On Fri, Jan 17, 2014 at 05:54:36PM +0000, Stephen Boyd wrote:

On 01/17/14 07:04, Will Deacon wrote:

quoted

Hi Stephen,

On Wed, Jan 15, 2014 at 08:54:27PM +0000, Stephen Boyd wrote:

quoted

On 01/15, Stephen Boyd wrote:

quoted

diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c
index 789d846a9184..e76750980b38 100644
--- a/arch/arm/kernel/perf_event.c
+++ b/arch/arm/kernel/perf_event.c

@@ -295,9 +297,15 @@ validate_group(struct perf_event *event)
 
 static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)
 {
-	struct arm_pmu *armpmu = (struct arm_pmu *) dev;
-	struct platform_device *plat_device = armpmu->plat_device;
-	struct arm_pmu_platdata *plat = dev_get_platdata(&plat_device->dev);
+	struct arm_pmu *armpmu;
+	struct platform_device *plat_device;
+	struct arm_pmu_platdata *plat;
+
+	if (irq_is_percpu(irq))
+		dev = *(struct arm_pmu_cpu **)dev;

Oh. I just realized that struct arm_pmu_cpu doesn't even exist. This
still compiles though because we're dealing with a void pointer.

Perhaps its better to just do

	dev = *(void **)dev;

here. Can you fix that up when applying? Otherwise I'll do it on
the next send if there are more comments.

Shouldn't that actually be some per_cpu accessor like this_cpu_ptr?

Nope. The genirq layer unwraps the per_cpu pointer and passes it to the
handler.

Ah yeah, I forget the dispatcher is what genirq sees as the handler. In
which case your idea looks right.

Sorry for the noise.

Will