Re: [PATCH] s390/pai: Simplify pai_init() and always return -ENODEV on failure

[Thomas Richter <tmricht@linux.ibm.com>]

On 9/21/26 15:01, Heiko Carstens wrote:
> On Mon, Sep 21, 2026 at 01:40:21PM +0200, Thomas Richter wrote:
>> On 9/14/26 14:14, Heiko Carstens wrote:
>>> On Thu, Sep 03, 2026 at 10:15:57AM +0200, Thomas Richter wrote:
>>>> Code simplification and remove variable state.
>>>> Return -ENODEV in all failures to indicate the PAI PMUs are not
>>>> available.
>>>
>>> Why do you want to hide different error / return codes? That makes
>>> debugging in case of an error more difficult.
>>
>> Well, hiding was not the intent.
>>
>> The kernel common code does a lot of checks at event creation,
>> then calls the PMU event_init() call back to install the event.
>> When this fails the return code is checked and further action
>> depends on this return code. See
>>
>>   perf_init_event()
>>   +--> perf_try_event_init()
>>        +--> PMU->event_init()
>>
>> See include/linux/perf_event.h for return code meanings.
>>
>> When we return the value from cpuhp_setup_state(), it must no be
>> -ENOENT and some other values. Because those are used for postprocessing.
>>
>> Thats why I always return -ENODEV, meaning event is valid for PMU, but
>> PMU not operational. I thouhgt this might fit best.
> 
> But what does have pai_init() to do with anything of the above?
> pai_init() is called via init_call() mechanism.


Ahh well, that is correct... lets drop it. Sorry for the noise.

-- 
Thomas Richter, Dept 3303, IBM s390 Linux Development, Boeblingen, Germany
--
IBM Deutschland Research & Development GmbH

Vorsitzender des Aufsichtsrats: Wolfgang Wendt

Geschäftsführung: David Faller

Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294