runtime: deadlock in TestWeakToStrongMarkTermination

[gopherbot]

#!watchflakes
default <- pkg ~ `runtime` && test == "TestWeakToStrongMarkTermination"

Issue created automatically to collect these failures.

Example (log):

=== RUN   TestAbortInCgo
=== PAUSE TestAbortInCgo

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "runtime:cpu4" && test == "TestAbortInCgo"

2024-11-23 00:14 gotip-linux-arm64-longtest go@8fb6a469 runtime:cpu4.TestAbortInCgo [ABORT] (log)

=== RUN   TestAbortInCgo
=== PAUSE TestAbortInCgo

— watchflakes

[mknyszek]

This is actually a failure in a new test, TestWeakToStrongMarkTermination. This appears to be a test-specific failure. As far as I can tell, the goroutine that is tasked with letting gcMarkDone proceed is the goroutine that calls gcMarkDone due to the timer allocation in time.Sleep. Whoops. I guess the sleep has to be lower-level than that, maybe a usleep?

[mknyszek]

This failure mode is incredibly rare, so if this happened during the development cycle, I don't think it would have a particularly high priority. But we should still fix it soon, since it was introduced in this release.

[prattmic]

Another case occurred in https://ci.chromium.org/b/8730523567768266401.

[gopherbot]

Change https://go.dev/cl/636155 mentions this issue: runtime: usleep in TestWeakToStrongMarkTermination