Re: [patch 1/4] add basic task isolation prctl interface

From: Marcelo Tosatti <mtosatti@redhat.com>
To: Frederic Weisbecker <frederic@kernel.org>
Cc: nsaenzju@redhat.com, linux-kernel@vger.kernel.org,
	Nitesh Lal <nilal@redhat.com>, Christoph Lameter <cl@linux.com>,
	Juri Lelli <juri.lelli@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Alex Belits <abelits@marvell.com>, Peter Xu <peterx@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>
Subject: Re: [patch 1/4] add basic task isolation prctl interface
Date: Tue, 27 Jul 2021 11:52:09 -0300	[thread overview]
Message-ID: <20210727145209.GA518735@fuller.cnet> (raw)
In-Reply-To: <20210727130930.GB283787@lothringen>

On Tue, Jul 27, 2021 at 03:09:30PM +0200, Frederic Weisbecker wrote:
> On Tue, Jul 27, 2021 at 02:38:15PM +0200, nsaenzju@redhat.com wrote:
> > Hi Marcelo,
> > 
> > On Tue, 2021-07-27 at 08:00 -0300, Marcelo Tosatti wrote:
> > OK, sorry if I'm being thick, but what is the benefit of having a distincnt
> > PR_ISOL_MODE instead expressing everything as PR_ISOL_FEATURES.
> > 
> >   PR_ISOL_MODE_NONE == Empty PR_ISOL_FEATURES bitmap
> > 
> >   PR_ISOL_MODE_NORMAL == Bitmap of commonly used PR_ISOL_FEATURES
> >   			      (we could introduce a define)
> > 
> >   PR_ISOL_MODE_NORMAL+PR_ISOL_VSYSCALLS == Custom bitmap
> > 
> > Other than that, my rationale is that if you extend PR_ISOL_MODE_NORMAL's
> > behaviour as new features are merged, wouldn't you be potentially breaking
> > userspace (i.e. older applications might not like the new default)?
> 
> I agree with Nicolas, and that was Thomas request too.
> Let's leave policy implementation to userspace and take
> only the individual isolation features to the kernel.
> 
> CPU/Task isolation is a relatively young feature and many users don't
> communicate much about their needs. We don't know exactly how finegrained
> the ABI will need to be so let's not make too many high level assumptions.
> 
> It's easy for userspace to set all isolation bits by itself.
> 
> Besides, those bits will be implemented one by one over time, this
> means that a prctl() bit saying "isolate everything" will have a different
> behaviour as those features get integrated. And we really want well defined
> behaviours.
> 
> Thanks.
> 
> 

OK, how about this:

...

The meaning of isolated is specified as follows:

Isolation features
==================

- prctl(PR_ISOL_GET, ISOL_SUP_FEATURES, 0, 0, 0) returns the supported
features as a return value.

- prctl(PR_ISOL_SET, ISOL_FEATURES, bitmask, 0, 0) enables the features in
the bitmask.

- prctl(PR_ISOL_GET, ISOL_FEATURES, 0, 0, 0) returns the currently
enabled features.

The supported features are:

ISOL_F_QUIESCE_ON_URET: quiesce deferred actions on return to userspace.
----------------------

Quiescing of different actions can be performed on return to userspace.

- prctl(PR_ISOL_GET, PR_ISOL_SUP_QUIESCE_CFG, 0, 0, 0) returns
the supported actions to be quiesced.

- prctl(PR_ISOL_SET, PR_ISOL_QUIESCE_CFG, quiesce_bitmask, 0, 0) returns
the currently supported actions to be quiesced.

- prctl(PR_ISOL_GET, PR_ISOL_QUIESCE_CFG, 0, 0, 0) returns
the currently enabled actions to be quiesced.

#define ISOL_F_QUIESCE_VMSTAT_SYNC      (1<<0)
#define ISOL_F_QUIESCE_NOHZ_FULL        (1<<1)
#define ISOL_F_QUIESCE_DEFER_TLB_FLUSH  (1<<2)
...