2023-01-24 15:00:14

by James Clark

Subject: [PATCH] perf mem/c2c: Document that SPE is used for mem and c2c on Arm

Setup is non-trivial so also link to the full SPE docs.

Signed-off-by: James Clark <[email protected]>
---
tools/perf/Documentation/perf-c2c.txt | 8 ++++++--
tools/perf/Documentation/perf-mem.txt | 7 ++++++-
2 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/tools/perf/Documentation/perf-c2c.txt b/tools/perf/Documentation/perf-c2c.txt
index af5c3106f468..4e8c263e1721 100644
--- a/tools/perf/Documentation/perf-c2c.txt
+++ b/tools/perf/Documentation/perf-c2c.txt
@@ -22,7 +22,11 @@ you to track down the cacheline contentions.
On Intel, the tool is based on load latency and precise store facility events
provided by Intel CPUs. On PowerPC, the tool uses random instruction sampling
with thresholding feature. On AMD, the tool uses IBS op pmu (due to hardware
-limitations, perf c2c is not supported on Zen3 cpus).
+limitations, perf c2c is not supported on Zen3 cpus). On Arm64 it uses SPE to
+sample load and store operations, therefore hardware and kernel support is
+required. See linkperf:perf-arm-spe[1] for a setup guide. Due to the
+statistical nature of Arm SPE sampling, not every memory operation will be
+sampled.

These events provide:
- memory address of the access
@@ -333,4 +337,4 @@ Check Joe's blog on c2c tool for detailed use case explanation:

SEE ALSO
--------
-linkperf:perf-record[1], linkperf:perf-mem[1]
+linkperf:perf-record[1], linkperf:perf-mem[1], linkperf:perf-arm-spe[1]
diff --git a/tools/perf/Documentation/perf-mem.txt b/tools/perf/Documentation/perf-mem.txt
index 005c95580b1e..19862572e3f2 100644
--- a/tools/perf/Documentation/perf-mem.txt
+++ b/tools/perf/Documentation/perf-mem.txt
@@ -23,6 +23,11 @@ Note that on Intel systems the memory latency reported is the use-latency,
not the pure load (or store latency). Use latency includes any pipeline
queueing delays in addition to the memory subsystem latency.

+On Arm64 this uses SPE to sample load and store operations, therefore hardware
+and kernel support is required. See linkperf:perf-arm-spe[1] for a setup guide.
+Due to the statistical nature of SPE sampling, not every memory operation will
+be sampled.
+
OPTIONS
-------
<command>...::
@@ -93,4 +98,4 @@ all perf record options.

SEE ALSO
--------
-linkperf:perf-record[1], linkperf:perf-report[1]
+linkperf:perf-record[1], linkperf:perf-report[1], linkperf:perf-arm-spe[1]

base-commit: 5670ebf54bd26482f57a094c53bdc562c106e0a9
--
2.39.1

2023-01-24 18:37:19

by Arnaldo Carvalho de Melo

Subject: Re: [PATCH] perf mem/c2c: Document that SPE is used for mem and c2c on Arm

Em Tue, Jan 24, 2023 at 02:59:29PM +0000, James Clark escreveu:
> Setup is non-trivial so also link to the full SPE docs.

Thanks, applied.

- Arnaldo


> Signed-off-by: James Clark <[email protected]>
> ---
> tools/perf/Documentation/perf-c2c.txt | 8 ++++++--
> tools/perf/Documentation/perf-mem.txt | 7 ++++++-
> 2 files changed, 12 insertions(+), 3 deletions(-)
>
> diff --git a/tools/perf/Documentation/perf-c2c.txt b/tools/perf/Documentation/perf-c2c.txt
> index af5c3106f468..4e8c263e1721 100644
> --- a/tools/perf/Documentation/perf-c2c.txt
> +++ b/tools/perf/Documentation/perf-c2c.txt
> @@ -22,7 +22,11 @@ you to track down the cacheline contentions.
> On Intel, the tool is based on load latency and precise store facility events
> provided by Intel CPUs. On PowerPC, the tool uses random instruction sampling
> with thresholding feature. On AMD, the tool uses IBS op pmu (due to hardware
> -limitations, perf c2c is not supported on Zen3 cpus).
> +limitations, perf c2c is not supported on Zen3 cpus). On Arm64 it uses SPE to
> +sample load and store operations, therefore hardware and kernel support is
> +required. See linkperf:perf-arm-spe[1] for a setup guide. Due to the
> +statistical nature of Arm SPE sampling, not every memory operation will be
> +sampled.
>
> These events provide:
> - memory address of the access
> @@ -333,4 +337,4 @@ Check Joe's blog on c2c tool for detailed use case explanation:
>
> SEE ALSO
> --------
> -linkperf:perf-record[1], linkperf:perf-mem[1]
> +linkperf:perf-record[1], linkperf:perf-mem[1], linkperf:perf-arm-spe[1]
> diff --git a/tools/perf/Documentation/perf-mem.txt b/tools/perf/Documentation/perf-mem.txt
> index 005c95580b1e..19862572e3f2 100644
> --- a/tools/perf/Documentation/perf-mem.txt
> +++ b/tools/perf/Documentation/perf-mem.txt
> @@ -23,6 +23,11 @@ Note that on Intel systems the memory latency reported is the use-latency,
> not the pure load (or store latency). Use latency includes any pipeline
> queueing delays in addition to the memory subsystem latency.
>
> +On Arm64 this uses SPE to sample load and store operations, therefore hardware
> +and kernel support is required. See linkperf:perf-arm-spe[1] for a setup guide.
> +Due to the statistical nature of SPE sampling, not every memory operation will
> +be sampled.
> +
> OPTIONS
> -------
> <command>...::
> @@ -93,4 +98,4 @@ all perf record options.
>
> SEE ALSO
> --------
> -linkperf:perf-record[1], linkperf:perf-report[1]
> +linkperf:perf-record[1], linkperf:perf-report[1], linkperf:perf-arm-spe[1]
>
> base-commit: 5670ebf54bd26482f57a094c53bdc562c106e0a9
> --
> 2.39.1
>

--

- Arnaldo