wsi/x11: Rewrite wait logic for ANI/WaitForPresentKHR.

When we need to poll the XCB connection with a non-trivial timeout,
be very careful to not hit an XCB bug where a poll() may hang for too
long even if an event is ready in the special event queue.

This is a pragmatic workaround, a wait_for_special_event_with_timeout()
is the only proper solution here.

Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19279>
This commit is contained in:
Hans-Kristian Arntzen
2022-10-26 14:17:22 +02:00
committed by Marge Bot
parent c31ffe63ae
commit c78257e8b0

View File

@@ -1162,14 +1162,65 @@ x11_handle_dri3_present_event(struct x11_swapchain *chain,
return VK_SUCCESS;
}
static uint64_t wsi_get_absolute_timeout(uint64_t timeout)
static VkResult
x11_poll_for_special_event(struct x11_swapchain *chain, uint64_t abs_timeout, xcb_generic_event_t **out_event)
{
uint64_t current_time = os_time_get_nano();
/* Start out with 1 ms intervals since that's what poll() supports. */
uint64_t poll_busywait_ns = 1000 * 1000;
xcb_generic_event_t *event;
uint64_t rel_timeout;
struct pollfd pfds;
timeout = MIN2(UINT64_MAX - current_time, timeout);
assert(abs_timeout != UINT64_MAX);
return current_time + timeout;
/* abs_timeout is assumed to be in timebase of os_time_get_absolute_timeout(). */
/* See comments in x11_manage_fifo_queues about problems with xcb_poll followed by poll().
* This path is suboptimal for scenarios where we're doing:
* - IMMEDIATE / MAILBOX (no acquire queue) and
* - Timeout that is neither 0 nor UINT64_MAX (very rare).
* The only real solution is a busy-poll scheme to ensure we don't sleep for too long.
* In a forward progress scenario, the XCB FD will be written at least once per frame,
* so we expect frequent wake-ups either way.
* This is a best-effort pragmatic solution until we have a proper solution in XCB.
*/
rel_timeout = abs_timeout;
*out_event = NULL;
event = NULL;
while (1) {
event = xcb_poll_for_special_event(chain->conn, chain->special_event);
if (event || rel_timeout == 0)
break;
/* If a non-special event happens, the fd will still
* poll. So recalculate the timeout now just in case.
*/
uint64_t current_time = os_time_get_nano();
if (abs_timeout > current_time)
rel_timeout = MIN2(poll_busywait_ns, abs_timeout - current_time);
else
rel_timeout = 0;
if (rel_timeout) {
pfds.fd = xcb_get_file_descriptor(chain->conn);
pfds.events = POLLIN;
int ret = poll(&pfds, 1, MAX2(rel_timeout / 1000 / 1000, 1u));
if (ret == -1)
return VK_ERROR_OUT_OF_DATE_KHR;
/* Gradually increase the poll duration if it takes a very long time to receive a poll event,
* since at that point, stutter isn't really the main concern anymore.
* We generally expect a special event to be received once every refresh duration. */
poll_busywait_ns += poll_busywait_ns / 2;
poll_busywait_ns = MIN2(10ull * 1000ull * 1000ull, poll_busywait_ns);
}
}
*out_event = event;
return event ? VK_SUCCESS : VK_TIMEOUT;
}
/**
@@ -1181,8 +1232,13 @@ x11_acquire_next_image_poll_x11(struct x11_swapchain *chain,
uint32_t *image_index, uint64_t timeout)
{
xcb_generic_event_t *event;
struct pollfd pfds;
uint64_t atimeout;
if (timeout == 0 || timeout == UINT64_MAX)
atimeout = timeout;
else
atimeout = os_time_get_absolute_timeout(timeout);
while (1) {
for (uint32_t i = 0; i < chain->base.image_count; i++) {
if (!chain->images[i].busy) {
@@ -1202,31 +1258,12 @@ x11_acquire_next_image_poll_x11(struct x11_swapchain *chain,
if (!event)
return x11_swapchain_result(chain, VK_ERROR_SURFACE_LOST_KHR);
} else {
event = xcb_poll_for_special_event(chain->conn, chain->special_event);
if (!event) {
int ret;
if (timeout == 0)
return x11_swapchain_result(chain, VK_NOT_READY);
atimeout = wsi_get_absolute_timeout(timeout);
pfds.fd = xcb_get_file_descriptor(chain->conn);
pfds.events = POLLIN;
ret = poll(&pfds, 1, timeout / 1000 / 1000);
if (ret == 0)
return x11_swapchain_result(chain, VK_TIMEOUT);
if (ret == -1)
return x11_swapchain_result(chain, VK_ERROR_OUT_OF_DATE_KHR);
/* If a non-special event happens, the fd will still
* poll. So recalculate the timeout now just in case.
*/
uint64_t current_time = os_time_get_nano();
if (atimeout > current_time)
timeout = atimeout - current_time;
else
timeout = 0;
continue;
VkResult result = x11_poll_for_special_event(chain, atimeout, &event);
if (result == VK_TIMEOUT) {
/* AcquireNextImageKHR reserves a special return value for 0 timeouts. */
return x11_swapchain_result(chain, timeout == 0 ? VK_NOT_READY : VK_TIMEOUT);
} else if (result != VK_SUCCESS) {
return x11_swapchain_result(chain, result);
}
}
@@ -2185,42 +2222,15 @@ static VkResult x11_wait_for_present_polled(
goto fail;
}
} else {
event = xcb_poll_for_special_event(chain->conn, chain->special_event);
if (!event) {
if (timeout == 0) {
result = x11_swapchain_result(chain, VK_TIMEOUT);
goto fail;
}
uint64_t current_time = os_time_get_nano();
if (abs_timeout_monotonic > current_time)
timeout = abs_timeout_monotonic - current_time;
else
timeout = 0;
struct pollfd pfds;
pfds.fd = xcb_get_file_descriptor(chain->conn);
pfds.events = POLLIN;
ret = poll(&pfds, 1, timeout / 1000 / 1000);
if (ret == 0) {
result = x11_swapchain_result(chain, VK_TIMEOUT);
goto fail;
}
if (ret == -1) {
result = x11_swapchain_result(chain, VK_ERROR_OUT_OF_DATE_KHR);
goto fail;
}
}
result = x11_poll_for_special_event(chain, abs_timeout_monotonic, &event);
if (result != VK_SUCCESS)
goto fail;
}
if (event) {
result = x11_handle_dri3_present_event(chain, (void *)event);
/* Ensure that VK_SUBOPTIMAL_KHR is reported to the application */
result = x11_swapchain_result(chain, result);
free(event);
}
result = x11_handle_dri3_present_event(chain, (void *)event);
/* Ensure that VK_SUBOPTIMAL_KHR is reported to the application */
result = x11_swapchain_result(chain, result);
free(event);
}
fail: