Re: [PATCH] mm: introduce sysctl file to flush per-cpu vmstat statistics

From: Marcelo Tosatti <mtosatti@redhat.com>
To: Frederic Weisbecker <frederic@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Arnaldo Carvalho de Melo <acme@redhat.com>,
	Christoph Lameter <cl@linux.com>,
	Matthew Wilcox <willy@infradead.org>,
	linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
	Alex Belits <abelits@marvell.com>, Phil Auld <pauld@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>
Subject: Re: [PATCH] mm: introduce sysctl file to flush per-cpu vmstat statistics
Date: Fri, 4 Dec 2020 10:31:37 -0300	[thread overview]
Message-ID: <20201204133137.GC5772@fuller.cnet> (raw)
In-Reply-To: <20201204002000.GA57685@lothringen>

On Fri, Dec 04, 2020 at 01:20:00AM +0100, Frederic Weisbecker wrote:
> On Wed, Dec 02, 2020 at 03:38:42PM -0300, Marcelo Tosatti wrote:
> > +Usage
> > +=====
> > +``PR_TASK_ISOLATION_FEATURES``:
> > +        Returns the supported features. Features are defined
> > +        at include/uapi/linux/isolation.h.
> > +
> > +        Usage::
> > +
> > +                prctl(PR_TASK_ISOLATION_FEATURES, ifeat, 0, 0, 0);
> > +
> > +        The 'ifeat' argument is a pointer to a struct isolation_features:
> > +
> > +                struct isolation_features {
> > +                        __u32   flags;
> > +                        __u32   pad[3];
> > +                };
> > +
> > +        Where flags contains bit sets for the features the kernel supports.
> > +
> > +        FIXME: describe error codes
> > +
> > +``PR_TASK_ISOLATION_SET``:
> > +        Enables task isolation mode. The type of task isolation is specified
> > +        via
> > +
> > +        struct isolation_control {
> > +                __u32   flags;
> > +                __u32   pad[15];
> > +        };
> > +
> > +``PR_TASK_ISOLATION_GET``:
> > +        Retrieves the currently configured task isolation mode and
> > +        parameters.
> 
> So, I guess we can think of two kinds of calls here:
> 
> * Modes, or isolation states: signal on disallowed kernel entry, arrange for not taking
>   some IPIs toward delayed processing, etc...
> 
>   For such "mode" it makes sense to have set() and get(), because the mode can be turned
>   on and off.
> 
> * Oneshot requests: quiescing (eg: vmstat), flush some pending events.
> 
>   For that it makes no sense to have a get().
> 
> 
> I would rather decouple the above with, for modes:
> 
>   PR_TASK_ISOLATION_SET
>   PR_TASK_ISOLATION_GET
> 
> And for oneshot requests:
> 
>   PR_TASK_ISOLATION_REQUEST
> 
> Thanks.

Yes, this was annoying.

Will extend the interface to split SET/GET per command as well (so 
its easier to know from which feature the error comes from),
and use REQUEST to finally enter isolated mode.

Thanks for the suggestion!