Re: [PATCH 1/2] ARM: cpuidle: fix !cpuidle_ops[cpu].init case during init

[Jisheng Zhang]

Hi Daniel,

On Wed, 30 Mar 2016 11:31:39 +0200 Daniel Lezcano wrote:

> On 03/30/2016 10:43 AM, Jisheng Zhang wrote:
> > On Wed, 30 Mar 2016 10:41:09 +0200 Daniel Lezcano wrote:
> >  
> >> On 03/30/2016 10:17 AM, Jisheng Zhang wrote:  
> >>> On Wed, 30 Mar 2016 10:09:12 +0200 Daniel Lezcano wrote:
> >>>  
> >>>> On 03/30/2016 09:16 AM, Jisheng Zhang wrote:  
> >>>>> Hi Daniel,  
> >>>>
> >>>> [ ... ]
> >>>>
> >>>> Added Lorenzo and Catalin.
> >>>>  
> >>>>>> Hi Jisheng,
> >>>>>>
> >>>>>> this should be handled in the arm_cpuidle_read_ops function.
> >>>>>>  
> >>>>>
> >>>>> Thanks for reviewing. After some consideration, I think this patch isn't correct
> >>>>> There may be platforms which doesn't need the init member at all, although
> >>>>> currently I don't see such platforms in mainline, So I'll drop this patch
> >>>>> and send out one v2 only does the optimization.  
> >>>>
> >>>> There is an inconsistency between ARM and ARM64. The 'cpu_get_ops', the
> >>>> arm_cpuidle_read_ops from the ARM64 side, returns -EOPNOTSUPP when the
> >>>> init function is not there for cpuidle.  
> >>>
> >>> yes.
> >>> arm64's arm_cpuidle_init() returns -EOPNOTSUPP if init callback isn't defined
> >>>  
> >>>>
> >>>> I don't think it is a problem, but as ARM/ARM64 are sharing the same
> >>>> cpuidle-arm.c driver it would make sense to unify the behavior between
> >>>> both archs.  
> >>>
> >>> yes, agree with you. From "unify" point of view, could I move back the suspend
> >>> callback check and init callback check into arm_cpuidle_init() for arm as V1 does?  
> >>
> >> Why ? To be consistent with ARM64 ?  
> >
> > Yes, that's my intention.  
> 
> Well, I don't have a strong opinion on that. ARM64 cpu_ops is slightly 
> different from cpuidle_ops as the cpu boot / hotplug operations are 
> placed in a different place and that explains why on ARM64 we can have 
> an successful 'get_ops' because we use the partially filled structure. 
> On ARM, it is cpuidle_ops only, so we can gracefully fail if the ops are 
> not defined.
> 
> IMO, it still make sense to keep the checks in arm_cpuidle_read_ops for ARM.
> 

Got your points. I'll send a v3 to add init check. These checks will be
in arm_cpuidle_read_ops.

Thanks,
Jisheng