wlroots/types/wlr_output.c

1597 lines
45 KiB
C

#define _POSIX_C_SOURCE 200809L
#include <assert.h>
#include <drm_fourcc.h>
#include <stdlib.h>
#include <string.h>
#include <tgmath.h>
#include <time.h>
#include <wayland-server-core.h>
#include <wlr/interfaces/wlr_output.h>
#include <wlr/render/interface.h>
#include <wlr/render/wlr_renderer.h>
#include <wlr/types/wlr_box.h>
#include <wlr/types/wlr_matrix.h>
#include <wlr/types/wlr_output.h>
#include <wlr/types/wlr_seat.h>
#include <wlr/types/wlr_surface.h>
#include <wlr/util/log.h>
#include <wlr/util/region.h>
#include "backend/backend.h"
#include "render/allocator.h"
#include "render/drm_format_set.h"
#include "render/swapchain.h"
#include "render/wlr_renderer.h"
#include "util/global.h"
#include "util/signal.h"
#define OUTPUT_VERSION 3
static void send_geometry(struct wl_resource *resource) {
struct wlr_output *output = wlr_output_from_resource(resource);
wl_output_send_geometry(resource, 0, 0,
output->phys_width, output->phys_height, output->subpixel,
output->make, output->model, output->transform);
}
static void send_current_mode(struct wl_resource *resource) {
struct wlr_output *output = wlr_output_from_resource(resource);
if (output->current_mode != NULL) {
struct wlr_output_mode *mode = output->current_mode;
wl_output_send_mode(resource, WL_OUTPUT_MODE_CURRENT,
mode->width, mode->height, mode->refresh);
} else {
// Output has no mode
wl_output_send_mode(resource, WL_OUTPUT_MODE_CURRENT, output->width,
output->height, output->refresh);
}
}
static void send_scale(struct wl_resource *resource) {
struct wlr_output *output = wlr_output_from_resource(resource);
uint32_t version = wl_resource_get_version(resource);
if (version >= WL_OUTPUT_SCALE_SINCE_VERSION) {
wl_output_send_scale(resource, (uint32_t)ceil(output->scale));
}
}
static void send_done(struct wl_resource *resource) {
uint32_t version = wl_resource_get_version(resource);
if (version >= WL_OUTPUT_DONE_SINCE_VERSION) {
wl_output_send_done(resource);
}
}
static void output_handle_resource_destroy(struct wl_resource *resource) {
wl_list_remove(wl_resource_get_link(resource));
}
static void output_handle_release(struct wl_client *client,
struct wl_resource *resource) {
wl_resource_destroy(resource);
}
static const struct wl_output_interface output_impl = {
.release = output_handle_release,
};
static void output_bind(struct wl_client *wl_client, void *data,
uint32_t version, uint32_t id) {
// `output` can be NULL if the output global is being destroyed
struct wlr_output *output = data;
struct wl_resource *resource = wl_resource_create(wl_client,
&wl_output_interface, version, id);
if (resource == NULL) {
wl_client_post_no_memory(wl_client);
return;
}
wl_resource_set_implementation(resource, &output_impl, output,
output_handle_resource_destroy);
if (output == NULL) {
wl_list_init(wl_resource_get_link(resource));
return;
}
wl_list_insert(&output->resources, wl_resource_get_link(resource));
send_geometry(resource);
send_current_mode(resource);
send_scale(resource);
send_done(resource);
struct wlr_output_event_bind evt = {
.output = output,
.resource = resource,
};
wlr_signal_emit_safe(&output->events.bind, &evt);
}
void wlr_output_create_global(struct wlr_output *output) {
if (output->global != NULL) {
return;
}
output->global = wl_global_create(output->display,
&wl_output_interface, OUTPUT_VERSION, output, output_bind);
if (output->global == NULL) {
wlr_log(WLR_ERROR, "Failed to allocate wl_output global");
}
}
void wlr_output_destroy_global(struct wlr_output *output) {
if (output->global == NULL) {
return;
}
// Make all output resources inert
struct wl_resource *resource, *tmp;
wl_resource_for_each_safe(resource, tmp, &output->resources) {
wl_resource_set_user_data(resource, NULL);
wl_list_remove(wl_resource_get_link(resource));
wl_list_init(wl_resource_get_link(resource));
}
wlr_global_destroy_safe(output->global, output->display);
output->global = NULL;
}
void wlr_output_update_enabled(struct wlr_output *output, bool enabled) {
if (output->enabled == enabled) {
return;
}
output->enabled = enabled;
wlr_signal_emit_safe(&output->events.enable, output);
}
static void output_update_matrix(struct wlr_output *output) {
wlr_matrix_identity(output->transform_matrix);
if (output->transform != WL_OUTPUT_TRANSFORM_NORMAL) {
int tr_width, tr_height;
wlr_output_transformed_resolution(output, &tr_width, &tr_height);
wlr_matrix_translate(output->transform_matrix,
output->width / 2.0, output->height / 2.0);
wlr_matrix_transform(output->transform_matrix, output->transform);
wlr_matrix_translate(output->transform_matrix,
- tr_width / 2.0, - tr_height / 2.0);
}
}
void wlr_output_enable(struct wlr_output *output, bool enable) {
if (output->enabled == enable) {
output->pending.committed &= ~WLR_OUTPUT_STATE_ENABLED;
return;
}
output->pending.committed |= WLR_OUTPUT_STATE_ENABLED;
output->pending.enabled = enable;
}
static void output_state_clear_mode(struct wlr_output_state *state) {
if (!(state->committed & WLR_OUTPUT_STATE_MODE)) {
return;
}
state->mode = NULL;
state->committed &= ~WLR_OUTPUT_STATE_MODE;
}
void wlr_output_set_mode(struct wlr_output *output,
struct wlr_output_mode *mode) {
output_state_clear_mode(&output->pending);
if (output->current_mode == mode) {
return;
}
output->pending.committed |= WLR_OUTPUT_STATE_MODE;
output->pending.mode_type = WLR_OUTPUT_STATE_MODE_FIXED;
output->pending.mode = mode;
}
void wlr_output_set_custom_mode(struct wlr_output *output, int32_t width,
int32_t height, int32_t refresh) {
output_state_clear_mode(&output->pending);
if (output->width == width && output->height == height &&
output->refresh == refresh) {
return;
}
output->pending.committed |= WLR_OUTPUT_STATE_MODE;
output->pending.mode_type = WLR_OUTPUT_STATE_MODE_CUSTOM;
output->pending.custom_mode.width = width;
output->pending.custom_mode.height = height;
output->pending.custom_mode.refresh = refresh;
}
void wlr_output_update_mode(struct wlr_output *output,
struct wlr_output_mode *mode) {
output->current_mode = mode;
if (mode != NULL) {
wlr_output_update_custom_mode(output, mode->width, mode->height,
mode->refresh);
} else {
wlr_output_update_custom_mode(output, 0, 0, 0);
}
}
void wlr_output_update_custom_mode(struct wlr_output *output, int32_t width,
int32_t height, int32_t refresh) {
if (output->width == width && output->height == height &&
output->refresh == refresh) {
return;
}
output->width = width;
output->height = height;
output_update_matrix(output);
output->refresh = refresh;
if (output->swapchain != NULL &&
(output->swapchain->width != output->width ||
output->swapchain->height != output->height)) {
wlr_swapchain_destroy(output->swapchain);
output->swapchain = NULL;
}
struct wl_resource *resource;
wl_resource_for_each(resource, &output->resources) {
send_current_mode(resource);
}
wlr_output_schedule_done(output);
wlr_signal_emit_safe(&output->events.mode, output);
}
void wlr_output_set_transform(struct wlr_output *output,
enum wl_output_transform transform) {
if (output->transform == transform) {
output->pending.committed &= ~WLR_OUTPUT_STATE_TRANSFORM;
return;
}
output->pending.committed |= WLR_OUTPUT_STATE_TRANSFORM;
output->pending.transform = transform;
}
void wlr_output_set_scale(struct wlr_output *output, float scale) {
if (output->scale == scale) {
output->pending.committed &= ~WLR_OUTPUT_STATE_SCALE;
return;
}
output->pending.committed |= WLR_OUTPUT_STATE_SCALE;
output->pending.scale = scale;
}
void wlr_output_enable_adaptive_sync(struct wlr_output *output, bool enabled) {
bool currently_enabled =
output->adaptive_sync_status != WLR_OUTPUT_ADAPTIVE_SYNC_DISABLED;
if (currently_enabled == enabled) {
output->pending.committed &= ~WLR_OUTPUT_STATE_ADAPTIVE_SYNC_ENABLED;
return;
}
output->pending.committed |= WLR_OUTPUT_STATE_ADAPTIVE_SYNC_ENABLED;
output->pending.adaptive_sync_enabled = enabled;
}
void wlr_output_set_subpixel(struct wlr_output *output,
enum wl_output_subpixel subpixel) {
if (output->subpixel == subpixel) {
return;
}
output->subpixel = subpixel;
struct wl_resource *resource;
wl_resource_for_each(resource, &output->resources) {
send_geometry(resource);
}
wlr_output_schedule_done(output);
}
void wlr_output_set_description(struct wlr_output *output, const char *desc) {
if (output->description != NULL && desc != NULL &&
strcmp(output->description, desc) == 0) {
return;
}
free(output->description);
if (desc != NULL) {
output->description = strdup(desc);
} else {
output->description = NULL;
}
wlr_signal_emit_safe(&output->events.description, output);
}
static void schedule_done_handle_idle_timer(void *data) {
struct wlr_output *output = data;
output->idle_done = NULL;
struct wl_resource *resource;
wl_resource_for_each(resource, &output->resources) {
uint32_t version = wl_resource_get_version(resource);
if (version >= WL_OUTPUT_DONE_SINCE_VERSION) {
wl_output_send_done(resource);
}
}
}
void wlr_output_schedule_done(struct wlr_output *output) {
if (output->idle_done != NULL) {
return; // Already scheduled
}
struct wl_event_loop *ev = wl_display_get_event_loop(output->display);
output->idle_done =
wl_event_loop_add_idle(ev, schedule_done_handle_idle_timer, output);
}
static void handle_display_destroy(struct wl_listener *listener, void *data) {
struct wlr_output *output =
wl_container_of(listener, output, display_destroy);
wlr_output_destroy_global(output);
}
void wlr_output_init(struct wlr_output *output, struct wlr_backend *backend,
const struct wlr_output_impl *impl, struct wl_display *display) {
assert(impl->commit);
if (impl->attach_render || impl->rollback_render) {
assert(impl->attach_render && impl->rollback_render);
}
if (impl->set_cursor || impl->move_cursor) {
assert(impl->set_cursor && impl->move_cursor);
}
output->backend = backend;
output->impl = impl;
output->display = display;
wl_list_init(&output->modes);
output->transform = WL_OUTPUT_TRANSFORM_NORMAL;
output->scale = 1;
output->commit_seq = 0;
wl_list_init(&output->cursors);
wl_list_init(&output->resources);
wl_signal_init(&output->events.frame);
wl_signal_init(&output->events.damage);
wl_signal_init(&output->events.needs_frame);
wl_signal_init(&output->events.precommit);
wl_signal_init(&output->events.commit);
wl_signal_init(&output->events.present);
wl_signal_init(&output->events.bind);
wl_signal_init(&output->events.enable);
wl_signal_init(&output->events.mode);
wl_signal_init(&output->events.description);
wl_signal_init(&output->events.destroy);
pixman_region32_init(&output->pending.damage);
const char *no_hardware_cursors = getenv("WLR_NO_HARDWARE_CURSORS");
if (no_hardware_cursors != NULL && strcmp(no_hardware_cursors, "1") == 0) {
wlr_log(WLR_DEBUG,
"WLR_NO_HARDWARE_CURSORS set, forcing software cursors");
output->software_cursor_locks = 1;
}
output->display_destroy.notify = handle_display_destroy;
wl_display_add_destroy_listener(display, &output->display_destroy);
}
static void output_clear_back_buffer(struct wlr_output *output);
void wlr_output_destroy(struct wlr_output *output) {
if (!output) {
return;
}
wl_list_remove(&output->display_destroy.link);
wlr_output_destroy_global(output);
output_clear_back_buffer(output);
wlr_signal_emit_safe(&output->events.destroy, output);
// The backend is responsible for free-ing the list of modes
struct wlr_output_cursor *cursor, *tmp_cursor;
wl_list_for_each_safe(cursor, tmp_cursor, &output->cursors, link) {
wlr_output_cursor_destroy(cursor);
}
wlr_swapchain_destroy(output->cursor_swapchain);
wlr_buffer_unlock(output->cursor_front_buffer);
wlr_swapchain_destroy(output->swapchain);
if (output->idle_frame != NULL) {
wl_event_source_remove(output->idle_frame);
}
if (output->idle_done != NULL) {
wl_event_source_remove(output->idle_done);
}
free(output->description);
pixman_region32_fini(&output->pending.damage);
if (output->impl && output->impl->destroy) {
output->impl->destroy(output);
} else {
free(output);
}
}
void wlr_output_transformed_resolution(struct wlr_output *output,
int *width, int *height) {
if (output->transform % 2 == 0) {
*width = output->width;
*height = output->height;
} else {
*width = output->height;
*height = output->width;
}
}
void wlr_output_effective_resolution(struct wlr_output *output,
int *width, int *height) {
wlr_output_transformed_resolution(output, width, height);
*width /= output->scale;
*height /= output->scale;
}
struct wlr_output_mode *wlr_output_preferred_mode(struct wlr_output *output) {
if (wl_list_empty(&output->modes)) {
return NULL;
}
struct wlr_output_mode *mode;
wl_list_for_each(mode, &output->modes, link) {
if (mode->preferred) {
return mode;
}
}
// No preferred mode, choose the last one
return wl_container_of(output->modes.prev, mode, link);
}
static void output_state_clear_buffer(struct wlr_output_state *state) {
if (!(state->committed & WLR_OUTPUT_STATE_BUFFER)) {
return;
}
wlr_buffer_unlock(state->buffer);
state->buffer = NULL;
state->committed &= ~WLR_OUTPUT_STATE_BUFFER;
}
static struct wlr_drm_format *output_pick_format(struct wlr_output *output,
const struct wlr_drm_format_set *display_formats);
static bool output_create_swapchain(struct wlr_output *output) {
if (output->swapchain != NULL) {
return true;
}
struct wlr_allocator *allocator = backend_get_allocator(output->backend);
if (allocator == NULL) {
wlr_log(WLR_ERROR, "Failed to get backend allocator");
return false;
}
const struct wlr_drm_format_set *display_formats = NULL;
if (output->impl->get_primary_formats) {
display_formats =
output->impl->get_primary_formats(output, allocator->buffer_caps);
if (display_formats == NULL) {
wlr_log(WLR_ERROR, "Failed to get primary display formats");
return false;
}
}
struct wlr_drm_format *format = output_pick_format(output, display_formats);
if (format == NULL) {
wlr_log(WLR_ERROR, "Failed to pick primary buffer format for output '%s'",
output->name);
return false;
}
wlr_log(WLR_DEBUG, "Choosing primary buffer format 0x%"PRIX32" for output '%s'",
format->format, output->name);
output->swapchain = wlr_swapchain_create(allocator, output->width,
output->height, format);
free(format);
if (output->swapchain == NULL) {
wlr_log(WLR_ERROR, "Failed to create output swapchain");
return false;
}
return true;
}
static bool output_attach_back_buffer(struct wlr_output *output,
int *buffer_age) {
assert(output->back_buffer == NULL);
if (!output_create_swapchain(output)) {
return false;
}
struct wlr_renderer *renderer = wlr_backend_get_renderer(output->backend);
assert(renderer != NULL);
struct wlr_buffer *buffer =
wlr_swapchain_acquire(output->swapchain, buffer_age);
if (buffer == NULL) {
return false;
}
if (!renderer_bind_buffer(renderer, buffer)) {
wlr_buffer_unlock(buffer);
return false;
}
output->back_buffer = buffer;
return true;
}
static void output_clear_back_buffer(struct wlr_output *output) {
if (output->back_buffer == NULL) {
return;
}
struct wlr_renderer *renderer = wlr_backend_get_renderer(output->backend);
assert(renderer != NULL);
renderer_bind_buffer(renderer, NULL);
wlr_buffer_unlock(output->back_buffer);
output->back_buffer = NULL;
}
bool wlr_output_attach_render(struct wlr_output *output, int *buffer_age) {
if (output->impl->attach_render) {
if (!output->impl->attach_render(output, buffer_age)) {
return false;
}
output_state_clear_buffer(&output->pending);
output->pending.committed |= WLR_OUTPUT_STATE_BUFFER;
output->pending.buffer_type = WLR_OUTPUT_STATE_BUFFER_RENDER;
} else {
if (!output_attach_back_buffer(output, buffer_age)) {
return false;
}
wlr_output_attach_buffer(output, output->back_buffer);
}
return true;
}
uint32_t wlr_output_preferred_read_format(struct wlr_output *output) {
struct wlr_renderer *renderer = wlr_backend_get_renderer(output->backend);
if (!renderer->impl->preferred_read_format || !renderer->impl->read_pixels) {
return DRM_FORMAT_INVALID;
}
if (output->impl->attach_render) {
if (!output->impl->attach_render(output, NULL)) {
return false;
}
} else {
if (!output_attach_back_buffer(output, NULL)) {
return false;
}
}
uint32_t fmt = renderer->impl->preferred_read_format(renderer);
if (output->impl->rollback_render) {
output->impl->rollback_render(output);
} else {
output_clear_back_buffer(output);
}
return fmt;
}
void wlr_output_set_damage(struct wlr_output *output,
pixman_region32_t *damage) {
pixman_region32_intersect_rect(&output->pending.damage, damage,
0, 0, output->width, output->height);
output->pending.committed |= WLR_OUTPUT_STATE_DAMAGE;
}
static void output_state_clear_gamma_lut(struct wlr_output_state *state) {
free(state->gamma_lut);
state->gamma_lut = NULL;
state->committed &= ~WLR_OUTPUT_STATE_GAMMA_LUT;
}
static void output_state_clear(struct wlr_output_state *state) {
output_state_clear_buffer(state);
output_state_clear_gamma_lut(state);
pixman_region32_clear(&state->damage);
state->committed = 0;
}
static void output_pending_resolution(struct wlr_output *output, int *width,
int *height) {
if (output->pending.committed & WLR_OUTPUT_STATE_MODE) {
switch (output->pending.mode_type) {
case WLR_OUTPUT_STATE_MODE_FIXED:
*width = output->pending.mode->width;
*height = output->pending.mode->height;
return;
case WLR_OUTPUT_STATE_MODE_CUSTOM:
*width = output->pending.custom_mode.width;
*height = output->pending.custom_mode.height;
return;
}
abort();
} else {
*width = output->width;
*height = output->height;
}
}
static bool output_basic_test(struct wlr_output *output) {
if (output->pending.committed & WLR_OUTPUT_STATE_BUFFER) {
