Merge branch 'dmabuf-wait-helper' into 'master'

Add helper to wait for DMA-BUFs to be ready on surface commit

Closes #3026

See merge request wlroots/wlroots!4040

backend/session/session.c

@@ -519,8 +519,6 @@ ssize_t wlr_session_find_gpus(struct wlr_session *session,
 			break;
 		}
 
-		bool is_boot_vga = false;
-
 		const char *path = udev_list_entry_get_name(entry);
 		struct udev_device *dev = udev_device_new_from_syspath(session->udev, path);
 		if (!dev) {
@@ -536,14 +534,20 @@ ssize_t wlr_session_find_gpus(struct wlr_session *session,
 			continue;
 		}
 
-		// This is owned by 'dev', so we don't need to free it
-		struct udev_device *pci =
-			udev_device_get_parent_with_subsystem_devtype(dev, "pci", NULL);
+		bool is_primary = false;
+		const char *boot_display = udev_device_get_sysattr_value(dev, "boot_display");
+		if (boot_display && strcmp(boot_display, "1") == 0) {
+		    is_primary = true;
+		} else {
+			// This is owned by 'dev', so we don't need to free it
+			struct udev_device *pci =
+				udev_device_get_parent_with_subsystem_devtype(dev, "pci", NULL);
 
-		if (pci) {
-			const char *id = udev_device_get_sysattr_value(pci, "boot_vga");
-			if (id && strcmp(id, "1") == 0) {
-				is_boot_vga = true;
+			if (pci) {
+				const char *id = udev_device_get_sysattr_value(pci, "boot_vga");
+				if (id && strcmp(id, "1") == 0) {
+					is_primary = true;
+				}
 			}
 		}
 
@@ -557,7 +561,7 @@ ssize_t wlr_session_find_gpus(struct wlr_session *session,
 		udev_device_unref(dev);
 
 		ret[i] = wlr_dev;
-		if (is_boot_vga) {
+		if (is_primary) {
 			struct wlr_device *tmp = ret[0];
 			ret[0] = ret[i];
 			ret[i] = tmp;

include/render/color.h

@@ -9,6 +9,7 @@ enum wlr_color_transform_type {
 	COLOR_TRANSFORM_INVERSE_EOTF,
 	COLOR_TRANSFORM_LCMS2,
 	COLOR_TRANSFORM_LUT_3X1D,
+	COLOR_TRANSFORM_PIPELINE,
 };
 
 struct wlr_color_transform {
@@ -39,6 +40,13 @@ struct wlr_color_transform_lut_3x1d {
 	size_t dim;
 };
 
+struct wlr_color_transform_pipeline {
+	struct wlr_color_transform base;
+
+	struct wlr_color_transform **transforms;
+	size_t len;
+};
+
 void wlr_color_transform_init(struct wlr_color_transform *tr,
 	enum wlr_color_transform_type type);
 
@@ -72,12 +80,6 @@ struct wlr_color_transform_inverse_eotf *wlr_color_transform_inverse_eotf_from_b
 struct wlr_color_transform_lut_3x1d *color_transform_lut_3x1d_from_base(
 	struct wlr_color_transform *tr);
 
-/**
- * Evaluate a 3x1D LUT color transform for a given RGB triplet.
- */
-void color_transform_lut_3x1d_eval(struct wlr_color_transform_lut_3x1d *tr,
-	float out[static 3], const float in[static 3]);
-
 /**
  * Obtain primaries values from a well-known primaries name.
  */

include/wlr/render/color.h

@@ -141,6 +141,13 @@ struct wlr_color_transform *wlr_color_transform_init_linear_to_inverse_eotf(
 struct wlr_color_transform *wlr_color_transform_init_lut_3x1d(size_t dim,
 	const uint16_t *r, const uint16_t *g, const uint16_t *b);
 
+/**
+ * Initialize a color transformation to apply a sequence of color transforms
+ * one after another.
+ */
+struct wlr_color_transform *wlr_color_transform_init_pipeline(
+	struct wlr_color_transform **transforms, size_t len);
+
 /**
  * Increase the reference count of the color transform by 1.
  */
@@ -152,4 +159,10 @@ struct wlr_color_transform *wlr_color_transform_ref(struct wlr_color_transform *
  */
 void wlr_color_transform_unref(struct wlr_color_transform *tr);
 
+/**
+ * Evaluate a color transform for a given RGB triplet.
+ */
+void wlr_color_transform_eval(struct wlr_color_transform *tr,
+	float out[static 3], const float in[static 3]);
+
 #endif

include/wlr/types/wlr_gamma_control_v1.h

@@ -10,6 +10,11 @@ struct wlr_gamma_control_manager_v1 {
 	struct wl_global *global;
 	struct wl_list controls; // wlr_gamma_control_v1.link
 
+	// Fallback to use when an struct wlr_output doesn't support gamma LUTs.
+	// Can be used to apply gamma LUTs via a struct wlr_renderer. Leave zero to
+	// indicate that the fallback is unsupported.
+	size_t fallback_gamma_size;
+
 	struct {
 		struct wl_signal destroy;
 		struct wl_signal set_gamma; // struct wlr_gamma_control_manager_v1_set_gamma_event
@@ -49,6 +54,8 @@ struct wlr_gamma_control_v1 *wlr_gamma_control_manager_v1_get_control(
 	struct wlr_gamma_control_manager_v1 *manager, struct wlr_output *output);
 bool wlr_gamma_control_v1_apply(struct wlr_gamma_control_v1 *gamma_control,
 	struct wlr_output_state *output_state);
+struct wlr_color_transform *wlr_gamma_control_v1_get_color_transform(
+	struct wlr_gamma_control_v1 *gamma_control);
 void wlr_gamma_control_v1_send_failed_and_destroy(struct wlr_gamma_control_v1 *gamma_control);
 
 #endif

include/wlr/types/wlr_linux_dmabuf_v1.h

@@ -16,6 +16,7 @@
 #include <wlr/render/dmabuf.h>
 #include <wlr/render/drm_format_set.h>
 
+struct wlr_compositor;
 struct wlr_surface;
 
 struct wlr_dmabuf_v1_buffer {
@@ -134,4 +135,40 @@ struct wlr_linux_dmabuf_feedback_v1_init_options {
 bool wlr_linux_dmabuf_feedback_v1_init_with_options(struct wlr_linux_dmabuf_feedback_v1 *feedback,
 	const struct wlr_linux_dmabuf_feedback_v1_init_options *options);
 
+/**
+ * A helper to wait for client DMA-BUFs to be ready.
+ *
+ * When attached to a surface, this helper will delay commits until the GPU
+ * work is done. In other words, wlr_surface.events.commit will only fire when
+ * GPU buffers attached to that commit are ready.
+ */
+struct wlr_surface_dmabuf_waiter {
+	struct wlr_surface *surface;
+
+	// private state
+
+	struct wl_list commits; // wlr_surface_dmabuf_waiter_commit.link
+	struct wl_listener client_commit;
+};
+
+/**
+ * Initialize a DMA-BUF waiter for a surface.
+ *
+ * Callers must call wlr_surface_dmabuf_waiter_finish() to unregister the waiter.
+ */
+void wlr_surface_dmabuf_waiter_init(struct wlr_surface_dmabuf_waiter *waiter,
+	struct wlr_surface *surface);
+/**
+ * Clean up a DMA-BUF waiter.
+ *
+ * Any pending commit waiting on GPU work to complete will be applied
+ * immediately.
+ */
+void wlr_surface_dmabuf_waiter_finish(struct wlr_surface_dmabuf_waiter *waiter);
+/**
+ * Initialize a compositor-wide DMA-BUF waiter, which will listen for new
+ * surfaces and attach DMA-BUF surface waiters to them.
+ */
+void wlr_compositor_dmabuf_waiter_create(struct wlr_compositor *compositor);
+
 #endif

include/wlr/types/wlr_output.h

@@ -267,6 +267,7 @@ struct wlr_output {
 	struct {
 		struct wl_listener display_destroy;
 		struct wlr_output_image_description image_description_value;
+		struct wlr_color_transform *color_transform;
 	} WLR_PRIVATE;
 };
 

include/wlr/types/wlr_scene.h

@@ -252,6 +252,11 @@ struct wlr_scene_output {
 
 		bool gamma_lut_changed;
 		struct wlr_gamma_control_v1 *gamma_lut;
+		struct wlr_color_transform *gamma_lut_color_transform;
+
+		struct wlr_color_transform *prev_gamma_lut_color_transform;
+		struct wlr_color_transform *prev_supplied_color_transform;
+		struct wlr_color_transform *prev_combined_color_transform;
 
 		struct wl_listener output_commit;
 		struct wl_listener output_damage;

include/xwayland/xwm.h

@@ -164,6 +164,7 @@ struct wlr_xwm {
 	struct wl_listener drop_focus_destroy;
 };
 
+// xwm_create takes ownership of wm_fd and will close it under all circumstances.
 struct wlr_xwm *xwm_create(struct wlr_xwayland *wlr_xwayland, int wm_fd);
 
 void xwm_destroy(struct wlr_xwm *xwm);

render/color.c

@@ -62,6 +62,33 @@ struct wlr_color_transform *wlr_color_transform_init_lut_3x1d(size_t dim,
 	return &tx->base;
 }
 
