[MODERATED] Re: [PATCH v4 3/8] L1TFv4 6

From: Andi Kleen <ak@linux.intel.com>
To: speck@linutronix.de
Subject: [MODERATED] Re: [PATCH v4 3/8] L1TFv4 6
Date: Thu, 10 May 2018 16:32:40 -0700	[thread overview]
Message-ID: <20180510233240.GL13616@tassilo.jf.intel.com> (raw)
In-Reply-To: <alpine.DEB.2.21.1805110101410.2101@nanos.tec.linutronix.de>

> > +static inline u64 flip_protnone_guard(u64 oldval, u64 val, u64 mask)
> > +{
> > +	/*
> > +	 * When a PTE transitions from NONE to !NONE or vice-versa
> > +	 * invert the PFN part to stop speculation.
> > +	 * pte_pfn undoes this when needed.
> > +	 */
> > +	if ((oldval & _PAGE_PROTNONE) != (val & _PAGE_PROTNONE))
> 
> I might be missing something, but. pte_pfn() uses protnone_mask() which in
> turn uses __pte_needs_invert(). __pte_needs_invert() checks for
> 
>      (val & (_PAGE_PRESENT|_PAGE_PROTNONE)) == _PAGE_PROTNONE
> 
> so the above condition is not matching the counterpart. So shouldn't that
> be:
> 
> 	if (__pte_needs_invert(oldval) != __pte_needs_invert(val))
> 
> Hmm?

_PAGE_PROTNONE aliases with global. For the general case you
have to check P to avoid mishandling kernel ptes.

I believe it's safe in that case because noone will change global
through pte_modify.

But can use the inlines too. It likely will generate a few more
instructions, but I doubt we need any micro optimization here.

-Andi