Sporadic failures of rt/tst-cputimer1 on ppc64le (POWER8 hardware)?

[Tulio Magno Quites Machado Filho]

Carlos O'Donell <carlos@redhat.com> writes:

> [ text/plain ]
> On ppc64le I see a sporadic failure in rt/tst-cputimer1:
>
> clock_gettime returned timespec = { 0, 405994 } 
> clock_getres returned timespec = { 0, 1 } 
> *** timer sig1 invoked too soon: 0.400715218 instead of expected 0.400761014 
> *** timer sig2 invoked too soon: 0.500710982 instead of expected 0.500761014 
>
> The kernel says the clock resolution is 1 nanosecond, that seems exceedingly
> prcise.

I believe it has been that way since http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=597bc5c00b666fe123abb0af64f6e86f7ab72a90

>
> The kernel invokes the signal ~46 micorseconds too early, which is not 
> allowed by the tests. You can invoke late, but not early.
>
> Have either of you seen this before?

I noticed this for the first time earlier today on a Linux 3.13 box, but I
still wasn't able to reproduce it on another server with a newer kernel.

I suspect it might be related to another failure on rt/tst-cputimer3 which I
can reproduce only on the same server that rt/tst-cputimer1 failed today.

Notice that I'm able to reproduce rt/tst-cputimer3 on glibc 2.23, while
rt/tst-cputimer1 failed with 2.24.

I wonder if these failures are related to this:
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=6e998916dfe327e785e7c2447959b2c1a3ea4930

Where did you reproduce it?

-- 
Tulio Magno