Re: [PATCH] fix ACPI induced voyager compile failure

[Ingo Molnar <mingo@elte.hu>]

* James Bottomley <James.Bottomley@HansenPartnership.com> wrote:

> On Thu, 2008-10-30 at 23:16 +0100, Ingo Molnar wrote: 
> > * James Bottomley <James.Bottomley@HansenPartnership.com> wrote:
> > 
> > > On Thu, 2008-10-30 at 22:58 +0100, Ingo Molnar wrote:
> > > > * James Bottomley <James.Bottomley@HansenPartnership.com> wrote:
> > > > 
> > > > > >From c339b7cdc39399855ec07dfbff67304f9c7fa49a Mon Sep 17 00:00:00 2001
> > > > > From: James Bottomley <James.Bottomley@HansenPartnership.com>
> > > > > Date: Wed, 29 Oct 2008 10:58:13 -0500
> > > > > Subject: [VOYAGER] x86: don't pull asm/acpi.h into fixmap_32.h
> > > > > 
> > > > > If it's not needed it causes compile failures.
> > > > > 
> > > > > Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
> > > > > ---
> > > > >  arch/x86/include/asm/fixmap_32.h |    2 ++
> > > > >  1 files changed, 2 insertions(+), 0 deletions(-)
> > > > > 
> > > > > diff --git a/arch/x86/include/asm/fixmap_32.h b/arch/x86/include/asm/fixmap_32.h
> > > > > index 09f29ab..c3302ee 100644
> > > > > --- a/arch/x86/include/asm/fixmap_32.h
> > > > > +++ b/arch/x86/include/asm/fixmap_32.h
> > > > > @@ -25,7 +25,9 @@ extern unsigned long __FIXADDR_TOP;
> > > > >  
> > > > >  #ifndef __ASSEMBLY__
> > > > >  #include <linux/kernel.h>
> > > > > +#ifdef CONFIG_ACPI
> > > > >  #include <asm/acpi.h>
> > > > > +#endif
> > > > 
> > > > hm, that's quite ugly - such headers are supposed to be safe even if 
> > > > CONFIG_APIC is not turned on.
> > > > 
> > > > What kind of compiler failures are you getting, could you send the 
> > > > .config that triggers it? I'm curious what the real cause of the build 
> > > > failure is - the #ifdef you are adding seems to fix a symptom, not a 
> > > > real bug - and maybe we can do a better fix.
> > > 
> > > It's the problem of voyager wanting its own definition of
> > > phys_cpu_present_map.  acpi.h pulls in mpspec.h which contaminates the
> > > voyager_smp.c build.
> > 
> > i guess voyager could use physid_mask_t just fine?
> > 
> > physid_mask_t is MAX_APICS derived - but it should work out fine 
> > because AFAICS voyager has a maximum of 4 cpus (but definitely less 
> > than say 32), while MAX_APICS never goes below 256.
> 
> Voyager has a max of 32 actually. [...]

(yes, <= 32 is what i meant.)

> [...] The CPU numbering is also fixed and identifies the processor 
> position (for FRU replacement), so the physical map is what we use 
> throughout the code.  The problem with the definition is that it 
> makes a rather pointless distinction between physical and logical 
> map types ... so there'd be a lot of code churn.  It would be really 
> nice if physmask_t and cpumask_t could become the same type ... then 
> we could do all this and no-one would notice.

if you mean physid_mask_t, i agree that it would be a tempting 
simplification if we got rid of the physid_mask_t distinction and made 
it equal to cpumask_t.

If the largest APIC ID in the bootup gets larger than NR_CPUS we 
should just clip it and print a warning about it in the syslog. (it 
would likely not boot up though, if that ID is an ioapic's ID)

OTOH, right now there's some practical complications: even on small 
systems we can easily get a relatively large APIC ID, needlessly 
driving up the NR_CPUS limit.

Yinghai, Peter, what do you think?

	Ingo