port-evbarm/53308: evbarm-earmv7hf performance regression under qemu

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

>Number:         53308
>Category:       port-evbarm
>Synopsis:       evbarm-earmv7hf performance regression under qemu
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    port-evbarm-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed May 23 13:10:01 +0000 2018
>Originator:     Andreas Gustafsson
>Release:        NetBSD-current, source date >= 2018.05.14.17.15.54
>Organization:
>Environment:
System: NetBSD
Architecture: arm
Machine: evbarm
>Description:

The evbarm-earmv7hf ATF tests on the TNF testbed have typically
completed in about 24 hours (unless they hit PR 53303), but their
performance recently worsened such that they no longer complete within
the testbed timeout of 40 hours.

The problem started with the following commits:

  2018.05.14.17.15.54 joerg src/sys/arch/aarch64/include/armreg.h 1.10
  2018.05.14.17.15.54 joerg src/sys/arch/arm/cortex/gtmr.c 1.27
  2018.05.14.17.15.54 joerg src/sys/arch/arm/include/armreg.h 1.121

Perhaps qemu is slow enough that each gtmr read returns a different
value?

>How-To-Repeat:

Inspect the test logs at

  http://releng.netbsd.org/b5reports/evbarm-earmv7hf/commits-2018.05.html#2018.05.14.17.15.54

>Fix:

-- 
Andreas Gustafsson, gson%gson.org@localhost