer_native_queue_mode_set_update (renderer_native,
                                                       kms_update);
-          clutter_frame_set_result (frame,
-                                    CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
           return;
         }
       else if (meta_renderer_native_has_pending_mode_set (renderer_native))
@@ -1468,8 +1467,6 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
 
           meta_frame_native_steal_kms_update (frame_native);
           meta_renderer_native_post_mode_set_updates (renderer_native);
-          clutter_frame_set_result (frame,
-                                    CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
           return;
         }
       break;
@@ -1485,8 +1482,6 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
                                                       kms_update);
 
           meta_renderer_native_post_mode_set_updates (renderer_native);
-          clutter_frame_set_result (frame,
-                                    CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
           return;
         }
       break;
@@ -1501,7 +1496,6 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
   kms_update = meta_frame_native_steal_kms_update (frame_native);
   meta_kms_device_post_update (kms_device, kms_update,
                                META_KMS_UPDATE_FLAG_NONE);
-  clutter_frame_set_result (frame, CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
 }
 
 gboolean
-- 
GitLab

From 35cbd749e3d27977e24ead76f3f730fe704e47a7 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Wed, 28 Jul 2021 16:29:27 +0800
Subject: [PATCH 09/32] onscreen/native: Split swap_buffers_with_damage into
 two functions

1. The EGL part: meta_onscreen_native_swap_buffers_with_damage
2. The KMS part: post_latest_swap
---
 src/backends/native/meta-onscreen-native.c | 66 +++++++++++++++++-----
 1 file changed, 53 insertions(+), 13 deletions(-)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 032b4b050db..b21e52b1ae7 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -124,6 +124,12 @@ struct _MetaOnscreenNative
   gulong privacy_screen_changed_handler_id;
   gulong color_space_changed_handler_id;
   gulong hdr_metadata_changed_handler_id;
+
+  struct {
+    int *rectangles;  /* 4 x n_rectangles */
+    int n_rectangles;
+    ClutterFrame *frame;
+  } next_post;
 };
 
 G_DEFINE_TYPE (MetaOnscreenNative, meta_onscreen_native,
@@ -131,6 +137,9 @@ G_DEFINE_TYPE (MetaOnscreenNative, meta_onscreen_native,
 
 static GQuark blit_source_quark = 0;
 
+static void
+post_latest_swap (CoglOnscreen *onscreen);
+
 static gboolean
 init_secondary_gpu_state (MetaRendererNative  *renderer_native,
                           CoglOnscreen        *onscreen,
@@ -1276,26 +1285,18 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
   CoglRendererEGL *cogl_renderer_egl = cogl_renderer->winsys;
   MetaRendererNativeGpuData *renderer_gpu_data = cogl_renderer_egl->platform;
   MetaRendererNative *renderer_native = renderer_gpu_data->renderer_native;
-  MetaRenderer *renderer = META_RENDERER (renderer_native);
-  MetaBackend *backend = meta_renderer_get_backend (renderer);
-  MetaMonitorManager *monitor_manager =
-    meta_backend_get_monitor_manager (backend);
   MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
   MetaGpuKms *render_gpu = onscreen_native->render_gpu;
   MetaDeviceFile *render_device_file;
   ClutterFrame *frame = user_data;
-  MetaFrameNative *frame_native = meta_frame_native_from_frame (frame);
-  MetaKmsUpdate *kms_update;
   CoglOnscreenClass *parent_class;
   gboolean egl_context_changed = FALSE;
-  MetaPowerSave power_save_mode;
   g_autoptr (GError) error = NULL;
   MetaDrmBufferFlags buffer_flags;
   MetaDrmBufferGbm *buffer_gbm;
   g_autoptr (MetaDrmBuffer) primary_gpu_fb = NULL;
   g_autoptr (MetaDrmBuffer) secondary_gpu_fb = NULL;
-  MetaKmsCrtc *kms_crtc;
-  MetaKmsDevice *kms_device;
+  size_t rectangles_size;
 
   COGL_TRACE_BEGIN_SCOPED (MetaRendererNativeSwapBuffers,
                            "Meta::OnscreenNative::swap_buffers_with_damage()");
@@ -1408,12 +1409,47 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
   if (egl_context_changed)
     _cogl_winsys_egl_ensure_current (cogl_display);
 
-  kms_crtc = meta_crtc_kms_get_kms_crtc (META_CRTC_KMS (onscreen_native->crtc));
-  kms_device = meta_kms_crtc_get_device (kms_crtc);
+  rectangles_size = n_rectangles * 4 * sizeof (int);
+  onscreen_native->next_post.rectangles =
+    g_realloc (onscreen_native->next_post.rectangles, rectangles_size);
+  memcpy (onscreen_native->next_post.rectangles, rectangles, rectangles_size);
+  onscreen_native->next_post.n_rectangles = n_rectangles;
+
+  g_clear_pointer (&onscreen_native->next_post.frame, clutter_frame_unref);
+  onscreen_native->next_post.frame = clutter_frame_ref (frame);
+
+  post_latest_swap (onscreen);
+}
+
+static void
+post_latest_swap (CoglOnscreen *onscreen)
+{
+  CoglFramebuffer *framebuffer = COGL_FRAMEBUFFER (onscreen);
+  CoglContext *cogl_context = cogl_framebuffer_get_context (framebuffer);
+  CoglRenderer *cogl_renderer = cogl_context->display->renderer;
+  CoglRendererEGL *cogl_renderer_egl = cogl_renderer->winsys;
+  MetaRendererNativeGpuData *renderer_gpu_data = cogl_renderer_egl->platform;
+  MetaRendererNative *renderer_native = renderer_gpu_data->renderer_native;
+  MetaRenderer *renderer = META_RENDERER (renderer_native);
+  MetaBackend *backend = meta_renderer_get_backend (renderer);
+  MetaMonitorManager *monitor_manager =
+    meta_backend_get_monitor_manager (backend);
+  MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
+  MetaPowerSave power_save_mode;
+  MetaCrtcKms *crtc_kms = META_CRTC_KMS (onscreen_native->crtc);
+  MetaKmsCrtc *kms_crtc = meta_crtc_kms_get_kms_crtc (crtc_kms);
+  MetaKmsDevice *kms_device = meta_kms_crtc_get_device (kms_crtc);
+  MetaKmsUpdate *kms_update;
+  g_autoptr (MetaKmsFeedback) kms_feedback = NULL;
+  g_autoptr (ClutterFrame) frame =
+    g_steal_pointer (&onscreen_native->next_post.frame);
+  MetaFrameNative *frame_native;
 
   power_save_mode = meta_monitor_manager_get_power_save_mode (monitor_manager);
   if (power_save_mode == META_POWER_SAVE_ON)
     {
+      frame_native = meta_frame_native_from_frame (frame);
+
       kms_update = meta_frame_native_ensure_kms_update (frame_native,
                                                         kms_device);
       meta_kms_update_add_result_listener (kms_update,
@@ -1428,8 +1464,8 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
                                       onscreen_native->crtc,
                                       kms_update,
                                       META_KMS_ASSIGN_PLANE_FLAG_NONE,
-                                      rectangles,
-                                      n_rectangles);
+                                      onscreen_native->next_post.rectangles,
+                                      onscreen_native->next_post.n_rectangles);
     }
   else
     {
@@ -2842,6 +2878,10 @@ meta_onscreen_native_dispose (GObject *object)
 
   g_clear_object (&onscreen_native->output);
   g_clear_object (&onscreen_native->crtc);
+
+  g_clear_pointer (&onscreen_native->next_post.rectangles, g_free);
+  g_clear_pointer (&onscreen_native->next_post.frame, clutter_frame_unref);
+  onscreen_native->next_post.n_rectangles = 0;
 }
 
 static void
-- 
GitLab

From 8a748c853d5fc6a2f4693102bba265ac90ad7aec Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 28 Oct 2021 17:24:11 +0800
Subject: [PATCH 10/32] onscreen/native: Add a missing frame notification

In the unlikely event that swap_buffers_with_damage drops the previous swap
it was forgetting to notify about the discarded frame. That could lead to
frame clock freezes.
---
 src/backends/native/meta-onscreen-native.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index b21e52b1ae7..7a520c4601c 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -1373,6 +1373,15 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
     {
     case META_RENDERER_NATIVE_MODE_GBM:
       g_warn_if_fail (onscreen_native->gbm.next_fb == NULL);
+      if (onscreen_native->gbm.next_fb != NULL)
+        {
+          CoglFrameInfo *frame_info;
+
+          frame_info = cogl_onscreen_peek_head_frame_info (onscreen);
+          frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
+          meta_onscreen_native_notify_frame_complete (onscreen);
+        }
+
       if (onscreen_native->secondary_gpu_state)
         g_set_object (&onscreen_native->gbm.next_fb, secondary_gpu_fb);
       else
-- 
GitLab

From 44c724164fbbca7a9aea1ce58a247ea58e19dc3f Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 5 Dec 2023 17:50:44 +0800
Subject: [PATCH 11/32] onscreen/native: Insert a 'posted' frame between 'next'
 and 'current'

This will allow us to keep track of up to two buffers that have been
swapped but not yet scanning out, for triple buffering.

This commit replaces mutter!1968.
---
 src/backends/native/meta-onscreen-native.c | 58 ++++++++++++----------
 1 file changed, 31 insertions(+), 27 deletions(-)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 7a520c4601c..0dc2d822e1a 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -97,8 +97,10 @@ struct _MetaOnscreenNative
   struct {
     struct gbm_surface *surface;
     MetaDrmBuffer *current_fb;
+    MetaDrmBuffer *posted_fb;
     MetaDrmBuffer *next_fb;
     CoglScanout *current_scanout;
+    CoglScanout *posted_scanout;
     CoglScanout *next_scanout;
   } gbm;
 
@@ -145,39 +147,30 @@ init_secondary_gpu_state (MetaRendererNative  *renderer_native,
                           CoglOnscreen        *onscreen,
                           GError             **error);
 
-static void
-free_current_bo (CoglOnscreen *onscreen)
-{
-  MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
-
-  g_clear_object (&onscreen_native->gbm.current_fb);
-  g_clear_object (&onscreen_native->gbm.current_scanout);
-}
-
 static void
 meta_onscreen_native_swap_drm_fb (CoglOnscreen *onscreen)
 {
   MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
 
-  if (!onscreen_native->gbm.next_fb)
+  if (!onscreen_native->gbm.posted_fb)
     return;
 
-  free_current_bo (onscreen);
+  g_set_object (&onscreen_native->gbm.current_fb,
+                onscreen_native->gbm.posted_fb);
+  g_clear_object (&onscreen_native->gbm.posted_fb);
 
-  g_set_object (&onscreen_native->gbm.current_fb, onscreen_native->gbm.next_fb);
-  g_clear_object (&onscreen_native->gbm.next_fb);
   g_set_object (&onscreen_native->gbm.current_scanout,
-                onscreen_native->gbm.next_scanout);
-  g_clear_object (&onscreen_native->gbm.next_scanout);
+                onscreen_native->gbm.posted_scanout);
+  g_clear_object (&onscreen_native->gbm.posted_scanout);
 }
 
 static void
-meta_onscreen_native_clear_next_fb (CoglOnscreen *onscreen)
+meta_onscreen_native_clear_posted_fb (CoglOnscreen *onscreen)
 {
   MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
 
-  g_clear_object (&onscreen_native->gbm.next_fb);
-  g_clear_object (&onscreen_native->gbm.next_scanout);
+  g_clear_object (&onscreen_native->gbm.posted_fb);
+  g_clear_object (&onscreen_native->gbm.posted_scanout);
 }
 
 static void
@@ -358,7 +351,7 @@ page_flip_feedback_discarded (MetaKmsCrtc  *kms_crtc,
   frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
 
   meta_onscreen_native_notify_frame_complete (onscreen);
-  meta_onscreen_native_clear_next_fb (onscreen);
+  meta_onscreen_native_clear_posted_fb (onscreen);
 }
 
 static const MetaKmsPageFlipListenerVtable page_flip_listener_vtable = {
@@ -529,13 +522,21 @@ meta_onscreen_native_flip_crtc (CoglOnscreen           *onscreen,
   switch (renderer_gpu_data->mode)
     {
     case META_RENDERER_NATIVE_MODE_GBM:
-      buffer = onscreen_native->gbm.next_fb;
+      g_set_object (&onscreen_native->gbm.posted_fb,
+                    onscreen_native->gbm.next_fb);
+      g_clear_object (&onscreen_native->gbm.next_fb);
+
+      buffer = onscreen_native->gbm.posted_fb;
+
+      g_set_object (&onscreen_native->gbm.posted_scanout,
+                    onscreen_native->gbm.next_scanout);
+      g_clear_object (&onscreen_native->gbm.next_scanout);
 
-      if (onscreen_native->gbm.next_scanout)
+      if (onscreen_native->gbm.posted_scanout)
         {
-          cogl_scanout_get_src_rect (onscreen_native->gbm.next_scanout,
+          cogl_scanout_get_src_rect (onscreen_native->gbm.posted_scanout,
                                      &src_rect);
-          cogl_scanout_get_dst_rect (onscreen_native->gbm.next_scanout,
+          cogl_scanout_get_dst_rect (onscreen_native->gbm.posted_scanout,
                                      &dst_rect);
         }
       else
@@ -1264,7 +1265,7 @@ swap_buffer_result_feedback (const MetaKmsFeedback *kms_feedback,
   frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
 
   meta_onscreen_native_notify_frame_complete (onscreen);
-  meta_onscreen_native_clear_next_fb (onscreen);
+  meta_onscreen_native_clear_posted_fb (onscreen);
 }
 
 static const MetaKmsResultListenerVtable swap_buffer_result_listener_vtable = {
@@ -1611,7 +1612,7 @@ scanout_result_feedback (const MetaKmsFeedback *kms_feedback,
 
       g_warning ("Direct scanout page flip failed: %s", error->message);
 
-      cogl_scanout_notify_failed (onscreen_native->gbm.next_scanout,
+      cogl_scanout_notify_failed (onscreen_native->gbm.posted_scanout,
                                   onscreen);
       clutter_stage_view_add_redraw_clip (view, NULL);
       clutter_stage_view_schedule_update_now (view);
@@ -1621,7 +1622,7 @@ scanout_result_feedback (const MetaKmsFeedback *kms_feedback,
   frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
 
   meta_onscreen_native_notify_frame_complete (onscreen);
-  meta_onscreen_native_clear_next_fb (onscreen);
+  meta_onscreen_native_clear_posted_fb (onscreen);
 }
 
 static const MetaKmsResultListenerVtable scanout_result_listener_vtable = {
@@ -2852,8 +2853,11 @@ meta_onscreen_native_dispose (GObject *object)
     {
     case META_RENDERER_NATIVE_MODE_GBM:
       g_clear_object (&onscreen_native->gbm.next_fb);
+      g_clear_object (&onscreen_native->gbm.posted_fb);
+      g_clear_object (&onscreen_native->gbm.current_fb);
       g_clear_object (&onscreen_native->gbm.next_scanout);
-      free_current_bo (onscreen);
+      g_clear_object (&onscreen_native->gbm.posted_scanout);
+      g_clear_object (&onscreen_native->gbm.current_scanout);
       break;
     case META_RENDERER_NATIVE_MODE_SURFACELESS:
       g_assert_not_reached ();
-- 
GitLab

From 2a3c02e6d3ba329f02185de11e75ff11bf35e05a Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Fri, 17 Sep 2021 17:59:28 +0800
Subject: [PATCH 12/32] onscreen/native: Defer posting if there's already a
 post in progress

And when the number of pending posts decreases we know it's safe to submit
a new one. Since KMS generally only supports one outstanding post right now,
"decreases" means equal to zero.
---
 src/backends/native/meta-onscreen-native.c | 192 ++++++++++++++++++---
 1 file changed, 164 insertions(+), 28 deletions(-)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 0dc2d822e1a..e7ff4db15b5 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -99,9 +99,11 @@ struct _MetaOnscreenNative
     MetaDrmBuffer *current_fb;
     MetaDrmBuffer *posted_fb;
     MetaDrmBuffer *next_fb;
+    MetaDrmBuffer *stalled_fb;
     CoglScanout *current_scanout;
     CoglScanout *posted_scanout;
     CoglScanout *next_scanout;
+    CoglScanout *stalled_scanout;
   } gbm;
 
 #ifdef HAVE_EGL_DEVICE
@@ -127,6 +129,10 @@ struct _MetaOnscreenNative
   gulong color_space_changed_handler_id;
   gulong hdr_metadata_changed_handler_id;
 
+  gboolean needs_flush;
+
+  unsigned int swaps_pending;
+
   struct {
     int *rectangles;  /* 4 x n_rectangles */
     int n_rectangles;
@@ -140,7 +146,11 @@ G_DEFINE_TYPE (MetaOnscreenNative, meta_onscreen_native,
 static GQuark blit_source_quark = 0;
 
 static void
-post_latest_swap (CoglOnscreen *onscreen);
+try_post_latest_swap (CoglOnscreen *onscreen);
+
+static void
+post_finish_frame (MetaOnscreenNative *onscreen_native,
+                   MetaKmsUpdate      *kms_update);
 
 static gboolean
 init_secondary_gpu_state (MetaRendererNative  *renderer_native,
@@ -244,6 +254,7 @@ notify_view_crtc_presented (MetaRendererView *view,
 
   meta_onscreen_native_notify_frame_complete (onscreen);
   meta_onscreen_native_swap_drm_fb (onscreen);
+  try_post_latest_swap (onscreen);
 }
 
 static void
@@ -293,15 +304,13 @@ page_flip_feedback_ready (MetaKmsCrtc *kms_crtc,
   CoglFramebuffer *framebuffer =
     clutter_stage_view_get_onscreen (CLUTTER_STAGE_VIEW (view));
   CoglOnscreen *onscreen = COGL_ONSCREEN (framebuffer);
-  MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
   CoglFrameInfo *frame_info;
 
   frame_info = cogl_onscreen_peek_head_frame_info (onscreen);
   frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
 
-  g_warn_if_fail (!onscreen_native->gbm.next_fb);
-
   meta_onscreen_native_notify_frame_complete (onscreen);
+  try_post_latest_swap (onscreen);
 }
 
 static void
@@ -352,6 +361,7 @@ page_flip_feedback_discarded (MetaKmsCrtc  *kms_crtc,
 
   meta_onscreen_native_notify_frame_complete (onscreen);
   meta_onscreen_native_clear_posted_fb (onscreen);
+  try_post_latest_swap (onscreen);
 }
 
 static const MetaKmsPageFlipListenerVtable page_flip_listener_vtable = {
@@ -412,18 +422,41 @@ custom_egl_stream_page_flip (gpointer custom_page_flip_data,
 }
 #endif /* HAVE_EGL_DEVICE */
 
-void
-meta_onscreen_native_dummy_power_save_page_flip (CoglOnscreen *onscreen)
+static void
+drop_stalled_swap (CoglOnscreen *onscreen)
 {
   CoglFrameInfo *frame_info;
+  MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
 
-  meta_onscreen_native_swap_drm_fb (onscreen);
+  /* Remember we can't compare stalled_fb because it's not used by
+   * META_RENDERER_NATIVE_MODE_EGL_DEVICE. So we judge stalled to be whenever
+   * swaps_pending > 1.
+   */
+  if (onscreen_native->swaps_pending <= 1)
+    return;
+
+  onscreen_native->swaps_pending--;
+
+  g_clear_object (&onscreen_native->gbm.stalled_fb);
+  g_clear_object (&onscreen_native->gbm.stalled_scanout);
 
   frame_info = cogl_onscreen_peek_tail_frame_info (onscreen);
   frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
   meta_onscreen_native_notify_frame_complete (onscreen);
 }
 
+void
+meta_onscreen_native_dummy_power_save_page_flip (CoglOnscreen *onscreen)
+{
+  drop_stalled_swap (onscreen);
+
+  /* If the monitor just woke up and the shell is fully idle (has nothing
+   * more to swap) then we just woke to an indefinitely black screen. Let's
+   * fix that using the last swap (which is never classified as "stalled").
+   */
+  try_post_latest_swap (onscreen);
+}
+
 static void
 apply_transform (MetaCrtcKms            *crtc_kms,
                  MetaKmsPlaneAssignment *kms_plane_assignment,
@@ -1262,9 +1295,16 @@ swap_buffer_result_feedback (const MetaKmsFeedback *kms_feedback,
     g_warning ("Page flip failed: %s", error->message);
 
   frame_info = cogl_onscreen_peek_head_frame_info (onscreen);
-  frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
 
-  meta_onscreen_native_notify_frame_complete (onscreen);
+  /* After resuming from suspend, drop_stalled_swap might have done this
+   * already and emptied the frame_info queue.
+   */
+  if (frame_info)
+    {
+      frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
+      meta_onscreen_native_notify_frame_complete (onscreen);
+    }
+
   meta_onscreen_native_clear_posted_fb (onscreen);
 }
 
@@ -1373,14 +1413,15 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
   switch (renderer_gpu_data->mode)
     {
     case META_RENDERER_NATIVE_MODE_GBM:
-      g_warn_if_fail (onscreen_native->gbm.next_fb == NULL);
       if (onscreen_native->gbm.next_fb != NULL)
         {
-          CoglFrameInfo *frame_info;
-
-          frame_info = cogl_onscreen_peek_head_frame_info (onscreen);
-          frame_info->flags |= COGL_FRAME_INFO_FLAG_SYMBOLIC;
-          meta_onscreen_native_notify_frame_complete (onscreen);
+          g_warn_if_fail (onscreen_native->gbm.stalled_fb == NULL);
+          drop_stalled_swap (onscreen);
+          g_assert (onscreen_native->gbm.stalled_fb == NULL);
+          onscreen_native->gbm.stalled_fb =
+            g_steal_pointer (&onscreen_native->gbm.next_fb);
+          onscreen_native->gbm.stalled_scanout =
+            g_steal_pointer (&onscreen_native->gbm.next_scanout);
         }
 
       if (onscreen_native->secondary_gpu_state)
@@ -1428,11 +1469,12 @@ meta_onscreen_native_swap_buffers_with_damage (CoglOnscreen  *onscreen,
   g_clear_pointer (&onscreen_native->next_post.frame, clutter_frame_unref);
   onscreen_native->next_post.frame = clutter_frame_ref (frame);
 
-  post_latest_swap (onscreen);
+  onscreen_native->swaps_pending++;
+  try_post_latest_swap (onscreen);
 }
 
 static void
-post_latest_swap (CoglOnscreen *onscreen)
+try_post_latest_swap (CoglOnscreen *onscreen)
 {
   CoglFramebuffer *framebuffer = COGL_FRAMEBUFFER (onscreen);
   CoglContext *cogl_context = cogl_framebuffer_get_context (framebuffer);
@@ -1451,15 +1493,41 @@ post_latest_swap (CoglOnscreen *onscreen)
   MetaKmsDevice *kms_device = meta_kms_crtc_get_device (kms_crtc);
   MetaKmsUpdate *kms_update;
   g_autoptr (MetaKmsFeedback) kms_feedback = NULL;
-  g_autoptr (ClutterFrame) frame =
-    g_steal_pointer (&onscreen_native->next_post.frame);
+  g_autoptr (ClutterFrame) frame = NULL;
   MetaFrameNative *frame_native;
 
+  if (onscreen_native->next_post.frame == NULL)
+    return;
+
   power_save_mode = meta_monitor_manager_get_power_save_mode (monitor_manager);
   if (power_save_mode == META_POWER_SAVE_ON)
     {
+      unsigned int frames_pending =
+        cogl_onscreen_count_pending_frames (onscreen);
+      unsigned int posts_pending;
+
+      g_assert (frames_pending >= onscreen_native->swaps_pending);
+      posts_pending = frames_pending - onscreen_native->swaps_pending;
+      if (posts_pending > 0)
+        return;  /* wait for the next frame notification and then try again */
+
+      frame = g_steal_pointer (&onscreen_native->next_post.frame);
       frame_native = meta_frame_native_from_frame (frame);
 
+      if (onscreen_native->swaps_pending == 0)
+        {
+          if (frame_native)
+            {
+              kms_update = meta_frame_native_steal_kms_update (frame_native);
+              if (kms_update)
+                post_finish_frame (onscreen_native, kms_update);
+            }
+          return;
+        }
+
+      drop_stalled_swap (onscreen);
+      onscreen_native->swaps_pending--;
+
       kms_update = meta_frame_native_ensure_kms_update (frame_native,
                                                         kms_device);
       meta_kms_update_add_result_listener (kms_update,
@@ -1674,6 +1742,18 @@ meta_onscreen_native_direct_scanout (CoglOnscreen   *onscreen,
       return FALSE;
     }
 
