Re: [PATCH] ARM: Fix SMP_ON_UP for non ARM ltd. implementations

[Will Deacon <will.deacon@arm.com>]

Hi Russell,

On Sun, 2011-01-30 at 12:20 +0000, Russell King - ARM Linux wrote:
> On Wed, Jan 26, 2011 at 03:04:12PM -0800, Stephen Boyd wrote:
> > @@ -393,20 +393,25 @@ ENDPROC(__turn_mmu_on)
> >  #ifdef CONFIG_SMP_ON_UP
> >  __fixup_smp:
> >       mov     r4, #0x00070000
> > -     orr     r3, r4, #0xff000000     @ mask 0xff070000
> > -     orr     r4, r4, #0x41000000     @ val 0x41070000
> > -     and     r0, r9, r3
> > -     teq     r0, r4                  @ ARM CPU and ARMv6/v7?
> > +     and     r0, r9, r4
> > +     teq     r0, r4                  @ ARMv6/v7?
> >       bne     __fixup_smp_on_up       @ no, assume UP
> > 
> > +     orr     r3, r4, #0xff000000     @ mask 0xff070000
> >       orr     r3, r3, #0x0000ff00
> >       orr     r3, r3, #0x000000f0     @ mask 0xff07fff0
> > +     orr     r4, r4, #0x41000000     @ val 0x41070000
> >       orr     r4, r4, #0x0000b000
> >       orr     r4, r4, #0x00000020     @ val 0x4107b020
> >       and     r0, r9, r3
> >       teq     r0, r4                  @ ARM 11MPCore?
> >       moveq   pc, lr                  @ yes, assume SMP
> > 
> > +     mov     r4, #0x00070000
> > +     and     r0, r9, #0x000f0000
> > +     teq     r0, r4                  @ ARMv6?
> > +     beq     __fixup_smp_on_up       @ yes, assume UP
> > +
> 
> Wouldn't it be better to check for CPUID presence first, then ARM11MPcore,
> and lastly preserve of MPIDR-flagged extensions?
> 

Yes, the v6 catch-all at the end is pretty horrible.


> Will - can you check whether the below is correct?
> 
>  arch/arm/kernel/head.S |   17 +++++++----------
>  1 files changed, 7 insertions(+), 10 deletions(-)
> 
> diff --git a/arch/arm/kernel/head.S b/arch/arm/kernel/head.S
> index 17a97b5..f6b31c4 100644
> --- a/arch/arm/kernel/head.S
> +++ b/arch/arm/kernel/head.S
> @@ -390,19 +390,16 @@ ENDPROC(__turn_mmu_on)
> 
>  #ifdef CONFIG_SMP_ON_UP
>  __fixup_smp:
> -       mov     r4, #0x00070000
> -       orr     r3, r4, #0xff000000     @ mask 0xff070000
> -       orr     r4, r4, #0x41000000     @ val 0x41070000
> -       and     r0, r9, r3
> -       teq     r0, r4                  @ ARM CPU and ARMv6/v7?
> +       and     r3, r9, #0x000f0000     @ architecture version
> +       teq     r3, #0x000f00000        @ CPU ID supported?


You've got an extra 0x0 on the end of that constant.


>         bne     __fixup_smp_on_up       @ no, assume UP
> 
> -       orr     r3, r3, #0x0000ff00
> -       orr     r3, r3, #0x000000f0     @ mask 0xff07fff0
> +       bic     r3, r9, #0x00ff0000
> +       bic     r3, r3, #0x0000000f     @ mask 0xff00fff0
> +       mov     r4, #0x41000000
>         orr     r4, r4, #0x0000b000
> -       orr     r4, r4, #0x00000020     @ val 0x4107b020
> -       and     r0, r9, r3
> -       teq     r0, r4                  @ ARM 11MPCore?
> +       orr     r4, r4, #0x00000020     @ val 0x4100b020
> +       teq     r3, r4                  @ ARM 11MPCore?
>         moveq   pc, lr                  @ yes, assume SMP
> 

Yup, that looks correct to me. 11MPCore has 0xf for the architecture
field so we'll end up identifying it correctly here.


>         mrc     p15, 0, r0, c0, c0, 5   @ read MPIDR


The code following this does:

tst     r0, #1 << 31
movne   pc, lr                  @ bit 31 => SMP

As an optimisation, we could also check that bit 30 is zero so that
we patch out the SMP stuff for single-core A5/A9/A15. Up to you.

With the typo fixed:

Acked-by: Will Deacon <will.deacon@arm.com>

Will