Re: [PATCH 2/2] intel_idle: Fix intel_idle() vs tracing

From: Rafael J. Wysocki
Date: Mon Nov 23 2020 - 08:55:23 EST


On Mon, Nov 23, 2020 at 2:46 PM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Mon, Nov 23, 2020 at 11:26:39AM +0100, Rafael J. Wysocki wrote:
> > On Fri, Nov 20, 2020 at 12:50 PM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > >
> > > cpuidle->enter() callbacks should not call into tracing because RCU
> > > has already been disabled. Instead of doing the broadcast thing
> > > itself, simply advertise to the cpuidle core that those states stop
> > > the timer.
> > >
> > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > > ---
> > > drivers/idle/intel_idle.c | 37 ++++++++++++++++++++-----------------
> > > 1 file changed, 20 insertions(+), 17 deletions(-)
> > >
> > > --- a/drivers/idle/intel_idle.c
> > > +++ b/drivers/idle/intel_idle.c
> > > @@ -126,26 +126,9 @@ static __cpuidle int intel_idle(struct c
> > > struct cpuidle_state *state = &drv->states[index];
> > > unsigned long eax = flg2MWAIT(state->flags);
> > > unsigned long ecx = 1; /* break on interrupt flag */
> > > - bool tick;
> > > -
> > > - if (!static_cpu_has(X86_FEATURE_ARAT)) {
> > > - /*
> > > - * Switch over to one-shot tick broadcast if the target C-state
> > > - * is deeper than C1.
> > > - */
> > > - if ((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) {
> > > - tick = true;
> > > - tick_broadcast_enter();
> > > - } else {
> > > - tick = false;
> > > - }
> > > - }
> > >
> > > mwait_idle_with_hints(eax, ecx);
> > >
> > > - if (!static_cpu_has(X86_FEATURE_ARAT) && tick)
> > > - tick_broadcast_exit();
> > > -
> > > return index;
> > > }
> > >
> > > @@ -1460,6 +1443,23 @@ static bool __init intel_idle_verify_cst
> > > return true;
> > > }
> > >
> > > +static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
> > > +{
> > > + unsigned long eax = flg2MWAIT(state->flags);
> > > +
> > > + if (boot_cpu_has(X86_FEATURE_ARAT))
> > > + return false;
> > > +
> > > + /*
> > > + * Switch over to one-shot tick broadcast if the target C-state
> > > + * is deeper than C1.
> > > + */
> > > + if ((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK)
> > > + return true;
> > > +
> > > + return false;
> > > +}
> > > +
> > > static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
> > > {
> > > int cstate;
> > > @@ -1507,6 +1507,9 @@ static void __init intel_idle_init_cstat
> > > !(cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
> > > drv->states[drv->state_count].flags |= CPUIDLE_FLAG_OFF;
> > >
> > > + if (intel_idle_state_needs_timer_stop(&drv->states[drv->state_count]))
> > > + drv->states[drv->state_count].flags |= CPUIDLE_FLAG_TIMER_STOP;
> > > +
> > > drv->state_count++;
> > > }
> >
> > This doesn't cover the ACPI case AFAICS.
>
> aa6b43d57f99 ("ACPI: processor: Use CPUIDLE_FLAG_TIMER_STOP")
>
> did that, no?

Nope.

intel_idle_init_cstates_acpi() needs to be updated too as it doesn't
pick up the flags automatically. It looks like CPUIDLE_FLAG_RCU_IDLE
needs to be copied too.