+  /* Our direct scanout frame counts as 1, so more than that means we would
+   * be jumping the queue (and post would fail).
+   */
+  if (cogl_onscreen_count_pending_frames (onscreen) > 1)
+    {
+      g_set_error_literal (error,
+                           COGL_SCANOUT_ERROR,
+                           COGL_SCANOUT_ERROR_INHIBITED,
+                           "Direct scanout is inhibited during triple buffering");
+      return FALSE;
+    }
+
   renderer_gpu_data = meta_renderer_native_get_gpu_data (renderer_native,
                                                          render_gpu);
 
@@ -1909,22 +1989,79 @@ meta_onscreen_native_finish_frame (CoglOnscreen *onscreen,
   MetaKmsDevice *kms_device = meta_kms_crtc_get_device (kms_crtc);
   MetaFrameNative *frame_native = meta_frame_native_from_frame (frame);
   MetaKmsUpdate *kms_update;
+  unsigned int frames_pending = cogl_onscreen_count_pending_frames (onscreen);
+  unsigned int swaps_pending = onscreen_native->swaps_pending;
+  unsigned int posts_pending = frames_pending - swaps_pending;
 
-  kms_update = meta_frame_native_steal_kms_update (frame_native);
-  if (!kms_update)
+  onscreen_native->needs_flush |= meta_kms_device_handle_flush (kms_device,
+                                                                kms_crtc);
+
+  if (!meta_frame_native_has_kms_update (frame_native))
     {
-      if (meta_kms_device_handle_flush (kms_device, kms_crtc))
-        {
-          kms_update = meta_kms_update_new (kms_device);
-          meta_kms_update_set_flushing (kms_update, kms_crtc);
-        }
-      else
+      if (!onscreen_native->needs_flush || posts_pending)
         {
           clutter_frame_set_result (frame, CLUTTER_FRAME_RESULT_IDLE);
           return;
         }
     }
 
+  if (posts_pending && !swaps_pending)
+    {
+      g_return_if_fail (meta_frame_native_has_kms_update (frame_native));
+      g_warn_if_fail (onscreen_native->next_post.frame == NULL);
+
+      g_clear_pointer (&onscreen_native->next_post.frame, clutter_frame_unref);
+      onscreen_native->next_post.frame = clutter_frame_ref (frame);
+      clutter_frame_set_result (frame, CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
+      return;
+    }
+
+  kms_update = meta_frame_native_steal_kms_update (frame_native);
+
+  if (posts_pending && swaps_pending)
+    {
+      MetaFrameNative *older_frame_native;
+      MetaKmsUpdate *older_kms_update;
+
+      g_return_if_fail (kms_update);
+      g_return_if_fail (onscreen_native->next_post.frame != NULL);
+
+      older_frame_native =
+        meta_frame_native_from_frame (onscreen_native->next_post.frame);
+      older_kms_update =
+        meta_frame_native_ensure_kms_update (older_frame_native, kms_device);
+      meta_kms_update_merge_from (older_kms_update, kms_update);
+      meta_kms_update_free (kms_update);
+      clutter_frame_set_result (frame, CLUTTER_FRAME_RESULT_IDLE);
+      return;
+    }
+
+  if (!kms_update)
+    {
+      kms_update = meta_kms_update_new (kms_device);
+      g_warn_if_fail (onscreen_native->needs_flush);
+    }
+
+  if (onscreen_native->needs_flush)
+    {
+      meta_kms_update_set_flushing (kms_update, kms_crtc);
+      onscreen_native->needs_flush = FALSE;
+    }
+
+  post_finish_frame (onscreen_native, kms_update);
+
+  clutter_frame_set_result (frame, CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
+}
+
+static void
+post_finish_frame (MetaOnscreenNative *onscreen_native,
+                   MetaKmsUpdate      *kms_update)
+{
+  MetaCrtc *crtc = onscreen_native->crtc;
+  MetaKmsCrtc *kms_crtc = meta_crtc_kms_get_kms_crtc (META_CRTC_KMS (crtc));
+  MetaKmsDevice *kms_device = meta_kms_crtc_get_device (kms_crtc);
+  g_autoptr (MetaKmsFeedback) kms_feedback = NULL;
+
   meta_kms_update_add_result_listener (kms_update,
                                        &finish_frame_result_listener_vtable,
                                        NULL,
@@ -1947,7 +2084,6 @@ meta_onscreen_native_finish_frame (CoglOnscreen *onscreen,
   meta_kms_update_set_flushing (kms_update, kms_crtc);
   meta_kms_device_post_update (kms_device, kms_update,
                                META_KMS_UPDATE_FLAG_NONE);
-  clutter_frame_set_result (frame, CLUTTER_FRAME_RESULT_PENDING_PRESENTED);
 }
 
 static gboolean
-- 
GitLab

From 0137bc8d0d03559001a71ad7ae87449591fb7e47 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Fri, 9 Dec 2022 14:22:31 +0800
Subject: [PATCH 13/32] onscreen/native: Increase secondary GPU dumb_fbs from 2
 to 3

So that they don't get overwritten prematurely during triple buffering
causing tearing.

https://launchpad.net/bugs/1999216
---
 src/backends/native/meta-onscreen-native.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index e7ff4db15b5..c4880a36f5b 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -75,7 +75,7 @@ typedef struct _MetaOnscreenNativeSecondaryGpuState
 
   struct {
     MetaDrmBufferDumb *current_dumb_fb;
-    MetaDrmBufferDumb *dumb_fbs[2];
+    MetaDrmBufferDumb *dumb_fbs[3];
   } cpu;
 
   gboolean noted_primary_gpu_copy_ok;
@@ -957,12 +957,17 @@ static MetaDrmBufferDumb *
 secondary_gpu_get_next_dumb_buffer (MetaOnscreenNativeSecondaryGpuState *secondary_gpu_state)
 {
   MetaDrmBufferDumb *current_dumb_fb;
+  const int n_dumb_fbs = G_N_ELEMENTS (secondary_gpu_state->cpu.dumb_fbs);
+  int i;
 
   current_dumb_fb = secondary_gpu_state->cpu.current_dumb_fb;
-  if (current_dumb_fb == secondary_gpu_state->cpu.dumb_fbs[0])
-    return secondary_gpu_state->cpu.dumb_fbs[1];
-  else
-    return secondary_gpu_state->cpu.dumb_fbs[0];
+  for (i = 0; i < n_dumb_fbs; i++)
+    {
+      if (current_dumb_fb == secondary_gpu_state->cpu.dumb_fbs[i])
+        return secondary_gpu_state->cpu.dumb_fbs[(i + 1) % n_dumb_fbs];
+    }
+
+  return secondary_gpu_state->cpu.dumb_fbs[0];
 }
 
 static MetaDrmBuffer *
-- 
GitLab

From e9a616eb224301bd67e591b61e525adfa546fd4b Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 4 Nov 2021 16:09:26 +0800
Subject: [PATCH 14/32] onscreen/native: Add function
 meta_onscreen_native_discard_pending_swaps

---
 src/backends/native/meta-onscreen-native.c | 13 +++++++++++++
 src/backends/native/meta-onscreen-native.h |  2 ++
 2 files changed, 15 insertions(+)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index c4880a36f5b..113c3b0fda0 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -2091,6 +2091,19 @@ post_finish_frame (MetaOnscreenNative *onscreen_native,
                                META_KMS_UPDATE_FLAG_NONE);
 }
 
+void
+meta_onscreen_native_discard_pending_swaps (CoglOnscreen *onscreen)
+{
+  MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
+
+  onscreen_native->swaps_pending = 0;
+
+  g_clear_object (&onscreen_native->gbm.stalled_fb);
+  g_clear_object (&onscreen_native->gbm.stalled_scanout);
+  g_clear_object (&onscreen_native->gbm.next_fb);
+  g_clear_object (&onscreen_native->gbm.next_scanout);
+}
+
 static gboolean
 should_surface_be_sharable (CoglOnscreen *onscreen)
 {
diff --git a/src/backends/native/meta-onscreen-native.h b/src/backends/native/meta-onscreen-native.h
index 0e1193325a9..e30357d19d1 100644
--- a/src/backends/native/meta-onscreen-native.h
+++ b/src/backends/native/meta-onscreen-native.h
@@ -48,6 +48,8 @@ void meta_onscreen_native_dummy_power_save_page_flip (CoglOnscreen *onscreen);
 gboolean meta_onscreen_native_is_buffer_scanout_compatible (CoglOnscreen *onscreen,
                                                             CoglScanout  *scanout);
 
+void meta_onscreen_native_discard_pending_swaps (CoglOnscreen *onscreen);
+
 void meta_onscreen_native_set_view (CoglOnscreen     *onscreen,
                                     MetaRendererView *view);
 
-- 
GitLab

From 0553e8ff43bc029928757bb12be3ee586b6efb66 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Wed, 11 May 2022 16:00:32 +0800
Subject: [PATCH 15/32] onscreen/native: Skip try_post_latest_swap if shutting
 down

Otherwise we could get:

meta_kms_prepare_shutdown ->
  flush_callbacks ->
  ... ->
  try_post_latest_swap ->
  post and queue more callbacks

So later in shutdown those callbacks would trigger an assertion failure
in meta_kms_impl_device_atomic_finalize:

g_hash_table_size (impl_device_atomic->page_flip_datas) == 0
---
 src/backends/native/meta-onscreen-native.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 113c3b0fda0..26a2cf90a32 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -1489,6 +1489,8 @@ try_post_latest_swap (CoglOnscreen *onscreen)
   MetaRendererNative *renderer_native = renderer_gpu_data->renderer_native;
   MetaRenderer *renderer = META_RENDERER (renderer_native);
   MetaBackend *backend = meta_renderer_get_backend (renderer);
+  MetaBackendNative *backend_native = META_BACKEND_NATIVE (backend);
+  MetaKms *kms = meta_backend_native_get_kms (backend_native);
   MetaMonitorManager *monitor_manager =
     meta_backend_get_monitor_manager (backend);
   MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
@@ -1501,7 +1503,8 @@ try_post_latest_swap (CoglOnscreen *onscreen)
   g_autoptr (ClutterFrame) frame = NULL;
   MetaFrameNative *frame_native;
 
-  if (onscreen_native->next_post.frame == NULL)
+  if (onscreen_native->next_post.frame == NULL ||
+      meta_kms_is_shutting_down (kms))
     return;
 
   power_save_mode = meta_monitor_manager_get_power_save_mode (monitor_manager);
-- 
GitLab

From e6b9e279c95266c96925e463f457ebcd21a64a5d Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 4 Nov 2021 16:09:52 +0800
Subject: [PATCH 16/32] renderer/native: Discard pending swaps when rebuilding
 views

It's analogous to discard_pending_page_flips but represents swaps that
might become flips after the next frame notification callbacks, thanks
to triple buffering. Since the views are being rebuilt and their onscreens
are about to be destroyed, turning those swaps into more flips/posts would
just lead to unexpected behaviour (like trying to flip on a half-destroyed
inactive CRTC).
---
 src/backends/native/meta-renderer-native.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/src/backends/native/meta-renderer-native.c b/src/backends/native/meta-renderer-native.c
index 25ee1ba35c7..c32a6caa62a 100644
--- a/src/backends/native/meta-renderer-native.c
+++ b/src/backends/native/meta-renderer-native.c
@@ -1533,6 +1533,26 @@ detach_onscreens (MetaRenderer *renderer)
     }
 }
 
+static void
+discard_pending_swaps (MetaRenderer *renderer)
+{
+  GList *views = meta_renderer_get_views (renderer);;
+  GList *l;
+
+  for (l = views; l; l = l->next)
+    {
+      ClutterStageView *stage_view = l->data;
+      CoglFramebuffer *fb = clutter_stage_view_get_onscreen (stage_view);
+      CoglOnscreen *onscreen;
+
+      if (!COGL_IS_ONSCREEN (fb))
+        continue;
+
+      onscreen = COGL_ONSCREEN (fb);
+      meta_onscreen_native_discard_pending_swaps (onscreen);
+    }
+}
+
 static void
 meta_renderer_native_rebuild_views (MetaRenderer *renderer)
 {
@@ -1543,6 +1563,7 @@ meta_renderer_native_rebuild_views (MetaRenderer *renderer)
   MetaRendererClass *parent_renderer_class =
     META_RENDERER_CLASS (meta_renderer_native_parent_class);
 
+  discard_pending_swaps (renderer);
   meta_kms_discard_pending_page_flips (kms);
   g_hash_table_remove_all (renderer_native->mode_set_updates);
 
-- 
GitLab

From f01e38384ef420d0cee042e61f70e78bade678d7 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 7 Dec 2023 16:28:20 +0800
Subject: [PATCH 17/32] clutter/frame-clock: Reuse existing idle period
 detection

While it was correct for double buffering, the test was wrong for
triple buffering and would lead to accidentally staying in triple
buffering for too long.

Fixes: 8f27ebf87eee6057992a90560d4118ab7bdf138d
---
 clutter/clutter/clutter-frame-clock.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 93e4c932902..1b212ee3fa9 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -491,6 +491,7 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
   int64_t min_render_time_allowed_us;
   int64_t max_render_time_allowed_us;
   int64_t next_presentation_time_us;
+  int64_t next_smooth_presentation_time_us;
   int64_t next_update_time_us;
 
   now_us = g_get_monotonic_time ();
@@ -535,7 +536,8 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
    *
    */
   last_presentation_time_us = frame_clock->last_presentation_time_us;
-  next_presentation_time_us = last_presentation_time_us + refresh_interval_us;
+  next_smooth_presentation_time_us = last_presentation_time_us + refresh_interval_us;
+  next_presentation_time_us = next_smooth_presentation_time_us;
 
   /*
    * However, the last presentation could have happened more than a frame ago.
@@ -601,7 +603,7 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
         }
     }
 
-  if (next_presentation_time_us != last_presentation_time_us + refresh_interval_us)
+  if (next_presentation_time_us != next_smooth_presentation_time_us)
     {
       /* There was an idle period since the last presentation, so there seems
        * be no constantly updating actor. In this case it's best to start
-- 
GitLab

From f8f543174e36137b82b2d495910c3d8db171a09f Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 10 Aug 2021 17:46:49 +0800
Subject: [PATCH 18/32] clutter/frame-clock: Lower the threshold for disabling
 error diffusion

Error diffusion was introduced in 0555a5bbc15 for Nvidia where last
presentation time is always unknown (zero). Dispatch times would drift
apart always being a fraction of a frame late, and accumulated to cause
periodic frame skips. So error diffusion corrected that precisely and
avoided the skips.

That works great with double buffering but less great with triple
buffering. It's certainly still needed with triple buffering but
correcting for a lateness of many milliseconds isn't a good idea. That's
because a dispatch being that late is not due to main loop jitter but due
to Nvidia's swap buffers blocking when the queue is full. So scheduling
the next frame even earlier using last_dispatch_lateness_us would just
perpetuate the problem of swap buffers blocking for too long.

So now we lower the threshold of when error diffusion gets disabled. It's
still high enough to fix the original smoothness problem it was for, but
now low enough to detect Nvidia's occasionally blocking swaps and backs
off in that case.

Since the average duration of a blocking swap is half a frame interval
and we want to distinguish between that and sub-millisecond jitter, the
logical threshold is halfway again: refresh_interval_us/4.
---
 clutter/clutter/clutter-frame-clock.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 1b212ee3fa9..48f07e1014e 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -924,7 +924,7 @@ clutter_frame_clock_dispatch (ClutterFrameClock *frame_clock,
                              frame_clock->refresh_interval_us;
 
   lateness_us = time_us - ideal_dispatch_time_us;
-  if (lateness_us < 0 || lateness_us >= frame_clock->refresh_interval_us)
+  if (lateness_us < 0 || lateness_us >= frame_clock->refresh_interval_us / 4)
     frame_clock->last_dispatch_lateness_us = 0;
   else
     frame_clock->last_dispatch_lateness_us = lateness_us;
-- 
GitLab

From 2fa613c9178333eaed43adbb91f8159016ff63ac Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Fri, 24 Jul 2020 14:13:11 +0800
Subject: [PATCH 19/32] clutter/frame-clock: Merge states DISPATCHING and
 PENDING_PRESENTED

Chronologically they already overlap in time as presentation may
complete in the middle of the dispatch function, otherwise they are
contiguous in time. And most switch statements treated the two states
the same already so they're easy to merge into a single `DISPATCHED`
state.

Having fewer states now will make life easier when we add more states
later.
---
 clutter/clutter/clutter-frame-clock.c | 31 ++++++++++-----------------
 1 file changed, 11 insertions(+), 20 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 48f07e1014e..384537e4178 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -70,8 +70,7 @@ typedef enum _ClutterFrameClockState
   CLUTTER_FRAME_CLOCK_STATE_IDLE,
   CLUTTER_FRAME_CLOCK_STATE_SCHEDULED,
   CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW,
-  CLUTTER_FRAME_CLOCK_STATE_DISPATCHING,
-  CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED,
+  CLUTTER_FRAME_CLOCK_STATE_DISPATCHED,
 } ClutterFrameClockState;
 
 struct _ClutterFrameClock
@@ -413,8 +412,7 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
       g_warn_if_reached ();
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHING:
-    case CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
       frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
       maybe_reschedule_update (frame_clock);
       break;
@@ -435,8 +433,7 @@ clutter_frame_clock_notify_ready (ClutterFrameClock *frame_clock)
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
       g_warn_if_reached ();
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHING:
-    case CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
       frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
       maybe_reschedule_update (frame_clock);
       break;
@@ -735,8 +732,7 @@ clutter_frame_clock_inhibit (ClutterFrameClock *frame_clock)
           frame_clock->pending_reschedule_now = TRUE;
           frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
           break;
-        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHING:
-        case CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED:
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
           break;
         }
 
@@ -775,8 +771,7 @@ clutter_frame_clock_schedule_update_now (ClutterFrameClock *frame_clock)
       break;
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
       return;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHING:
-    case CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
       frame_clock->pending_reschedule = TRUE;
       frame_clock->pending_reschedule_now = TRUE;
       return;
@@ -831,8 +826,7 @@ clutter_frame_clock_schedule_update (ClutterFrameClock *frame_clock)
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
       return;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHING:
-    case CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
       frame_clock->pending_reschedule = TRUE;
       return;
     }
@@ -887,8 +881,7 @@ clutter_frame_clock_set_mode (ClutterFrameClock     *frame_clock,
       frame_clock->pending_reschedule_now = TRUE;
       frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHING:
-    case CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
       break;
     }
 
@@ -948,7 +941,7 @@ clutter_frame_clock_dispatch (ClutterFrameClock *frame_clock,
   frame_clock->last_dispatch_time_us = time_us;
   g_source_set_ready_time (frame_clock->source, -1);
 
-  frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHING;
+  frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED;
 
   frame_count = frame_clock->frame_count++;
 
@@ -982,20 +975,20 @@ clutter_frame_clock_dispatch (ClutterFrameClock *frame_clock,
   switch (frame_clock->state)
     {
     case CLUTTER_FRAME_CLOCK_STATE_INIT:
-    case CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED:
       g_warn_if_reached ();
       break;
     case CLUTTER_FRAME_CLOCK_STATE_IDLE:
+      /* Presentation completed synchronously in the above listener */
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHING:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
       switch (result)
         {
         case CLUTTER_FRAME_RESULT_PENDING_PRESENTED:
-          frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_PENDING_PRESENTED;
           break;
         case CLUTTER_FRAME_RESULT_IDLE:
+          /* The frame was aborted; nothing to paint/present */
           frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
           maybe_reschedule_update (frame_clock);
           break;
@@ -1212,8 +1205,6 @@ clutter_frame_clock_dispose (GObject *object)
 {
   ClutterFrameClock *frame_clock = CLUTTER_FRAME_CLOCK (object);
 
-  g_warn_if_fail (frame_clock->state != CLUTTER_FRAME_CLOCK_STATE_DISPATCHING);
-
   if (frame_clock->source)
     {
       g_signal_emit (frame_clock, signals[DESTROY], 0);
-- 
GitLab

From 982550cfc36ed7ec20b6883b43d3b538c961ae12 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Mon, 13 Dec 2021 17:03:44 +0800
Subject: [PATCH 20/32] clutter/frame-clock: Clamp render times to two frames,
 not one

Because the buffer queue may grow by an extra frame in triple buffering,
we need measurements exceeding one frame to indicate when and by how much
the scaling is necessary.
---
 clutter/clutter/clutter-frame-clock.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 384537e4178..6a11d4b66a0 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -385,7 +385,7 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
         CLAMP (frame_clock->last_dispatch_lateness_us + dispatch_to_swap_us +
                MAX (swap_to_rendering_done_us, swap_to_flip_us),
                frame_clock->shortterm_max_update_duration_us,
-               frame_clock->refresh_interval_us);
+               2 * frame_clock->refresh_interval_us);
 
       maybe_update_longterm_max_duration_us (frame_clock, frame_info);
 
@@ -471,7 +471,7 @@ clutter_frame_clock_compute_max_render_time_us (ClutterFrameClock *frame_clock)
     frame_clock->vblank_duration_us +
     clutter_max_render_time_constant_us;
 
-  max_render_time_us = CLAMP (max_render_time_us, 0, refresh_interval_us);
+  max_render_time_us = CLAMP (max_render_time_us, 0, 2 * refresh_interval_us);
 
   return max_render_time_us;
 }
-- 
GitLab

From 8cb2307647029a588b83b0fae2bcc8225c3d8439 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 10 Sep 2020 16:34:53 +0800
Subject: [PATCH 21/32] clutter/frame-clock: Add triple buffering support

---
 clutter/clutter/clutter-frame-clock.c | 198 ++++++++++++++++++++++----
 1 file changed, 170 insertions(+), 28 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 6a11d4b66a0..32d2b2d874d 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -70,7 +70,10 @@ typedef enum _ClutterFrameClockState
   CLUTTER_FRAME_CLOCK_STATE_IDLE,
   CLUTTER_FRAME_CLOCK_STATE_SCHEDULED,
   CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW,
-  CLUTTER_FRAME_CLOCK_STATE_DISPATCHED,
+  CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE,
+  CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED,
+  CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW,
+  CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO,
 } ClutterFrameClockState;
 
 struct _ClutterFrameClock
@@ -91,6 +94,7 @@ struct _ClutterFrameClock
   ClutterFrameClockMode mode;
 
   int64_t last_dispatch_time_us;
+  int64_t prev_last_dispatch_time_us;
   int64_t last_dispatch_lateness_us;
   int64_t last_presentation_time_us;
   int64_t next_update_time_us;
@@ -110,6 +114,7 @@ struct _ClutterFrameClock
   int64_t vblank_duration_us;
   /* Last KMS buffer submission time. */
   int64_t last_flip_time_us;
+  int64_t prev_last_flip_time_us;
 
   /* Last time we promoted short-term maximum to long-term one */
   int64_t longterm_promotion_us;
@@ -364,14 +369,35 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
       frame_info->has_valid_gpu_rendering_duration)
     {
       int64_t dispatch_to_swap_us, swap_to_rendering_done_us, swap_to_flip_us;
+      int64_t dispatch_time_us = 0, flip_time_us = 0;
+
+      switch (frame_clock->state)
+        {
+        case CLUTTER_FRAME_CLOCK_STATE_INIT:
+        case CLUTTER_FRAME_CLOCK_STATE_IDLE:
+        case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
+        case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
+          g_warn_if_reached ();
+          G_GNUC_FALLTHROUGH;
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+          dispatch_time_us = frame_clock->last_dispatch_time_us;
+          flip_time_us = frame_clock->last_flip_time_us;
+          break;
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
+          dispatch_time_us = frame_clock->prev_last_dispatch_time_us;
+          flip_time_us = frame_clock->prev_last_flip_time_us;
+          break;
+        }
 
       dispatch_to_swap_us =
         frame_info->cpu_time_before_buffer_swap_us -
-        frame_clock->last_dispatch_time_us;
+        dispatch_time_us;
       swap_to_rendering_done_us =
         frame_info->gpu_rendering_duration_ns / 1000;
       swap_to_flip_us =
-        frame_clock->last_flip_time_us -
+        flip_time_us -
         frame_info->cpu_time_before_buffer_swap_us;
 
       CLUTTER_NOTE (FRAME_TIMINGS,
@@ -412,10 +438,22 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
       g_warn_if_reached ();
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
       frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
       maybe_reschedule_update (frame_clock);
       break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED;
+      maybe_reschedule_update (frame_clock);
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW;
+      maybe_reschedule_update (frame_clock);
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
+      maybe_reschedule_update (frame_clock);
+      break;
     }
 }
 
@@ -433,10 +471,22 @@ clutter_frame_clock_notify_ready (ClutterFrameClock *frame_clock)
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
       g_warn_if_reached ();
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
       frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
       maybe_reschedule_update (frame_clock);
       break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED;
+      maybe_reschedule_update (frame_clock);
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW;
+      maybe_reschedule_update (frame_clock);
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
+      maybe_reschedule_update (frame_clock);
+      break;
     }
 }
 
@@ -451,7 +501,14 @@ clutter_frame_clock_compute_max_render_time_us (ClutterFrameClock *frame_clock)
   if (!frame_clock->ever_got_measurements ||
       G_UNLIKELY (clutter_paint_debug_flags &
                   CLUTTER_DEBUG_DISABLE_DYNAMIC_MAX_RENDER_TIME))
-    return refresh_interval_us * SYNC_DELAY_FALLBACK_FRACTION;
+    {
+      int64_t ret = refresh_interval_us * SYNC_DELAY_FALLBACK_FRACTION;
+
+      if (frame_clock->state == CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE)
+        ret += refresh_interval_us;
+
+      return ret;
+    }
 
   /* Max render time shows how early the frame clock needs to be dispatched
    * to make it to the predicted next presentation time. It is an estimate of
@@ -488,7 +545,7 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
   int64_t min_render_time_allowed_us;
   int64_t max_render_time_allowed_us;
   int64_t next_presentation_time_us;
-  int64_t next_smooth_presentation_time_us;
+  int64_t next_smooth_presentation_time_us = 0;
   int64_t next_update_time_us;
 
   now_us = g_get_monotonic_time ();
@@ -533,7 +590,27 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
    *
    */
   last_presentation_time_us = frame_clock->last_presentation_time_us;
-  next_smooth_presentation_time_us = last_presentation_time_us + refresh_interval_us;
+  switch (frame_clock->state)
+    {
+    case CLUTTER_FRAME_CLOCK_STATE_INIT:
+    case CLUTTER_FRAME_CLOCK_STATE_IDLE:
+    case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
+    case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
+      next_smooth_presentation_time_us = last_presentation_time_us +
+                                         refresh_interval_us;
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+      next_smooth_presentation_time_us = last_presentation_time_us +
+                                         2 * refresh_interval_us;
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
+      next_smooth_presentation_time_us = last_presentation_time_us +
+                                         3 * refresh_interval_us;
+      break;
+    }
+
   next_presentation_time_us = next_smooth_presentation_time_us;
 
   /*
@@ -732,7 +809,17 @@ clutter_frame_clock_inhibit (ClutterFrameClock *frame_clock)
           frame_clock->pending_reschedule_now = TRUE;
           frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
           break;
-        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+          frame_clock->pending_reschedule = TRUE;
+          frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
+          break;
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+          frame_clock->pending_reschedule = TRUE;
+          frame_clock->pending_reschedule_now = TRUE;
+          frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
+          break;
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
           break;
         }
 
@@ -768,10 +855,18 @@ clutter_frame_clock_schedule_update_now (ClutterFrameClock *frame_clock)
     case CLUTTER_FRAME_CLOCK_STATE_INIT:
     case CLUTTER_FRAME_CLOCK_STATE_IDLE:
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW;
       break;
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
       return;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+      next_update_time_us = g_get_monotonic_time ();
+      frame_clock->state =
+        CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW;
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
       frame_clock->pending_reschedule = TRUE;
       frame_clock->pending_reschedule_now = TRUE;
       return;
@@ -800,7 +895,6 @@ clutter_frame_clock_schedule_update_now (ClutterFrameClock *frame_clock)
 
   frame_clock->next_update_time_us = next_update_time_us;
   g_source_set_ready_time (frame_clock->source, next_update_time_us);
-  frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW;
 }
 
 void
@@ -822,11 +916,23 @@ clutter_frame_clock_schedule_update (ClutterFrameClock *frame_clock)
       frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED;
       return;
     case CLUTTER_FRAME_CLOCK_STATE_IDLE:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED;
       break;
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
       return;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+      calculate_next_update_time_us (frame_clock,
+                                     &next_update_time_us,
+                                     &frame_clock->next_presentation_time_us,
+                                     &frame_clock->next_frame_deadline_us);
+      frame_clock->is_next_presentation_time_valid =
+        (frame_clock->next_presentation_time_us != 0);
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED;
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
       frame_clock->pending_reschedule = TRUE;
       return;
     }
@@ -855,7 +961,6 @@ clutter_frame_clock_schedule_update (ClutterFrameClock *frame_clock)
 
   frame_clock->next_update_time_us = next_update_time_us;
   g_source_set_ready_time (frame_clock->source, next_update_time_us);
-  frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED;
 }
 
 void
@@ -871,6 +976,8 @@ clutter_frame_clock_set_mode (ClutterFrameClock     *frame_clock,
     {
     case CLUTTER_FRAME_CLOCK_STATE_INIT:
     case CLUTTER_FRAME_CLOCK_STATE_IDLE:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
       break;
     case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
       frame_clock->pending_reschedule = TRUE;
@@ -881,7 +988,14 @@ clutter_frame_clock_set_mode (ClutterFrameClock     *frame_clock,
       frame_clock->pending_reschedule_now = TRUE;
       frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+      frame_clock->pending_reschedule = TRUE;
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+      frame_clock->pending_reschedule = TRUE;
+      frame_clock->pending_reschedule_now = TRUE;
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
       break;
     }
 
@@ -938,10 +1052,27 @@ clutter_frame_clock_dispatch (ClutterFrameClock *frame_clock,
     }
 #endif
 
+  frame_clock->prev_last_dispatch_time_us = frame_clock->last_dispatch_time_us;
   frame_clock->last_dispatch_time_us = time_us;
   g_source_set_ready_time (frame_clock->source, -1);
 
-  frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED;
+  switch (frame_clock->state)
+    {
+    case CLUTTER_FRAME_CLOCK_STATE_INIT:
+    case CLUTTER_FRAME_CLOCK_STATE_IDLE:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
+      g_warn_if_reached ();
+      return;
+    case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
+    case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
+      break;
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO;
+      break;
+    }
 
   frame_count = frame_clock->frame_count++;
 
@@ -972,26 +1103,36 @@ clutter_frame_clock_dispatch (ClutterFrameClock *frame_clock,
   result = iface->frame (frame_clock, frame, frame_clock->listener.user_data);
   COGL_TRACE_END (ClutterFrameClockFrame);
 
-  switch (frame_clock->state)
+  switch (result)
     {
-    case CLUTTER_FRAME_CLOCK_STATE_INIT:
-      g_warn_if_reached ();
+    case CLUTTER_FRAME_RESULT_PENDING_PRESENTED:
       break;
-    case CLUTTER_FRAME_CLOCK_STATE_IDLE:
-      /* Presentation completed synchronously in the above listener */
-    case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
-    case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
-      break;
-    case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED:
-      switch (result)
+    case CLUTTER_FRAME_RESULT_IDLE:
+      /* The frame was aborted; nothing to paint/present */
+      switch (frame_clock->state)
         {
-        case CLUTTER_FRAME_RESULT_PENDING_PRESENTED:
+        case CLUTTER_FRAME_CLOCK_STATE_INIT:
+        case CLUTTER_FRAME_CLOCK_STATE_IDLE:
+        case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED:
+        case CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW:
+          g_warn_if_reached ();
           break;
-        case CLUTTER_FRAME_RESULT_IDLE:
-          /* The frame was aborted; nothing to paint/present */
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
           frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_IDLE;
           maybe_reschedule_update (frame_clock);
           break;
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED:
+          frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED;
+          maybe_reschedule_update (frame_clock);
+          break;
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
+          frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_SCHEDULED_NOW;
+          maybe_reschedule_update (frame_clock);
+          break;
+        case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
+          frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE;
+          maybe_reschedule_update (frame_clock);
+          break;
         }
       break;
     }
@@ -1027,6 +1168,7 @@ void
 clutter_frame_clock_record_flip_time (ClutterFrameClock *frame_clock,
                                       int64_t            flip_time_us)
 {
+  frame_clock->prev_last_flip_time_us = frame_clock->last_flip_time_us;
   frame_clock->last_flip_time_us = flip_time_us;
 }
 
-- 
GitLab

From a790d7e8638db72eac5d3d2308db8deb805ddd01 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Fri, 10 Dec 2021 16:28:04 +0800
Subject: [PATCH 22/32] clutter/frame-clock: Log N-buffers in
 CLUTTTER_DEBUG=frame-timings

---
 clutter/clutter/clutter-frame-clock.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 32d2b2d874d..e3fb2308640 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -279,6 +279,12 @@ void
 clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
                                       ClutterFrameInfo  *frame_info)
 {
+#ifdef CLUTTER_ENABLE_DEBUG
+  const char *debug_state =
+    frame_clock->state == CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO ?
+    "Triple buffering" : "Double buffering";
+#endif
+
   COGL_TRACE_BEGIN_SCOPED (ClutterFrameClockNotifyPresented,
                            "Clutter::FrameClock::presented()");
   COGL_TRACE_DESCRIBE (ClutterFrameClockNotifyPresented,
@@ -401,7 +407,8 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
         frame_info->cpu_time_before_buffer_swap_us;
 
       CLUTTER_NOTE (FRAME_TIMINGS,
-                    "update2dispatch %ld µs, dispatch2swap %ld µs, swap2render %ld µs, swap2flip %ld µs",
+                    "%s: update2dispatch %ld µs, dispatch2swap %ld µs, swap2render %ld µs, swap2flip %ld µs",
+                    debug_state,
                     frame_clock->last_dispatch_lateness_us,
                     dispatch_to_swap_us,
                     swap_to_rendering_done_us,
@@ -420,7 +427,8 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
     }
   else
     {
-      CLUTTER_NOTE (FRAME_TIMINGS, "update2dispatch %ld µs",
+      CLUTTER_NOTE (FRAME_TIMINGS, "%s: update2dispatch %ld µs",
+                    debug_state,
                     frame_clock->last_dispatch_lateness_us);
     }
 
-- 
GitLab

From 004f9f7d7b52bffbdacf6579a41331b351276d98 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 7 Sep 2021 19:08:15 +0800
Subject: [PATCH 23/32] clutter/frame: Add ClutterFrameHint to ClutterFrame

This will allow the backend to provide performance hints to the frame
clock in future.
---
 clutter/clutter/clutter-frame-clock.h   |  6 ++++++
 clutter/clutter/clutter-frame-private.h |  1 +
 clutter/clutter/clutter-frame.c         | 13 +++++++++++++
 clutter/clutter/clutter-frame.h         |  7 +++++++
 4 files changed, 27 insertions(+)

diff --git a/clutter/clutter/clutter-frame-clock.h b/clutter/clutter/clutter-frame-clock.h
index a7be5ef3167..5adf995098b 100644
--- a/clutter/clutter/clutter-frame-clock.h
+++ b/clutter/clutter/clutter-frame-clock.h
@@ -33,6 +33,12 @@ typedef enum _ClutterFrameResult
   CLUTTER_FRAME_RESULT_IDLE,
 } ClutterFrameResult;
 
+typedef enum _ClutterFrameHint
+{
+  CLUTTER_FRAME_HINT_NONE                     = 0,
+  CLUTTER_FRAME_HINT_DIRECT_SCANOUT_ATTEMPTED = 1 << 0,
+} ClutterFrameHint;
+
 #define CLUTTER_TYPE_FRAME_CLOCK (clutter_frame_clock_get_type ())
 CLUTTER_EXPORT
 G_DECLARE_FINAL_TYPE (ClutterFrameClock, clutter_frame_clock,
diff --git a/clutter/clutter/clutter-frame-private.h b/clutter/clutter/clutter-frame-private.h
index ef66b874edf..ce140560a89 100644
--- a/clutter/clutter/clutter-frame-private.h
+++ b/clutter/clutter/clutter-frame-private.h
@@ -36,6 +36,7 @@ struct _ClutterFrame
 
   gboolean has_result;
   ClutterFrameResult result;
+  ClutterFrameHint hints;
 };
 
