Merge branch 'fix/usleep_returns_early' into 'master'

fix(newlib): Fixed an issue where usleep() could consume more CPU cycles

Closes IDFGH-14342

See merge request espressif/esp-idf!38812

Author: sudeep-mohanty

components/driver/test_apps/legacy_mcpwm_driver/main/test_legacy_mcpwm.c

@@ -172,7 +172,7 @@ static uint32_t pcnt_get_pulse_number(pcnt_unit_handle_t pwm_pcnt_unit, int capt
     int count_value = 0;
     TEST_ESP_OK(pcnt_unit_clear_count(pwm_pcnt_unit));
     TEST_ESP_OK(pcnt_unit_start(pwm_pcnt_unit));
-    usleep(capture_window_ms * 1000);
+    vTaskDelay(pdMS_TO_TICKS(capture_window_ms));
     TEST_ESP_OK(pcnt_unit_stop(pwm_pcnt_unit));
     TEST_ESP_OK(pcnt_unit_get_count(pwm_pcnt_unit, &count_value));
     printf("count value: %d\r\n", count_value);

components/newlib/src/time.c

@@ -219,12 +219,13 @@ int usleep(useconds_t us)
         do {
             vTaskDelay((((target_us - now_us) + us_per_tick - 1) / us_per_tick));
             now_us = esp_time_impl_get_time();
-            /* If the time left until the target is less than 1 tick, then we use ROM delay to fill the gap */
-            uint64_t time_left = target_us - now_us;
-            if (time_left != 0 && time_left < us_per_tick) {
-                esp_rom_delay_us(time_left);
-                break;
-            }
+            /* It is possible that the time left until the target time is less
+             * than a tick period. However, we let usleep() to sleep for an
+             * entire tick period. This, could result in usleep() sleeping for
+             * a longer time than the requested time but that does not violate
+             * the spec of usleep(). Additionally, it allows FreeRTOS to schedule
+             * other tasks while the current task is sleeping.
+             */
         } while (now_us < target_us);
     }
     return 0;