lkml.org 
[lkml]   [2015]   [Jul]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [RESEND PATCH v3] ARM: tegra124: pmu support
From
On Fri, Jul 17, 2015 at 4:59 PM, Thierry Reding
<thierry.reding@gmail.com> wrote:
> On Mon, Jul 13, 2015 at 10:35:45AM -0700, Kyle Huey wrote:
>> This patch modifies the device tree for tegra124 based devices to enable
>> the Cortex A15 PMU. The interrupt numbers are taken from NVIDIA TRM
>> DP-06905-001_v03p. This patch was tested on a Jetson TK1.
>>
>> Updated for proper ordering and to add interrupt-affinity values.
>>
>> Signed-off-by: Kyle Huey <khuey@kylehuey.com>
>> ---
>> arch/arm/boot/dts/tegra124.dtsi | 17 +++++++++++++----
>> 1 file changed, 13 insertions(+), 4 deletions(-)
>
> Is there any way to test this? What are the effects of adding this?

Yes. This enables the ARM PMU driver for the Cortex A15, which allows
one to use hardware performance counters via the perf_event_open API.
For a simple test program, see
https://github.com/khuey/perf-counter-test/. Without this patch, the
perf_event_open syscall will fail. With this patch, the program will
print out the performance counter value for each iteration of the
loop. (IIRC on the A15 the branch counter was removed, so you may want
to replace 0xD with 0x8 which counts instructions executed if you want
to see a non-zero number there). You also will see a message about
the PMU in the kernel log at startup after applying this patch.

I have also tested this extensively (including the interrupt features
of the PMU) on a more complex program.

> Does it enable using perf for profiling?

I have not tested it, but I believe you can use perf without this
patch if you do not use features that require hardware performance
counter support. This patch would enable those features.

>> diff --git a/arch/arm/boot/dts/tegra124.dtsi b/arch/arm/boot/dts/tegra124.dtsi
>> index 13cc7ca..de07d7e 100644
>> --- a/arch/arm/boot/dts/tegra124.dtsi
>> +++ b/arch/arm/boot/dts/tegra124.dtsi
>> @@ -918,31 +918,40 @@
>> #address-cells = <1>;
>> #size-cells = <0>;
>>
>> - cpu@0 {
>> + A15_0: cpu@0 {
>> device_type = "cpu";
>> compatible = "arm,cortex-a15";
>> reg = <0>;
>> };
>>
>> - cpu@1 {
>> + A15_1: cpu@1 {
>> device_type = "cpu";
>> compatible = "arm,cortex-a15";
>> reg = <1>;
>> };
>>
>> - cpu@2 {
>> + A15_2: cpu@2 {
>> device_type = "cpu";
>> compatible = "arm,cortex-a15";
>> reg = <2>;
>> };
>>
>> - cpu@3 {
>> + A15_3: cpu@3 {
>> device_type = "cpu";
>> compatible = "arm,cortex-a15";
>> reg = <3>;
>> };
>> };
>>
>> + pmu {
>> + compatible = "arm,cortex-a15-pmu";
>> + interrupts = <GIC_SPI 144 IRQ_TYPE_LEVEL_HIGH>,
>> + <GIC_SPI 145 IRQ_TYPE_LEVEL_HIGH>,
>> + <GIC_SPI 146 IRQ_TYPE_LEVEL_HIGH>,
>> + <GIC_SPI 147 IRQ_TYPE_LEVEL_HIGH>;
>> + interrupt-affinity = <&A15_0>, <&A15_1>, <&A15_2>, <&A15_3>;
>
> These labels look somewhat artificial to me, perhaps we could do
> something like the following instead?
>
> interrupt-affinity = <&{/cpus/cpu@0}>, ...;
>
> That's slightly more obvious and avoids the need to "invent" labels for
> the CPUs.
>
> No need to respin, I can fix that up when applying if nobody objects to
> using the alternative notation.
>
> Thierry

I have no objections. I was not aware that the device tree syntax
supported that. FWIW I cargo-culted my way to victory from
vexpress-v2p-ca9.dts here.

- Kyle


\
 
 \ /
  Last update: 2015-07-18 16:21    [W:0.082 / U:0.948 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site