 CLUTTER_EXPORT
diff --git a/clutter/clutter/clutter-frame.c b/clutter/clutter/clutter-frame.c
index 7436f9f182d..53c289b2c5a 100644
--- a/clutter/clutter/clutter-frame.c
+++ b/clutter/clutter/clutter-frame.c
@@ -115,3 +115,16 @@ clutter_frame_set_result (ClutterFrame       *frame,
   frame->result = result;
   frame->has_result = TRUE;
 }
+
+void
+clutter_frame_set_hint (ClutterFrame     *frame,
+                        ClutterFrameHint  hint)
+{
+  frame->hints |= hint;
+}
+
+ClutterFrameHint
+clutter_frame_get_hints (ClutterFrame *frame)
+{
+  return frame->hints;
+}
diff --git a/clutter/clutter/clutter-frame.h b/clutter/clutter/clutter-frame.h
index 34f0770bd7d..c7b3d02acb9 100644
--- a/clutter/clutter/clutter-frame.h
+++ b/clutter/clutter/clutter-frame.h
@@ -54,4 +54,11 @@ void clutter_frame_set_result (ClutterFrame       *frame,
 CLUTTER_EXPORT
 gboolean clutter_frame_has_result (ClutterFrame *frame);
 
+CLUTTER_EXPORT
+void clutter_frame_set_hint (ClutterFrame     *frame,
+                             ClutterFrameHint  hint);
+
+CLUTTER_EXPORT
+ClutterFrameHint clutter_frame_get_hints (ClutterFrame *frame);
+
 G_DEFINE_AUTOPTR_CLEANUP_FUNC (ClutterFrame, clutter_frame_unref)
-- 
GitLab

From ef4e11b6420f27375855815f679a6194e4f34835 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 7 Sep 2021 19:10:26 +0800
Subject: [PATCH 24/32] backends: Flag that the frame attempted direct scanout

We need this hint whether direct scanout succeeds or fails because it's
the mechanism by which we will tell the clock to enforce double buffering,
thus making direct scanout possible on future frames. Triple buffering
will be disabled until such time that direct scanout is not being attempted.
---
 src/backends/meta-stage-impl.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/backends/meta-stage-impl.c b/src/backends/meta-stage-impl.c
index 7aa24439d73..727e1a5f30b 100644
--- a/src/backends/meta-stage-impl.c
+++ b/src/backends/meta-stage-impl.c
@@ -774,6 +774,8 @@ meta_stage_impl_redraw_view (ClutterStageWindow *stage_window,
     {
       g_autoptr (GError) error = NULL;
 
+      clutter_frame_set_hint (frame, CLUTTER_FRAME_HINT_DIRECT_SCANOUT_ATTEMPTED);
+
       if (meta_stage_impl_scanout_view (stage_impl,
                                         stage_view,
                                         scanout,
-- 
GitLab

From 7663ee635af88c75ca539938f26d639a37239355 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 7 Sep 2021 19:15:18 +0800
Subject: [PATCH 25/32] clutter: Pass ClutterFrameHint(s) to the frame clock

---
 clutter/clutter/clutter-frame-clock.c |  8 ++++++--
 clutter/clutter/clutter-frame-clock.h |  5 +++--
 clutter/clutter/clutter-stage-view.c  | 11 +++++++++--
 3 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index e3fb2308640..84d2f1f3113 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -116,6 +116,8 @@ struct _ClutterFrameClock
   int64_t last_flip_time_us;
   int64_t prev_last_flip_time_us;
 
+  ClutterFrameHint last_flip_hints;
+
   /* Last time we promoted short-term maximum to long-term one */
   int64_t longterm_promotion_us;
   /* Long-term maximum update duration */
@@ -1173,11 +1175,13 @@ frame_clock_source_dispatch (GSource     *source,
 }
 
 void
-clutter_frame_clock_record_flip_time (ClutterFrameClock *frame_clock,
-                                      int64_t            flip_time_us)
+clutter_frame_clock_record_flip (ClutterFrameClock *frame_clock,
+                                 int64_t            flip_time_us,
+                                 ClutterFrameHint   hints)
 {
   frame_clock->prev_last_flip_time_us = frame_clock->last_flip_time_us;
   frame_clock->last_flip_time_us = flip_time_us;
+  frame_clock->last_flip_hints = hints;
 }
 
 GString *
diff --git a/clutter/clutter/clutter-frame-clock.h b/clutter/clutter/clutter-frame-clock.h
index 5adf995098b..bfc89bde091 100644
--- a/clutter/clutter/clutter-frame-clock.h
+++ b/clutter/clutter/clutter-frame-clock.h
@@ -108,7 +108,8 @@ void clutter_frame_clock_remove_timeline (ClutterFrameClock *frame_clock,
 CLUTTER_EXPORT
 float clutter_frame_clock_get_refresh_rate (ClutterFrameClock *frame_clock);
 
-void clutter_frame_clock_record_flip_time (ClutterFrameClock *frame_clock,
-                                           int64_t            flip_time_us);
+void clutter_frame_clock_record_flip (ClutterFrameClock *frame_clock,
+                                      int64_t            flip_time_us,
+                                      ClutterFrameHint   hints);
 
 GString * clutter_frame_clock_get_max_render_time_debug_info (ClutterFrameClock *frame_clock);
diff --git a/clutter/clutter/clutter-stage-view.c b/clutter/clutter/clutter-stage-view.c
index b503ef839a0..1fbe3ae7a33 100644
--- a/clutter/clutter/clutter-stage-view.c
+++ b/clutter/clutter/clutter-stage-view.c
@@ -902,14 +902,21 @@ handle_frame_clock_frame (ClutterFrameClock *frame_clock,
 
       _clutter_stage_window_redraw_view (stage_window, view, frame);
 
-      clutter_frame_clock_record_flip_time (frame_clock,
-                                            g_get_monotonic_time ());
+      clutter_frame_clock_record_flip (frame_clock,
+                                       g_get_monotonic_time (),
+                                       clutter_frame_get_hints (frame));
 
       clutter_stage_emit_after_paint (stage, view, frame);
 
       if (_clutter_context_get_show_fps ())
         end_frame_timing_measurement (view);
     }
+  else
+    {
+      clutter_frame_clock_record_flip (frame_clock,
+                                       g_get_monotonic_time (),
+                                       clutter_frame_get_hints (frame));
+    }
 
   _clutter_stage_window_finish_frame (stage_window, view, frame);
 
-- 
GitLab

From d43d23079900ff0e07f17f48236d1d3594cd1c4c Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 7 Sep 2021 19:15:55 +0800
Subject: [PATCH 26/32] clutter/frame-clock: Throttle back to double buffering
 for direct scanout

There's no compositing during direct scanout so the "render" time is zero.
Thus there is no need to implement triple buffering for direct scanouts.
Stick to double buffering and enjoy the lower latency.
---
 clutter/clutter/clutter-frame-clock.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 84d2f1f3113..c02eb85e26d 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -934,6 +934,13 @@ clutter_frame_clock_schedule_update (ClutterFrameClock *frame_clock)
     case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
       return;
     case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
+      if (frame_clock->last_flip_hints & CLUTTER_FRAME_HINT_DIRECT_SCANOUT_ATTEMPTED)
+        {
+          /* Force double buffering, disable triple buffering */
+          frame_clock->pending_reschedule = TRUE;
+          return;
+        }
+
       calculate_next_update_time_us (frame_clock,
                                      &next_update_time_us,
                                      &frame_clock->next_presentation_time_us,
-- 
GitLab

From 3a756aa16a4576789fc5b1c614cb0a135fb78528 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 10 Mar 2022 16:44:14 +0800
Subject: [PATCH 27/32] clutter/frame-clock: Add environment variable
 MUTTER_DEBUG_TRIPLE_BUFFERING

With possible values {never, auto, always} where auto is the default.

This also allows some unification with automatic throttling for
direct scanout.
---
 clutter/clutter/clutter-frame-clock.c | 55 ++++++++++++++++++++++-----
 1 file changed, 45 insertions(+), 10 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index c02eb85e26d..142f670f519 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -42,6 +42,15 @@ enum
 
 static guint signals[N_SIGNALS];
 
+typedef enum
+{
+  TRIPLE_BUFFERING_MODE_NEVER,
+  TRIPLE_BUFFERING_MODE_AUTO,
+  TRIPLE_BUFFERING_MODE_ALWAYS,
+} TripleBufferingMode;
+
+static TripleBufferingMode triple_buffering_mode = TRIPLE_BUFFERING_MODE_AUTO;
+
 #define SYNC_DELAY_FALLBACK_FRACTION 0.875
 
 #define MINIMUM_REFRESH_RATE 30.f
@@ -911,6 +920,12 @@ void
 clutter_frame_clock_schedule_update (ClutterFrameClock *frame_clock)
 {
   int64_t next_update_time_us = -1;
+  TripleBufferingMode current_mode = triple_buffering_mode;
+
+  if (current_mode == TRIPLE_BUFFERING_MODE_AUTO &&
+      (frame_clock->last_flip_hints &
+       CLUTTER_FRAME_HINT_DIRECT_SCANOUT_ATTEMPTED))
+    current_mode = TRIPLE_BUFFERING_MODE_NEVER;
 
   if (frame_clock->inhibit_count > 0)
     {
@@ -934,20 +949,31 @@ clutter_frame_clock_schedule_update (ClutterFrameClock *frame_clock)
     case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED_NOW:
       return;
     case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE:
-      if (frame_clock->last_flip_hints & CLUTTER_FRAME_HINT_DIRECT_SCANOUT_ATTEMPTED)
+      switch (current_mode)
         {
-          /* Force double buffering, disable triple buffering */
+        case TRIPLE_BUFFERING_MODE_NEVER:
           frame_clock->pending_reschedule = TRUE;
           return;
+        case TRIPLE_BUFFERING_MODE_AUTO:
+          calculate_next_update_time_us (frame_clock,
+                                         &next_update_time_us,
+                                         &frame_clock->next_presentation_time_us,
+                                         &frame_clock->next_frame_deadline_us);
+          frame_clock->is_next_presentation_time_valid =
+            (frame_clock->next_presentation_time_us != 0);
+          frame_clock->has_next_frame_deadline =
+            (frame_clock->next_frame_deadline_us != 0);
+          frame_clock->state =
+            CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED;
+          break;
+        case TRIPLE_BUFFERING_MODE_ALWAYS:
+          next_update_time_us = g_get_monotonic_time ();
+          frame_clock->next_presentation_time_us = 0;
+          frame_clock->is_next_presentation_time_valid = FALSE;
+          frame_clock->state =
+            CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED;
+          break;
         }
-
-      calculate_next_update_time_us (frame_clock,
-                                     &next_update_time_us,
-                                     &frame_clock->next_presentation_time_us,
-                                     &frame_clock->next_frame_deadline_us);
-      frame_clock->is_next_presentation_time_valid =
-        (frame_clock->next_presentation_time_us != 0);
-      frame_clock->state = CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE_AND_SCHEDULED;
       break;
     case CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_TWO:
       frame_clock->pending_reschedule = TRUE;
@@ -1389,6 +1415,15 @@ static void
 clutter_frame_clock_class_init (ClutterFrameClockClass *klass)
 {
   GObjectClass *object_class = G_OBJECT_CLASS (klass);
+  const char *mode_str;
+
+  mode_str = g_getenv ("MUTTER_DEBUG_TRIPLE_BUFFERING");
+  if (!g_strcmp0 (mode_str, "never"))
+    triple_buffering_mode = TRIPLE_BUFFERING_MODE_NEVER;
+  else if (!g_strcmp0 (mode_str, "auto"))
+    triple_buffering_mode = TRIPLE_BUFFERING_MODE_AUTO;
+  else if (!g_strcmp0 (mode_str, "always"))
+    triple_buffering_mode = TRIPLE_BUFFERING_MODE_ALWAYS;
 
   object_class->dispose = clutter_frame_clock_dispose;
 
-- 
GitLab

From aa692ddfb6588309460bca7341d474429a6551d4 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 30 Jun 2022 18:56:06 +0800
Subject: [PATCH 28/32] tests/native-kms-render: Fix failing client-scanout
 test