+struct wlr_color_transform *wlr_color_transform_init_pipeline(
+		struct wlr_color_transform **transforms, size_t len) {
+	assert(len > 0);
+
+	struct wlr_color_transform **copy = calloc(len, sizeof(copy[0]));
+	if (copy == NULL) {
+		return NULL;
+	}
+
+	struct wlr_color_transform_pipeline *tx = calloc(1, sizeof(*tx));
+	if (!tx) {
+		free(copy);
+		return NULL;
+	}
+	wlr_color_transform_init(&tx->base, COLOR_TRANSFORM_PIPELINE);
+
+	// TODO: flatten nested pipeline transforms
+	for (size_t i = 0; i < len; i++) {
+		copy[i] = wlr_color_transform_ref(transforms[i]);
+	}
+
+	tx->transforms = copy;
+	tx->len = len;
+
+	return &tx->base;
+}
+
 static void color_transform_destroy(struct wlr_color_transform *tr) {
 	switch (tr->type) {
 	case COLOR_TRANSFORM_INVERSE_EOTF:
@@ -73,6 +100,14 @@ static void color_transform_destroy(struct wlr_color_transform *tr) {
 		struct wlr_color_transform_lut_3x1d *lut_3x1d = color_transform_lut_3x1d_from_base(tr);
 		free(lut_3x1d->lut_3x1d);
 		break;
+	case COLOR_TRANSFORM_PIPELINE:;
+		struct wlr_color_transform_pipeline *pipeline =
+			wl_container_of(tr, pipeline, base);
+		for (size_t i = 0; i < pipeline->len; i++) {
+			wlr_color_transform_unref(pipeline->transforms[i]);
+		}
+		free(pipeline->transforms);
+		break;
 	}
 	wlr_addon_set_finish(&tr->addons);
 	free(tr);
@@ -108,6 +143,65 @@ struct wlr_color_transform_lut_3x1d *color_transform_lut_3x1d_from_base(
 	return lut_3x1d;
 }
 
+static float srgb_eval_inverse_eotf(float x) {
+	// See https://www.w3.org/Graphics/Color/srgb
+	if (x <= 0.0031308) {
+		return 12.92 * x;
+	} else {
+		return 1.055 * powf(x, 1.0 / 2.4) - 0.055;
+	}
+}
+
+static float st2084_pq_eval_inverse_eotf(float x) {
+	// H.273 TransferCharacteristics code point 16
+	float c1 = 0.8359375;
+	float c2 = 18.8515625;
+	float c3 = 18.6875;
+	float m = 78.84375;
+	float n = 0.1593017578125;
+	if (x < 0) {
+		x = 0;
+	}
+	if (x > 1) {
+		x = 1;
+	}
+	float pow_n = powf(x, n);
+	return powf((c1 + c2 * pow_n) / (1 + c3 * pow_n), m);
+}
+
+static float bt1886_eval_inverse_eotf(float x) {
+	float lb = powf(0.0001, 1.0 / 2.4);
+	float lw = powf(1.0, 1.0 / 2.4);
+	float a  = powf(lw - lb, 2.4);
+	float b  = lb / (lw - lb);
+	return powf(x / a, 1.0 / 2.4) - b;
+}
+
+static float transfer_function_eval_inverse_eotf(
+		enum wlr_color_transfer_function tf, float x) {
+	switch (tf) {
+	case WLR_COLOR_TRANSFER_FUNCTION_SRGB:
+		return srgb_eval_inverse_eotf(x);
+	case WLR_COLOR_TRANSFER_FUNCTION_ST2084_PQ:
+		return st2084_pq_eval_inverse_eotf(x);
+	case WLR_COLOR_TRANSFER_FUNCTION_EXT_LINEAR:
+		return x;
+	case WLR_COLOR_TRANSFER_FUNCTION_GAMMA22:
+		return powf(x, 1.0 / 2.2);
+	case WLR_COLOR_TRANSFER_FUNCTION_BT1886:
+		return bt1886_eval_inverse_eotf(x);
+	}
+	abort(); // unreachable
+}
+
+static void color_transform_inverse_eotf_eval(
+		struct wlr_color_transform_inverse_eotf *tr,
+		float out[static 3], const float in[static 3]) {
+	for (size_t i = 0; i < 3; i++) {
+		out[i] = transfer_function_eval_inverse_eotf(tr->tf, in[i]);
+	}
+}
+
 static float lut_1d_get(const uint16_t *lut, size_t len, size_t i) {
 	if (i >= len) {
 		i = len - 1;
@@ -125,13 +219,38 @@ static float lut_1d_eval(const uint16_t *lut, size_t len, float x) {
 	return a * (1 - frac_part) + b * frac_part;
 }
 
-void color_transform_lut_3x1d_eval(struct wlr_color_transform_lut_3x1d *tr,
+static void color_transform_lut_3x1d_eval(struct wlr_color_transform_lut_3x1d *tr,
 		float out[static 3], const float in[static 3]) {
 	for (size_t i = 0; i < 3; i++) {
 		out[i] = lut_1d_eval(&tr->lut_3x1d[tr->dim * i], tr->dim, in[i]);
 	}
 }
 
