mesa/src/vulkan/runtime/vk_sync_timeline.c
Faith Ekstrand 4a634a56bb vulkan/sync: Return early in vk_sync_timeline_wait() if wait_value == 0
This is a trivial case so we don't even need to take the lock.

Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36827>
2025-08-22 23:04:56 +00:00

598 lines
19 KiB
C

/*
* Copyright © 2021 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
#include "vk_sync_timeline.h"
#include <inttypes.h>
#include "util/os_time.h"
#include "util/timespec.h"
#include "vk_alloc.h"
#include "vk_device.h"
#include "vk_log.h"
static struct vk_sync_timeline *
to_vk_sync_timeline(struct vk_sync *sync)
{
assert(sync->type->init == vk_sync_timeline_init);
return container_of(sync, struct vk_sync_timeline, sync);
}
static struct vk_sync_timeline_state *
to_vk_sync_timeline_state(struct vk_sync *sync)
{
struct vk_sync_timeline *timeline = to_vk_sync_timeline(sync);
return timeline->state;
}
static void
vk_sync_timeline_type_validate(const struct vk_sync_timeline_type *ttype)
{
ASSERTED const enum vk_sync_features req_features =
VK_SYNC_FEATURE_BINARY |
VK_SYNC_FEATURE_GPU_WAIT |
VK_SYNC_FEATURE_GPU_MULTI_WAIT |
VK_SYNC_FEATURE_CPU_WAIT |
VK_SYNC_FEATURE_CPU_RESET;
assert(!(req_features & ~ttype->point_sync_type->features));
}
static void
vk_sync_timeline_state_ref(struct vk_sync_timeline_state *state)
{
p_atomic_inc(&state->refcount);
}
static void
vk_sync_timeline_state_unref(struct vk_device *device,
struct vk_sync_timeline_state *state)
{
if (p_atomic_dec_return(&state->refcount))
return;
list_for_each_entry_safe(struct vk_sync_timeline_point, point,
&state->free_points, link) {
list_del(&point->link);
vk_sync_finish(device, &point->sync);
vk_free(&device->alloc, point);
}
list_for_each_entry_safe(struct vk_sync_timeline_point, point,
&state->pending_points, link) {
list_del(&point->link);
vk_sync_finish(device, &point->sync);
vk_free(&device->alloc, point);
}
u_cnd_monotonic_destroy(&state->cond);
mtx_destroy(&state->mutex);
vk_free(&device->alloc, state);
}
VkResult
vk_sync_timeline_init(struct vk_device *device,
struct vk_sync *sync,
uint64_t initial_value)
{
struct vk_sync_timeline *timeline = to_vk_sync_timeline(sync);
struct vk_sync_timeline_state *state;
int ret;
ASSERTED const struct vk_sync_timeline_type *ttype =
container_of(timeline->sync.type, struct vk_sync_timeline_type, sync);
vk_sync_timeline_type_validate(ttype);
state = vk_zalloc(&device->alloc, sizeof(*state), 8,
VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!state)
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
ret = mtx_init(&state->mutex, mtx_plain);
if (ret != thrd_success) {
vk_free(&device->alloc, state);
return vk_errorf(device, VK_ERROR_UNKNOWN, "mtx_init failed");
}
ret = u_cnd_monotonic_init(&state->cond);
if (ret != thrd_success) {
mtx_destroy(&state->mutex);
vk_free(&device->alloc, state);
return vk_errorf(device, VK_ERROR_UNKNOWN, "cnd_init failed");
}
state->highest_past = state->highest_pending = initial_value;
list_inithead(&state->pending_points);
list_inithead(&state->free_points);
p_atomic_set(&state->refcount, 1);
timeline->state = state;
return VK_SUCCESS;
}
static VkResult
vk_sync_timeline_gc_locked(struct vk_device *device,
struct vk_sync_timeline_state *state,
bool drain);
static void
vk_sync_timeline_finish(struct vk_device *device,
struct vk_sync *sync)
{
struct vk_sync_timeline_state *state = to_vk_sync_timeline_state(sync);
/* We need to garbage collect to get rid of any pending points so that the
* vk_sync_timeline_state_unref() at the end drops the final reference
* held by the vk_sync_timeline. It's up to the client to ensure that
* there are no vk_sync in-flight when this is called so this should get
* rid of all pending time points. The only time point references left are
* those held by waits or vk_queue_submit().
