[Intel-gfx] [igt-dev] [PATCH i-g-t] tests/perf_pmu: Handle CPU hotplug failures better

Tomi Sarvela tomi.p.sarvela at intel.com
Mon Feb 26 10:14:08 UTC 2018


On 02/26/2018 12:03 PM, Petri Latvala wrote:
> On Fri, Feb 23, 2018 at 02:20:59PM +0000, Tvrtko Ursulin wrote:
>>
>> On 23/02/2018 11:58, Petri Latvala wrote:
>>> We have IGT_HANG and IGT_HANG_WITHOUT_RESET so the users can opt
>>> in/out of some fatal behaviour already. I'm fine with auto-rebooting,
>>> even as the default, if users can opt out of it with
>>> IGT_NO_REBOOT_PRETTY_PLEASE or so.
>>
>> I am fine with something like that. Just lets define how to call the env
>> variable and what the default should be?
> 
> 
> IGT_REBOOT_ON_FATAL_ERROR=1, default behaviour is not to reboot?

Default behaviour is not to reboot on panic either, but we want to have 
that.

So something that adds forced panic / reboot when test spectacularly 
fails is probably the sanest default: only if we have asked for that 
behaviour.

Turning that on with ENV or piglit.conf entry are both ok, either way 
it's something that user needs to know and explicitly turn on.

Tomi

> 
> Tomi, preference?
> 
> 
> 
>> Do we have a return code from a test which stops the test runner?
>>
>> I am thinking that the best approach would be not to reboot but to halt
>> testing, unless this environment option is set.
> 
> 
> Yeah, that would be the best out of all the options. Unfortunately
> that's not going to materialize in the near future. I have an
> implementation of aborting support hacked up on my test machines, but
> it's quite ugly and requires hacking deeply into the bowels of piglit.
> 
> Piglit is a hammer and we have screws...
> 
> 
> 


Tomi
-- 
Intel Finland Oy - BIC 0357606-4 - Westendinkatu 7, 02160 Espoo


More information about the Intel-gfx mailing list