[PATCH] ARM: Add check for Cortex-A15 errata 798181 ECO

[linux@arm.linux.org.uk (Russell King - ARM Linux)]

On Mon, Jul 15, 2013 at 01:59:24PM -0700, Olof Johansson wrote:
> On Mon, Jul 15, 2013 at 1:42 PM, Rob Herring <robherring2@gmail.com> wrote:
> > diff --git a/arch/arm/kernel/smp_tlb.c b/arch/arm/kernel/smp_tlb.c
> > index 9a52a07..7fd64b7 100644
> > --- a/arch/arm/kernel/smp_tlb.c
> > +++ b/arch/arm/kernel/smp_tlb.c
> > @@ -74,10 +74,21 @@ static inline void ipi_flush_bp_all(void *ignored)
> >  static int erratum_a15_798181(void)
> >  {
> >         unsigned int midr = read_cpuid_id();
> > +       unsigned int revidr;
> >
> >         /* Cortex-A15 r0p0..r3p2 affected */
> >         if ((midr & 0xff0ffff0) != 0x410fc0f0 || midr > 0x413fc0f2)
> >                 return 0;
> > +
> > +       /* Check for Cortex A15 <= r3p2 with ECO fix */
> > +       revidr = read_cpuid(CPUID_REVIDR);
> > +       if ((revidr & 0x210) == 0x210)
> > +               return 0;
> 
> Reading and evaluating all this on every invalidate seems suboptimal.
> It should be possible to read just once and cache the result.
> 
> I would have had the same comment about the original patch, but ARM
> pushes those to Russell in secret so it wasn't on the list. :-)

For a single read, there's probably not much between reading the CP15
register and doing the tests, vs reading a bool from memory to decide
whether we need to do it.

For more than that (as the above patch is growing) it probably makes
sense to think about using a bool in memory now.