Re: [PATCH resend 2/8] sched: core scheduling tagging infrastructure

[Josh Don]

Hi Peter,

On Fri, Mar 26, 2021 at 5:10 PM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Wed, Mar 24, 2021 at 05:40:14PM -0400, Joel Fernandes (Google) wrote:
> > From: Josh Don <joshdon@xxxxxxxxxx>
> >
> > A single unsigned long is insufficient as a cookie value for core
> > scheduling. We will minimally have cookie values for a per-task and a
> > per-group interface, which must be combined into an overall cookie.
> >
> > This patch adds the infrastructure necessary for setting task and group
> > cookie. Namely, it reworks the core_cookie into a struct, and provides
> > interfaces for setting task and group cookie, as well as other
> > operations (i.e. compare()). Subsequent patches will use these hooks to
> > provide an API for setting these cookies.
> >
>
> *urgh*... so I specifically wanted the task interface first to avoid /
> get-rid of all this madness. And then you keep it :-(

Sorry, I misunderstood the ask here :/ I had separated out the cgroup
interface parts of the patch, leaving (mostly) the parts which
introduced a compound cookie structure. I see now that you just wanted
the plain task interface to start, with no notion of group cookie.

> I've spend the past few hours rewriting patches #2 and #3, and adapting
> #4. The thing was working before I added SHARE_FROM back and introduced
> GET, but now I'm seeing a few FAILs from the selftest.
>
> I'm too tired to make sense of anything much, or even focus my eyes
> consistently, so I'll have to prod at it some more next week, but I've
> pushed out the lot to my queue.git:
>
>   https://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git/log/?h=sched/core-sched

Thanks, I'll take a look next week.

> Also, we really need a better name than coretag.c.

Yea, we don't really otherwise use the phrase "tagging". core_sched.c
is probably too confusing given we have sched/core.c.