[PATCH 1/2] genirq: Provide an accessor for IRQ_PER_CPU flag.

[PATCH 1/2] genirq: Provide an accessor for IRQ_PER_CPU flag.

[Chris Smith]

Adds an accessor function, irq_is_per_cpu(), for the flag IRQ_PER_CPU.

This is useful if you have an IRQ that is per-cpu in some hardware
implementations, and not in others (example: ARM PMU Interrupts), and
you want to handle both cases.

Signed-off-by: Chris Smith <chris.smith@st.com>
---
 include/linux/irqdesc.h |    8 ++++++++
 1 files changed, 8 insertions(+), 0 deletions(-)

diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
index f1e2527..771cb35 100644
--- a/include/linux/irqdesc.h
+++ b/include/linux/irqdesc.h
@@ -154,6 +154,14 @@ static inline int irq_balancing_disabled(unsigned int irq)
 	return desc->status_use_accessors & IRQ_NO_BALANCING_MASK;
 }
 
+static inline int irq_is_per_cpu(unsigned int irq)
+{
+	struct irq_desc *desc;
+
+	desc = irq_to_desc(irq);
+	return desc->status_use_accessors & IRQ_PER_CPU;
+}
+
 static inline void
 irq_set_lockdep_class(unsigned int irq, struct lock_class_key *class)
 {
-- 
1.7.6.5

[PATCH 2/2] arm: pmu: add support for per-cpu PMU interrupts.

[Chris Smith]

We have systems where the PMU interrupts are PPIs, and so per-cpu. This
patch adds support for per-cpu PMU interrupts to the perf event code.

Note that because the dev_id passed to request_percpu_irq has to be
percpu data, we now pass cpu_hw_events as the dev_id rather than armpmu.
This only place the dev_id was used was in armpmu_platform_irq, which is
fixed up accordingly.

Signed-off-by: Chris Smith <chris.smith@st.com>
---
 arch/arm/kernel/perf_event.c    |   91 +++++++++++++++++++++++++++++---------
 arch/arm/kernel/perf_event_v6.c |    3 +-
 arch/arm/kernel/perf_event_v7.c |    3 +-
 3 files changed, 71 insertions(+), 26 deletions(-)

diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c
index b2abfa1..add9064 100644
--- a/arch/arm/kernel/perf_event.c
+++ b/arch/arm/kernel/perf_event.c
@@ -14,6 +14,7 @@
 
 #include <linux/bitmap.h>
 #include <linux/interrupt.h>
+#include <linux/irq.h>
 #include <linux/kernel.h>
 #include <linux/export.h>
 #include <linux/perf_event.h>
@@ -355,13 +356,27 @@ validate_group(struct perf_event *event)
 
 static irqreturn_t armpmu_platform_irq(int irq, void *dev)
 {
-	struct arm_pmu *armpmu = (struct arm_pmu *) dev;
+	struct pmu_hw_events *cpuc  = (struct pmu_hw_events *)dev;
+	struct arm_pmu *armpmu = to_arm_pmu((*(cpuc->events))->pmu);
+
 	struct platform_device *plat_device = armpmu->plat_device;
 	struct arm_pmu_platdata *plat = dev_get_platdata(&plat_device->dev);
 
 	return plat->handle_irq(irq, dev, armpmu->handle_irq);
 }
 
+/* Wrap enable_percpu_irq up as a smp_call_func_t */
+static void armpmu_enable_percpu_irq(void *irq)
+{
+	enable_percpu_irq((int)irq, IRQ_TYPE_NONE);
+}
+
+/* Wrap disable_percpu_irq up as a smp_call_func_t */
+static void armpmu_disable_percpu_irq(void *irq)
+{
+	disable_percpu_irq((int)irq);
+}
+
 static void
 armpmu_release_hardware(struct arm_pmu *armpmu)
 {
@@ -379,7 +394,14 @@ armpmu_release_hardware(struct arm_pmu *armpmu)
 		if (irq >= 0) {
 			if (plat && plat->disable_irq)
 				plat->disable_irq(irq);
-			free_irq(irq, armpmu);
+			else if (irq_is_per_cpu(irq))
+				on_each_cpu(armpmu_disable_percpu_irq,
+						(void *)irq, 1);
+
+			if (irq_is_per_cpu(irq))
+				free_percpu_irq(irq, &cpu_hw_events);
+			else
+				free_irq(irq, &per_cpu(cpu_hw_events, i));
 		}
 	}
 
@@ -421,28 +443,53 @@ armpmu_reserve_hardware(struct arm_pmu *armpmu)
 		if (irq < 0)
 			continue;
 
