2023-01-20 15:55:05

by James Clark

[permalink] [raw]
Subject: [PATCH v5 0/8] perf cs_etm: Basic support for virtual/kernel timestamps

Changes since v4:

* Rebase onto perf/core
* Convert new perf_pmu__cpu_slots_per_cycle() function to use
new helper functions

===========================
Changes since v3:

* Scale time estimates by INSTR_PER_NS, rather than assuming 1
instruction = 1ns
* Add a new commit that fixes some issues around timestamps going
backwards
* Use nanoseconds inside cs-etm-decoder.c, rather than storing the
raw time values and converting when a sample is synthesized. This
simplifies some of the code like estimating the first timestamp.

===========================
Changes since v2:

* Remove const to non-const change and copy strings where needed
instead.
* Use sizeof() instead of PATH_MAX
* Append "will not be set accurately." to new error message
* Remove unneeded stat() call
* Rebase on perf/core

==========================

Changes since v1:

* Add 3 refactor commits for sysfs reading around pmu.c as suggested
by Arnaldo here [1]
* The dependency on [2] has now reached mainline so is no longer
blocking
* Rebase on perf/core

[1]: https://lore.kernel.org/all/YnqVqq5QW%[email protected]/
[2]: https://lore.kernel.org/all/[email protected]/

German Gomez (4):
perf pmu: Add function to check if a pmu file exists
perf cs_etm: Keep separate symbols for ETMv4 and ETE parameters
perf cs_etm: Record ts_source in AUXTRACE_INFO for ETMv4 and ETE
perf cs_etm: Set the time field in the synthetic samples

James Clark (4):
perf: Remove duplication around EVENT_SOURCE_DEVICE_PATH
perf: Use perf_pmu__open_file() and perf_pmu__scan_file()
perf: Remove remaining duplication of bus/event_source/devices/...
perf: cs-etm: Ensure that Coresight timestamps don't go backwards

tools/perf/arch/arm/util/auxtrace.c | 5 +-
tools/perf/arch/arm/util/cs-etm.c | 91 ++++++++-
tools/perf/arch/arm64/util/pmu.c | 4 +-
tools/perf/arch/x86/util/pmu.c | 12 +-
tools/perf/util/cputopo.c | 9 +-
tools/perf/util/cs-etm-base.c | 34 +++-
.../perf/util/cs-etm-decoder/cs-etm-decoder.c | 68 +++++--
tools/perf/util/cs-etm.c | 95 +++++++++-
tools/perf/util/cs-etm.h | 16 +-
tools/perf/util/pmu-hybrid.c | 27 +--
tools/perf/util/pmu.c | 177 +++++++-----------
tools/perf/util/pmu.h | 10 +-
12 files changed, 351 insertions(+), 197 deletions(-)


base-commit: 1962ab6f6e0b39e4216206205bda14aff87705f3
prerequisite-patch-id: 9722bf86e3e6d16d177ff9a1411992a795a7dcbd
prerequisite-patch-id: b05dbef439c2ea8465f3321532257b0ca29f21f9
prerequisite-patch-id: 92680a4781cbcf010fcb007e6ea030f59e9eaefc
prerequisite-patch-id: 8e3a73a04e4b89b503377b5fac1d89d551159393
prerequisite-patch-id: 09980d8fedcdaa70b220a7802428109f48448a58
--
2.25.1


2023-01-20 18:08:48

by Tanmay Jagdale

[permalink] [raw]
Subject: Re: [PATCH v5 0/8] perf cs_etm: Basic support for virtual/kernel timestamps

Hi James,

> Changes since v4:
>
> * Rebase onto perf/core
> * Convert new perf_pmu__cpu_slots_per_cycle() function to use
> new helper functions
>
> ===========================
> Changes since v3:
>
> * Scale time estimates by INSTR_PER_NS, rather than assuming 1
> instruction = 1ns
> * Add a new commit that fixes some issues around timestamps going
> backwards
> * Use nanoseconds inside cs-etm-decoder.c, rather than storing the
> raw time values and converting when a sample is synthesized. This
> simplifies some of the code like estimating the first timestamp.
>
> ===========================
> Changes since v2:
>
> * Remove const to non-const change and copy strings where needed
> instead.
> * Use sizeof() instead of PATH_MAX
> * Append "will not be set accurately." to new error message
> * Remove unneeded stat() call
> * Rebase on perf/core
>
> ==========================
>
> Changes since v1:
>
> * Add 3 refactor commits for sysfs reading around pmu.c as suggested
> by Arnaldo here [1]
> * The dependency on [2] has now reached mainline so is no longer
> blocking
> * Rebase on perf/core
>
> [1]: https://lore.kernel.org/all/YnqVqq5QW%[email protected]/
> [2]: https://lore.kernel.org/all/[email protected]/
>
> German Gomez (4):
> perf pmu: Add function to check if a pmu file exists
> perf cs_etm: Keep separate symbols for ETMv4 and ETE parameters
> perf cs_etm: Record ts_source in AUXTRACE_INFO for ETMv4 and ETE
> perf cs_etm: Set the time field in the synthetic samples
>
> James Clark (4):
> perf: Remove duplication around EVENT_SOURCE_DEVICE_PATH
> perf: Use perf_pmu__open_file() and perf_pmu__scan_file()
> perf: Remove remaining duplication of bus/event_source/devices/...
> perf: cs-etm: Ensure that Coresight timestamps don't go backwards
>
> tools/perf/arch/arm/util/auxtrace.c | 5 +-
> tools/perf/arch/arm/util/cs-etm.c | 91 ++++++++-
> tools/perf/arch/arm64/util/pmu.c | 4 +-
> tools/perf/arch/x86/util/pmu.c | 12 +-
> tools/perf/util/cputopo.c | 9 +-
> tools/perf/util/cs-etm-base.c | 34 +++-
> .../perf/util/cs-etm-decoder/cs-etm-decoder.c | 68 +++++--
> tools/perf/util/cs-etm.c | 95 +++++++++-
> tools/perf/util/cs-etm.h | 16 +-
> tools/perf/util/pmu-hybrid.c | 27 +--
> tools/perf/util/pmu.c | 177 +++++++-----------
> tools/perf/util/pmu.h | 10 +-
> 12 files changed, 351 insertions(+), 197 deletions(-)
>
>
> base-commit: 1962ab6f6e0b39e4216206205bda14aff87705f3
> prerequisite-patch-id: 9722bf86e3e6d16d177ff9a1411992a795a7dcbd
> prerequisite-patch-id: b05dbef439c2ea8465f3321532257b0ca29f21f9
> prerequisite-patch-id: 92680a4781cbcf010fcb007e6ea030f59e9eaefc
> prerequisite-patch-id: 8e3a73a04e4b89b503377b5fac1d89d551159393
> prerequisite-patch-id: 09980d8fedcdaa70b220a7802428109f48448a58
For the series,

Tested-by: Tanmay Jagdale <[email protected]>

Thanks,
Tanmay
> --
> 2.25.1