Re: [PATCH v4 4/4] cpufreq: Use arch specific feedback for cpuinfo_cur_freq

[Vanshidhar Konda]

On Fri, Apr 05, 2024 at 02:33:19PM +0100, Beata Michalska wrote:
> Some architectures provide a way to determine an average frequency over
> a certain period of time based on available performance monitors (AMU on
> ARM or APERF/MPERf on x86). With those at hand, enroll arch_freq_get_on_cpu
> into cpuinfo_cur_freq policy sysfs attribute handler, which is expected to
> represent the current frequency of a given CPU, as obtained by the hardware.
> This is the type of feedback that counters do provide.

--- snip ---

While testing this patch series on AmpereOne system, I simulated CPU
frequency throttling when the system is under power or thermal
constraints.

In this scenario, based on the user guilde, I expect scaling_cur_freq
is the frequency the kernel requests from the hardware; cpuinfo_cur_freq
is the actual frequency that the hardware is able to run at during the
power or thermal constraints.

The AmpereOne system I'm testing on has the following configuration:
- Max frequency is 3000000
- Support for AMU registers
- ACPI CPPC feedback counters use PCC register space
- Fedora 39 with 6.7.5 kernel
- Fedora 39 with 6.9.0-rc3 + this patch series

With 6.7.5 kernel:
Core        scaling_cur_freq        cpuinfo_cur_freq
----        ----------------        ----------------
0             3000000                 2593000
1             3000000                 2613000
2             3000000                 2625000
3             3000000                 2632000

With 6.9.0-rc3 + this patch series:
Core        scaling_cur_freq        cpuinfo_cur_freq
----        ----------------        ----------------
0             2671875                 2671875
1             2589632                 2589632
2             2648437                 2648437
3             2698242                 2698242

In the second case we can't identify that the CPU frequency is
being throttled by the hardware. I noticed this behavior with
or without this patch.

Thanks,
Vanshi