[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <86a6c5c7-47b7-49af-af57-b1d2171ba97c@igalia.com>
Date: Sat, 1 Mar 2025 08:33:36 +0900
From: Changwoo Min <changwoo@...lia.com>
To: Tejun Heo <tj@...nel.org>, Andrea Righi <arighi@...dia.com>
Cc: void@...ifault.com, kernel-dev@...lia.com, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] sched_ext: Add trace point to track sched_ext core
events
Hi Tejun and Andrea,
On 25. 3. 1. 02:31, Tejun Heo wrote:
> On Fri, Feb 28, 2025 at 11:03:54AM +0100, Andrea Righi wrote:
>>> +TRACE_EVENT(sched_ext_event,
>>> + TP_PROTO(const char *name, __u64 delta),
>>> + TP_ARGS(name, delta),
>>> +
>>> + TP_STRUCT__entry(
>>> + __string(name, name)
>>> + __field( __u64, delta )
>>
>> I'm wondering if we should use a __s64 here (and %lld below). We don't have
>> negative deltas right now, but in the future who knows...
That makes sense. I will change it to __s64 in the tracepoint.
Also, I will make corresponding changes (u64 -> s64) in other
places, including struct scx_event_stats.
>
> And let's also print out the updated value.
You might have two options here: 1) returning per-CPU event
counter or 2) returning aggregated event counter. The first opion
will be fast but less meaningful from user's point of view
compared to the second option. Assuming the tracepoint are not in
the hot path, I think the second option will be better choice.
I will add an @event field and a special version of
scx_bpf_events() for faster aggregation.
Thanks!
Changwoo Min
Powered by blists - more mailing lists