runtime/pprof: TestCPUProfileMultithreaded is flaky on ppc64le

[mwhudson]

For example from http://build.golang.org/log/a58e91460d1b6bec7380bd3f9f86f57483aa2e22:

--- FAIL: TestCPUProfileMultithreaded (0.25s)
    pprof_test.go:165: total 12 CPU profile samples collected
    pprof_test.go:181: runtime/pprof_test.cpuHog1: 0
    pprof_test.go:181: runtime/pprof_test.cpuHog2: 12
    pprof_test.go:195: runtime/pprof_test.cpuHog1 has 0 samples out of 12, want at least 1, ideally 6
FAIL

Very old neurons recall that the timer on powerpc has pretty low resolution, but I don't know if that is true or the problem here.

[aclements]

@davecheney, can you provide the kernel config for the linux-ppc64le-canonical builder? In particular, if CONFIG_HIGH_RES_TIMERS is not set, this is probably the same as #13405.

[aclements]

I'm guessing it's this one: http://kernel.ubuntu.com/~kernel-ppa/configs/trusty/ppc64el-config.flavour.generic, which does have CONFIG_HIGH_RES_TIMERS set.

[rsc]

@aclements, what's the magic /proc file to look in to find out timer resolution?

[aclements]

grep resolution: /proc/timer_list

[rsc]

FWIW I got lots of '1 nsecs' with that grep on a ppc64le machine (but not the Canonical builder). So at least some hardware does have good timers.

[davecheney]

@rsc from the canonical ppc64le machine

winton-01(~) % grep resolution: /proc/timer_list
  .resolution: 1 nsecs
  .resolution: 1 nsecs
  .resolution: 1 nsecs
  .resolution: 1 nsecs
  .resolution: 1 nsecs
  .resolution: 1 nsecs
  .resolution: 1 nsecs
  .resolution: 1 nsecs

[gopherbot]

CL https://golang.org/cl/18683 mentions this issue.