在 2024/11/21 21:43, Steven Rostedt 写道:
On Thu, 21 Nov 2024 19:34:31 +0800
Shuai Xue <xueshuai@xxxxxxxxxxxxxxxxx> wrote:
Sure, I can reorganize the directory. You have two optional proposals:
1. /sys/kernel/debug/tracing/events/pci_hp_event
I'm guessing the above has TRACING_SYSTEM = pci_hp_event ? That is, the
above is a system and not an event.
2. /sys/kernel/debug/tracing/events/pci/pci_hp_event
Whereas here, it's an event.
I personally prefer the second approach. However, I'll defer the final decision
to the tracing subsystem maintainer, considering their expertise and
familiarity with the existing conventions.
Normally the system is determined by the users of the tracing
infrastructure and not the tracing maintainers. But I can give you some
guidelines.
The "system" level:
/sys/kernel/tracing/events/<system>
is just a way to group like events making it easier to enable them all at once:
echo 1 > /sys/kernel/tracing/events/<system>/enable
or
trace-cmd start -e <system>
The name of the "system" should be something that all the events underneath
represent. Note, although events in two different systems can have the same
name, it's best to try to keep them all unique. That's because if you have:
systemA/foo and systemB/foo
If you add to the kernel command line: trace_event=foo
it will enable both events. Although that could be fixed with: trace_event=systemA:foo
Note: trace_event=systemA would enable all systemA events at boot.
That said, is all these events going to be related to hotplug? If so, you
probably want "hotplug" or "hp" in the system name. If you make it just
"pci", then it will be expected that all the events will be related to PCI
in general.
I think Lukas is asking for a system for pci related tracepoint.
Does that help?
Yes, it really does.
-- Steve
Thanks a lot.
Best Regards,
Shuai