-		/*
-		 * If we have a single PMU interrupt that we can't shift,
-		 * assume that we're running on a uniprocessor machine and
-		 * continue. Otherwise, continue without this interrupt.
-		 */
-		if (irq_set_affinity(irq, cpumask_of(i)) && irqs > 1) {
-			pr_warning("unable to set irq affinity (irq=%d, cpu=%u)\n",
-				    irq, i);
-			continue;
+		if (irq_is_per_cpu(irq)) {
+			/*
+			 * We assume that if the PMU IRQ is per-cpu, then
+			 * there is only one.
+			 */
+			WARN_ON(irqs > 1);
+
+			err = request_percpu_irq(irq, handle_irq,
+					"arm-pmu", &cpu_hw_events);
+			if (err) {
+				pr_err("unable to request IRQ%d for ARM PMU counters\n",
+					irq);
+				armpmu_release_hardware(armpmu);
+				return err;
+			} else if (plat && plat->enable_irq) {
+				plat->enable_irq(irq);
+			} else {
+				on_each_cpu(armpmu_enable_percpu_irq,
+						(void *)irq, 1);
+			}
+
+		} else {
+			/*
+			 * If we have a single PMU interrupt that we can't
+			 * shift, assume that we're running on a uniprocessor
+			 * machine and continue. Otherwise, continue without
+			 * this interrupt.
+			 */
+			if (irq_set_affinity(irq, cpumask_of(i)) && irqs > 1) {
+				pr_warning("unable to set irq affinity (irq=%d, cpu=%u)\n",
+					    irq, i);
+				continue;
+			}
+
+			err = request_irq(irq, handle_irq,
+					IRQF_DISABLED | IRQF_NOBALANCING,
+					"arm-pmu", &per_cpu(cpu_hw_events, i));
+
+			if (err) {
+				pr_err("unable to request IRQ%d for ARM PMU counters\n",
+					irq);
+				armpmu_release_hardware(armpmu);
+				return err;
+			} else if (plat && plat->enable_irq)
+				plat->enable_irq(irq);
 		}
 
-		err = request_irq(irq, handle_irq,
-				  IRQF_DISABLED | IRQF_NOBALANCING,
-				  "arm-pmu", armpmu);
-		if (err) {
-			pr_err("unable to request IRQ%d for ARM PMU counters\n",
-				irq);
-			armpmu_release_hardware(armpmu);
-			return err;
-		} else if (plat && plat->enable_irq)
-			plat->enable_irq(irq);
-
 		cpumask_set_cpu(i, &armpmu->active_irqs);
 	}
 
diff --git a/arch/arm/kernel/perf_event_v6.c b/arch/arm/kernel/perf_event_v6.c
index b78af0c..6b348f5 100644
--- a/arch/arm/kernel/perf_event_v6.c
+++ b/arch/arm/kernel/perf_event_v6.c
@@ -473,7 +473,7 @@ armv6pmu_handle_irq(int irq_num,
 {
 	unsigned long pmcr = armv6_pmcr_read();
 	struct perf_sample_data data;
-	struct pmu_hw_events *cpuc;
+	struct pmu_hw_events *cpuc = (struct pmu_hw_events *)dev;
 	struct pt_regs *regs;
 	int idx;
 
@@ -491,7 +491,6 @@ armv6pmu_handle_irq(int irq_num,
 
 	perf_sample_data_init(&data, 0);
 
-	cpuc = &__get_cpu_var(cpu_hw_events);
 	for (idx = 0; idx < cpu_pmu->num_events; ++idx) {
 		struct perf_event *event = cpuc->events[idx];
 		struct hw_perf_event *hwc;
diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c
index 4d7095a..56ddad7 100644
--- a/arch/arm/kernel/perf_event_v7.c
+++ b/arch/arm/kernel/perf_event_v7.c
@@ -933,7 +933,7 @@ static irqreturn_t armv7pmu_handle_irq(int irq_num, void *dev)
 {
 	u32 pmnc;
 	struct perf_sample_data data;
-	struct pmu_hw_events *cpuc;
+	struct pmu_hw_events *cpuc  = (struct pmu_hw_events *)dev;
 	struct pt_regs *regs;
 	int idx;
 
@@ -955,7 +955,6 @@ static irqreturn_t armv7pmu_handle_irq(int irq_num, void *dev)
 
 	perf_sample_data_init(&data, 0);
 
-	cpuc = &__get_cpu_var(cpu_hw_events);
 	for (idx = 0; idx < cpu_pmu->num_events; ++idx) {
 		struct perf_event *event = cpuc->events[idx];
 		struct hw_perf_event *hwc;
-- 
1.7.6.5

[PATCH 2/2] arm: pmu: add support for per-cpu PMU interrupts.

[Will Deacon]

Hi Chris,

On Mon, Jul 30, 2012 at 03:32:12PM +0100, Chris Smith wrote:
> We have systems where the PMU interrupts are PPIs, and so per-cpu. This
> patch adds support for per-cpu PMU interrupts to the perf event code.
> 
> Note that because the dev_id passed to request_percpu_irq has to be
> percpu data, we now pass cpu_hw_events as the dev_id rather than armpmu.
> This only place the dev_id was used was in armpmu_platform_irq, which is
> fixed up accordingly.
> 
> Signed-off-by: Chris Smith <chris.smith@st.com>
> ---

I'm actually in the process of moving request/free IRQ into an armpmu
callback, so this patch would fit more neatly into that.

I should have something available this week that you can take a look at.

Will