wlroots-hyprland/backend/drm/libliftoff.c
Simon Ser 835208db98 output-layer: add support for scaling buffers
This allows callers to set a destination size different from the
buffer size to scale them.

The DRM backend supports this. The Wayland backend doesn't yet
(we'd need to wire up viewporter).
2023-04-04 19:39:38 +02:00

479 lines
14 KiB
C

#define _POSIX_C_SOURCE 200809L
#include <fcntl.h>
#include <libliftoff.h>
#include <sys/stat.h>
#include <unistd.h>
#include <wlr/util/log.h>
#include "backend/drm/drm.h"
#include "backend/drm/iface.h"
static bool init(struct wlr_drm_backend *drm) {
// TODO: lower log level
liftoff_log_set_priority(LIFTOFF_DEBUG);
int drm_fd = fcntl(drm->fd, F_DUPFD_CLOEXEC, 0);
if (drm_fd < 0) {
wlr_log_errno(WLR_ERROR, "fcntl(F_DUPFD_CLOEXEC) failed");
return false;
}
drm->liftoff = liftoff_device_create(drm_fd);
if (!drm->liftoff) {
wlr_log(WLR_ERROR, "Failed to create liftoff device");
close(drm_fd);
return false;
}
for (size_t i = 0; i < drm->num_planes; i++) {
struct wlr_drm_plane *plane = &drm->planes[i];
if (plane->initial_crtc_id != 0) {
continue;
}
plane->liftoff = liftoff_plane_create(drm->liftoff, plane->id);
if (plane->liftoff == NULL) {
wlr_log(WLR_ERROR, "Failed to create liftoff plane");
return false;
}
}
for (size_t i = 0; i < drm->num_crtcs; i++) {
struct wlr_drm_crtc *crtc = &drm->crtcs[i];
crtc->liftoff = liftoff_output_create(drm->liftoff, crtc->id);
if (!crtc->liftoff) {
wlr_log(WLR_ERROR, "Failed to create liftoff output");
return false;
}
crtc->liftoff_composition_layer = liftoff_layer_create(crtc->liftoff);
if (!crtc->liftoff_composition_layer) {
wlr_log(WLR_ERROR, "Failed to create liftoff composition layer");
return false;
}
liftoff_output_set_composition_layer(crtc->liftoff,
crtc->liftoff_composition_layer);
if (crtc->primary) {
crtc->primary->liftoff_layer = liftoff_layer_create(crtc->liftoff);
if (!crtc->primary->liftoff_layer) {
wlr_log(WLR_ERROR, "Failed to create liftoff layer for primary plane");
return false;
}
}
if (crtc->cursor) {
crtc->cursor->liftoff_layer = liftoff_layer_create(crtc->liftoff);
if (!crtc->cursor->liftoff_layer) {
wlr_log(WLR_ERROR, "Failed to create liftoff layer for cursor plane");
return false;
}
}
}
return true;
}
static bool register_planes_for_crtc(struct wlr_drm_backend *drm,
struct wlr_drm_crtc *crtc) {
// When performing the first modeset on a CRTC, we need to be a bit careful
// when it comes to planes: we don't want to allow libliftoff to make use
// of planes currently already in-use on another CRTC. We need to wait for
// a modeset to happen on the other CRTC before being able to use these.
for (size_t i = 0; i < drm->num_planes; i++) {
struct wlr_drm_plane *plane = &drm->planes[i];
if (plane->liftoff != NULL || plane->initial_crtc_id != crtc->id) {
continue;
}
plane->liftoff = liftoff_plane_create(drm->liftoff, plane->id);
if (plane->liftoff == NULL) {
wlr_log(WLR_ERROR, "Failed to create liftoff plane");
return false;
}
}
return true;
}
static void finish(struct wlr_drm_backend *drm) {
for (size_t i = 0; i < drm->num_crtcs; i++) {
struct wlr_drm_crtc *crtc = &drm->crtcs[i];
if (crtc->primary) {
liftoff_layer_destroy(crtc->primary->liftoff_layer);
}
if (crtc->cursor) {
liftoff_layer_destroy(crtc->cursor->liftoff_layer);
}
liftoff_layer_destroy(crtc->liftoff_composition_layer);
liftoff_output_destroy(crtc->liftoff);
}
liftoff_device_destroy(drm->liftoff);
}
static bool add_prop(drmModeAtomicReq *req, uint32_t obj,
uint32_t prop, uint64_t val) {
if (drmModeAtomicAddProperty(req, obj, prop, val) < 0) {
wlr_log_errno(WLR_ERROR, "drmModeAtomicAddProperty failed");
return false;
}
return true;
}
static void commit_blob(struct wlr_drm_backend *drm,
uint32_t *current, uint32_t next) {
if (*current == next) {
return;
}
if (*current != 0) {
drmModeDestroyPropertyBlob(drm->fd, *current);
}
*current = next;
}
static void rollback_blob(struct wlr_drm_backend *drm,
uint32_t *current, uint32_t next) {
if (*current == next) {
return;
}
if (next != 0) {
drmModeDestroyPropertyBlob(drm->fd, next);
}
}
static bool set_plane_props(struct wlr_drm_plane *plane,
struct liftoff_layer *layer, struct wlr_drm_fb *fb, int32_t x, int32_t y, uint64_t zpos) {
if (fb == NULL) {
wlr_log(WLR_ERROR, "Failed to acquire FB for plane %"PRIu32, plane->id);
return false;
}
uint32_t width = fb->wlr_buf->width;
uint32_t height = fb->wlr_buf->height;
// The SRC_* properties are in 16.16 fixed point
return liftoff_layer_set_property(layer, "zpos", zpos) == 0 &&
liftoff_layer_set_property(layer, "SRC_X", 0) == 0 &&
liftoff_layer_set_property(layer, "SRC_Y", 0) == 0 &&
liftoff_layer_set_property(layer, "SRC_W", (uint64_t)width << 16) == 0 &&
liftoff_layer_set_property(layer, "SRC_H", (uint64_t)height << 16) == 0 &&
liftoff_layer_set_property(layer, "CRTC_X", (uint64_t)x) == 0 &&
liftoff_layer_set_property(layer, "CRTC_Y", (uint64_t)y) == 0 &&
liftoff_layer_set_property(layer, "CRTC_W", width) == 0 &&
liftoff_layer_set_property(layer, "CRTC_H", height) == 0 &&
liftoff_layer_set_property(layer, "FB_ID", fb->id) == 0;
}
static bool disable_plane(struct wlr_drm_plane *plane) {
return liftoff_layer_set_property(plane->liftoff_layer, "FB_ID", 0) == 0;
}
static uint64_t to_fp16(double v) {
return (uint64_t)round(v * (1 << 16));
}
static bool set_layer_props(struct wlr_drm_backend *drm,
const struct wlr_output_layer_state *state, uint64_t zpos) {
struct wlr_drm_layer *layer = get_drm_layer(drm, state->layer);
uint32_t width = 0, height = 0;
if (state->buffer != NULL) {
width = state->buffer->width;
height = state->buffer->height;
}
struct wlr_drm_fb *fb = layer->pending_fb;
int ret = 0;
if (state->buffer == NULL) {
ret = liftoff_layer_set_property(layer->liftoff, "FB_ID", 0);
} else if (fb == NULL) {
liftoff_layer_set_fb_composited(layer->liftoff);
} else {
ret = liftoff_layer_set_property(layer->liftoff, "FB_ID", fb->id);
}
if (ret != 0) {
return false;
}
uint64_t crtc_x = (uint64_t)state->dst_box.x;
uint64_t crtc_y = (uint64_t)state->dst_box.y;
uint64_t crtc_w = (uint64_t)state->dst_box.width;
uint64_t crtc_h = (uint64_t)state->dst_box.height;
uint64_t src_x = to_fp16(0);
uint64_t src_y = to_fp16(0);
uint64_t src_w = to_fp16(width);
uint64_t src_h = to_fp16(height);
return
liftoff_layer_set_property(layer->liftoff, "zpos", zpos) == 0 &&
liftoff_layer_set_property(layer->liftoff, "CRTC_X", crtc_x) == 0 &&
liftoff_layer_set_property(layer->liftoff, "CRTC_Y", crtc_y) == 0 &&
liftoff_layer_set_property(layer->liftoff, "CRTC_W", crtc_w) == 0 &&
liftoff_layer_set_property(layer->liftoff, "CRTC_H", crtc_h) == 0 &&
liftoff_layer_set_property(layer->liftoff, "SRC_X", src_x) == 0 &&
liftoff_layer_set_property(layer->liftoff, "SRC_Y", src_y) == 0 &&
liftoff_layer_set_property(layer->liftoff, "SRC_W", src_w) == 0 &&
liftoff_layer_set_property(layer->liftoff, "SRC_H", src_h) == 0;
}
static bool devid_from_fd(int fd, dev_t *devid) {
struct stat stat;
if (fstat(fd, &stat) != 0) {
wlr_log_errno(WLR_ERROR, "fstat failed");
return false;
}
*devid = stat.st_rdev;
return true;
}
static void update_layer_feedback(struct wlr_drm_backend *drm,
struct wlr_drm_layer *layer) {
bool changed = false;
for (size_t i = 0; i < drm->num_planes; i++) {
struct wlr_drm_plane *plane = &drm->planes[i];
bool is_candidate = liftoff_layer_is_candidate_plane(layer->liftoff,
plane->liftoff);
if (layer->candidate_planes[i] != is_candidate) {
layer->candidate_planes[i] = is_candidate;
changed = true;
}
}
if (!changed) {
return;
}
dev_t target_device;
if (!devid_from_fd(drm->fd, &target_device)) {
return;
}
struct wlr_drm_format_set formats = {0};
for (size_t i = 0; i < drm->num_planes; i++) {
struct wlr_drm_plane *plane = &drm->planes[i];
if (!layer->candidate_planes[i]) {
continue;
}
for (size_t j = 0; j < plane->formats.len; j++) {
const struct wlr_drm_format *format = plane->formats.formats[j];
for (size_t k = 0; k < format->len; k++) {
wlr_drm_format_set_add(&formats, format->format,
format->modifiers[k]);
}
}
}
struct wlr_output_layer_feedback_event event = {
.target_device = target_device,
.formats = &formats,
};
wl_signal_emit_mutable(&layer->wlr->events.feedback, &event);
wlr_drm_format_set_finish(&formats);
}
static bool crtc_commit(struct wlr_drm_connector *conn,
const struct wlr_drm_connector_state *state, uint32_t flags,
bool test_only) {
struct wlr_drm_backend *drm = conn->backend;
struct wlr_output *output = &conn->output;
struct wlr_drm_crtc *crtc = conn->crtc;
bool modeset = state->modeset;
bool active = state->active;
if (modeset && !register_planes_for_crtc(drm, crtc)) {
return false;
}
uint32_t mode_id = crtc->mode_id;
if (modeset) {
if (!create_mode_blob(drm, conn, state, &mode_id)) {
return false;
}
}
uint32_t gamma_lut = crtc->gamma_lut;
if (state->base->committed & WLR_OUTPUT_STATE_GAMMA_LUT) {
// Fallback to legacy gamma interface when gamma properties are not
// available (can happen on older Intel GPUs that support gamma but not
// degamma).
if (crtc->props.gamma_lut == 0) {
if (!drm_legacy_crtc_set_gamma(drm, crtc,
state->base->gamma_lut_size,
state->base->gamma_lut)) {
return false;
}
} else {
if (!create_gamma_lut_blob(drm, state->base->gamma_lut_size,
state->base->gamma_lut, &gamma_lut)) {
return false;
}
}
}
uint32_t fb_damage_clips = 0;
if ((state->base->committed & WLR_OUTPUT_STATE_DAMAGE) &&
pixman_region32_not_empty(&state->base->damage) &&
crtc->primary->props.fb_damage_clips != 0) {
int rects_len;
const pixman_box32_t *rects = pixman_region32_rectangles(&state->base->damage, &rects_len);
if (drmModeCreatePropertyBlob(drm->fd, rects,
sizeof(*rects) * rects_len, &fb_damage_clips) != 0) {
wlr_log_errno(WLR_ERROR, "Failed to create FB_DAMAGE_CLIPS property blob");
}
}
bool prev_vrr_enabled =
output->adaptive_sync_status == WLR_OUTPUT_ADAPTIVE_SYNC_ENABLED;
bool vrr_enabled = prev_vrr_enabled;
if ((state->base->committed & WLR_OUTPUT_STATE_ADAPTIVE_SYNC_ENABLED) &&
drm_connector_supports_vrr(conn)) {
vrr_enabled = state->base->adaptive_sync_enabled;
}
if (test_only) {
flags |= DRM_MODE_ATOMIC_TEST_ONLY;
}
if (modeset) {
flags |= DRM_MODE_ATOMIC_ALLOW_MODESET;
} else if (!test_only && (state->base->committed & WLR_OUTPUT_STATE_BUFFER)) {
// The wlr_output API requires non-modeset commits with a new buffer to
// wait for the frame event. However compositors often perform
// non-modesets commits without a new buffer without waiting for the
// frame event. In that case we need to make the KMS commit blocking,
// otherwise the kernel will error out with EBUSY.
