[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAJZ5v0jenM_pYUkTv-qPV21tok15R+KfT497itPO=fLUywDKqw@mail.gmail.com>
Date: Wed, 23 Apr 2025 16:59:30 +0200
From: "Rafael J. Wysocki" <rafael@...nel.org>
To: Viresh Kumar <viresh.kumar@...aro.org>
Cc: Lifeng Zheng <zhenglifeng1@...wei.com>, linux-pm@...r.kernel.org,
Vincent Guittot <vincent.guittot@...aro.org>, Nicholas Chin <nic.c3.14@...il.com>,
"Rafael J. Wysocki" <rafael.j.wysocki@...el.com>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/6] cpufreq: acpi: Re-sync CPU boost state on system resume
On Wed, Apr 23, 2025 at 4:40 PM Rafael J. Wysocki <rafael@...nel.org> wrote:
>
> On Wed, Apr 23, 2025 at 4:26 PM Rafael J. Wysocki <rafael@...nel.org> wrote:
> >
> > On Tue, Apr 22, 2025 at 11:54 AM Viresh Kumar <viresh.kumar@...aro.org> wrote:
> > >
> > > During suspend/resume cycles, platform firmware may alter the CPU boost
> > > state.
> > >
> > > If boost is disabled before suspend, it correctly remains off after
> > > resume. However, if firmware re-enables boost during suspend, the system
> > > may resume with boost frequencies enabled—even when the boost flag was
> > > originally disabled. This violates expected behavior.
> > >
> > > Ensure the boost state is re-synchronized with the kernel policy during
> > > system resume to maintain consistency.
> > >
> > > Fixes: 2b16c631832d ("cpufreq: ACPI: Remove set_boost in acpi_cpufreq_cpu_init()")
> > > Reported-by: Nicholas Chin <nic.c3.14@...il.com>
> > > Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220013
> > > Signed-off-by: Viresh Kumar <viresh.kumar@...aro.org>
> > > ---
> > > drivers/cpufreq/acpi-cpufreq.c | 15 +++++++++++++--
> > > 1 file changed, 13 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
> > > index 7002e8de8098..0ffabf740ff5 100644
> > > --- a/drivers/cpufreq/acpi-cpufreq.c
> > > +++ b/drivers/cpufreq/acpi-cpufreq.c
> > > @@ -893,8 +893,19 @@ 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)
> > > - policy->boost_supported = true;
> > > + if (acpi_cpufreq_driver.set_boost) {
> > > + if (policy->boost_supported) {
> > > + /*
> > > + * The firmware may have altered boost state while the
> > > + * CPU was offline (for example during a suspend-resume
> > > + * cycle).
> > > + */
> > > + if (policy->boost_enabled != boost_state(cpu))
> > > + set_boost(policy, policy->boost_enabled);
> > > + } else {
> > > + policy->boost_supported = true;
> >
> > IIUC policy->boost_enabled is false at this point, so say that
> > boost_state(cpu) returns true and say cpufreq_boost_enabled() returns
> > false.
>
> This cannot happen for CPU 0 because of acpi_cpufreq_boost_init() ->
>
> > cpufreq_online() will see policy->boost_enabled ==
> > cpufreq_boost_enabled(), so it won't do anything regarding boost, and
> > say that this happens for all online CPUs.
>
> -> so if boost_state(0) returns true, policy->boost_enabled will be
> set for all policies to start with due to the code in
> cpufreq_online(), but this is far from obvious.
>
> I would at least say in the changelog that set_boost() need not be
> called directly at the policy initialization time because of the
> above.
I also think that acpi_cpufreq_resume() may be a better place for
re-syncing the boost state with policy->boost_enabled because it may
do that for CPU 0 as well as for the non-boot CPUs.
Powered by blists - more mailing lists