On Thu, Jul 21, 2016 at 04:30:03PM -0700, Viresh Kumar wrote:
> On 21-07-16, 16:21, Steve Muckle wrote:
> > On Thu, Jul 21, 2016 at 01:30:41PM -0700, Viresh Kumar wrote:
> > > Okay, but in that case shouldn't we do something like this:
> > >
> > > unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
> > > unsigned int target_freq)
> > > {
> > > target_freq = clamp_val(target_freq, policy->min, policy->max);
> > > policy->cached_target_freq = target_freq;
> > >
> > > if (cpufreq_driver->target_index) {
> > > policy->cached_resolved_idx =
> > > cpufreq_frequency_table_target(policy,
> > > target_freq,
> > >
> > > CPUFREQ_RELATION_L);
> > > return
> > > policy->freq_table[policy->cached_resolved_idx].frequency;
> > > }
> > >
> > > if (cpufreq_driver->resolve_freq)
> > > return cpufreq_driver->resolve_freq(policy, target_freq);
> > > }
> >
> > Thanks for the review.
> >
> > My thinking (noted in the commit text) was that the caller of
> > cpufreq_driver_resolve_freq() would verify that the driver supported the
> > proper calls before using this API.
>
> Okay, but the caller isn't doing that today. Right?
There is no caller yet.
> > This way it can be checked once,
> > presumably in a governor's init routine. Checking the pointer over and
> > over again in a fast path is wasteful.
>
> But we just can not assume the callers to always check that the driver
> has a ->target() and no ->resolve_freq(), and in that case not to call
> this routine. We would be forced to add a WARN_ON() in that case here
> to make sure we aren't trying to access a NULL ->resolve_freq.
Why not? Can we not catch that in code review?
If somehow this slips past and someone tries to use a driver with
schedutil that doesn't provide either target_index or resolve_freq, it's
not like it'll be a rare crash. It'll die immediately and in a very
obvious way.
> Over that, it will be used for a very small number of drivers which
> still use the ->target() callback and anyway we are going to do a
> function call for them. We can add a likely() here if that helps, but
> some sort of checking is surely required IMO.
>
> And, this is a core API, which can be used for other governor's
> tomorrow :)
As another alternative, this could be caught in cpufreq driver
initialization? I believe you suggested that originally, but I avoided
it as I didn't want to have to implement resolve_freq() for every
target() style driver. It sounds like there aren't many though.