if (output->frame_pending) {
wlr_log(WLR_DEBUG, "Tried to commit a buffer while a frame is pending");
return false;
}
if (output->pending.buffer_type == WLR_OUTPUT_STATE_BUFFER_SCANOUT &&
output->back_buffer == NULL) {
if (output->attach_render_locks > 0) {
wlr_log(WLR_DEBUG, "Direct scan-out disabled by lock");
return false;
}
// If the output has at least one software cursor, refuse to attach the
// buffer
struct wlr_output_cursor *cursor;
wl_list_for_each(cursor, &output->cursors, link) {
if (cursor->enabled && cursor->visible &&
cursor != output->hardware_cursor) {
wlr_log(WLR_DEBUG,
"Direct scan-out disabled by software cursor");
return false;
}
}
// If the size doesn't match, reject buffer (scaling is not
// supported)
int pending_width, pending_height;
output_pending_resolution(output, &pending_width, &pending_height);
if (output->pending.buffer->width != pending_width ||
output->pending.buffer->height != pending_height) {
wlr_log(WLR_DEBUG, "Direct scan-out buffer size mismatch");
return false;
}
}
}
bool enabled = output->enabled;
if (output->pending.committed & WLR_OUTPUT_STATE_ENABLED) {
enabled = output->pending.enabled;
}
if (!enabled && output->pending.committed & WLR_OUTPUT_STATE_BUFFER) {
wlr_log(WLR_DEBUG, "Tried to commit a buffer on a disabled output");
return false;
}
if (!enabled && output->pending.committed & WLR_OUTPUT_STATE_MODE) {
wlr_log(WLR_DEBUG, "Tried to modeset a disabled output");
return false;
}
if (!enabled && output->pending.committed & WLR_OUTPUT_STATE_ADAPTIVE_SYNC_ENABLED) {
wlr_log(WLR_DEBUG, "Tried to enable adaptive sync on a disabled output");
return false;
}
if (!enabled && output->pending.committed & WLR_OUTPUT_STATE_GAMMA_LUT) {
wlr_log(WLR_DEBUG, "Tried to set the gamma lut on a disabled output");
return false;
}
return true;
}
bool wlr_output_test(struct wlr_output *output) {
if (!output_basic_test(output)) {
return false;
}
if (!output->impl->test) {
return true;
}
return output->impl->test(output);
}
bool wlr_output_commit(struct wlr_output *output) {
if (!output_basic_test(output)) {
wlr_log(WLR_ERROR, "Basic output test failed for %s", output->name);
return false;
}
if ((output->pending.committed & WLR_OUTPUT_STATE_BUFFER) &&
output->idle_frame != NULL) {
wl_event_source_remove(output->idle_frame);
output->idle_frame = NULL;
}
struct timespec now;
clock_gettime(CLOCK_MONOTONIC, &now);
struct wlr_output_event_precommit pre_event = {
.output = output,
.when = &now,
};
wlr_signal_emit_safe(&output->events.precommit, &pre_event);
// output_clear_back_buffer detaches the buffer from the renderer. This is
// important to do before calling impl->commit(), because this marks an
// implicit rendering synchronization point. The backend needs it to avoid
// displaying a buffer when asynchronous GPU work isn't finished.
struct wlr_buffer *back_buffer = NULL;
if ((output->pending.committed & WLR_OUTPUT_STATE_BUFFER) &&
output->back_buffer != NULL) {
back_buffer = wlr_buffer_lock(output->back_buffer);
output_clear_back_buffer(output);
}
if (!output->impl->commit(output)) {
wlr_buffer_unlock(back_buffer);
output_state_clear(&output->pending);
return false;
}
if (output->pending.committed & WLR_OUTPUT_STATE_BUFFER) {
struct wlr_output_cursor *cursor;
wl_list_for_each(cursor, &output->cursors, link) {
if (!cursor->enabled || !cursor->visible || cursor->surface == NULL) {
continue;
}
wlr_surface_send_frame_done(cursor->surface, &now);
}
}
output->commit_seq++;
bool scale_updated = output->pending.committed & WLR_OUTPUT_STATE_SCALE;
if (scale_updated) {
output->scale = output->pending.scale;
}
if (output->pending.committed & WLR_OUTPUT_STATE_TRANSFORM) {
output->transform = output->pending.transform;
output_update_matrix(output);
}
bool geometry_updated = output->pending.committed &
(WLR_OUTPUT_STATE_MODE | WLR_OUTPUT_STATE_TRANSFORM);
if (geometry_updated || scale_updated) {
struct wl_resource *resource;
wl_resource_for_each(resource, &output->resources) {
if (geometry_updated) {
send_geometry(resource);
}
if (scale_updated) {
send_scale(resource);
}
}
wlr_output_schedule_done(output);
}
if (output->pending.committed & WLR_OUTPUT_STATE_BUFFER) {
output->frame_pending = true;
output->needs_frame = false;
}
if (back_buffer != NULL) {
wlr_swapchain_set_buffer_submitted(output->swapchain, back_buffer);
wlr_buffer_unlock(back_buffer);
}
uint32_t committed = output->pending.committed;
output_state_clear(&output->pending);
struct wlr_output_event_commit event = {
.output = output,
.committed = committed,
.when = &now,
};
wlr_signal_emit_safe(&output->events.commit, &event);
return true;
}
void wlr_output_rollback(struct wlr_output *output) {
if (output->impl->rollback_render &&
(output->pending.committed & WLR_OUTPUT_STATE_BUFFER) &&
output->pending.buffer_type == WLR_OUTPUT_STATE_BUFFER_RENDER) {
output->impl->rollback_render(output);
}
output_clear_back_buffer(output);
output_state_clear(&output->pending);
}
void wlr_output_attach_buffer(struct wlr_output *output,
struct wlr_buffer *buffer) {
output_state_clear_buffer(&output->pending);
output->pending.committed |= WLR_OUTPUT_STATE_BUFFER;
output->pending.buffer_type = WLR_OUTPUT_STATE_BUFFER_SCANOUT;
output->pending.buffer = wlr_buffer_lock(buffer);
}
void wlr_output_send_frame(struct wlr_output *output) {
output->frame_pending = false;
wlr_signal_emit_safe(&output->events.frame, output);
}
static void schedule_frame_handle_idle_timer(void *data) {
struct wlr_output *output = data;
output->idle_frame = NULL;
if (!output->frame_pending) {
wlr_output_send_frame(output);
}
}
void wlr_output_schedule_frame(struct wlr_output *output) {
// Make sure the compositor commits a new frame. This is necessary to make
// clients which ask for frame callbacks without submitting a new buffer
// work.
