Isolation might be broken for SoCs due to different arch-timers

[viresh kumar]

Currently isolation scripts strictly depend on output from cat /proc/interrupts and search for a pattern with this string:

cat /proc/interrupts | grep arch_timer | grep 30 | sed 's/\s\+/ /g' | sed 's/^\s//g' | cut -d' ' -f$((2+$ISOL_CPU))

And this strictly works only if arch_timer is present, otherwise it might report wrong results as well.. Somebody needs to look into improving this part..

[Hongbo Zhang (hongbo-zhang)]

On Snoball with dual A9 cores, this works:
cat /proc/interrupts | grep twd | sed 's/\s\+/ /g' | sed 's/^\s//g' | cut -d' ' -f$((2+$ISOL_CPU))

[Hongbo Zhang (hongbo-zhang)]

I guess A9 uses twd, while A15 uses arch_timer, this needs to be verified.

[viresh kumar (viresh.kumar)]

Yes, ofcourse. I knew it. I am updating my scripts to get rid of this dependency.

[Mike Holmes (mike-holmes)]

New version up for review https://review.linaro.org/2414

[Mike Holmes (mike-holmes)]

This has been merged

[Mike Holmes (mike-holmes)]

This patch may now be correct but the result appears to be a reduction in the passing rate, The Arndale LE run on the 3rd had the patch and the isolation time is down to 72 seconds in this one case, this appears to be reflected on the other platforms as well.

https://validation.linaro.org/dashboard/image-charts/LNG-NO_HZ

[viresh kumar (viresh.kumar)]

Something surely happened with Arndale as it has started getting the timeout interruptions as the timer-counter overflows after 90 seconds.

For all other platforms: x86, amarilo, etc, the kernel version is incorrect. We must use 3.14 as the patches are present there.

--
viresh

[Anders Roxell (aroxell)]

x86 are running kernel v3.14.3, see: https://validation.linaro.org/scheduler/job/139763/log_file#L_5_13

[Mike Holmes (mike-holmes)]

X86 was apparently working prior to this patch, the failure after this patch is applied is ok.