Re: [PATCH][RFC] ACPI: add "processor.broadcast_ppc" hook to broadcast _PPC to all online CPUs

From: Rafael J. Wysocki
Date: Thu Feb 14 2019 - 05:21:28 EST


On Wed, Feb 13, 2019 at 5:44 PM Yu Chen <yu.c.chen@xxxxxxxxx> wrote:
>
> Hi Rafael,
> On Mon, Feb 11, 2019 at 11:41:26AM +0100, Rafael J. Wysocki wrote:
> > On Sat, Feb 9, 2019 at 12:54 PM Chen Yu <yu.c.chen@xxxxxxxxx> wrote:
> > >
> > > On Dell Inc. XPS13 9333, the BIOS changes the value of
> > > MSR_IA32_MISC_ENABLE_TURBO_DISABLE at runtime (e.g., when
> > > the power source changes), the maximum frequency of the
> > > CPU is not updated accordingly. This is due to the policy's
> > > cpuinfo.max is not updated when _PPC notifier fires.
> > >
> > > Fix this problem by updating the policy's cpuinfo.max
> > > and broadcast the _PPC notifier to all online CPUs.
> > >
> > > Link: https://bugzilla.kernel.org/show_bug.cgi?id=200759
> > > Reported-and-tested-by: Gabriele Mazzotta <gabriele.mzt@xxxxxxxxx>
> > > Originally-by: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
> > > Signed-off-by: Chen Yu <yu.c.chen@xxxxxxxxx>
> > > ---
> > > drivers/acpi/processor_perflib.c | 16 ++++++++++++++--
> > > drivers/cpufreq/cpufreq.c | 2 ++
> > > drivers/cpufreq/intel_pstate.c | 15 ++++++++++++++-
> > > 3 files changed, 30 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/acpi/processor_perflib.c b/drivers/acpi/processor_perflib.c
> > > index a303fd0e108c..737dbf5aa7f7 100644
> > > --- a/drivers/acpi/processor_perflib.c
> > > +++ b/drivers/acpi/processor_perflib.c
> > > @@ -63,6 +63,10 @@ module_param(ignore_ppc, int, 0644);
> > > MODULE_PARM_DESC(ignore_ppc, "If the frequency of your machine gets wrongly" \
> > > "limited by BIOS, this should help");
> > >
> > > +static int broadcast_ppc;
> > > +module_param(broadcast_ppc, int, 0644);
> > > +MODULE_PARM_DESC(broadcast_ppc, "Broadcast the ppc to all online CPUs");
> > > +
> > > #define PPC_REGISTERED 1
> > > #define PPC_IN_USE 2
> > >
> > > @@ -180,8 +184,16 @@ void acpi_processor_ppc_has_changed(struct acpi_processor *pr, int event_flag)
> > > else
> > > acpi_processor_ppc_ost(pr->handle, 0);
> > > }
> > > - if (ret >= 0)
> > > - cpufreq_update_policy(pr->id);
> > > + if (ret >= 0) {
> > > + if (broadcast_ppc) {
> > > + int cpu;
> > > +
> > > + for_each_possible_cpu(cpu)
> > > + cpufreq_update_policy(cpu);
> > > + } else {
> > > + cpufreq_update_policy(pr->id);
> > > + }
> > > + }
> > > }
> > >
> > > int acpi_processor_get_bios_limit(int cpu, unsigned int *limit)
> > > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> > > index e35a886e00bc..95e08816b512 100644
> > > --- a/drivers/cpufreq/cpufreq.c
> > > +++ b/drivers/cpufreq/cpufreq.c
> > > @@ -2237,6 +2237,8 @@ static int cpufreq_set_policy(struct cpufreq_policy *policy,
> > >
> > > policy->min = new_policy->min;
> > > policy->max = new_policy->max;
> > > + policy->cpuinfo.max_freq = new_policy->cpuinfo.max_freq;
> > > + policy->cpuinfo.min_freq = new_policy->cpuinfo.min_freq;
> > > trace_cpu_frequency_limits(policy);
> > >
> > > policy->cached_target_freq = UINT_MAX;
> > > diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
> > > index dd66decf2087..e1881313c396 100644
> > > --- a/drivers/cpufreq/intel_pstate.c
> > > +++ b/drivers/cpufreq/intel_pstate.c
> > > @@ -2081,11 +2081,24 @@ static void intel_pstate_adjust_policy_max(struct cpufreq_policy *policy,
> > >
> > > static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
> > > {
> > > + int max_freq;
> > > struct cpudata *cpu = all_cpu_data[policy->cpu];
> > >
> > > update_turbo_state();
> >
> > Well, update_turbo_state() should handle the case at hand already.
> >
> > That's what it's for actually: It checks if
> > MSR_IA32_MISC_ENABLE_TURBO_DISABLE is set and sets
> > global.turbo_disabled is that's the case.
> >
> > Why isn't that sufficient?
> >
> update_turbo_state() changes the flag of global.turbo_diabled but we
> need to also leverage it to adjust the policy.max accordingly. This is why
> we add intel_pstate_get_max_freq() to get the updated max freq in
> intel_pstate_verify_policy().

