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

[Marcelo Tosatti <mtosatti@redhat.com>]

On Wed, Jul 28, 2021 at 01:55:33PM +0200, nsaenzju@redhat.com wrote:
> Hi Marcelo,
> 
> On Wed, 2021-07-28 at 06:37 -0300, Marcelo Tosatti wrote:
> > On Wed, Jul 28, 2021 at 01:45:39AM +0200, Frederic Weisbecker wrote:
> > > On Tue, Jul 27, 2021 at 11:52:09AM -0300, Marcelo Tosatti wrote:
> > > > 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.
> > > 
> > > So what are the ISOL_FEATURES here? A mode that we enter such as flush
> > > vmstat _everytime_ we resume to userpace after (and including) this prctl() ?
> > 
> > ISOL_FEATURES is just the "command" type (which you can get and set).
> > 
> > The bitmask would include ISOL_F_QUIESCE_ON_URET, so:
> > 
> > - bitmask = ISOL_F_QUIESCE_ON_URET;
> > - prctl(PR_ISOL_SET, ISOL_FEATURES, bitmask, 0, 0) enables the features in
> > the bitmask.
> > 
> > - quiesce_bitmap = prctl(PR_ISOL_GET, PR_ISOL_SUP_QUIESCE_CFG, 0, 0, 0)
> >   (1)
> > 
> >   (returns the supported actions to be quiesced).
> > 
> > - prctl(PR_ISOL_SET, PR_ISOL_QUIESCE_CFG, quiesce_bitmask, 0, 0) _sets_
> > the actions to be quiesced (2)
> > 
> > If an application does not modify "quiesce_bitmask" between 
> > points (1) and (2) above, it will enable quiescing of all
> > "features" the kernel supports.
> 
> I think this pattern of enabling all by default might be prone to subtly
> breaking things.

The reasoning behind this pattern is that many latency sensitive applications
(as far as i am aware) prefer "as few interruptions as possible, no
interruptions is preferred".

In that case, the pattern makes sense.

> For example, let's say we introduce ISOL_F_QUIESCE_DEFER_TLB_FLUSH, this will
> defer relatively short IPIs on isolated CPUs in exchange for a longer flush
> whenever we enter the kernel (syscall, IRQs, NMI, etc...). 

Why the flush has to be longer when you enter the kernel?

ISOL_F_QUIESCE_DEFER_TLB_FLUSH might collapse multiple IPIs 
into a single IPI, so the behaviour might be beneficial 
for "standard" types of application as well.

> A latency sensitive
> application might be OK with the former but not with the latter.

Two alternatives:

1) The pattern above, where particular subsystems that might interrupt 
the kernel are enabled automatically if the kernel supports it.

Pros: 
Applications which implement this only need to be changed once,
and can benefit from new kernel features.

Applications can disable particular features if they turn
out to be problematic.

Cons: 
New features might break applications.

2) Force applications to enable each new feature individually.

Pros: Won't cause regressions, kernel behaviour is explicitly 
controlled by userspace.

Cons: Apps won't benefit from new features automatically.

---

It seems to me 1) is preferred. Can also add a sysfs control to
have a "default_isolation_feature" flag, which can be changed
by a sysadmin in case a new feature is undesired.

Thoughts?