kern/55229: cond_timedwait_race now fails randomly on real hardware

[gson%gson.org@localhost (Andreas Gustafsson)]

>Number:         55229
>Category:       kern
>Synopsis:       cond_timedwait_race now fails randomly on real hardware
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun May 03 09:55:00 +0000 2020
>Originator:     Andreas Gustafsson
>Release:        NetBSD-current
>Organization:
>Environment:
System: NetBSD
Architecture: x86_64
Machine: amd64
>Description:

The lib/libpthread/t_cond:cond_timedwait_race test case has started
randomly failing on my bare metal amd64 testbed, where it used to
reliably pass.  There were no failures in 2019, and none in 2020 until
late April, when the test started to fail in about half the runs:

   $ zgrep -a cond_timedwait_race: 2020/*/test.log.gz
   [...]
   2020/2020.04.23.09.01.33/test.log.gz:    cond_timedwait_race: [0.995759s] Passed.
   2020/2020.04.23.09.16.21/test.log.gz:    cond_timedwait_race: [0.993798s] Passed.
   2020/2020.04.23.09.40.08/test.log.gz:    cond_timedwait_race: [1.015245s] Passed.
   2020/2020.04.23.20.38.33/test.log.gz:    cond_timedwait_race: [1.018188s] Passed.
   2020/2020.04.24.19.47.03/test.log.gz:    cond_timedwait_race: [300.153806s] Failed: Test case timed out after 300 seconds
   2020/2020.04.25.20.05.25/test.log.gz:    cond_timedwait_race: [1.331791s] Passed.
   2020/2020.04.26.16.16.13/test.log.gz:    cond_timedwait_race: [1.388661s] Passed.
   2020/2020.04.26.20.41.30/test.log.gz:    cond_timedwait_race: [300.133543s] Failed: Test case timed out after 300 seconds
   2020/2020.04.27.20.46.01/test.log.gz:    cond_timedwait_race: [300.164854s] Failed: Test case timed out after 300 seconds
   2020/2020.04.28.19.45.58/test.log.gz:    cond_timedwait_race: [300.142224s] Failed: Test case timed out after 300 seconds
   2020/2020.04.29.21.00.42/test.log.gz:    cond_timedwait_race: [1.273650s] Passed.
   2020/2020.04.30.15.12.25/test.log.gz:    cond_timedwait_race: [1.413426s] Passed.
   2020/2020.05.01.01.31.17/test.log.gz:    cond_timedwait_race: [300.163612s] Failed: Test case timed out after 300 seconds
   2020/2020.05.01.08.32.50/test.log.gz:    cond_timedwait_race: [300.124285s] Failed: Test case timed out after 300 seconds
   2020/2020.05.01.08.45.01/test.log.gz:    cond_timedwait_race: [1.371221s] Passed.
   2020/2020.05.01.10.38.59/test.log.gz:    cond_timedwait_race: [1.447664s] Passed.
   2020/2020.05.01.15.52.38/test.log.gz:    cond_timedwait_race: [300.126774s] Failed: Test case timed out after 300 seconds
   2020/2020.05.01.16.29.34/test.log.gz:    cond_timedwait_race: [300.127535s] Failed: Test case timed out after 300 seconds
   2020/2020.05.01.19.59.47/test.log.gz:    cond_timedwait_race: [1.356858s] Passed.
   2020/2020.05.02.19.46.58/test.log.gz:    cond_timedwait_race: [300.154863s] Failed: Test case timed out after 300 seconds

The qemu-based TNF i386 testbed does not appear to be affected.

>How-To-Repeat:

Run the ATF tests on real hardware.

>Fix: