| commit | d39b0128985042eda1c2f0a1cb89d16b996f6ba9 | [log] [tgz] |
|---|---|---|
| author | Venkatesh Srinivas <venkateshs@fuchsia.infra.roller.chromium.org> | Mon Aug 16 23:12:12 2021 +0000 |
| committer | Copybara-Service <copybara-worker@google.com> | Mon Aug 16 16:13:19 2021 -0700 |
| tree | c0850067f29f7d6266444db511db5796373425f2 | |
| parent | 0c645ed94708877c31d8e9a5ec6dcc0cdf556f13 [diff] |
[roll] Roll fuchsia [perf_mon][x86] Support for Intel Goldmont CPUs
Add support for Intel Goldmont / Apollo Lake CPUs to perf_mon.
Intel Goldmont CPUs have a version 4 architectural PMU along
with well-documented non-architectural (model specific) per-core
and offcore counters. Intel Goldmont performance counter events
are fairly different than Intel 'big core' events; this will
provide a good model for support for Goldmont Plus / Tremont
cores and a 'shake out' Skylake-specifics in the existing
perf_mon system.
All perf_mon events require globally unique IDs; start Goldmont
IDs at 0x100; Intel Skylake core events start at 0x0.
1) Add support for fixed-function and architectural counters
2) Add support for a handful of non-architectural (model specific)
performance counters, both so we can add the model-specific
category and to provide a model of how to add non-architectural
performance counters.
3) Also add basic-cpu.cpspec - an example cpspec file that
uses only fixed-function and architectural performance counters,
which should work on every Intel CPU with an architectural PMU.
Tested:
host$ fx set bringup.x64 --auto-dir \
--with-base //bundles/bringup:tests \
--with-base //garnet/bin/cpuperf
...
fuchsia$ cpuperf --spec-file=/boot/data/basic-cpu.cpspec
[00039.625] 03839:03841> arch_perfmon_start:1499: Enabling perfmon, 2 fixed, 4 programmable, 0 misc
[00040.625] 03839:03841> arch_perfmon_stop_locked:1646: Disabling perfmon
| llc_misses|llc_references|instructions_retired|branch_instructions_retired|unhalted_reference_cycles|branch_misses_retired
Trace 0: | 3,083| 45,121| 1,480,754| 358,869| 7,784,244| 2,000
Trace 1: | 5,285| 54,922| 1,816,260| 436,853| 9,329,892| 2,931
Trace 2: | 8,360| 43,048| 1,103,370| 260,514| 5,699,616| 2,696
Trace 3: | 384| 2,508| 81,077| 19,765| 444,834| 135
Limitations:
* Only a handful of non-architectural events, not all required for
TopDown analysis, were added
* No offcore events were added
* cpuperf list-event hardcodes "skylake" and does not list
goldmont events.
Original-Bug: 33981 Fuchsia should run on Intel NUC6CAYH
Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/540121
Original-Revision: 5709cbdcc818475bd76f041aa535231fba11c5e4
GitOrigin-RevId: 7886df92f6a0a777277d911fd866bc5a25123d4e
Change-Id: Ieab6170a23f8665abed3c304c01c21dd521b8f3c
This repository contains Fuchsia's Global Integration manifest files.
All changes should be made to the internal version of this repository. Our infrastructure automatically updates this version when the internal one changes.
Currently all changes must be made by a Google employee. Non-Google employees wishing to make a change can ask for assistance via the IRC channel #fuchsia on Freenode.
First install Jiri.
Next run:
$ jiri init $ jiri import minimal https://fuchsia.googlesource.com/integration $ jiri update
Third party projects should have their own subdirectory in ./third_party.