Hello,
On 29/11/2018 17:57:05-0200, Rafael David Tinoco wrote:
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@bootlin.com
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.
I would think that enabling HR timers would actually make things better, not matter how many CPUs are in the system. At least, this was the case for AT91.
Would you mind if we change the default for up to 32Hz ? Or, do you have any other suggestion ?
I must admit that the whole point of moving this test out of rtctest was that I didn't want to maintain it. John is the one responsible for the timers so he will have to take that decision.