1807030 - Linux power profiles show extremely low values

Status: RESOLVED FIXED

(Core :: Gecko Profiler, defect, P1)

Description

[Florian Quèze [:florian]]

mozregression points to bug 1794941 being the cause.

Example profile without the bug: https://share.firefox.dev/3vdBJvA
Example profile with the bug: https://share.firefox.dev/3VnW0ZH

https://searchfox.org/mozilla-central/rev/e6b709df9b93858364f02ab89f40d78762693db8/tools/profiler/core/PowerCounters-linux.cpp

// > users must then scale back: count * 1/(1e9*2^32) to get Joules
#define PERF_EVENT_SCALE_NANOJOULES 2.3283064365386962890625e-1

Note the 1e9 factor there.

After bug 1794941, we read the scale from the "/sys/bus/event_source/devices/power/events/energy-{type}.scale" file, which typically contains:
2.3283064365386962890625e-10
This is 1/(2^32).

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1794941

Comment 2

[Razvan Cojocaru]

Attachment: Bug 1807030 - Correct final perf event scale magnitude. r?dlrobertson

While at it, decrease the size of the GetRaplPerfEventConfig()
buffer to the minimum required length (as discussed in a
previous review).

Also, the logging of the read and final values is now more
generous (to aid with future debugging).

Comment 3

[Dan Robertson (:dlrobertson)]

(In reply to Florian Quèze [:florian] from comment #0)

Note the 1e9 factor there.

Good catch!

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/1672a1e0a261
Correct final perf event scale magnitude. r=dlrobertson

Comment 5

[Norisz Fay [:noriszfay]]

https://hg.mozilla.org/mozilla-central/rev/1672a1e0a261