*/
mtx_lock(&state->mutex);
vk_sync_timeline_gc_locked(device, state, true);
assert(list_is_empty(&state->pending_points));
mtx_unlock(&state->mutex);
vk_sync_timeline_state_unref(device, state);
}
static struct vk_sync_timeline_point *
vk_sync_timeline_first_point(struct vk_sync_timeline_state *state)
{
struct vk_sync_timeline_point *point =
list_first_entry(&state->pending_points,
struct vk_sync_timeline_point, link);
assert(point->value <= state->highest_pending);
assert(point->value > state->highest_past);
return point;
}
static VkResult
vk_sync_timeline_alloc_point_locked(struct vk_device *device,
struct vk_sync_timeline *timeline,
uint64_t value,
struct vk_sync_timeline_point **point_out)
{
struct vk_sync_timeline_state *state = timeline->state;
struct vk_sync_timeline_point *point;
VkResult result;
result = vk_sync_timeline_gc_locked(device, state, false);
if (unlikely(result != VK_SUCCESS))
return result;
if (list_is_empty(&state->free_points)) {
const struct vk_sync_timeline_type *ttype =
container_of(timeline->sync.type, struct vk_sync_timeline_type, sync);
const struct vk_sync_type *point_sync_type = ttype->point_sync_type;
size_t size = offsetof(struct vk_sync_timeline_point, sync) +
point_sync_type->size;
point = vk_zalloc(&device->alloc, size, 8,
VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!point)
return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
point->timeline_state = state;
result = vk_sync_init(device, &point->sync, point_sync_type,
0 /* flags */, 0 /* initial_value */);
if (unlikely(result != VK_SUCCESS)) {
vk_free(&device->alloc, point);
return result;
}
} else {
point = list_first_entry(&state->free_points,
struct vk_sync_timeline_point, link);
if (point->sync.type->reset) {
result = vk_sync_reset(device, &point->sync);
if (unlikely(result != VK_SUCCESS))
return result;
}
list_del(&point->link);
}
point->value = value;
assert(point->refcount == 0);
point->refcount++;
*point_out = point;
vk_sync_timeline_state_ref(state);
return VK_SUCCESS;
}
VkResult
vk_sync_timeline_alloc_point(struct vk_device *device,
struct vk_sync_timeline *timeline,
uint64_t value,
struct vk_sync_timeline_point **point_out)
{
VkResult result;
mtx_lock(&timeline->state->mutex);
result = vk_sync_timeline_alloc_point_locked(device, timeline, value, point_out);
mtx_unlock(&timeline->state->mutex);
return result;
}
static void
vk_sync_timeline_ref_point_locked(struct vk_sync_timeline_point *point)
{
point->refcount++;
}
/* Returns true if this was the last reference to point.
*
* DO NOT call this helper directly. You should call vk_sync_timeline_unref_point_locked()
* or vk_sync_timeline_point_unref() instead.
*/
static bool
vk_sync_timeline_unref_point_no_unref_state_locked(struct vk_sync_timeline_point *point)
{
struct vk_sync_timeline_state *state = point->timeline_state;
assert(point->refcount > 0);
point->refcount--;
if (point->refcount > 0)
return false;
assert(point->refcount == 0);
/* The pending list also takes a reference so this can't be pending */
assert(!point->pending);
list_add(&point->link, &state->free_points);
return true;
}
static void
vk_sync_timeline_unref_point_locked(struct vk_device *device,
struct vk_sync_timeline_state *state,
struct vk_sync_timeline_point *point)
{
/* The caller needs to have its own reference to the state, not just the
* one implicit in point, because it's also holding the lock.
*/
assert(p_atomic_read(&state->refcount) > 1);
if (vk_sync_timeline_unref_point_no_unref_state_locked(point))
vk_sync_timeline_state_unref(device, state);
}
void
vk_sync_timeline_point_unref(struct vk_device *device,
struct vk_sync_timeline_point *point)
{
struct vk_sync_timeline_state *state = point->timeline_state;
mtx_lock(&state->mutex);
const bool last_ref =
vk_sync_timeline_unref_point_no_unref_state_locked(point);
mtx_unlock(&state->mutex);
/* Drop the state reference outside the mutex so we don't free the state
* and then try to unlock the mutex.
