runtime/pprof: TestCPUProfileLabel is flaky on aix-ppc64

[bcmills]

2020-04-08T18:37:38-d5e1b7c/aix-ppc64
2020-03-17T01:24:51-7ec4adb/aix-ppc64
2019-12-20T20:12:18-b234fdb/aix-ppc64
2019-11-19T19:21:57-e0306c1/aix-ppc64
2019-09-26T17:37:02-a37f2b4/aix-ppc64

Not clear to me whether this is a builder issue, an AIX platform issue, or a timing-sensitive test.

CC @hyangah for pprof, @trex58 @Helflym for aix, @andybons @toothrot @cagedmantis @dmitshur for builders.

[Helflym]

I think that the same than #36084. The test seems to be stuck in a usleep called by svmon as shown by the samples taken.

184: 0x10002f304 (runtime.usleep:509) 0x10003fe33 (runtime.sysmon:4333) 0x1000373e3 (runtime.mstart1:1238) 0x1000372cb (runtime.mstart:1203)

It's not just the builder as I was able to get the same failure on my local machine.
Is it possible that a deadlock occurs between sysmon and pprof functions ?

[dmitshur]

This flaky TestCPUProfileLabel failure happened at least once also on linux/amd64:

--- FAIL: TestCPUProfileLabel (5.00s)
    pprof_test.go:514: total 4 CPU profile samples collected:
        3: 0x466db7 (runtime.madvise:539) 0x416f74 (runtime.sysUnused:111) 0x4211d3 (runtime.(*pageAlloc).scavengeRangeLocked:734) 0x420cf6 (runtime.(*pageAlloc).scavengeOne:628) 0x4205cd (runtime.(*pageAlloc).scavenge:419) 0x420504 (runtime.bgscavenge.func2:310) 0x462e88 (runtime.systemstack:477) 0x4202af (runtime.bgscavenge:298)
        
        1: 0x442f0f (runtime.runqget:5975) 0x43b430 (runtime.findrunnable:2728) 0x43cad8 (runtime.schedule:3361) 0x43d02c (runtime.park_m:3510) 0x462e02 (runtime.mcall:433)
        
    pprof_test.go:530: too few samples; got 4, want at least 125, ideally 500
    pprof_test.go:549: runtime/pprof.cpuHogger;key=value: 0
    pprof_test.go:552: no samples in expected functions
    pprof_test.go:562: runtime/pprof.cpuHogger;key=value has 0 samples out of 0, want at least 1, ideally 0

(Build log.)

From a slowbot run on CL 361294. It didn't happen on the first retry.