riot-ci/RIOT
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

test/xtimer_hang: DEBUG_PINS

Browse Source 
Add the option to use debug pins to investigate timing issues.
This commit is contained in:
josar 2018-10-16 21:35:15 +02:00
parent 6ed11de354
commit 162d17c5a2
3 changed files with 49 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
tests/xtimer_hang/Makefile
View File

@ -6,4 +6,18 @@ USEMODULE += xtimer
TEST_ON_CI_WHITELIST += all
# Port and pin configuration for probing with oscilloscope
# Define Test pin for hardware timer interrupt, hardware dependent
# For all ATmega Platforms
#CFLAGS += -DDEBUG_TIMER_PORT=PORTF
#CFLAGS += -DDEBUG_TIMER_DDR=DDRF
#CFLAGS += -DDEBUG_TIMER_PIN=PORTF4
# Define test probing pins GPIO API based.
# Port number should be found in port enum e.g in cpu/include/periph_cpu.h
#FEATURES_REQUIRED += periph_gpio
# Jiminy probing Pins
#CFLAGS += -DWORKER_THREAD_PIN=GPIO_PIN\(5,7\)
#CFLAGS += -DMAIN_THREAD_PIN=GPIO_PIN\(5,6\)
include $(RIOTBASE)/Makefile.include

11
tests/xtimer_hang/README.md
View File

@ -9,3 +9,14 @@ the test has failed.
Please note (again), this is a runtime test to check if xtimer runs into a
deadlock, it is not about clock stability nor accuracy of timing intervals.
When debug pins are used and observed the expected output is as follows:
The MAIN_THREAD_PIN is on for ca. 100ms and that in a regular interval. If this interval is not regular or has gaps this
is an error. The WORKER_THREAD_PIN should toggle after 1ms and 1.1ms. (As this might fall in the XTIMER_ISR_BACKOFF the
first pin toggle is delayed untill the second is ready. Thus the time interval can be longer.)
If the Timer fall in the same interrupt there might be a interrupt before the second worker thread timer is set.
This leads to a separation of the timer interrupts until they again fall in the same interrupt.
It might happen that from the seperation of the interrupts till the merge there is only a uneven count of
WORKER_THREAD_PIN toggles, which means a loss of one worker time 2, which is expected as it has a lower priority then
worker timer 1. And thus in the moment when the hardware interrupt for the 2 worker is executet it has to wait for the
1 worker timer because of the XTIMER_ISR_BACKOFF and so the first timer is executed first as it has the higher priority.

24
tests/xtimer_hang/main.c
View File

@ -29,6 +29,11 @@
#include "thread.h"
#include "log.h"
#if defined(MAIN_THREAD_PIN) || defined(WORKER_THREAD_PIN)
#include "board.h"
#include "periph/gpio.h"
#endif
#define TEST_TIME_S (10LU)
#define TEST_INTERVAL_MS (100LU)
#define TEST_TIMER_STACKSIZE (THREAD_STACKSIZE_DEFAULT)
@ -38,14 +43,27 @@ char stack_timer2[TEST_TIMER_STACKSIZE];
void* timer_func(void* arg)
{
#if defined(WORKER_THREAD_PIN)
gpio_t worker_pin = WORKER_THREAD_PIN;
gpio_init(worker_pin, GPIO_OUT);
#endif
LOG_DEBUG("run thread %" PRIkernel_pid "\n", thread_getpid());
while(1) {
#if defined(WORKER_THREAD_PIN)
gpio_set(worker_pin);
gpio_clear(worker_pin);
#endif
xtimer_usleep(*(uint32_t *)(arg));
}
}
int main(void)
{
#if defined(MAIN_THREAD_PIN)
gpio_t main_pin = MAIN_THREAD_PIN;
gpio_init(main_pin, GPIO_OUT);
#endif
LOG_DEBUG("[INIT]\n");
uint32_t sleep_timer1 = 1000;
uint32_t sleep_timer2 = 1100;
@ -63,7 +81,13 @@ int main(void)
puts("[START]");
while((now = xtimer_now_usec()) < until) {
unsigned percent = (100 * (now - start)) / (until - start);
#if defined(MAIN_THREAD_PIN)
gpio_set(main_pin);
#endif
xtimer_usleep(TEST_INTERVAL_MS * US_PER_MS);
#if defined(MAIN_THREAD_PIN)
gpio_clear(main_pin);
#endif
printf("Testing (%3u%%)\n", percent);
}
puts("Testing (100%)");