*/
if (last_ref)
vk_sync_timeline_state_unref(device, state);
}
static void
vk_sync_timeline_complete_point_locked(struct vk_device *device,
struct vk_sync_timeline_state *state,
struct vk_sync_timeline_point *point)
{
if (!point->pending)
return;
assert(state->highest_past < point->value);
state->highest_past = point->value;
point->pending = false;
list_del(&point->link);
/* Drop the pending reference */
vk_sync_timeline_unref_point_locked(device, state, point);
}
static VkResult
vk_sync_timeline_gc_locked(struct vk_device *device,
struct vk_sync_timeline_state *state,
bool drain)
{
list_for_each_entry_safe(struct vk_sync_timeline_point, point,
&state->pending_points, link) {
/* state->highest_pending is only incremented once submission has
* happened. If this point has a greater serial, it means the point
* hasn't been submitted yet.
*/
if (point->value > state->highest_pending)
return VK_SUCCESS;
/* If someone is waiting on this time point, consider it busy and don't
* try to recycle it. There's a slim possibility that it's no longer
* busy by the time we look at it but we would be recycling it out from
* under a waiter and that can lead to weird races.
*
* We walk the list in-order so if this time point is still busy so is
* every following time point
*/
assert(point->refcount > 0);
if (point->refcount > 1 && !drain)
return VK_SUCCESS;
/* Garbage collect any signaled point. */
VkResult result = vk_sync_wait(device, &point->sync, 0,
VK_SYNC_WAIT_COMPLETE,
0 /* abs_timeout_ns */);
if (result == VK_TIMEOUT) {
/* We walk the list in-order so if this time point is still busy so
* is every following time point
*/
return VK_SUCCESS;
} else if (result != VK_SUCCESS) {
return result;
}
vk_sync_timeline_complete_point_locked(device, state, point);
}
return VK_SUCCESS;
}
VkResult
vk_sync_timeline_point_install(struct vk_device *device,
struct vk_sync_timeline_point *point)
{
struct vk_sync_timeline_state *state = point->timeline_state;
mtx_lock(&state->mutex);
assert(point->value > state->highest_pending);
state->highest_pending = point->value;
/* Adding to the pending list implicitly takes a reference but also this
* function is documented to consume the reference to point so we don't
* need to do anything to the reference count here.
*/
point->pending = true;
list_addtail(&point->link, &state->pending_points);
int ret = u_cnd_monotonic_broadcast(&state->cond);
mtx_unlock(&state->mutex);
if (ret == thrd_error)
return vk_errorf(device, VK_ERROR_UNKNOWN, "cnd_broadcast failed");
return VK_SUCCESS;
}
static VkResult
vk_sync_timeline_get_point_locked(struct vk_device *device,
struct vk_sync_timeline_state *state,
uint64_t wait_value,
struct vk_sync_timeline_point **point_out)
{
if (state->highest_past >= wait_value) {
/* Nothing to wait on */
*point_out = NULL;
return VK_SUCCESS;
}
list_for_each_entry(struct vk_sync_timeline_point, point,
&state->pending_points, link) {
if (point->value >= wait_value) {
vk_sync_timeline_ref_point_locked(point);
*point_out = point;
return VK_SUCCESS;
}
}
return VK_NOT_READY;
}
VkResult
vk_sync_timeline_get_point(struct vk_device *device,
struct vk_sync_timeline *timeline,
uint64_t wait_value,
struct vk_sync_timeline_point **point_out)
{
struct vk_sync_timeline_state *state = timeline->state;
mtx_lock(&state->mutex);
VkResult result = vk_sync_timeline_get_point_locked(device, state,
wait_value, point_out);
mtx_unlock(&state->mutex);
return result;
}
static VkResult
vk_sync_timeline_signal_locked(struct vk_device *device,
struct vk_sync_timeline_state *state,
uint64_t value)
{
VkResult result = vk_sync_timeline_gc_locked(device, state, true);
if (unlikely(result != VK_SUCCESS))
return result;
