[PATCH v3] drivers: psci: PSCI checker module

Kevin Brodsky kevin.brodsky at arm.com
Thu Oct 27 05:51:57 PDT 2016


On 27/10/16 10:13, Lorenzo Pieralisi wrote:
> On Wed, Oct 26, 2016 at 11:11:48AM -0700, Paul E. McKenney wrote:
>> On Wed, Oct 26, 2016 at 06:35:34PM +0100, Lorenzo Pieralisi wrote:
>>> On Wed, Oct 26, 2016 at 10:22:52AM -0700, Paul E. McKenney wrote:
>>>> On Wed, Oct 26, 2016 at 06:10:06PM +0100, Lorenzo Pieralisi wrote:
>> [ . . . ]
>>
>>>>> Thanks a lot for your feedback, thoughts appreciated.
>>>> Let me ask the question more directly.
>>>>
>>>> Why on earth are we trying to run these tests concurrently?
>>> We must prevent that, no question about that, that's why I started
>>> this discussion. It is not fine to enable this checker and the
>>> RCU/LOCK torture hotplug tests at the same time.
>>>
>>>> After all, if we just run one at a time in isolation, there is no
>>>> problem.
>>> Fine by me, it was to understand if the current assumptions we made
>>> are correct and they are definitely not. If we enable the PSCI checker
>>> we must disable the torture rcu/lock hotplug tests either statically or
>>> dynamically.
>> What rcutorture, locktorture, and rcuperf do is to invoke
>> torture_init_begin(), which returns false if one of these tests
>> is already running.
>>
>> Perhaps we should extract this torture-test-exclusion and require
>> than conflicting torture tests invoke it?
> Yes if it can be extracted as a check (but it should also prevent the
> torture tests from running and vice versa), either that or Kconfig
> dependency (which we could do as a first step, waiting to add the
> required interface to the torture test code ?).
>
> Thanks !
> Lorenzo

That sounds like a reasonable idea, but then that would mean that the PSCI checker
would have to wait until the torture test is finished if it is already running (and
the other way around).

I wasn't aware that torture tests were hotplugging CPUs. I think that the most
sensible thing to do right now is to make CONFIG_PSCI_CHECKER depend on
!CONFIG_TORTURE_TEST (or maybe specifically !CONFIG_RCU_TORTURE_TEST &&
!CONFIG_LOCK_TORTURE_TEST). We can try to make them work together afterwards, but for
the sake of getting this patch merged in a reasonable amount of time, I think we
should just exclude the conflicting tests at the build level in this patch. I'll also
update the comment accordingly.

Thanks,
Kevin
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.




More information about the linux-arm-kernel mailing list