[PATCH V3 0/4] TopDown metrics support for Ice Lake (perf tool)

From: kan . liang
Date: Fri Sep 11 2020 - 13:09:35 EST

From: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>

Changes since V2:
- Even topdown_metric_attrs() is failed to be found in the previous
platforms, the 'str' variable is still allocated.
zfree it before reuse it in Patch 3.

Changes since V1:
- Fix a grammar error in the printf message in Patch 3
- Add the "Acked-by" tag from Jirka.

The kernel patches have been merged into the tip's perf/core branch.
The patch set is on top of commit 2cb5383b30d4 ("perf/x86/intel: Support
per-thread RDPMC TopDown metrics") of the tip's perf/core branch.

The changes for the perf tool include:
- Extend --topdown option to support per thread TopDown metrics
- Support sample-read topdown metric group
- Add a complete document for the TopDown usage.

Ice Lake has support for measuring the level 1 TopDown metrics
directly in hardware. This is implemented by an additional METRICS
register, and a new Fixed Counter 3 that measures pipeline SLOTS.

New in Icelake
- Do not require generic counters. This allows to collect TopDown always
in addition to other events.
- Measuring TopDown per thread/process instead of only per core

For the Ice Lake implementation of performance metrics, the values in
PERF_METRICS MSR are derived from fixed counter 3. Software should start
both registers, PERF_METRICS and fixed counter 3, from zero.
Additionally, software is recommended to periodically clear both
registers in order to maintain accurate measurements. The latter is
required for certain scenarios that involve sampling metrics at high
rates. Software should always write fixed counter 3 before write to

IA32_PERF_GLOBAL_STATUS. OVF_PERF_METRICS[48]: If this bit is set,
it indicates that some PERF_METRICS-related counter has overflowed and
a PMI is triggered. Software has to synchronize, e.g. re-start,
PERF_METRICS as well as fixed counter 3. Otherwise, PERF_METRICS may
return invalid values.

- To get accurate result and avoid reading the METRICS register multiple
times, the TopDown metrics events and SLOTS event have to be in the
same group.
- METRICS and SLOTS registers have to be cleared after each read by SW.
That is to prevent the lose of precision.
- Cannot do sampling read SLOTS and TopDown metric events

Please refer SDM Vol3, Performance Metrics for the details of
TopDown metrics.

Andi Kleen (2):
perf stat: Support new per thread TopDown metrics
perf, tools: Add documentation for topdown metrics

Kan Liang (2):
perf tools: Rename group to topdown
perf record: Support sample-read topdown metric group

tools/perf/Documentation/perf-stat.txt | 7 +-
tools/perf/Documentation/topdown.txt | 256 +++++++++++++++++++++++++
tools/perf/arch/x86/util/Build | 2 +-
tools/perf/arch/x86/util/group.c | 28 ---
tools/perf/arch/x86/util/topdown.c | 63 ++++++
tools/perf/builtin-stat.c | 81 +++-----
tools/perf/util/Build | 1 +
tools/perf/util/group.h | 8 -
tools/perf/util/record.c | 3 +-
tools/perf/util/stat-shadow.c | 89 +++++++++
tools/perf/util/stat.c | 4 +
tools/perf/util/stat.h | 8 +
tools/perf/util/topdown.c | 58 ++++++
tools/perf/util/topdown.h | 12 ++
14 files changed, 528 insertions(+), 92 deletions(-)
create mode 100644 tools/perf/Documentation/topdown.txt
delete mode 100644 tools/perf/arch/x86/util/group.c
create mode 100644 tools/perf/arch/x86/util/topdown.c
delete mode 100644 tools/perf/util/group.h
create mode 100644 tools/perf/util/topdown.c
create mode 100644 tools/perf/util/topdown.h