Re: [PATCH 1/2] x86/mm/pat: Change pat_disable() to emulate PAT table

[Toshi Kani <toshi.kani@hpe.com>]

On Fri, 2016-03-11 at 09:12 +0000, Borislav Petkov wrote:
> On Thu, Mar 10, 2016 at 09:45:45PM -0700, Toshi Kani wrote:
 :
> >  
> > -static inline void pat_disable(const char *reason)
> > +void pat_disable(const char *reason)
> >  {
> > +	if (boot_cpu_done) {
> > +		pr_info("x86/PAT: PAT cannot be disabled after
> > initialized\n");
> 
> 		pr_err()

Will do.

> 
> > +		return;
> > +	}
> > +
> >  	__pat_enabled = 0;
> >  	pr_info("x86/PAT: %s\n", reason);
> > +
> > +	pat_disable_init();
> 
> Why can't you call pat_init() here simply? It checks pat_enabled(). You
> can call it pat_setup() or so if it looks confusing to call an init
> function in a disable function...

How about pat_disable_setup()?  It's only used for the disabled case, so
I'd prefer to keep the word "disable".

Yes, calling pat_init() from pat_disable() works too. I changed it in this
way because:
 - pat_bsp_init() calls pat_disabled() in an error case. It is simpler to
avoid a recursive call to pat_init().
 - pat_bsp_init() has two different error paths, 1) call pat_disable() and
return, and 2) goto done and call pat_init_cache_modes(). We can remove
case 2) to keep the error handling consistent in this way.

> Then you don't have to add yet another static disable_init_done but rely
> on boot_cpu_done which gets set in pat_init().

Right, but it will do 'boot_cpu_done = true' twice, and this implicit
recursive call may cause an issue in future if someone makes change
carelessly.

> Also, I don't see the static_cpu_has() check I suggested yesterday - we
> need to check the feature bits if PAT gets disabled early on some old
> Intels.

Sorry, I should have mentioned it. I ended up not needing this change. The
table in patch 2/2 covers this case as:

  MTRR   PAT    ACTION
  ====================================================================
  E      D      MTRR calls pat_init() -> PAT disabled per cpu_has_pat

That is, the check with cpu_has_pat in pat_bsp_init() calls pat_disable()
in this case. I preferred this way because it will continue to log a
message "PAT not supported by CPU.", and keeps __pat_enabled as the single
variable to manage the PAT state.

Thanks,
-Toshi