Re: [PATCH] cpufreq: acpi: Don't enable boost on policy exit

[Viresh Kumar]

On Thu, 17 Apr 2025 at 21:23, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> But the changelog isn't because the patch really doesn't address the
> issue at hand, which is most likely related to the platform firmware
> clearing the "boost disable" bit.

I think the patch and changelog were still correct as the driver was also
enabling the boost at exit(). So it fixes the problem, but not fully.

> Frankly, I'd rather get back to the state from before commit
> 2b16c631832d ("cpufreq: ACPI: Remove set_boost in
> acpi_cpufreq_cpu_init()") and start over with the knowledge that the
> platform firmware may scribble on the MSR before the kernel gets
> control back.
>
> On a way back from system suspend the MSR needs to be put back in sync
> with the boost settings in the kernel.

What about something like this instead ? Nicholas, can you give this a try
along with the $Subject patch (both patches should be applied) ?

diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
index 924314cdeebc..71557f2ac22a 100644
--- a/drivers/cpufreq/acpi-cpufreq.c
+++ b/drivers/cpufreq/acpi-cpufreq.c
@@ -909,8 +909,10 @@ static int acpi_cpufreq_cpu_init(struct
cpufreq_policy *policy)
        if (perf->states[0].core_frequency * 1000 != freq_table[0].frequency)
                pr_warn(FW_WARN "P-state 0 is not max freq\n");

-       if (acpi_cpufreq_driver.set_boost)
+       if (acpi_cpufreq_driver.set_boost) {
                policy->boost_supported = true;
+               policy->boost_enabled = boost_state(cpu);
+       }

        return result;