fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 05:10:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

venus: fix racy semaphore feedback counter update

Browse source 
Previously, we update the sfb dst slot upon vn_SignalSemaphore so that
vn_GetSemaphoreCounterValue can poll just the feedback slot itself.
However, that can race with pending sfb cmds that are going to update
the slot value, ending up with stuck sync progression.

This change fixes it by disallowing vn_SignalSemaphore to touch the sfb
dst slot. To ensure counter query being monotonic, vn_GetSemaphoreCounterValue
now takes the greater of signaled counter and the sfb counter read.

Test with dEQP-VK.synchronization* group:
- w/o this: stuck shows up within 2 min with 8 parallel deqp runs
- with this: no stuck for multiple full runs of the same

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14304
Fixes: 5c7e60362c ("venus: enable timeline semaphore feedback")
(cherry picked from commit 829bd406c0)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38803>
This commit is contained in:
Yiwei Zhang 2025-11-18 19:20:42 -08:00
parent e6a8e1321b
commit 68d31c28d4
1 changed files with 11 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
src/virtio/vulkan/vn_queue.c
View file

@ -2158,7 +2158,7 @@ vn_GetSemaphoreCounterValue(VkDevice device,
if (sem->feedback.slot) {
simple_mtx_lock(&sem->feedback.async_wait_mtx);
const uint64_t counter = vn_feedback_get_counter(sem->feedback.slot);
uint64_t counter = vn_feedback_get_counter(sem->feedback.slot);
if (sem->feedback.signaled_counter < counter) {
/* When the timeline semaphore feedback slot gets signaled, the real
* semaphore signal operation follows after but the signaling isr can
@ -2206,6 +2206,11 @@ vn_GetSemaphoreCounterValue(VkDevice device,
sem->feedback.signaled_counter = counter;
}
/* vn_SignalSemaphore writes the sfb signaled_counter without updating
* the slot. So the semaphore counter query here must consider both.
*/
counter = MAX2(counter, sem->feedback.signaled_counter);
simple_mtx_unlock(&sem->feedback.async_wait_mtx);
*pValue = counter;
@ -2227,9 +2232,13 @@ vn_SignalSemaphore(VkDevice device, const VkSemaphoreSignalInfo *pSignalInfo)
vn_async_vkSignalSemaphore(dev->primary_ring, device, pSignalInfo);
if (sem->feedback.slot) {
/* Must not update the sfb dst slot here because there's no followed
* submission to flush the cache (implicit sync guarantee) before the
* pending sfb cmd to update the slot. Otherwise, the slot update can be
* written by the racy update here.
*/
simple_mtx_lock(&sem->feedback.async_wait_mtx);
vn_feedback_set_counter(sem->feedback.slot, pSignalInfo->value);
/* Update async counters. Since we're signaling, we're aligned with
* the renderer.
*/