commit | c53ec7eedd2c07538d262fc39abb76574af6e483 | [log] [tgz] |
---|---|---|
author | Ryan Savitski <rsavitski@google.com> | Thu Nov 07 17:04:50 2019 +0000 |
committer | Ryan Savitski <rsavitski@google.com> | Fri Nov 08 12:11:01 2019 +0000 |
tree | b8cf24d2a6f026a48d2e31b917ac25380abf84d3 | |
parent | 6953471e0d3bfbc2bbccf6ca6a82679f3de5a036 [diff] |
compact sched_waking: event decoding in trace_processor Compact waking events are now handled by the sched_event_tracker, which populates the raw & instants tables. Handling of the default-encoded events hasn't changed. The emitting pid is inferred from the [pending] scheduling slices, in other words, sched_switch events. So if you have compact sched_waking, but no sched_switch, we skip all waking events during parsing. I think this is fine, and just needs some extra documentation. We could instead start encoding the emitting pid in the trace to remove this limitation, but I'm not sure who'd want *compact* sched_waking without any sched_switch events. Sorter: I've kept the approach of having all events in the same per-cpu queue. We can theoretically do better with multiple queues, but that also requires further changes to the sorter that are outside the scope of this cl. In practice, this approach does not give significant overheads on typical traces (especially if there's no write_to_file, as then we sort each queue at most once at the end). Change-Id: I1deda1d433b696dd16a5f127604c9ab8f6f8cdc9
Perfetto is an open-source project for performance instrumentation and tracing of Linux/Android/Chrome platforms and user-space apps.
See www.perfetto.dev for docs.
You can reach us on our Discord channel. If you prefer using IRC we have an experimental Discord <> IRC bridge synced with #perfetto-dev
on Freenode.