pthread: Fix pthread_cond_timedwait returning early from timeout

The reason timeouts would sometimes happen before the abstime deadline was due
to rounding errors converting the timeout to milliseconds, and also because
vTaskDelay(1) only delays until the next tick which is less than one full tick
period.

Closes https://github.com/espressif/esp-idf/issues/6901
This commit is contained in:
Angus Gratton
2021-05-21 19:09:33 +10:00
committed by bot
parent 3326cc852f
commit 262107691a
3 changed files with 73 additions and 7 deletions

View File

@@ -128,7 +128,8 @@ int pthread_cond_timedwait(pthread_cond_t *cv, pthread_mutex_t *mut, const struc
gettimeofday(&cur_time, NULL);
abs_time.tv_sec = to->tv_sec;
abs_time.tv_usec = to->tv_nsec / 1000;
// Round up nanoseconds to the next microsecond
abs_time.tv_usec = (to->tv_nsec + 1000 - 1) / 1000;
if (timercmp(&abs_time, &cur_time, <)) {
/* As per the pthread spec, if the time has already
@@ -137,14 +138,27 @@ int pthread_cond_timedwait(pthread_cond_t *cv, pthread_mutex_t *mut, const struc
timeout_msec = 0;
} else {
timersub(&abs_time, &cur_time, &diff_time);
timeout_msec = (diff_time.tv_sec * 1000) + (diff_time.tv_usec / 1000);
// Round up timeout microseconds to the next millisecond
timeout_msec = (diff_time.tv_sec * 1000) +
((diff_time.tv_usec + 1000 - 1) / 1000);
}
if (timeout_msec <= 0) {
return ETIMEDOUT;
}
timeout_ticks = timeout_msec / portTICK_PERIOD_MS;
// Round up milliseconds to the next tick
timeout_ticks = (timeout_msec + portTICK_PERIOD_MS - 1) / portTICK_PERIOD_MS;
/* We have to add 1 more tick of delay
The reason for this is that vTaskDelay(1) will sleep until the start of the next tick,
which can be any amount of time up to one tick period. So if we don't add one more tick,
we're likely to timeout a small time (< 1 tick period) before the requested timeout.
If we add 1 tick then we will timeout a small time (< 1 tick period) after the
requested timeout.
*/
timeout_ticks += 1;
}
esp_pthread_cond_waiter_t w;