It was assuming an immediate transition from compositing (triple
buffering) to direct scanout (double buffering), whereas there is
a one frame delay in that transition as the buffer queue shrinks.
We don't lose any frames in the transition.
---
 src/tests/native-kms-render.c | 106 ++++++++++++++++++++++++++++------
 1 file changed, 87 insertions(+), 19 deletions(-)

diff --git a/src/tests/native-kms-render.c b/src/tests/native-kms-render.c
index f5ebc23fec7..2f870fdc331 100644
--- a/src/tests/native-kms-render.c
+++ b/src/tests/native-kms-render.c
@@ -39,6 +39,8 @@
 #include "tests/meta-wayland-test-driver.h"
 #include "tests/meta-wayland-test-utils.h"
 
+#define N_FRAMES_PER_TEST 30
+
 typedef struct
 {
   int number_of_frames_left;
@@ -46,12 +48,15 @@ typedef struct
 
   struct {
     int n_paints;
-    uint32_t fb_id;
+    int n_presentations;
+    int n_direct_scanouts;
+    GList *fb_ids;
   } scanout;
 
   gboolean wait_for_scanout;
 
   struct {
+    int scanouts_attempted;
     gboolean scanout_sabotaged;
     gboolean fallback_painted;
     guint repaint_guard_id;
@@ -101,7 +106,7 @@ meta_test_kms_render_basic (void)
   gulong handler_id;
 
   test = (KmsRenderingTest) {
-    .number_of_frames_left = 10,
+    .number_of_frames_left = N_FRAMES_PER_TEST,
     .loop = g_main_loop_new (NULL, FALSE),
   };
   handler_id = g_signal_connect (stage, "after-update",
@@ -123,7 +128,6 @@ on_scanout_before_update (ClutterStage     *stage,
                           KmsRenderingTest *test)
 {
   test->scanout.n_paints = 0;
-  test->scanout.fb_id = 0;
 }
 
 static void
@@ -135,6 +139,7 @@ on_scanout_before_paint (ClutterStage     *stage,
   CoglScanout *scanout;
   CoglScanoutBuffer *scanout_buffer;
   MetaDrmBuffer *buffer;
+  uint32_t fb_id;
 
   scanout = clutter_stage_view_peek_scanout (stage_view);
   if (!scanout)
@@ -143,8 +148,13 @@ on_scanout_before_paint (ClutterStage     *stage,
   scanout_buffer = cogl_scanout_get_buffer (scanout);
   g_assert_true (META_IS_DRM_BUFFER (scanout_buffer));
   buffer = META_DRM_BUFFER (scanout_buffer);
-  test->scanout.fb_id = meta_drm_buffer_get_fb_id (buffer);
-  g_assert_cmpuint (test->scanout.fb_id, >, 0);
+  fb_id = meta_drm_buffer_get_fb_id (buffer);
+  g_assert_cmpuint (fb_id, >, 0);
+  test->scanout.fb_ids = g_list_append (test->scanout.fb_ids,
+                                        GUINT_TO_POINTER (fb_id));
+
+  /* Triple buffering, but no higher */
+  g_assert_cmpuint (g_list_length (test->scanout.fb_ids), <=, 2);
 }
 
 static void
@@ -173,12 +183,12 @@ on_scanout_presented (ClutterStage     *stage,
   MetaDeviceFile *device_file;
   GError *error = NULL;
   drmModeCrtc *drm_crtc;
+  uint32_t first_fb_id_expected;
 
-  if (test->wait_for_scanout && test->scanout.n_paints > 0)
+  if (test->wait_for_scanout && test->scanout.fb_ids == NULL)
     return;
 
-  if (test->wait_for_scanout && test->scanout.fb_id == 0)
-    return;
+  test->scanout.n_presentations++;
 
   device_pool = meta_backend_native_get_device_pool (backend_native);
 
@@ -197,15 +207,41 @@ on_scanout_presented (ClutterStage     *stage,
   drm_crtc = drmModeGetCrtc (meta_device_file_get_fd (device_file),
                              meta_kms_crtc_get_id (kms_crtc));
   g_assert_nonnull (drm_crtc);
-  if (test->scanout.fb_id == 0)
-    g_assert_cmpuint (drm_crtc->buffer_id, !=, test->scanout.fb_id);
+
+  if (test->scanout.fb_ids)
+    {
+      test->scanout.n_direct_scanouts++;
+      first_fb_id_expected = GPOINTER_TO_UINT (test->scanout.fb_ids->data);
+      test->scanout.fb_ids = g_list_delete_link (test->scanout.fb_ids,
+                                                 test->scanout.fb_ids);
+    }
   else
-    g_assert_cmpuint (drm_crtc->buffer_id, ==, test->scanout.fb_id);
+    {
+      first_fb_id_expected = 0;
+    }
+
+  /* The buffer ID won't match on the first frame because switching from
+   * triple buffered compositing to double buffered direct scanout takes
+   * an extra frame to drain the queue. Thereafter we are in direct scanout
+   * mode and expect the buffer IDs to match.
+   */
+  if (test->scanout.n_presentations > 1)
+    {
+      if (first_fb_id_expected == 0)
+        g_assert_cmpuint (drm_crtc->buffer_id, !=, first_fb_id_expected);
+      else
+        g_assert_cmpuint (drm_crtc->buffer_id, ==, first_fb_id_expected);
+    }
+
   drmModeFreeCrtc (drm_crtc);
 
   meta_device_file_release (device_file);
 
-  g_main_loop_quit (test->loop);
+  test->number_of_frames_left--;
+  if (test->number_of_frames_left <= 0)
+    g_main_loop_quit (test->loop);
+  else
+    clutter_actor_queue_redraw (CLUTTER_ACTOR (stage));
 }
 
 typedef enum
@@ -244,7 +280,9 @@ meta_test_kms_render_client_scanout (void)
   g_assert_nonnull (wayland_test_client);
 
   test = (KmsRenderingTest) {
+    .number_of_frames_left = N_FRAMES_PER_TEST,
     .loop = g_main_loop_new (NULL, FALSE),
+    .scanout = {0},
     .wait_for_scanout = TRUE,
   };
 
@@ -270,7 +308,8 @@ meta_test_kms_render_client_scanout (void)
   clutter_actor_queue_redraw (CLUTTER_ACTOR (stage));
   g_main_loop_run (test.loop);
 
-  g_assert_cmpuint (test.scanout.fb_id, >, 0);
+  g_assert_cmpint (test.scanout.n_presentations, ==, N_FRAMES_PER_TEST);
+  g_assert_cmpint (test.scanout.n_direct_scanouts, ==, N_FRAMES_PER_TEST);
 
   g_debug ("Unmake fullscreen");
   window = meta_find_window_from_title (test_context, "dma-buf-scanout-test");
@@ -292,10 +331,15 @@ meta_test_kms_render_client_scanout (void)
   g_assert_cmpint (buffer_rect.y, ==, 10);
 
   test.wait_for_scanout = FALSE;
+  test.number_of_frames_left = N_FRAMES_PER_TEST;
+  test.scanout.n_presentations = 0;
+  test.scanout.n_direct_scanouts = 0;
+
   clutter_actor_queue_redraw (CLUTTER_ACTOR (stage));
   g_main_loop_run (test.loop);
 
-  g_assert_cmpuint (test.scanout.fb_id, ==, 0);
+  g_assert_cmpint (test.scanout.n_presentations, ==, N_FRAMES_PER_TEST);
+  g_assert_cmpint (test.scanout.n_direct_scanouts, ==, 0);
 
   g_debug ("Moving back to 0, 0");
   meta_window_move_frame (window, TRUE, 0, 0);
@@ -307,10 +351,15 @@ meta_test_kms_render_client_scanout (void)
   g_assert_cmpint (buffer_rect.y, ==, 0);
 
   test.wait_for_scanout = TRUE;
+  test.number_of_frames_left = N_FRAMES_PER_TEST;
+  test.scanout.n_presentations = 0;
+  test.scanout.n_direct_scanouts = 0;
+
   clutter_actor_queue_redraw (CLUTTER_ACTOR (stage));
   g_main_loop_run (test.loop);
 
-  g_assert_cmpuint (test.scanout.fb_id, >, 0);
+  g_assert_cmpint (test.scanout.n_presentations, ==, N_FRAMES_PER_TEST);
+  g_assert_cmpint (test.scanout.n_direct_scanouts, ==, N_FRAMES_PER_TEST);
 
   g_signal_handler_disconnect (stage, before_update_handler_id);
   g_signal_handler_disconnect (stage, before_paint_handler_id);
@@ -364,6 +413,15 @@ on_scanout_fallback_before_paint (ClutterStage     *stage,
   if (!scanout)
     return;
 
+  test->scanout_fallback.scanouts_attempted++;
+
+  /* The first scanout candidate frame will get composited due to triple
+   * buffering draining the queue to drop to double buffering. So don't
+   * sabotage that first frame.
+   */
+  if (test->scanout_fallback.scanouts_attempted < 2)
+    return;
+
   g_assert_false (test->scanout_fallback.scanout_sabotaged);
 
   if (is_atomic_mode_setting (kms_device))
@@ -401,6 +459,15 @@ on_scanout_fallback_paint_view (ClutterStage     *stage,
       g_clear_handle_id (&test->scanout_fallback.repaint_guard_id,
                          g_source_remove);
       test->scanout_fallback.fallback_painted = TRUE;
+      test->scanout_fallback.scanout_sabotaged = FALSE;
+    }
+  else if (test->scanout_fallback.scanouts_attempted == 1)
+    {
+      /* Now that we've seen the first scanout attempt that was inhibited by
+       * triple buffering, try a second frame. The second one should scanout
+       * and will be sabotaged.
+       */
+      clutter_actor_queue_redraw (CLUTTER_ACTOR (stage));
     }
 }
 
@@ -410,11 +477,11 @@ on_scanout_fallback_presented (ClutterStage     *stage,
                                ClutterFrameInfo *frame_info,
                                KmsRenderingTest *test)
 {
-  if (!test->scanout_fallback.scanout_sabotaged)
-    return;
+  if (test->scanout_fallback.fallback_painted)
+    g_main_loop_quit (test->loop);
 
-  g_assert_true (test->scanout_fallback.fallback_painted);
-  g_main_loop_quit (test->loop);
+  test->number_of_frames_left--;
+  g_assert_cmpint (test->number_of_frames_left, >, 0);
 }
 
 static void
@@ -443,6 +510,7 @@ meta_test_kms_render_client_scanout_fallback (void)
   g_assert_nonnull (wayland_test_client);
 
   test = (KmsRenderingTest) {
+    .number_of_frames_left = N_FRAMES_PER_TEST,
     .loop = g_main_loop_new (NULL, FALSE),
   };
 
-- 
GitLab

From 707bf11d6c4bdde75573498b9df28b02d93d56c5 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu, 22 Jun 2023 15:19:53 +0800
Subject: [PATCH 29/32] onscreen/native: Avoid callbacks on "detached"
 onscreens

Detached onscreens have no valid view so avoid servicing callbacks on
them during/after sleep mode. As previously mentioned in 45bda2d969f.

Closes: https://launchpad.net/bugs/2020049
---
 src/backends/native/meta-onscreen-native.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 26a2cf90a32..20cb96e4eb7 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -1504,6 +1504,7 @@ try_post_latest_swap (CoglOnscreen *onscreen)
   MetaFrameNative *frame_native;
 
   if (onscreen_native->next_post.frame == NULL ||
+      onscreen_native->view == NULL ||
       meta_kms_is_shutting_down (kms))
     return;
 
-- 
GitLab

From 3d1a2183b454d4a04705295072db39a2f3be7dd7 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Tue, 18 Jul 2023 16:08:25 +0800
Subject: [PATCH 30/32] clutter/frame-clock: Record measurements of zero for
 cursor-only updates

But only if we've ever got actual swap measurements
(COGL_FEATURE_ID_TIMESTAMP_QUERY). If it's supported then we now drop to
double buffering and get optimal latency on a burst of cursor-only
updates.

Closes: https://launchpad.net/bugs/2023363
---
 clutter/clutter/clutter-frame-clock.c | 28 +++++++++++++++------------
 1 file changed, 16 insertions(+), 12 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index 142f670f519..eed8fd2a8ea 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -260,10 +260,6 @@ static void
 maybe_update_longterm_max_duration_us (ClutterFrameClock *frame_clock,
                                        ClutterFrameInfo  *frame_info)
 {
-  /* Do not update long-term max if there has been no measurement */
-  if (!frame_clock->shortterm_max_update_duration_us)
-    return;
-
   if ((frame_info->presentation_time - frame_clock->longterm_promotion_us) <
       G_USEC_PER_SEC)
     return;
@@ -382,8 +378,9 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
 
   frame_clock->got_measurements_last_frame = FALSE;
 
-  if (frame_info->cpu_time_before_buffer_swap_us != 0 &&
-      frame_info->has_valid_gpu_rendering_duration)
+  if ((frame_info->cpu_time_before_buffer_swap_us != 0 &&
+       frame_info->has_valid_gpu_rendering_duration) ||
+      frame_clock->ever_got_measurements)
     {
       int64_t dispatch_to_swap_us, swap_to_rendering_done_us, swap_to_flip_us;
       int64_t dispatch_time_us = 0, flip_time_us = 0;
@@ -408,14 +405,21 @@ clutter_frame_clock_notify_presented (ClutterFrameClock *frame_clock,
           break;
         }
 
-      dispatch_to_swap_us =
-        frame_info->cpu_time_before_buffer_swap_us -
-        dispatch_time_us;
+      if (frame_info->cpu_time_before_buffer_swap_us == 0)
+        {
+          /* Cursor-only updates with no "swap" or "flip" */
+          dispatch_to_swap_us = 0;
+          swap_to_flip_us = 0;
+        }
+      else
+        {
+          dispatch_to_swap_us = frame_info->cpu_time_before_buffer_swap_us -
+                                dispatch_time_us;
+          swap_to_flip_us = flip_time_us -
+                            frame_info->cpu_time_before_buffer_swap_us;
+        }
       swap_to_rendering_done_us =
         frame_info->gpu_rendering_duration_ns / 1000;
-      swap_to_flip_us =
-        flip_time_us -
-        frame_info->cpu_time_before_buffer_swap_us;
 
       CLUTTER_NOTE (FRAME_TIMINGS,
                     "%s: update2dispatch %ld µs, dispatch2swap %ld µs, swap2render %ld µs, swap2flip %ld µs",
-- 
GitLab

From 9cdbdaacb5ba9ce23ffc0e56723eb0384e968274 Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Wed, 26 Jul 2023 17:27:10 +0800
Subject: [PATCH 31/32] backends/native: Make await_flush later and inlined

"Later" as in after GL is finished rendering so as to not block the deadline
timer from updating the cursor smoothly during long GL renders. Indeed this
could make a primary plane frame late in some cases but that's only when
going from idle to active, so there is no previous frame to observe stutter
within the primary plane. It does however fix observable stutter in the
cursor plane by not dropping/stalling an otherwise good cursor update in a
continuous stream of cursor updates.

"Inlined" as in we don't need most of `meta_kms_device_await_flush`. The
important part is that it calls `disarm_crtc_frame_deadline_timer` to avoid
attempting two posts at once. So that's all that is kept.

This also fixes a deadlock in triple buffering. By not cancelling the
cursor update during a GL render we're also not cancelling a primary plane
update that might have already been piggybacked onto it by `queue_update`.

Cherry picked from !3149
---
 src/backends/native/meta-kms-impl-device.c | 4 +++-
 src/backends/native/meta-onscreen-native.c | 4 ----
 2 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/src/backends/native/meta-kms-impl-device.c b/src/backends/native/meta-kms-impl-device.c
index b15eee14d59..05bc89e8372 100644
--- a/src/backends/native/meta-kms-impl-device.c
+++ b/src/backends/native/meta-kms-impl-device.c
@@ -1559,9 +1559,11 @@ meta_kms_impl_device_handle_update (MetaKmsImplDevice *impl_device,
       meta_kms_update_merge_from (crtc_frame->pending_update, update);
       meta_kms_update_free (update);
       update = g_steal_pointer (&crtc_frame->pending_update);
-      disarm_crtc_frame_deadline_timer (crtc_frame);
     }
 
+  if (crtc_frame->deadline.armed)
+    disarm_crtc_frame_deadline_timer (crtc_frame);
+
   meta_kms_device_handle_flush (priv->device, latch_crtc);
 
   feedback = do_process (impl_device, latch_crtc, update, flags);
diff --git a/src/backends/native/meta-onscreen-native.c b/src/backends/native/meta-onscreen-native.c
index 20cb96e4eb7..dd96024f420 100644
--- a/src/backends/native/meta-onscreen-native.c
+++ b/src/backends/native/meta-onscreen-native.c
@@ -1878,11 +1878,7 @@ meta_onscreen_native_before_redraw (CoglOnscreen *onscreen,
                                     ClutterFrame *frame)
 {
   MetaOnscreenNative *onscreen_native = META_ONSCREEN_NATIVE (onscreen);
-  MetaCrtcKms *crtc_kms = META_CRTC_KMS (onscreen_native->crtc);
-  MetaKmsCrtc *kms_crtc = meta_crtc_kms_get_kms_crtc (crtc_kms);
 
-  meta_kms_device_await_flush (meta_kms_crtc_get_device (kms_crtc),
-                               kms_crtc);
   maybe_update_frame_sync (onscreen_native, frame);
 }
 
-- 
GitLab

From 350429237619f2f6d1d49b853ee5e1a3a556095b Mon Sep 17 00:00:00 2001
From: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Wed, 17 Jan 2024 17:21:03 +0800
Subject: [PATCH 32/32] clutter/frame-clock: Optimize latency for platforms
 missing TIMESTAMP_QUERY

Previously if we had no measurements then `compute_max_render_time_us`
would pessimise its answer to ensure triple buffering could be reached:
```
if (frame_clock->state == CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE)
  ret += refresh_interval_us;
```
But that also meant entering triple buffering even when not required.

Now we make `compute_max_render_time_us` more honest and return failure
if the answer isn't known (or is disabled). This in turn allows us to
optimize `calculate_next_update_time_us` for this special case, ensuring
triple buffering can be used, but isn't blindly always used.

This makes a visible difference to the latency when dragging windows in
Xorg, but will also help Wayland sessions on platforms lacking
TIMESTAMP_QUERY such as Raspberry Pi.
---
 clutter/clutter/clutter-frame-clock.c | 69 +++++++++++++++++----------
 1 file changed, 45 insertions(+), 24 deletions(-)

diff --git a/clutter/clutter/clutter-frame-clock.c b/clutter/clutter/clutter-frame-clock.c
index eed8fd2a8ea..a6c7ecea21f 100644
--- a/clutter/clutter/clutter-frame-clock.c
+++ b/clutter/clutter/clutter-frame-clock.c
@@ -513,25 +513,18 @@ clutter_frame_clock_notify_ready (ClutterFrameClock *frame_clock)
     }
 }
 
-static int64_t
-clutter_frame_clock_compute_max_render_time_us (ClutterFrameClock *frame_clock)
+static gboolean
+clutter_frame_clock_compute_max_render_time_us (ClutterFrameClock *frame_clock,
+                                                int64_t           *max_render_time_us)
 {
   int64_t refresh_interval_us;
-  int64_t max_render_time_us;
 
   refresh_interval_us = frame_clock->refresh_interval_us;
 
   if (!frame_clock->ever_got_measurements ||
       G_UNLIKELY (clutter_paint_debug_flags &
                   CLUTTER_DEBUG_DISABLE_DYNAMIC_MAX_RENDER_TIME))
-    {
-      int64_t ret = refresh_interval_us * SYNC_DELAY_FALLBACK_FRACTION;
-
-      if (frame_clock->state == CLUTTER_FRAME_CLOCK_STATE_DISPATCHED_ONE)
-        ret += refresh_interval_us;
-
-      return ret;
-    }
+    return FALSE;
 
   /* Max render time shows how early the frame clock needs to be dispatched
    * to make it to the predicted next presentation time. It is an estimate of
@@ -545,15 +538,15 @@ clutter_frame_clock_compute_max_render_time_us (ClutterFrameClock *frame_clock)
    * - The duration of vertical blank.
    * - A constant to account for variations in the above estimates.
    */
-  max_render_time_us =
+  *max_render_time_us =
     MAX (frame_clock->longterm_max_update_duration_us,
          frame_clock->shortterm_max_update_duration_us) +
     frame_clock->vblank_duration_us +
     clutter_max_render_time_constant_us;
 
-  max_render_time_us = CLAMP (max_render_time_us, 0, 2 * refresh_interval_us);
+  *max_render_time_us = CLAMP (*max_render_time_us, 0, 2 * refresh_interval_us);
 
-  return max_render_time_us;
+  return TRUE;
 }
 
 static void
@@ -570,6 +563,7 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
   int64_t next_presentation_time_us;
   int64_t next_smooth_presentation_time_us = 0;
   int64_t next_update_time_us;
+  gboolean max_render_time_is_known;
 
   now_us = g_get_monotonic_time ();
 
@@ -589,10 +583,13 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
     }
 
   min_render_time_allowed_us = refresh_interval_us / 2;
-  max_render_time_allowed_us =
-    clutter_frame_clock_compute_max_render_time_us (frame_clock);
 
-  if (min_render_time_allowed_us > max_render_time_allowed_us)
+  max_render_time_is_known =
+    clutter_frame_clock_compute_max_render_time_us (frame_clock,
+                                                    &max_render_time_allowed_us);
+
+  if (max_render_time_is_known &&
+      min_render_time_allowed_us > max_render_time_allowed_us)
     min_render_time_allowed_us = max_render_time_allowed_us;
 
   /*
@@ -712,6 +709,24 @@ calculate_next_update_time_us (ClutterFrameClock *frame_clock,
     }
   else
     {
+      /* If the max render time isn't known then using the current value of
+       * next_presentation_time_us is suboptimal. Targeting always one frame
+       * prior to that we'd lose the ability to scale up to triple buffering
+       * on late presentation. But targeting two frames prior we would be
+       * always triple buffering even when not required.
+       *   So the algorithm for deciding when to scale up to triple buffering
+       * in the absence of render time measurements is to simply target full
+       * frame rate. If we're keeping up then we'll stay double buffering. If
+       * we're not keeping up then this will switch us to triple buffering.
+       */
+      if (!max_render_time_is_known)
+        {
+          max_render_time_allowed_us =
+            refresh_interval_us * SYNC_DELAY_FALLBACK_FRACTION;
+          next_presentation_time_us =
+            last_presentation_time_us + refresh_interval_us;
+        }
+
       while (next_presentation_time_us - min_render_time_allowed_us < now_us)
         next_presentation_time_us += refresh_interval_us;
 
@@ -743,7 +758,9 @@ calculate_next_variable_update_time_us (ClutterFrameClock *frame_clock,
 
   refresh_interval_us = frame_clock->refresh_interval_us;
 
-  if (frame_clock->last_presentation_time_us == 0)
+  if (frame_clock->last_presentation_time_us == 0 ||
+      !clutter_frame_clock_compute_max_render_time_us (frame_clock,
+                                                       &max_render_time_allowed_us))
     {
       *out_next_update_time_us =
         frame_clock->last_dispatch_time_us ?
@@ -756,9 +773,6 @@ calculate_next_variable_update_time_us (ClutterFrameClock *frame_clock,
       return;
     }
 
-  max_render_time_allowed_us =
-    clutter_frame_clock_compute_max_render_time_us (frame_clock);
-
   last_presentation_time_us = frame_clock->last_presentation_time_us;
   next_presentation_time_us = last_presentation_time_us + refresh_interval_us;
 
@@ -1224,12 +1238,19 @@ clutter_frame_clock_record_flip (ClutterFrameClock *frame_clock,
 GString *
 clutter_frame_clock_get_max_render_time_debug_info (ClutterFrameClock *frame_clock)
 {
+  int64_t max_render_time_us;
   int64_t max_update_duration_us;
   GString *string;
 
-  string = g_string_new (NULL);
-  g_string_append_printf (string, "Max render time: %ld µs",
-                          clutter_frame_clock_compute_max_render_time_us (frame_clock));
+  string = g_string_new ("Max render time: ");
+  if (!clutter_frame_clock_compute_max_render_time_us (frame_clock,
+                                                       &max_render_time_us))
+    {
+      g_string_append (string, "unknown");
+      return string;
+    }
+
+  g_string_append_printf (string, "%ld µs", max_render_time_us);
 
   if (frame_clock->got_measurements_last_frame)
     g_string_append_printf (string, " =");
-- 
GitLab

Places

File _service:download_url:1441.patch of Package mutter (Revision e09bea664856ffb3db92802976cbb216)

Places