[PATCH v3 2/5] cpufreq: move invariance setter calls in cpufreq core

[Viresh Kumar]

On 24-08-20, 22:02, Ionela Voinescu wrote:
> To properly scale its per-entity load-tracking signals, the task scheduler
> needs to be given a frequency scale factor, i.e. some image of the current
> frequency the CPU is running at. Currently, this scale can be computed
> either by using counters (APERF/MPERF on x86, AMU on arm64), or by
> piggy-backing on the frequency selection done by cpufreq.
> 
> For the latter, drivers have to explicitly set the scale factor
> themselves, despite it being purely boiler-plate code: the required
> information depends entirely on the kind of frequency switch callback
> implemented by the driver, i.e. either of: target_index(), target(),
> fast_switch() and setpolicy().
> 
> The fitness of those callbacks with regard to driving the Frequency
> Invariance Engine (FIE) is studied below:
> 
> target_index()
> ==============
> Documentation states that the chosen frequency "must be determined by
> freq_table[index].frequency". It isn't clear if it *has* to be that
> frequency, or if it can use that frequency value to do some computation
> that ultimately leads to a different frequency selection. All drivers
> go for the former, while the vexpress-spc-cpufreq has an atypical
> implementation which is handled separately.
> 
> Therefore, the hook works on the assumption the core can use
> freq_table[index].frequency.
> 
> target()
> =======
> This has been flagged as deprecated since:
> 
>   commit 9c0ebcf78fde ("cpufreq: Implement light weight ->target_index() routine")
> 
> It also doesn't have that many users:
> 
>   gx-suspmod.c:439:       .target = cpufreq_gx_target,
>   s3c24xx-cpufreq.c:428:  .target = s3c_cpufreq_target,
>   intel_pstate.c:2528:    .target = intel_cpufreq_target,
>   cppc_cpufreq.c:401:     .target = cppc_cpufreq_set_target,
>   cpufreq-nforce2.c:371:  .target = nforce2_target,
>   sh-cpufreq.c:163:       .target = sh_cpufreq_target,
>   pcc-cpufreq.c:573:      .target = pcc_cpufreq_target,
> 
> Similarly to the path taken for target_index() calls in the cpufreq core
> during a frequency change, all of the drivers above will mark the end of a
> frequency change by a call to cpufreq_freq_transition_end().
> 
> Therefore, cpufreq_freq_transition_end() can be used as the location for
> the arch_set_freq_scale() call to potentially inform the scheduler of the
> frequency change.
> 
> This change maintains the previous functionality for the drivers that
> implement the target_index() callback, while also adding support for the
> few drivers that implement the deprecated target() callback.
> 
> fast_switch()
> =============
> This callback *has* to return the frequency that was selected.
> 
> setpolicy()
> ===========
> This callback does not have any designated way of informing what was the
> end choice. But there are only two drivers using setpolicy(), and none
> of them have current FIE support:
> 
>   drivers/cpufreq/longrun.c:281:	.setpolicy	= longrun_set_policy,
>   drivers/cpufreq/intel_pstate.c:2215:	.setpolicy	= intel_pstate_set_policy,
> 
> The intel_pstate is known to use counter-driven frequency invariance.
> 
> Conclusion
> ==========
> 
> Given that the significant majority of current FIE enabled drivers use
> callbacks that lend themselves to triggering the setting of the FIE scale
> factor in a generic way, move the invariance setter calls to cpufreq core.
> 
> As a result of setting the frequency scale factor in cpufreq core, after
> callbacks that lend themselves to trigger it, remove this functionality
> from the driver side.
> 
> To be noted that despite marking a successful frequency change, many
> cpufreq drivers will consider the new frequency as the requested
> frequency, although this is might not be the one granted by the hardware.
> 
> Therefore, the call to arch_set_freq_scale() is a "best effort" one, and
> it is up to the architecture if the new frequency is used in the new
> frequency scale factor setting (determined by the implementation of
> arch_set_freq_scale()) or eventually used by the scheduler (determined
> by the implementation of arch_scale_freq_capacity()). The architecture
> is in a better position to decide if it has better methods to obtain
> more accurate information regarding the current frequency and use that
> information instead (for example, the use of counters).
> 
> Also, the implementation to arch_set_freq_scale() will now have to handle
> error conditions (current frequency == 0) in order to prevent the
> overhead in cpufreq core when the default arch_set_freq_scale()
> implementation is used.
> 
> Signed-off-by: Ionela Voinescu <ionela.voinescu at arm.com>
> Suggested-by: Valentin Schneider <valentin.schneider at arm.com>
> Cc: Rafael J. Wysocki <rjw at rjwysocki.net>
> Cc: Viresh Kumar <viresh.kumar at linaro.org>
> ---
>  drivers/cpufreq/cpufreq-dt.c           | 10 +---------
>  drivers/cpufreq/cpufreq.c              | 12 +++++++++++-
>  drivers/cpufreq/qcom-cpufreq-hw.c      |  9 +--------
>  drivers/cpufreq/scmi-cpufreq.c         | 12 ++----------
>  drivers/cpufreq/scpi-cpufreq.c         |  6 +-----
>  drivers/cpufreq/vexpress-spc-cpufreq.c | 12 ++----------
>  6 files changed, 18 insertions(+), 43 deletions(-)

Acked-by: Viresh Kumar <viresh.kumar at linaro.org>

-- 
viresh