[MODERATED] Re: SBB V10 Bundle

[Ingo Molnar <mingo@kernel.org>]

* speck for Thomas Gleixner <speck@linutronix.de> wrote:

> > Another question is:
> > 
> > +static int ssb_prctl_set(unsigned long ctrl)
> > +{
> > +       bool rds = !!test_tsk_thread_flag(current, TIF_RDS);
> > +
> > +       if (ssb_mode != SPEC_STORE_BYPASS_PRCTL)
> > +               return -ENXIO;
> > 
> > This might be overly permissive in the 'auto' case which defaulted to 'on', where 
> > an application might still want to disable it for good reasons?
> 
> Permissive? Its restrictive.

I mean when 'auto' set it to 'off', i.e. turned off mitigation. (Darn all the 
negatives ...)

> Also the default is 'prctl' now and I don't think that you want applications
> let override the admin decision, which might be 'on' == global mitigation or
> 'off' == 'no mitigation at all'.

Sure, explicit 'on' and 'off' is an admin choice - but 'auto' isn't really: it's 
the admin saying "whatever you think is right".

> So when the mitigation mode is 'prctl' which is the default then
> applications can fiddle with it. If the admin decided global on or off then
> it's rightfully rejected.
> 
> Hmm?

Correct, the only question is when 'auto' decides 'off' and app uses the prctl().

Do we even care, or do we think if the kernel thinks 'off' is correct then it 
should be forced?

I'd be fine with "we don't care" if "prctl" is the expected distro default.

Thanks,

	Ingo