[LTP] [PATCH v1] Correctly check setitimer params in setitimer01

[Li Wang]

Btw, these are the final changes I propose to fix the problem,
but I haven't figured out why the resolution from CLOCK_*_COARSE worked.

--- a/testcases/kernel/syscalls/setitimer/setitimer01.c
+++ b/testcases/kernel/syscalls/setitimer/setitimer01.c
@@ -76,7 +76,7 @@ static void verify_setitimer(unsigned int i)
                        ovalue->it_value.tv_sec,
                        ovalue->it_value.tv_usec);

-               if (ovalue->it_value.tv_sec != 0 ||
ovalue->it_value.tv_usec > usec)
+               if (ovalue->it_value.tv_sec != 0 ||
ovalue->it_value.tv_usec - time_step > usec)
                        tst_res(TFAIL, "Ending counters are out of range");

                for (;;)
@@ -95,11 +95,9 @@ static void setup(void)
 {
        struct timespec res;

-       SAFE_CLOCK_GETRES(CLOCK_MONOTONIC, &res);
+       SAFE_CLOCK_GETRES(CLOCK_MONOTONIC_COARSE, &res);

        time_step = res.tv_nsec / 1000;
-       if (time_step < 10000)
-               time_step = 10000;

        tst_res(TINFO, "clock resolution: %luns, time step: %luus",
                res.tv_nsec,


On Fri, Nov 4, 2022 at 2:48 PM Li Wang <liwang@redhat.com> wrote:

> Hi Andrea,
>
> Andrea Cervesato via ltp <ltp@lists.linux.it> wrote:
>
> Last test rewrite didn't consider the right expected boundaries when
>> setitimer syscall was tested. We also introduced counter times as
>> multiple of clock resolution, to avoid kernel rounding during setitimer
>> counter increase.
>>
>
> Good catch, but I'm afraid this can not solve the problem thoroughly.
> According failure log on "ITIMER_VIRTUAL/PROF" with running this patch.:
>
> setitimer01.c:64: TINFO: tc->which = ITIMER_VIRTUAL
> setitimer01.c:69: TPASS: sys_setitimer(tc->which, value, NULL) passed
> setitimer01.c:72: TPASS: sys_setitimer(tc->which, value, ovalue) passed
> setitimer01.c:76: TINFO: tv_sec=0, tv_usec=31000
> setitimer01.c:79: TFAIL: Ending counters are out of range
> setitimer01.c:88: TPASS: Child received signal: SIGVTALRM
>
> setitimer01.c:64: TINFO: tc->which = ITIMER_PROF
> setitimer01.c:69: TPASS: sys_setitimer(tc->which, value, NULL) passed
> setitimer01.c:72: TPASS: sys_setitimer(tc->which, value, ovalue) passed
> setitimer01.c:76: TINFO: tv_sec=0, tv_usec=31000
> setitimer01.c:79: TFAIL: Ending counters are out of range
> setitimer01.c:88: TPASS: Child received signal: SIGPROF
>
> It seems the tv_usec always increase 1000us, I highly suspect
> this increase comes from kernel function set_cpu_itimer() that
> explicitly add TICK_NSEC when 'nval' is larger than zero:
>
> $ cat kernel/time/itimer.c -n
>    ...
>    168 static void set_cpu_itimer(struct task_struct *tsk, unsigned int
> clock_id,
>    169                 const struct itimerspec64 *const value,
>    170                 struct itimerspec64 *const ovalue)
>    171 {
>                    ...
>    182          if (oval || nval) {
>    183                  if (nval > 0)
>    184                          nval += TICK_NSEC;
>    185                  set_process_cpu_timer(tsk, clock_id, &nval, &oval);
>    186          }
>    187          ...
>    198 }
>
> To verify my guess, I do a modification based on your patch and
> then easily get the result in pass.
>
> --- a/testcases/kernel/syscalls/setitimer/setitimer01.c
> +++ b/testcases/kernel/syscalls/setitimer/setitimer01.c
> @@ -76,7 +76,7 @@ static void verify_setitimer(unsigned int i)
>                         ovalue->it_value.tv_sec,
>                         ovalue->it_value.tv_usec);
>
> -               if (ovalue->it_value.tv_sec != 0 ||
> ovalue->it_value.tv_usec > usec)
> +               if (ovalue->it_value.tv_sec != 0 ||
> ovalue->it_value.tv_usec - time_step > usec)
>                         tst_res(TFAIL, "Ending counters are out of range");
>
>                 for (;;)
> @@ -98,8 +98,8 @@ static void setup(void)
>         SAFE_CLOCK_GETRES(CLOCK_MONOTONIC, &res);
>
>         time_step = res.tv_nsec / 1000;
> -       if (time_step < 10000)
> -               time_step = 10000;
> +       if (time_step < 1000)
> +               time_step = 1000;
>
>         tst_res(TINFO, "clock resolution: %luns, time step: %luus",
>                 res.tv_nsec,
>
>
> But after trying this with my RasberryPi4, it fails again with an increase
> 4000us each time. So it might related to the system use different time
> resolutions. When I shift to use `CLOCK_MONOTONIC_COARSE`
> then test gets passed on all my platforms.
>
> Any comments?
>
>
> ----------CLOCK_MONOTONIC-------------
> setitimer01.c:65: TINFO: tc->which = ITIMER_VIRTUAL
> setitimer01.c:70: TPASS: sys_setitimer(tc->which, value, NULL) passed
> setitimer01.c:73: TPASS: sys_setitimer(tc->which, value, ovalue) passed
> setitimer01.c:77: TINFO: tv_sec=0, tv_usec=7000
> setitimer01.c:80: TFAIL: Ending counters are out of range
>
> -----------CLOCK_MONOTONIC_COARSE-------
> setitimer01.c:65: TINFO: tc->which = ITIMER_VIRTUAL
> setitimer01.c:70: TPASS: sys_setitimer(tc->which, value, NULL) passed
> setitimer01.c:73: TPASS: sys_setitimer(tc->which, value, ovalue) passed
> setitimer01.c:77: TINFO: tv_sec=0, tv_usec=16000
> setitimer01.c:89: TPASS: Child received signal: SIGVTALRM
>
> # lscpu
> Architecture:        aarch64
> Byte Order:          Little Endian
> CPU(s):              4
> On-line CPU(s) list: 0-3
> Thread(s) per core:  1
> Core(s) per cluster: 4
> Socket(s):           -
> Cluster(s):          1
> Vendor ID:           ARM
> Model:               3
> Model name:          Cortex-A72
> Stepping:            r0p3
> CPU max MHz:         1500.0000
> CPU min MHz:         600.0000
> BogoMIPS:            108.00
> Flags:               fp asimd evtstrm crc32 cpuid
>
> --
> Regards,
> Li Wang
>


-- 
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20221104/48c8dd3e/attachment-0001.htm>