On 4/19/18 9:50 AM, Alexandre Belloni wrote:
> Since commit 6610e0893b8bc ("RTC: Rework RTC code to use timerqueue for
> events"), PIE are completely handled using hrtimers, without actually using
> any underlying hardware RTC.
>
> Move PIE testing out of rtctest. It still depends on the presence of an RTC
> (to access the device file) but doesn't depend on it actually working.
>
> Signed-off-by: Alexandre Belloni <alexandre.belloni@xxxxxxxxxxx>
> ---
> tools/testing/selftests/timers/.gitignore | 1 +
> tools/testing/selftests/timers/Makefile | 2 +-
> tools/testing/selftests/timers/rtcpie.c | 132 ++++++++++++++++++++++
> tools/testing/selftests/timers/rtctest.c | 83 +-------------
> 4 files changed, 138 insertions(+), 80 deletions(-)
> create mode 100644 tools/testing/selftests/timers/rtcpie.c
>
...
> + /* The frequencies 128Hz, 256Hz, ... 8192Hz are only allowed for root. */
> + for (tmp=2; tmp<=64; tmp*=2) {
> +
> + retval = ioctl(fd, RTC_IRQP_SET, tmp);
> + if (retval == -1) {
> + /* not all RTCs can change their periodic IRQ rate */
> + if (errno == EINVAL) {
> + fprintf(stderr,
> + "\n...Periodic IRQ rate is fixed\n");
> + goto done;
> + }
> + perror("RTC_IRQP_SET ioctl");
> + exit(errno);
> + }
Hello Alexandre,
In our tests, having failures under 64Hz is quite common in embedded
systems with few number of CPUs/Cores:
--------
root@bug3402:opt$ find /sys -name rtc0
/sys/devices/platform/9010000.pl031/rtc/rtc0
/sys/class/rtc/rtc0
selftests: timers: rtcpie
Periodic IRQ rate is 1Hz.
Counting 20 interrupts at:
2Hz: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
4Hz: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
8Hz: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
16Hz: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
32Hz: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
64Hz:
PIE delta error: 0.017697 should be close to 0.015625
not ok 1..9 selftests: timers: rtcpie [FAIL]
--------
Mainly because 64Hz gives us.. ~16ms in the test, and the variation
being taken in consideration for an error, for this test, is 10%, which
is ~1.6ms... pretty close to scheduler limit for lower number of CPUs in
a functional testing environment.
Would you mind if we change the default for up to 32Hz ? Or, do you have
any other suggestion ?
Best Rgds
Rafael
--
Rafael D. Tinoco
Linaro - Kernel Validation