wlr_output_update_needs_frame(output);
if (output->frame_pending || output->idle_frame != NULL) {
return;
}
// We're using an idle timer here in case a buffer swap happens right after
// this function is called
struct wl_event_loop *ev = wl_display_get_event_loop(output->display);
output->idle_frame =
wl_event_loop_add_idle(ev, schedule_frame_handle_idle_timer, output);
}
void wlr_output_send_present(struct wlr_output *output,
struct wlr_output_event_present *event) {
struct wlr_output_event_present _event = {0};
if (event == NULL) {
event = &_event;
event->commit_seq = output->commit_seq + 1;
}
event->output = output;
struct timespec now;
if (event->when == NULL) {
clockid_t clock = wlr_backend_get_presentation_clock(output->backend);
errno = 0;
if (clock_gettime(clock, &now) != 0) {
wlr_log_errno(WLR_ERROR, "failed to send output present event: "
"failed to read clock");
return;
}
event->when = &now;
}
wlr_signal_emit_safe(&output->events.present, event);
}
void wlr_output_set_gamma(struct wlr_output *output, size_t size,
const uint16_t *r, const uint16_t *g, const uint16_t *b) {
output_state_clear_gamma_lut(&output->pending);
output->pending.gamma_lut_size = size;
output->pending.gamma_lut = malloc(3 * size * sizeof(uint16_t));
if (output->pending.gamma_lut == NULL) {
wlr_log_errno(WLR_ERROR, "Allocation failed");
return;
}
memcpy(output->pending.gamma_lut, r, size * sizeof(uint16_t));
memcpy(output->pending.gamma_lut + size, g, size * sizeof(uint16_t));
memcpy(output->pending.gamma_lut + 2 * size, b, size * sizeof(uint16_t));
output->pending.committed |= WLR_OUTPUT_STATE_GAMMA_LUT;
}
size_t wlr_output_get_gamma_size(struct wlr_output *output) {
if (!output->impl->get_gamma_size) {
return 0;
}
return output->impl->get_gamma_size(output);
}
bool wlr_output_export_dmabuf(struct wlr_output *output,
struct wlr_dmabuf_attributes *attribs) {
if (!output->impl->export_dmabuf) {
return false;
}
return output->impl->export_dmabuf(output, attribs);
}
void wlr_output_update_needs_frame(struct wlr_output *output) {
if (output->needs_frame) {
return;
}
output->needs_frame = true;
wlr_signal_emit_safe(&output->events.needs_frame, output);
}
void wlr_output_damage_whole(struct wlr_output *output) {
int width, height;
wlr_output_transformed_resolution(output, &width, &height);
pixman_region32_t damage;
pixman_region32_init_rect(&damage, 0, 0, width, height);
struct wlr_output_event_damage event = {
.output = output,
.damage = &damage,
};
wlr_signal_emit_safe(&output->events.damage, &event);
pixman_region32_fini(&damage);
}
struct wlr_output *wlr_output_from_resource(struct wl_resource *resource) {
assert(wl_resource_instance_of(resource, &wl_output_interface,
&output_impl));
return wl_resource_get_user_data(resource);
}
void wlr_output_lock_attach_render(struct wlr_output *output, bool lock) {
if (lock) {
++output->attach_render_locks;
} else {
assert(output->attach_render_locks > 0);
--output->attach_render_locks;
}
wlr_log(WLR_DEBUG, "%s direct scan-out on output '%s' (locks: %d)",
lock ? "Disabling" : "Enabling", output->name,
output->attach_render_locks);
}
static void output_cursor_damage_whole(struct wlr_output_cursor *cursor);
void wlr_output_lock_software_cursors(struct wlr_output *output, bool lock) {
if (lock) {
++output->software_cursor_locks;
} else {
assert(output->software_cursor_locks > 0);
--output->software_cursor_locks;
}
wlr_log(WLR_DEBUG, "%s hardware cursors on output '%s' (locks: %d)",
lock ? "Disabling" : "Enabling", output->name,
output->software_cursor_locks);
if (output->software_cursor_locks > 0 && output->hardware_cursor != NULL) {
assert(output->impl->set_cursor);
output->impl->set_cursor(output, NULL, 0, 0);
output_cursor_damage_whole(output->hardware_cursor);
output->hardware_cursor = NULL;
}
// If it's possible to use hardware cursors again, don't switch immediately
// since a recorder is likely to lock software cursors for the next frame
// again.
}
static void output_scissor(struct wlr_output *output, pixman_box32_t *rect) {
struct wlr_renderer *renderer = wlr_backend_get_renderer(output->backend);
assert(renderer);
struct wlr_box box = {
.x = rect->x1,
.y = rect->y1,
.width = rect->x2 - rect->x1,
.height = rect->y2 - rect->y1,
};
int ow, oh;
wlr_output_transformed_resolution(output, &ow, &oh);
enum wl_output_transform transform =
wlr_output_transform_invert(output->transform);
wlr_box_transform(&box, &box, transform, ow, oh);
wlr_renderer_scissor(renderer, &box);
}
static void output_cursor_get_box(struct wlr_output_cursor *cursor,
struct wlr_box *box);
static void output_cursor_render(struct wlr_output_cursor *cursor,
pixman_region32_t *damage) {
struct wlr_renderer *renderer =
wlr_backend_get_renderer(cursor->output->backend);
assert(renderer);
struct wlr_texture *texture = cursor->texture;
if (cursor->surface != NULL) {
texture = wlr_surface_get_texture(cursor->surface);
}
if (texture == NULL) {
return;
}
struct wlr_box box;
output_cursor_get_box(cursor, &box);
pixman_region32_t surface_damage;
pixman_region32_init(&surface_damage);
pixman_region32_union_rect(&surface_damage, &surface_damage, box.x, box.y,
box.width, box.height);
pixman_region32_intersect(&surface_damage, &surface_damage, damage);
if (!pixman_region32_not_empty(&surface_damage)) {
goto surface_damage_finish;
}
float matrix[9];
wlr_matrix_project_box(matrix, &box, WL_OUTPUT_TRANSFORM_NORMAL, 0,
cursor->output->transform_matrix);
int nrects;
pixman_box32_t *rects = pixman_region32_rectangles(&surface_damage, &nrects);
for (int i = 0; i < nrects; ++i) {
output_scissor(cursor->output, &rects[i]);
wlr_render_texture_with_matrix(renderer, texture, matrix, 1.0f);
}
wlr_renderer_scissor(renderer, NULL);
surface_damage_finish:
pixman_region32_fini(&surface_damage);
}
void wlr_output_render_software_cursors(struct wlr_output *output,
pixman_region32_t *damage) {
int width, height;
wlr_output_transformed_resolution(output, &width, &height);
pixman_region32_t render_damage;
pixman_region32_init(&render_damage);
pixman_region32_union_rect(&render_damage, &render_damage, 0, 0,
width, height);
if (damage != NULL) {
// Damage tracking supported
pixman_region32_intersect(&render_damage, &render_damage, damage);
}
if (pixman_region32_not_empty(&render_damage)) {
struct wlr_output_cursor *cursor;
wl_list_for_each(cursor, &output->cursors, link) {
if (!cursor->enabled || !cursor->visible ||
output->hardware_cursor == cursor) {
continue;
}
output_cursor_render(cursor, &render_damage);
}
}
pixman_region32_fini(&render_damage);
}
/**
* Returns the cursor box, scaled for its output.
*/
static void output_cursor_get_box(struct wlr_output_cursor *cursor,
struct wlr_box *box) {
box->x = cursor->x - cursor->hotspot_x;
box->y = cursor->y - cursor->hotspot_y;
box->width = cursor->width;
box->height = cursor->height;
}
static void output_cursor_damage_whole(struct wlr_output_cursor *cursor) {
struct wlr_box box;
output_cursor_get_box(cursor, &box);
pixman_region32_t damage;
pixman_region32_init_rect(&damage, box.x, box.y, box.width, box.height);
struct wlr_output_event_damage event = {
.output = cursor->output,
.damage = &damage,
};
wlr_signal_emit_safe(&cursor->output->events.damage, &event);
pixman_region32_fini(&damage);
}
static void output_cursor_reset(struct wlr_output_cursor *cursor) {
if (cursor->output->hardware_cursor != cursor) {
output_cursor_damage_whole(cursor);
}
if (cursor->surface != NULL) {
wl_list_remove(&cursor->surface_commit.link);
wl_list_remove(&cursor->surface_destroy.link);
if (cursor->visible) {
wlr_surface_send_leave(cursor->surface, cursor->output);
}
cursor->surface = NULL;
}
}
static void output_cursor_update_visible(struct wlr_output_cursor *cursor) {
struct wlr_box output_box;
output_box.x = output_box.y = 0;
wlr_output_transformed_resolution(cursor->output, &output_box.width,
&output_box.height);
struct wlr_box cursor_box;
output_cursor_get_box(cursor, &cursor_box);
struct wlr_box intersection;
bool visible =
wlr_box_intersection(&intersection, &output_box, &cursor_box);
if (cursor->surface != NULL) {
if (cursor->visible && !visible) {
wlr_surface_send_leave(cursor->surface, cursor->output);
}
if (!cursor->visible && visible) {
wlr_surface_send_enter(cursor->surface, cursor->output);
}
}
cursor->visible = visible;
}
static struct wlr_drm_format *output_pick_format(struct wlr_output *output,
const struct wlr_drm_format_set *display_formats) {
struct wlr_renderer *renderer = wlr_backend_get_renderer(output->backend);
struct wlr_allocator *allocator = backend_get_allocator(output->backend);
assert(renderer != NULL && allocator != NULL);
const struct wlr_drm_format_set *render_formats =
wlr_renderer_get_render_formats(renderer);
if (render_formats == NULL) {
wlr_log(WLR_ERROR, "Failed to get render formats");
return NULL;
}
uint32_t fmt = DRM_FORMAT_ARGB8888;
const struct wlr_drm_format *render_format =
wlr_drm_format_set_get(render_formats, fmt);
if (render_format == NULL) {
wlr_log(WLR_DEBUG, "Renderer doesn't support format 0x%"PRIX32, fmt);
return NULL;
}
const struct wlr_drm_format *display_format;
if (display_formats != NULL) {
display_format = wlr_drm_format_set_get(display_formats, fmt);
if (display_format == NULL) {
wlr_log(WLR_DEBUG, "Output doesn't support format 0x%"PRIX32, fmt);
return NULL;
}
} else {
// The output can display any format
display_format = render_format;
}
struct wlr_drm_format *format =
wlr_drm_format_intersect(display_format, render_format);
if (format == NULL) {
wlr_log(WLR_DEBUG, "Failed to intersect display and render "
"modifiers for format 0x%"PRIX32, fmt);
return NULL;
}
return format;
}
static struct wlr_drm_format *output_pick_cursor_format(struct wlr_output *output) {
struct wlr_allocator *allocator = backend_get_allocator(output->backend);
assert(allocator != NULL);
const struct wlr_drm_format_set *display_formats = NULL;