Yes, that's why intel_pstate_verify_policy() passes the return value
of intel_pstate_get_max_freq() as the second arg
cpufreq_verify_within_limits(), so really my question was about why
cpuinfo.max_freq needed to be updated (below).

> > > + max_freq = intel_pstate_get_max_freq(cpu);
> > > +
> > > + if (acpi_ppc && policy->max == policy->cpuinfo.max_freq &&
> > > + max_freq != policy->cpuinfo.max_freq) {
> > > + /*
> > > + * System was not running under any constraints, but the
> > > + * current max possible frequency is changed. So reset
> > > + * policy limits.
> > > + */
> > > + policy->cpuinfo.max_freq = policy->max = max_freq;
> > > + }
> >
> > Why does policy->cpuinfo.max_freq need to be updated?
> >
> This is my understanding:
> There's a corner case that, if the system boots with battery,
> the max cpu frequency will not scale up if we plug the AC later.

I see. The *initial* cpuinfo.max_freq may be too low. This part is
missing from your patch changelog.

The driver is not expected to update cpuinfo.max_freq after init.
That may not actually break anything, even though it is racy in
principle, but if it is done, it needs to be done in the "passive"
mode too and that may be more problematic.

Anyway, this is more fundamental than you seem to be thinking.

> According to the log provided by Gabriele Mazzotta, if the system
> boot up with battery, then plug the AC after boot up, the max perf ratio
> and policy->cpuinfo.max will remain 17 rather than increasing to
> 30(when AC plugged thus turbo enabled):
>
> [ 52.158810] CPU 0: _PPC is 6 - frequency limited
> [ 52.158822] intel_pstate: set_policy cpuinfo.max 1700000 policy->max 1700000
> [ 52.158825] intel_pstate: cpu:0 max_state 30 min_policy_perf:8 max_policy_perf:17
> [ 52.158827] intel_pstate: cpu:0 global_min:8 global_max:30
> [ 52.158829] intel_pstate: cpu:0 max_perf_ratio:17 min_perf_ratio:8
>
> This is caused by:
> In current intel_pstate, there's only one chance for policy.cpuinfo.max to get updated
> which is during boot up in __intel_pstate_cpu_init(). If the turbo status changes,
> we might need to also update the policy->cpuinfo.max to tell user that the hardware
> status has changed.
>
> So we give it a chance to adjust the policy.cpuinfo.max and policy.max in
> cpufreq_driver->verify() according to turbo status, this is what this patch mainly
> aims to do.
>
> Besides, since on this platform there's only one _PPC notification for one CPU, it is
> necessary to broadcast the notification to all CPUs on this package. And this patch
> broadcast it to all online CPUs to make the change simpler.

You're trying to make two substantial changes in one go, broadcasting
_PPC and updating cpuinfo.max_freq. Don't do that, they need to be
separate changes.

Moreover, we may want to address the initial cpuinfo.max_freq issue in
a different way.

Thanks,
Rafael