+void wlr_color_transform_eval(struct wlr_color_transform *tr,
+		float out[static 3], const float in[static 3]) {
+	switch (tr->type) {
+	case COLOR_TRANSFORM_INVERSE_EOTF:
+		color_transform_inverse_eotf_eval(wlr_color_transform_inverse_eotf_from_base(tr), out, in);
+		break;
+	case COLOR_TRANSFORM_LCMS2:
+		color_transform_lcms2_eval(color_transform_lcms2_from_base(tr), out, in);
+		break;
+	case COLOR_TRANSFORM_LUT_3X1D:
+		color_transform_lut_3x1d_eval(color_transform_lut_3x1d_from_base(tr), out, in);
+		break;
+	case COLOR_TRANSFORM_PIPELINE:;
+		struct wlr_color_transform_pipeline *pipeline =
+			wl_container_of(tr, pipeline, base);
+		float color[3];
+		memcpy(color, in, sizeof(color));
+		for (size_t i = 0; i < pipeline->len; i++) {
+			wlr_color_transform_eval(pipeline->transforms[i], color, color);
+		}
+		memcpy(out, color, sizeof(color));
+		break;
+	}
+}
+
 void wlr_color_primaries_from_named(struct wlr_color_primaries *out,
 		enum wlr_color_named_primaries named) {
 	switch (named) {

render/vulkan/pass.c

@@ -964,19 +964,6 @@ static bool create_3d_lut_image(struct wlr_vk_renderer *renderer,
 	*ds = VK_NULL_HANDLE;
 	*ds_pool = NULL;
 
-	struct wlr_color_transform_lcms2 *tr_lcms2 = NULL;
-	struct wlr_color_transform_lut_3x1d *tr_lut_3x1d = NULL;
-	switch (tr->type) {
-	case COLOR_TRANSFORM_INVERSE_EOTF:
-		abort(); // unreachable
-	case COLOR_TRANSFORM_LCMS2:
-		tr_lcms2 = color_transform_lcms2_from_base(tr);
-		break;
-	case COLOR_TRANSFORM_LUT_3X1D:
-		tr_lut_3x1d = color_transform_lut_3x1d_from_base(tr);
-		break;
-	}
-
 	// R32G32B32 is not a required Vulkan format
 	// TODO: use it when available
 	VkFormat format = VK_FORMAT_R32G32B32A32_SFLOAT;
@@ -1074,11 +1061,7 @@ static bool create_3d_lut_image(struct wlr_vk_renderer *renderer,
 					b_index * sample_range,
 				};
 				float rgb_out[3];
-				if (tr_lcms2 != NULL) {
-					color_transform_lcms2_eval(tr_lcms2, rgb_out, rgb_in);
-				} else {
-					color_transform_lut_3x1d_eval(tr_lut_3x1d, rgb_out, rgb_in);
-				}
+				wlr_color_transform_eval(tr, rgb_out, rgb_in);
 
 				dst[dst_offset] = rgb_out[0];
 				dst[dst_offset + 1] = rgb_out[1];

types/output/output.c

@@ -242,6 +242,15 @@ static void output_apply_state(struct wlr_output *output,
 		}
 	}
 
+	if (state->committed & WLR_OUTPUT_STATE_COLOR_TRANSFORM) {
+		wlr_color_transform_unref(output->color_transform);
+		if (state->color_transform != NULL) {
+			output->color_transform = wlr_color_transform_ref(state->color_transform);
+		} else {
+			output->color_transform = NULL;
+		}
+	}
+
 	bool geometry_updated = state->committed &
 		(WLR_OUTPUT_STATE_MODE | WLR_OUTPUT_STATE_TRANSFORM |
 		WLR_OUTPUT_STATE_SUBPIXEL);
@@ -407,6 +416,7 @@ void wlr_output_finish(struct wlr_output *output) {
 
 	wlr_swapchain_destroy(output->cursor_swapchain);
 	wlr_buffer_unlock(output->cursor_front_buffer);
+	wlr_color_transform_unref(output->color_transform);
 
 	wlr_swapchain_destroy(output->swapchain);
 
@@ -515,8 +525,7 @@ const struct wlr_output_image_description *output_pending_image_description(
  * Returns a bitfield of the unchanged fields.
  *
  * Some fields are not checked: damage always changes in-between frames, the
- * gamma LUT is too expensive to check, the contents of the buffer might have
- * changed, etc.
+ * contents of the buffer might have changed, etc.
  */
 static uint32_t output_compare_state(struct wlr_output *output,
 		const struct wlr_output_state *state) {
@@ -562,6 +571,10 @@ static uint32_t output_compare_state(struct wlr_output *output,
 			output->subpixel == state->subpixel) {
 		fields |= WLR_OUTPUT_STATE_SUBPIXEL;
 	}
+	if ((state->committed & WLR_OUTPUT_STATE_COLOR_TRANSFORM) &&
+			output->color_transform == state->color_transform) {
+		fields |= WLR_OUTPUT_STATE_COLOR_TRANSFORM;
+	}
 	return fields;
 }
 

types/scene/wlr_scene.c

@@ -1530,6 +1530,8 @@ static void scene_handle_gamma_control_manager_v1_set_gamma(struct wl_listener *
 
 	output->gamma_lut_changed = true;
 	output->gamma_lut = event->control;
+	wlr_color_transform_unref(output->gamma_lut_color_transform);
+	output->gamma_lut_color_transform = wlr_gamma_control_v1_get_color_transform(event->control);
 	wlr_output_schedule_frame(output->output);
 }
 