if (output->impl->get_cursor_formats) {
display_formats =
output->impl->get_cursor_formats(output, allocator->buffer_caps);
if (display_formats == NULL) {
wlr_log(WLR_ERROR, "Failed to get cursor display formats");
return NULL;
}
}
return output_pick_format(output, display_formats);
}
static struct wlr_buffer *render_cursor_buffer(struct wlr_output_cursor *cursor) {
struct wlr_output *output = cursor->output;
float scale = output->scale;
enum wl_output_transform transform = WL_OUTPUT_TRANSFORM_NORMAL;
struct wlr_texture *texture = cursor->texture;
if (cursor->surface != NULL) {
texture = wlr_surface_get_texture(cursor->surface);
scale = cursor->surface->current.scale;
transform = cursor->surface->current.transform;
}
if (texture == NULL) {
return NULL;
}
struct wlr_renderer *renderer = wlr_backend_get_renderer(output->backend);
if (renderer == NULL) {
wlr_log(WLR_ERROR, "Failed to get backend renderer");
return NULL;
}
struct wlr_allocator *allocator = backend_get_allocator(output->backend);
if (allocator == NULL) {
wlr_log(WLR_ERROR, "Failed to get backend allocator");
return NULL;
}
int width = texture->width;
int height = texture->height;
if (output->impl->get_cursor_size) {
// Apply hardware limitations on buffer size
output->impl->get_cursor_size(cursor->output, &width, &height);
if ((int)texture->width > width || (int)texture->height > height) {
wlr_log(WLR_DEBUG, "Cursor texture too large (%dx%d), "
"exceeds hardware limitations (%dx%d)", texture->width,
texture->height, width, height);
return NULL;
}
}
if (output->cursor_swapchain == NULL ||
output->cursor_swapchain->width != width ||
output->cursor_swapchain->height != height) {
struct wlr_drm_format *format =
output_pick_cursor_format(output);
if (format == NULL) {
wlr_log(WLR_ERROR, "Failed to pick cursor format");
return NULL;
}
wlr_swapchain_destroy(output->cursor_swapchain);
output->cursor_swapchain = wlr_swapchain_create(allocator,
width, height, format);
if (output->cursor_swapchain == NULL) {
wlr_log(WLR_ERROR, "Failed to create cursor swapchain");
return NULL;
}
}
struct wlr_buffer *buffer =
wlr_swapchain_acquire(output->cursor_swapchain, NULL);
if (buffer == NULL) {
return NULL;
}
struct wlr_box cursor_box = {
.width = texture->width * output->scale / scale,
.height = texture->height * output->scale / scale,
};
float output_matrix[9];
wlr_matrix_identity(output_matrix);
if (output->transform != WL_OUTPUT_TRANSFORM_NORMAL) {
struct wlr_box tr_size = {
.width = buffer->width,
.height = buffer->height,
};
wlr_box_transform(&tr_size, &tr_size, output->transform, 0, 0);
wlr_matrix_translate(output_matrix, buffer->width / 2.0,
buffer->height / 2.0);
wlr_matrix_transform(output_matrix, output->transform);
wlr_matrix_translate(output_matrix, - tr_size.width / 2.0,
- tr_size.height / 2.0);
}
float matrix[9];
wlr_matrix_project_box(matrix, &cursor_box, transform, 0, output_matrix);
if (!wlr_renderer_begin_with_buffer(renderer, buffer)) {
wlr_buffer_unlock(buffer);
return NULL;
}
wlr_renderer_clear(renderer, (float[]){ 0.0, 0.0, 0.0, 0.0 });
wlr_render_texture_with_matrix(renderer, texture, matrix, 1.0);
wlr_renderer_end(renderer);
return buffer;
}
static bool output_cursor_attempt_hardware(struct wlr_output_cursor *cursor) {
struct wlr_output *output = cursor->output;
if (!output->impl->set_cursor ||
output->software_cursor_locks > 0) {
return false;
}
struct wlr_output_cursor *hwcur = output->hardware_cursor;
if (hwcur != NULL && hwcur != cursor) {
return false;
}
struct wlr_texture *texture = cursor->texture;
if (cursor->surface != NULL) {
// TODO: try using the surface buffer directly
texture = wlr_surface_get_texture(cursor->surface);
}
// If the cursor was hidden or was a software cursor, the hardware
// cursor position is outdated
output->impl->move_cursor(cursor->output,
(int)cursor->x, (int)cursor->y);
struct wlr_buffer *buffer = NULL;
if (texture != NULL) {
buffer = render_cursor_buffer(cursor);
if (buffer == NULL) {
wlr_log(WLR_ERROR, "Failed to render cursor buffer");
return false;
}
}
struct wlr_box hotspot = {
.x = cursor->hotspot_x,
.y = cursor->hotspot_y,
};
wlr_box_transform(&hotspot, &hotspot,
wlr_output_transform_invert(output->transform),
buffer ? buffer->width : 0, buffer ? buffer->height : 0);
bool ok = output->impl->set_cursor(cursor->output, buffer,
hotspot.x, hotspot.y);
if (ok) {
wlr_buffer_unlock(output->cursor_front_buffer);
output->cursor_front_buffer = buffer;
output->hardware_cursor = cursor;
} else {
wlr_buffer_unlock(buffer);
}
return ok;
}
bool wlr_output_cursor_set_image(struct wlr_output_cursor *cursor,
const uint8_t *pixels, int32_t stride, uint32_t width, uint32_t height,
int32_t hotspot_x, int32_t hotspot_y) {
struct wlr_renderer *renderer =
wlr_backend_get_renderer(cursor->output->backend);
if (!renderer) {
// if the backend has no renderer, we can't draw a cursor, but this is
// actually okay, for ex. with the noop backend
return true;
}
output_cursor_reset(cursor);
cursor->width = width;
cursor->height = height;
cursor->hotspot_x = hotspot_x;
cursor->hotspot_y = hotspot_y;
output_cursor_update_visible(cursor);
wlr_texture_destroy(cursor->texture);
cursor->texture = NULL;
cursor->enabled = false;
if (pixels != NULL) {
cursor->texture = wlr_texture_from_pixels(renderer,
DRM_FORMAT_ARGB8888, stride, width, height, pixels);
if (cursor->texture == NULL) {
return false;
}
cursor->enabled = true;
}
if (output_cursor_attempt_hardware(cursor)) {
return true;
}
wlr_log(WLR_DEBUG, "Falling back to software cursor on output '%s'",
cursor->output->name);
output_cursor_damage_whole(cursor);
return true;
}
static void output_cursor_commit(struct wlr_output_cursor *cursor,
bool update_hotspot) {
if (cursor->output->hardware_cursor != cursor) {
output_cursor_damage_whole(cursor);
}
struct wlr_surface *surface = cursor->surface;
assert(surface != NULL);
// Some clients commit a cursor surface with a NULL buffer to hide it.