if (unlikely(value <= state->highest_past)) {
return vk_device_set_lost(device, "Timeline values must only ever "
"strictly increase.");
}
assert(list_is_empty(&state->pending_points));
assert(state->highest_pending == state->highest_past);
state->highest_pending = state->highest_past = value;
int ret = u_cnd_monotonic_broadcast(&state->cond);
if (ret == thrd_error)
return vk_errorf(device, VK_ERROR_UNKNOWN, "cnd_broadcast failed");
return VK_SUCCESS;
}
static VkResult
vk_sync_timeline_signal(struct vk_device *device,
struct vk_sync *sync,
uint64_t value)
{
struct vk_sync_timeline_state *state = to_vk_sync_timeline_state(sync);
mtx_lock(&state->mutex);
VkResult result = vk_sync_timeline_signal_locked(device, state, value);
mtx_unlock(&state->mutex);
return result;
}
static VkResult
vk_sync_timeline_get_value(struct vk_device *device,
struct vk_sync *sync,
uint64_t *value)
{
struct vk_sync_timeline_state *state = to_vk_sync_timeline_state(sync);
mtx_lock(&state->mutex);
VkResult result = vk_sync_timeline_gc_locked(device, state, true);
mtx_unlock(&state->mutex);
if (result != VK_SUCCESS)
return result;
*value = state->highest_past;
return VK_SUCCESS;
}
static VkResult
vk_sync_timeline_wait_locked(struct vk_device *device,
struct vk_sync_timeline_state *state,
uint64_t wait_value,
enum vk_sync_wait_flags wait_flags,
uint64_t abs_timeout_ns)
{
struct timespec abs_timeout_ts;
timespec_from_nsec(&abs_timeout_ts, abs_timeout_ns);
/* Wait on the queue_submit condition variable until the timeline has a
* time point pending that's at least as high as wait_value.
*/
while (state->highest_pending < wait_value) {
int ret = u_cnd_monotonic_timedwait(&state->cond, &state->mutex,
&abs_timeout_ts);
if (ret == thrd_timedout)
return VK_TIMEOUT;
if (ret != thrd_success)
return vk_errorf(device, VK_ERROR_UNKNOWN, "cnd_timedwait failed");
}
if (wait_flags & VK_SYNC_WAIT_PENDING)
return VK_SUCCESS;
VkResult result = vk_sync_timeline_gc_locked(device, state, false);
if (result != VK_SUCCESS)
return result;
while (state->highest_past < wait_value) {
struct vk_sync_timeline_point *point = vk_sync_timeline_first_point(state);
/* Drop the lock while we wait. */
vk_sync_timeline_ref_point_locked(point);
mtx_unlock(&state->mutex);
result = vk_sync_wait(device, &point->sync, 0,
VK_SYNC_WAIT_COMPLETE,
abs_timeout_ns);
/* Pick the mutex back up */
mtx_lock(&state->mutex);
vk_sync_timeline_unref_point_locked(device, state, point);
/* This covers both VK_TIMEOUT and VK_ERROR_DEVICE_LOST */
if (result != VK_SUCCESS)
return result;
vk_sync_timeline_complete_point_locked(device, state, point);
}
return VK_SUCCESS;
}
static VkResult
vk_sync_timeline_wait(struct vk_device *device,
struct vk_sync *sync,
uint64_t wait_value,
enum vk_sync_wait_flags wait_flags,
uint64_t abs_timeout_ns)
{
struct vk_sync_timeline_state *state = to_vk_sync_timeline_state(sync);
if (wait_value == 0)
return VK_SUCCESS;
mtx_lock(&state->mutex);
VkResult result = vk_sync_timeline_wait_locked(device, state,
wait_value, wait_flags,
abs_timeout_ns);
mtx_unlock(&state->mutex);
return result;
}
struct vk_sync_timeline_type
vk_sync_timeline_get_type(const struct vk_sync_type *point_sync_type)
{
return (struct vk_sync_timeline_type) {
.sync = {
.size = sizeof(struct vk_sync_timeline),
.features = VK_SYNC_FEATURE_TIMELINE |
VK_SYNC_FEATURE_GPU_WAIT |
VK_SYNC_FEATURE_CPU_WAIT |
VK_SYNC_FEATURE_CPU_SIGNAL |
VK_SYNC_FEATURE_WAIT_ANY |
VK_SYNC_FEATURE_WAIT_PENDING,
.init = vk_sync_timeline_init,
.finish = vk_sync_timeline_finish,
.signal = vk_sync_timeline_signal,
.get_value = vk_sync_timeline_get_value,
.wait = vk_sync_timeline_wait,
},
.point_sync_type = point_sync_type,
};
}