@@ -1547,6 +1549,8 @@ static void scene_handle_gamma_control_manager_v1_destroy(struct wl_listener *li
 	wl_list_for_each(output, &scene->outputs, link) {
 		output->gamma_lut_changed = false;
 		output->gamma_lut = NULL;
+		wlr_color_transform_unref(output->gamma_lut_color_transform);
+		output->gamma_lut_color_transform = NULL;
 	}
 }
 
@@ -1766,6 +1770,10 @@ void wlr_scene_output_destroy(struct wlr_scene_output *scene_output) {
 	wl_list_remove(&scene_output->output_damage.link);
 	wl_list_remove(&scene_output->output_needs_frame.link);
 	wlr_drm_syncobj_timeline_unref(scene_output->in_timeline);
+	wlr_color_transform_unref(scene_output->gamma_lut_color_transform);
+	wlr_color_transform_unref(scene_output->prev_gamma_lut_color_transform);
+	wlr_color_transform_unref(scene_output->prev_supplied_color_transform);
+	wlr_color_transform_unref(scene_output->prev_combined_color_transform);
 	wl_array_release(&scene_output->render_list);
 	free(scene_output);
 }
@@ -2104,16 +2112,15 @@ static void scene_output_state_attempt_gamma(struct wlr_scene_output *scene_outp
 		return;
 	}
 
-	if (!wlr_gamma_control_v1_apply(scene_output->gamma_lut, &gamma_pending)) {
-		wlr_output_state_finish(&gamma_pending);
-		return;
-	}
-
+	wlr_output_state_set_color_transform(&gamma_pending, scene_output->gamma_lut_color_transform);
 	scene_output->gamma_lut_changed = false;
+
 	if (!wlr_output_test_state(scene_output->output, &gamma_pending)) {
 		wlr_gamma_control_v1_send_failed_and_destroy(scene_output->gamma_lut);
 
 		scene_output->gamma_lut = NULL;
+		wlr_color_transform_unref(scene_output->gamma_lut_color_transform);
+		scene_output->gamma_lut_color_transform = NULL;
 		wlr_output_state_finish(&gamma_pending);
 		return;
 	}
@@ -2122,6 +2129,41 @@ static void scene_output_state_attempt_gamma(struct wlr_scene_output *scene_outp
 	wlr_output_state_finish(&gamma_pending);
 }
 
+static struct wlr_color_transform *scene_output_combine_color_transforms(
+		struct wlr_scene_output *scene_output, struct wlr_color_transform *supplied) {
+	struct wlr_color_transform *gamma_lut = scene_output->gamma_lut_color_transform;
+	assert(gamma_lut != NULL);
+
+	if (gamma_lut == scene_output->prev_gamma_lut_color_transform &&
+			supplied == scene_output->prev_supplied_color_transform) {
+		return wlr_color_transform_ref(scene_output->prev_combined_color_transform);
+	}
+
+	struct wlr_color_transform *combined;
+	if (supplied == NULL) {
+		combined = wlr_color_transform_ref(gamma_lut);
+	} else {
+		struct wlr_color_transform *transforms[] = {
+			gamma_lut,
+			supplied,
+		};
+		size_t transforms_len = sizeof(transforms) / sizeof(transforms[0]);
+		combined = wlr_color_transform_init_pipeline(transforms, transforms_len);
+		if (combined == NULL) {
+			return NULL;
+		}
+	}
+
+	wlr_color_transform_unref(scene_output->prev_gamma_lut_color_transform);
+	scene_output->prev_gamma_lut_color_transform = wlr_color_transform_ref(gamma_lut);
+	wlr_color_transform_unref(scene_output->prev_supplied_color_transform);
+	scene_output->prev_supplied_color_transform = supplied ? wlr_color_transform_ref(supplied) : NULL;
+	wlr_color_transform_unref(scene_output->prev_combined_color_transform);
+	scene_output->prev_combined_color_transform = wlr_color_transform_ref(combined);
+
+	return combined;
+}
+
 bool wlr_scene_output_build_state(struct wlr_scene_output *scene_output,
 		struct wlr_output_state *state, const struct wlr_scene_output_state_options *options) {
 	struct wlr_scene_output_state_options default_options = {0};
@@ -2145,6 +2187,16 @@ bool wlr_scene_output_build_state(struct wlr_scene_output *scene_output,
 	enum wlr_scene_debug_damage_option debug_damage =
 		scene_output->scene->debug_damage_option;
 
+	bool render_gamma_lut = false;
+	if (wlr_output_get_gamma_size(output) == 0 && output->renderer->features.output_color_transform) {
+		if (scene_output->gamma_lut_color_transform != scene_output->prev_gamma_lut_color_transform) {
+			scene_output_damage_whole(scene_output);
+		}
+		if (scene_output->gamma_lut_color_transform != NULL) {
+			render_gamma_lut = true;
+		}
+	}
+
 	struct render_data render_data = {
 		.transform = output->transform,
 		.scale = output->scale,
@@ -2245,7 +2297,7 @@ bool wlr_scene_output_build_state(struct wlr_scene_output *scene_output,
 	// - There are no color transforms that need to be applied
 	// - Damage highlight debugging is not enabled
 	enum scene_direct_scanout_result scanout_result = SCANOUT_INELIGIBLE;
-	if (options->color_transform == NULL && list_len == 1
+	if (options->color_transform == NULL && !render_gamma_lut && list_len == 1
 			&& debug_damage != WLR_SCENE_DEBUG_DAMAGE_HIGHLIGHT) {
 		scanout_result = scene_entry_try_direct_scanout(&list_data[0], state, &render_data);
 	}
@@ -2319,6 +2371,17 @@ bool wlr_scene_output_build_state(struct wlr_scene_output *scene_output,
 		color_transform = wlr_color_transform_ref(options->color_transform);
 	}
 
+	if (render_gamma_lut) {
+		struct wlr_color_transform *combined =
+			scene_output_combine_color_transforms(scene_output, color_transform);
+		wlr_color_transform_unref(color_transform);
+		if (combined == NULL) {
+			wlr_buffer_unlock(buffer);
+			return false;
+		}
+		color_transform = combined;
+	}
+
 	scene_output->in_point++;
 	struct wlr_render_pass *render_pass = wlr_renderer_begin_buffer_pass(output->renderer, buffer,
 			&(struct wlr_buffer_pass_options){
@@ -2441,7 +2504,9 @@ bool wlr_scene_output_build_state(struct wlr_scene_output *scene_output,
 			scene_output->in_point);
 	}
 
-	scene_output_state_attempt_gamma(scene_output, state);
+	if (!render_gamma_lut) {
+		scene_output_state_attempt_gamma(scene_output, state);
+	}
 
 	return true;
 }

types/wlr_gamma_control_v1.c

@@ -157,6 +157,9 @@ static void gamma_control_manager_get_gamma_control(struct wl_client *client,
 	}
 
 	size_t gamma_size = wlr_output_get_gamma_size(output);
+	if (gamma_size == 0) {
+		gamma_size = manager->fallback_gamma_size;
+	}
 	if (gamma_size == 0) {
 		zwlr_gamma_control_v1_send_failed(resource);
 		return;
@@ -262,15 +265,24 @@ struct wlr_gamma_control_v1 *wlr_gamma_control_manager_v1_get_control(
 	return NULL;
 }
 
+struct wlr_color_transform *wlr_gamma_control_v1_get_color_transform(
+		struct wlr_gamma_control_v1 *gamma_control) {
+	if (gamma_control == NULL || gamma_control->table == NULL) {
+		return NULL;
+	}
+
+	const uint16_t *r = gamma_control->table;
+	const uint16_t *g = gamma_control->table + gamma_control->ramp_size;
+	const uint16_t *b = gamma_control->table + 2 * gamma_control->ramp_size;
+
+	return wlr_color_transform_init_lut_3x1d(gamma_control->ramp_size, r, g, b);
+}
+
 bool wlr_gamma_control_v1_apply(struct wlr_gamma_control_v1 *gamma_control,
 		struct wlr_output_state *output_state) {
 	struct wlr_color_transform *tr = NULL;
 	if (gamma_control != NULL && gamma_control->table != NULL) {
-		const uint16_t *r = gamma_control->table;
-		const uint16_t *g = gamma_control->table + gamma_control->ramp_size;
-		const uint16_t *b = gamma_control->table + 2 * gamma_control->ramp_size;
-
-		tr = wlr_color_transform_init_lut_3x1d(gamma_control->ramp_size, r, g, b);
+		tr = wlr_gamma_control_v1_get_color_transform(gamma_control);
 		if (tr == NULL) {
 			return false;
 		}

