On 01/10, Will Deacon wrote:

On Thu, Jan 09, 2014 at 07:57:12PM +0000, Stephen Boyd wrote:

quoted

(Adding DT reviewers)

On 01/09/14 03:04, Will Deacon wrote:

quoted

On Wed, Jan 08, 2014 at 10:59:40PM +0000, Stephen Boyd wrote:

quoted

+static int krait_pmu_init(struct arm_pmu *cpu_pmu)
+{
+	u32 id = read_cpuid_id() & 0xffffff00;
+
+	armv7pmu_init(cpu_pmu);
+	cpu_pmu->name		= "ARMv7 Krait";
+	/* Some early versions of Krait don't support PC write events */
+	if (id == 0x511f0400 || id == 0x510f0600)
+		cpu_pmu->map_event	= krait_map_event_no_branch;

Hmm, I'd really rather this information came via the DT. In fact, you could
just drop the branch event from your main map_event_function and keep things
simple. It depends how badly you want to advertise it in perf list :)

Not every version of Krait is missing support for this event, so I'd
like to keep it so things like perf stat show branch counts. How about I
add a bool property to the pmu node indicating that this PMU is missing
support for the PC write events? Something like "no-pc-write"?

Perhaps, although I think it should be qualcomm-specific, so something like
"qcom,krait-no-pc-write"? Again, I'd be glad to hear something from a DT
reviewer on this.

Yes a vendor prefix is probably a good idea. I'll add that in.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation