[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Z5tvnjpNeuFZzdwy@slm.duckdns.org>
Date: Thu, 30 Jan 2025 02:25:02 -1000
From: Tejun Heo <tj@...nel.org>
To: Changwoo Min <changwoo@...lia.com>
Cc: void@...ifault.com, arighi@...dia.com, kernel-dev@...lia.com,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 08/11] sched_ext: Add an event, BYPASS_DISPATCH
On Thu, Jan 30, 2025 at 03:26:16PM +0900, Changwoo Min wrote:
> Hello,
>
> On 25. 1. 28. 05:05, Tejun Heo wrote:
> > On Sun, Jan 26, 2025 at 07:16:11PM +0900, Changwoo Min wrote:
> > ...
> > > has_tasks:
> > > + if (scx_rq_bypassing(rq))
> > > + __scx_add_event(BYPASS_DISPATCH, 1);
> >
> > Can we do this at the sources even if that's a bit more code?
>
> Sure. I will remove scx_add_event() and __scx_add_event() and will use
> this_cpu_add() and __this_cpu_add() directly.
Oh, that's not what I meant. I meant that in the code quoted above, the stat
is being incremented in a spot where most code paths converge and then the
specific stat condition is being tested again. It'd be better to update the
stat where the condition is detected initially.
Thanks.
--
tejun
Powered by blists - more mailing lists