cursor->enabled = wlr_surface_has_buffer(surface);
cursor->width = surface->current.width * cursor->output->scale;
cursor->height = surface->current.height * cursor->output->scale;
output_cursor_update_visible(cursor);
if (update_hotspot) {
cursor->hotspot_x -= surface->current.dx * cursor->output->scale;
cursor->hotspot_y -= surface->current.dy * cursor->output->scale;
}
if (output_cursor_attempt_hardware(cursor)) {
return;
}
// Fallback to software cursor
output_cursor_damage_whole(cursor);
}
static void output_cursor_handle_commit(struct wl_listener *listener,
void *data) {
struct wlr_output_cursor *cursor =
wl_container_of(listener, cursor, surface_commit);
output_cursor_commit(cursor, true);
}
static void output_cursor_handle_destroy(struct wl_listener *listener,
void *data) {
struct wlr_output_cursor *cursor = wl_container_of(listener, cursor,
surface_destroy);
output_cursor_reset(cursor);
}
void wlr_output_cursor_set_surface(struct wlr_output_cursor *cursor,
struct wlr_surface *surface, int32_t hotspot_x, int32_t hotspot_y) {
hotspot_x *= cursor->output->scale;
hotspot_y *= cursor->output->scale;
if (surface && surface == cursor->surface) {
// Only update the hotspot: surface hasn't changed
if (cursor->output->hardware_cursor != cursor) {
output_cursor_damage_whole(cursor);
}
cursor->hotspot_x = hotspot_x;
cursor->hotspot_y = hotspot_y;
if (cursor->output->hardware_cursor != cursor) {
output_cursor_damage_whole(cursor);
} else {
struct wlr_buffer *buffer = cursor->output->cursor_front_buffer;
struct wlr_box hotspot = {
.x = cursor->hotspot_x,
.y = cursor->hotspot_y,
};
wlr_box_transform(&hotspot, &hotspot,
wlr_output_transform_invert(cursor->output->transform),
buffer ? buffer->width : 0, buffer ? buffer->height : 0);
assert(cursor->output->impl->set_cursor);
cursor->output->impl->set_cursor(cursor->output,
buffer, hotspot.x, hotspot.y);
}
return;
}
output_cursor_reset(cursor);
cursor->surface = surface;
cursor->hotspot_x = hotspot_x;
cursor->hotspot_y = hotspot_y;
if (surface != NULL) {
wl_signal_add(&surface->events.commit, &cursor->surface_commit);
wl_signal_add(&surface->events.destroy, &cursor->surface_destroy);
cursor->visible = false;
output_cursor_commit(cursor, false);
} else {
cursor->enabled = false;
cursor->width = 0;
cursor->height = 0;
if (cursor->output->hardware_cursor == cursor) {
assert(cursor->output->impl->set_cursor);
cursor->output->impl->set_cursor(cursor->output, NULL, 0, 0);
}
}
}
bool wlr_output_cursor_move(struct wlr_output_cursor *cursor,
double x, double y) {
if (cursor->x == x && cursor->y == y) {
return true;
}
if (cursor->output->hardware_cursor != cursor) {
output_cursor_damage_whole(cursor);
}
bool was_visible = cursor->visible;
x *= cursor->output->scale;
y *= cursor->output->scale;
cursor->x = x;
cursor->y = y;
output_cursor_update_visible(cursor);
if (!was_visible && !cursor->visible) {
// Cursor is still hidden, do nothing
return true;
}
if (cursor->output->hardware_cursor != cursor) {
output_cursor_damage_whole(cursor);
return true;
}
assert(cursor->output->impl->move_cursor);
return cursor->output->impl->move_cursor(cursor->output, (int)x, (int)y);
}
struct wlr_output_cursor *wlr_output_cursor_create(struct wlr_output *output) {
struct wlr_output_cursor *cursor =
calloc(1, sizeof(struct wlr_output_cursor));
if (cursor == NULL) {
return NULL;
}
cursor->output = output;
wl_signal_init(&cursor->events.destroy);
wl_list_init(&cursor->surface_commit.link);
cursor->surface_commit.notify = output_cursor_handle_commit;
wl_list_init(&cursor->surface_destroy.link);
cursor->surface_destroy.notify = output_cursor_handle_destroy;
wl_list_insert(&output->cursors, &cursor->link);
cursor->visible = true; // default position is at (0, 0)
return cursor;
}
void wlr_output_cursor_destroy(struct wlr_output_cursor *cursor) {
if (cursor == NULL) {
return;
}
output_cursor_reset(cursor);
wlr_signal_emit_safe(&cursor->events.destroy, cursor);
if (cursor->output->hardware_cursor == cursor) {
// If this cursor was the hardware cursor, disable it
if (cursor->output->impl->set_cursor) {
cursor->output->impl->set_cursor(cursor->output, NULL, 0, 0);
}
cursor->output->hardware_cursor = NULL;
}
wlr_texture_destroy(cursor->texture);
wl_list_remove(&cursor->link);
free(cursor);
}
enum wl_output_transform wlr_output_transform_invert(
enum wl_output_transform tr) {
if ((tr & WL_OUTPUT_TRANSFORM_90) && !(tr & WL_OUTPUT_TRANSFORM_FLIPPED)) {
tr ^= WL_OUTPUT_TRANSFORM_180;
}
return tr;
}
enum wl_output_transform wlr_output_transform_compose(
enum wl_output_transform tr_a, enum wl_output_transform tr_b) {
uint32_t flipped = (tr_a ^ tr_b) & WL_OUTPUT_TRANSFORM_FLIPPED;
uint32_t rotation_mask = WL_OUTPUT_TRANSFORM_90 | WL_OUTPUT_TRANSFORM_180;
uint32_t rotated;
if (tr_b & WL_OUTPUT_TRANSFORM_FLIPPED) {
// When a rotation of k degrees is followed by a flip, the
// equivalent transform is a flip followed by a rotation of
// -k degrees.
rotated = (tr_b - tr_a) & rotation_mask;
} else {
rotated = (tr_a + tr_b) & rotation_mask;
}
return flipped | rotated;
}