flags |= DRM_MODE_ATOMIC_NONBLOCK;
}
drmModeAtomicReq *req = drmModeAtomicAlloc();
if (req == NULL) {
wlr_log(WLR_ERROR, "drmModeAtomicAlloc failed");
return false;
}
bool ok = add_prop(req, conn->id, conn->props.crtc_id,
active ? crtc->id : 0);
if (modeset && active && conn->props.link_status != 0) {
ok = ok && add_prop(req, conn->id, conn->props.link_status,
DRM_MODE_LINK_STATUS_GOOD);
}
if (active && conn->props.content_type != 0) {
ok = ok && add_prop(req, conn->id, conn->props.content_type,
DRM_MODE_CONTENT_TYPE_GRAPHICS);
}
// TODO: set "max bpc"
ok = ok &&
add_prop(req, crtc->id, crtc->props.mode_id, mode_id) &&
add_prop(req, crtc->id, crtc->props.active, active);
if (active) {
if (crtc->props.gamma_lut != 0) {
ok = ok && add_prop(req, crtc->id, crtc->props.gamma_lut, gamma_lut);
}
if (crtc->props.vrr_enabled != 0) {
ok = ok && add_prop(req, crtc->id, crtc->props.vrr_enabled, vrr_enabled);
}
ok = ok &&
set_plane_props(crtc->primary, crtc->primary->liftoff_layer, state->primary_fb, 0, 0, 0) &&
set_plane_props(crtc->primary, crtc->liftoff_composition_layer, state->primary_fb, 0, 0, 0);
liftoff_layer_set_property(crtc->primary->liftoff_layer,
"FB_DAMAGE_CLIPS", fb_damage_clips);
liftoff_layer_set_property(crtc->liftoff_composition_layer,
"FB_DAMAGE_CLIPS", fb_damage_clips);
if (state->base->committed & WLR_OUTPUT_STATE_LAYERS) {
for (size_t i = 0; i < state->base->layers_len; i++) {
const struct wlr_output_layer_state *layer_state = &state->base->layers[i];
ok = ok && set_layer_props(drm, layer_state, i + 1);
}
}
if (crtc->cursor) {
if (drm_connector_is_cursor_visible(conn)) {
ok = ok && set_plane_props(crtc->cursor, crtc->cursor->liftoff_layer,
get_next_cursor_fb(conn), conn->cursor_x, conn->cursor_y,
wl_list_length(&crtc->layers) + 1);
} else {
ok = ok && disable_plane(crtc->cursor);
}
}
} else {
ok = ok && disable_plane(crtc->primary);
if (crtc->cursor) {
ok = ok && disable_plane(crtc->cursor);
}
}
if (!ok) {
goto out;
}
int ret = liftoff_output_apply(crtc->liftoff, req, flags);
if (ret != 0) {
wlr_drm_conn_log(conn, test_only ? WLR_DEBUG : WLR_ERROR,
"liftoff_output_apply failed: %s", strerror(-ret));
ok = false;
goto out;
}
if (crtc->cursor &&
liftoff_layer_needs_composition(crtc->cursor->liftoff_layer)) {
wlr_drm_conn_log(conn, WLR_DEBUG, "Failed to scan-out cursor plane");
ok = false;
goto out;
}
ret = drmModeAtomicCommit(drm->fd, req, flags, drm);
if (ret != 0) {
wlr_drm_conn_log_errno(conn, test_only ? WLR_DEBUG : WLR_ERROR,
"Atomic commit failed");
ok = false;
goto out;
}
if (state->base->committed & WLR_OUTPUT_STATE_LAYERS) {
for (size_t i = 0; i < state->base->layers_len; i++) {
struct wlr_output_layer_state *layer_state = &state->base->layers[i];
struct wlr_drm_layer *layer = get_drm_layer(drm, layer_state->layer);
layer_state->accepted =
!liftoff_layer_needs_composition(layer->liftoff);
if (!test_only && !layer_state->accepted) {
update_layer_feedback(drm, layer);
}
}
}
out:
drmModeAtomicFree(req);
if (ok && !test_only) {
commit_blob(drm, &crtc->mode_id, mode_id);
commit_blob(drm, &crtc->gamma_lut, gamma_lut);
if (vrr_enabled != prev_vrr_enabled) {
output->adaptive_sync_status = vrr_enabled ?
WLR_OUTPUT_ADAPTIVE_SYNC_ENABLED :
WLR_OUTPUT_ADAPTIVE_SYNC_DISABLED;
wlr_drm_conn_log(conn, WLR_DEBUG, "VRR %s",
vrr_enabled ? "enabled" : "disabled");
}
} else {
rollback_blob(drm, &crtc->mode_id, mode_id);
rollback_blob(drm, &crtc->gamma_lut, gamma_lut);
}
if (fb_damage_clips != 0 &&
drmModeDestroyPropertyBlob(drm->fd, fb_damage_clips) != 0) {
wlr_log_errno(WLR_ERROR, "Failed to destroy FB_DAMAGE_CLIPS property blob");
}
return ok;
}
const struct wlr_drm_interface liftoff_iface = {
.init = init,
.finish = finish,
.crtc_commit = crtc_commit,
};