Re: [PATCH 09/11] arm: perf: parse cpu affinity from dt

[Mark Rutland <mark.rutland@arm.com>]

On Mon, Nov 17, 2014 at 11:20:35AM +0000, Will Deacon wrote:
> On Fri, Nov 07, 2014 at 04:25:34PM +0000, Mark Rutland wrote:
> > The current way we read interrupts form devicetree assumes that
> > interrupts are in increasing order of logical cpu id (MPIDR.Aff{2,1,0}),
> > and that these logical ids are in a contiguous block. This may not be
> > the case in general - after a kexec cpu ids may be arbitrarily assigned,
> > and multi-cluster systems do not have a contiguous range of cpu ids.
> > 
> > This patch parses cpu affinity information for interrupts from an
> > optional "interrupts-affinity" devicetree property described in the
> > devicetree binding document. Support for existing dts and board files
> > remains.
> > 
> > Signed-off-by: Mark Rutland <mark.rutland@arm.com>
> > ---
> >  arch/arm/include/asm/pmu.h       |  12 +++
> >  arch/arm/kernel/perf_event_cpu.c | 196 +++++++++++++++++++++++++++++----------
> >  2 files changed, 161 insertions(+), 47 deletions(-)
> > 
> > diff --git a/arch/arm/include/asm/pmu.h b/arch/arm/include/asm/pmu.h
> > index b630a44..92fc1da 100644
> > --- a/arch/arm/include/asm/pmu.h
> > +++ b/arch/arm/include/asm/pmu.h
> > @@ -12,6 +12,7 @@
> >  #ifndef __ARM_PMU_H__
> >  #define __ARM_PMU_H__
> >  
> > +#include <linux/cpumask.h>
> >  #include <linux/interrupt.h>
> >  #include <linux/perf_event.h>
> >  
> > @@ -89,6 +90,15 @@ struct pmu_hw_events {
> >  	struct arm_pmu		*percpu_pmu;
> >  };
> >  
> > +/*
> > + * For systems with heterogeneous PMUs, we need to know which CPUs each
> > + * (possibly percpu) IRQ targets. Map between them with an array of these.
> > + */
> > +struct cpu_irq {
> > +	cpumask_t cpus;
> > +	int irq;
> > +};
> > +
> >  struct arm_pmu {
> >  	struct pmu	pmu;
> >  	cpumask_t	active_irqs;
> > @@ -118,6 +128,8 @@ struct arm_pmu {
> >  	struct platform_device	*plat_device;
> >  	struct pmu_hw_events	__percpu *hw_events;
> >  	struct notifier_block	hotplug_nb;
> > +	int		nr_irqs;
> > +	struct cpu_irq *irq_map;
> >  };
> >  
> >  #define to_arm_pmu(p) (container_of(p, struct arm_pmu, pmu))
> > diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c
> > index dfcaba5..f09c8a0 100644
> > --- a/arch/arm/kernel/perf_event_cpu.c
> > +++ b/arch/arm/kernel/perf_event_cpu.c
> > @@ -85,20 +85,27 @@ static void cpu_pmu_free_irq(struct arm_pmu *cpu_pmu)
> >  	struct platform_device *pmu_device = cpu_pmu->plat_device;
> >  	struct pmu_hw_events __percpu *hw_events = cpu_pmu->hw_events;
> >  
> > -	irqs = min(pmu_device->num_resources, num_possible_cpus());
> > +	irqs = cpu_pmu->nr_irqs;
> >  
> > -	irq = platform_get_irq(pmu_device, 0);
> > -	if (irq >= 0 && irq_is_percpu(irq)) {
> > -		on_each_cpu(cpu_pmu_disable_percpu_irq, &irq, 1);
> > -		free_percpu_irq(irq, &hw_events->percpu_pmu);
> > -	} else {
> > -		for (i = 0; i < irqs; ++i) {
> > -			if (!cpumask_test_and_clear_cpu(i, &cpu_pmu->active_irqs))
> > -				continue;
> > -			irq = platform_get_irq(pmu_device, i);
> > -			if (irq >= 0)
> > -				free_irq(irq, per_cpu_ptr(&hw_events->percpu_pmu, i));
> > +	for (i = 0; i < irqs; i++) {
> > +		struct cpu_irq *map = &cpu_pmu->irq_map[i];
> > +		irq = map->irq;
> > +
> > +		if (irq <= 0)
> > +			continue;
> > +
> > +		if (irq_is_percpu(irq)) {
> > +			on_each_cpu(cpu_pmu_disable_percpu_irq, &irq, 1);
> 
> Hmm, ok, so we're assuming that all the PMUs will be wired with PPIs in this
> case. I have a patch allowing per-cpu interrupts to be requested for a
> cpumask, but I suppose that can wait until it's actually needed.

I wasn't too keen on assuming all CPUs, but I didn't have the facility
to request a PPI on a subset of CPUs. If you can point me at your patch,
I'd be happy to take a look.

I should have the target CPU mask decoded from whatever the binding
settles on, so at this point it's just plumbing.

Thanks,
Mark.