cpufreq: Make policy min/max hard requirements
authorVincent Donnefort <vincent.donnefort@arm.com>
Wed, 8 Sep 2021 14:05:26 +0000 (15:05 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 18 Nov 2021 18:16:14 +0000 (19:16 +0100)
commit9053ab4dfc1b4e2b41f623f0efc69afeb38b0ec0
treea3bd6bf83ba5d11aec90d62266a6863b44726351
parente8e98aa12411604360c4da14b83753cba581b5ca
cpufreq: Make policy min/max hard requirements

[ Upstream commit 15171769069408789a72f9aa9a52cc931b839b56 ]

When applying the policy min/max limits, the requested frequency is
simply clamped to not be out of range. It means, however, if one of the
boundaries isn't an available frequency, the frequency resolution can
return a value out of those limits, depending on the relation used.

e.g. freq{0,1,2} being available frequencies.

          freq0  policy->min  freq1  policy->max   freq2
            |        |          |        |           |
          17kHz     18kHz     19kHz     20kHz      21kHz

     __resolve_freq(21kHz, CPUFREQ_RELATION_L) -> 21kHz (out of bounds)
     __resolve_freq(17kHz, CPUFREQ_RELATION_H) -> 17kHz (out of bounds)

If, during the policy init, we resolve the requested min/max to existing
frequencies, we ensure that any CPUFREQ_RELATION_* would resolve to a
frequency which is inside the policy min/max range.

Making the policy limits rigid helps to introduce the inefficient
frequencies support. Resolving an inefficient frequency to an efficient
one should not transgress policy->max (which can be set for thermal
reason) and having a value we can trust simplify this comparison.

Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/cpufreq/cpufreq.c