types/wlr_linux_dmabuf_v1.c

@@ -1,6 +1,7 @@
 #include <assert.h>
 #include <drm_fourcc.h>
 #include <fcntl.h>
+#include <poll.h>
 #include <stdlib.h>
 #include <sys/mman.h>
 #include <unistd.h>
@@ -1173,3 +1174,210 @@ error:
 	wlr_linux_dmabuf_feedback_v1_finish(feedback);
 	return false;
 }
+
+struct wlr_surface_dmabuf_waiter_commit {
+	struct wlr_surface_dmabuf_waiter *waiter;
+	uint32_t surface_lock_seq;
+	int fds[WLR_DMABUF_MAX_PLANES]; // not owned by us
+	struct wl_event_source *event_sources[WLR_DMABUF_MAX_PLANES];
+	struct wl_list link; // wlr_surface_dmabuf_waiter.commits
+};
+
+static void surface_dmabuf_waiter_commit_destroy(
+		struct wlr_surface_dmabuf_waiter_commit *waiter_commit) {
+	for (size_t i = 0; i < WLR_DMABUF_MAX_PLANES; i++) {
+		if (waiter_commit->event_sources[i] != NULL) {
+			wl_event_source_remove(waiter_commit->event_sources[i]);
+		}
+	}
+
+	wlr_surface_unlock_cached(waiter_commit->waiter->surface, waiter_commit->surface_lock_seq);
+
+	wl_list_remove(&waiter_commit->link);
+	free(waiter_commit);
+}
+
+static int surface_dmabuf_waiter_handle_fd_event(int fd, uint32_t mask, void *data) {
+	struct wlr_surface_dmabuf_waiter_commit *waiter_commit = data;
+
+	if (mask & (WL_EVENT_HANGUP | WL_EVENT_ERROR)) {
+		wlr_log(WLR_ERROR, "Got hangup/error while polling on DMA-BUF");
+	}
+
+	bool found = false;
+	bool need_wait = false;
+	for (size_t i = 0; i < WLR_DMABUF_MAX_PLANES; i++) {
+		if (waiter_commit->fds[i] == fd) {
+			wl_event_source_remove(waiter_commit->event_sources[i]);
+			waiter_commit->event_sources[i] = NULL;
+			found = true;
+		} else if (waiter_commit->event_sources[i] != NULL) {
+			need_wait = true;
+		}
+	}
+	if (!found) {
+		wlr_log(WLR_ERROR, "Got event for unknown DMA-BUF FD");
+		return 0;
+	} else if (need_wait) {
+		return 0;
+	}
+
+	surface_dmabuf_waiter_commit_destroy(waiter_commit);
+	return 0;
+}
+
+static void surface_dmabuf_waiter_handle_client_commit(struct wl_listener *listener, void *data) {
+	struct wlr_surface_dmabuf_waiter *waiter = wl_container_of(listener, waiter, client_commit);
+	struct wlr_surface *surface = waiter->surface;
+
+	if (!(surface->pending.committed & WLR_SURFACE_STATE_BUFFER)) {
+		return;
+	}
+
+	struct wlr_buffer *buffer = surface->pending.buffer;
+	struct wlr_dmabuf_attributes dmabuf = {0};
+	if (buffer == NULL || !wlr_buffer_get_dmabuf(buffer, &dmabuf)) {
+		return;
+	}
+
+	// First check whether all DMA-BUF FDs are already ready
+	bool need_wait = false;
+	bool ready[WLR_DMABUF_MAX_PLANES] = {0};
+	for (int i = 0; i < dmabuf.n_planes; i++) {
+		struct pollfd pollfd = {
+			.fd = dmabuf.fd[i],
+			.events = POLLIN,
+		};
+		if (poll(&pollfd, 1, 0) < 0) {
+			wlr_log_errno(WLR_ERROR, "poll() failed");
+			return;
+		} else if (pollfd.revents & (POLLHUP | POLLERR)) {
+			wlr_log(WLR_ERROR, "Got hangup/error while polling on DMA-BUF");
+			return;
+		}
+
+		if (pollfd.revents & POLLIN) {
+			ready[i] = true;
+		} else {
+			need_wait = true;
+		}
+	}
+	if (!need_wait) {
+		return;
+	}
+
+	struct wlr_surface_dmabuf_waiter_commit *waiter_commit = calloc(1, sizeof(*waiter_commit));
+	if (waiter_commit == NULL) {
+		wlr_log_errno(WLR_ERROR, "Allocation failed");
+		return;
+	}
+	waiter_commit->waiter = waiter;
+
+	struct wl_client *client = wl_resource_get_client(surface->resource);
+	struct wl_display *display = wl_client_get_display(client);
+	struct wl_event_loop *event_loop = wl_display_get_event_loop(display);
+
+	for (int i = 0; i < dmabuf.n_planes; i++) {
+		if (ready[i]) {
+			continue;
+		}
+		struct wl_event_source *event_source = wl_event_loop_add_fd(event_loop, dmabuf.fd[i],
+			WL_EVENT_READABLE, surface_dmabuf_waiter_handle_fd_event, waiter_commit);
+		if (event_source == NULL) {
+			wlr_log(WLR_ERROR, "wl_event_loop_add_fd() failed");
+			goto error;
+		}
+		waiter_commit->fds[i] = dmabuf.fd[i];
+		waiter_commit->event_sources[i] = event_source;
+	}
+
+	// wlr_compositor ensures the wlr_buffer will remain alive (IOW, the
+	// DMA-BUF FDs will remain opened) while we have a lock
+	waiter_commit->surface_lock_seq = wlr_surface_lock_pending(waiter->surface);
+	wl_list_insert(&waiter->commits, &waiter_commit->link);
+
+	return;
+
+error:
+	for (int i = 0; i < dmabuf.n_planes; i++) {
+		if (waiter_commit->event_sources[i] != NULL) {
+			wl_event_source_remove(waiter_commit->event_sources[i]);
+		}
+	}
+	free(waiter_commit);
+}
+
+void wlr_surface_dmabuf_waiter_init(struct wlr_surface_dmabuf_waiter *waiter,
+		struct wlr_surface *surface) {
+	*waiter = (struct wlr_surface_dmabuf_waiter){0};
+
+	waiter->surface = surface;
+	wl_list_init(&waiter->commits);
+
+	waiter->client_commit.notify = surface_dmabuf_waiter_handle_client_commit;
+	wl_signal_add(&surface->events.client_commit, &waiter->client_commit);
+}
+
+void wlr_surface_dmabuf_waiter_finish(struct wlr_surface_dmabuf_waiter *waiter) {
+	struct wlr_surface_dmabuf_waiter_commit *waiter_commit, *waiter_commit_tmp;
+	wl_list_for_each_safe(waiter_commit, waiter_commit_tmp, &waiter->commits, link) {
+		surface_dmabuf_waiter_commit_destroy(waiter_commit);
+	}
+
+	wl_list_remove(&waiter->commits);
+	wl_list_remove(&waiter->client_commit.link);
+}
+
+struct wlr_compositor_dmabuf_waiter {
+	struct wl_listener new_surface;
+	struct wl_listener destroy;
+};
+
+struct wlr_compositor_dmabuf_waiter_surface {
+	struct wlr_surface_dmabuf_waiter base;
+	struct wl_listener destroy;
+};
+
+static void compositor_dmabuf_waiter_surface_handle_destroy(struct wl_listener *listener, void *data) {
+	struct wlr_compositor_dmabuf_waiter_surface *waiter_surface =
+		wl_container_of(listener, waiter_surface, destroy);
+	wlr_surface_dmabuf_waiter_finish(&waiter_surface->base);
+	wl_list_remove(&waiter_surface->destroy.link);
+	free(waiter_surface);
+}
+
+static void compositor_dmabuf_waiter_handle_new_surface(struct wl_listener *listener, void *data) {
+	struct wlr_surface *surface = data;
+
+	struct wlr_compositor_dmabuf_waiter_surface *waiter_surface = calloc(1, sizeof(*waiter_surface));
+	if (waiter_surface == NULL) {
+		wlr_log(WLR_ERROR, "Allocation failed");
+		return;
+	}
+
+	wlr_surface_dmabuf_waiter_init(&waiter_surface->base, surface);
+
+	waiter_surface->destroy.notify = compositor_dmabuf_waiter_surface_handle_destroy;
+	wl_signal_add(&surface->events.destroy, &waiter_surface->destroy);
+}
+
+static void compositor_dmabuf_waiter_handle_destroy(struct wl_listener *listener, void *data) {
+	struct wlr_compositor_dmabuf_waiter *waiter =
+		wl_container_of(listener, waiter, destroy);
+	wl_list_remove(&waiter->new_surface.link);
+	wl_list_remove(&waiter->destroy.link);
+	free(waiter);
+}
+
+void wlr_compositor_dmabuf_waiter_create(struct wlr_compositor *compositor) {
+	struct wlr_compositor_dmabuf_waiter *waiter = calloc(1, sizeof(*waiter));
+	if (waiter == NULL) {
+		wlr_log(WLR_ERROR, "Allocation failed");
+		return;
+	}
+
+	waiter->new_surface.notify = compositor_dmabuf_waiter_handle_new_surface;
+	wl_signal_add(&compositor->events.new_surface, &waiter->new_surface);
+	waiter->destroy.notify = compositor_dmabuf_waiter_handle_destroy;
+	wl_signal_add(&compositor->events.destroy, &waiter->destroy);
+}

xwayland/xwayland.c

@@ -42,6 +42,9 @@ static void handle_server_start(struct wl_listener *listener, void *data) {
 static void xwayland_mark_ready(struct wlr_xwayland *xwayland) {
 	assert(xwayland->server->wm_fd[0] >= 0);
 	xwayland->xwm = xwm_create(xwayland, xwayland->server->wm_fd[0]);
+	// xwm_create takes ownership of wm_fd[0] under all circumstances
+	xwayland->server->wm_fd[0] = -1;
+
 	if (!xwayland->xwm) {
 		return;
 	}

xwayland/xwm.c

@@ -2530,6 +2530,7 @@ void xwm_set_cursor(struct wlr_xwm *xwm, const uint8_t *pixels, uint32_t stride,
 struct wlr_xwm *xwm_create(struct wlr_xwayland *xwayland, int wm_fd) {
 	struct wlr_xwm *xwm = calloc(1, sizeof(*xwm));
 	if (xwm == NULL) {
+		close(wm_fd);
 		return NULL;
 	}
 
@@ -2544,11 +2545,13 @@ struct wlr_xwm *xwm_create(struct wlr_xwayland *xwayland, int wm_fd) {
 
 	xwm->ping_timeout = 10000;
 
+	// xcb_connect_to_fd takes ownership of the FD regardless of success/failure
 	xwm->xcb_conn = xcb_connect_to_fd(wm_fd, NULL);
 
 	int rc = xcb_connection_has_error(xwm->xcb_conn);
 	if (rc) {
 		wlr_log(WLR_ERROR, "xcb connect failed: %d", rc);
+		xcb_disconnect(xwm->xcb_conn);
 		free(xwm);
 		return NULL;
 	}