On Mon, Dec 16, 2013 at 10:07:02AM -0800, David Cohen wrote:
> On Mon, Dec 16, 2013 at 11:13:59AM +0100, Ingo Molnar wrote:
> > 
> > * David Cohen <david.a.cohen@linux.intel.com> wrote:
> > 
> > > Despite our intention to not make Intel MID dependent of SFI, it does
> > > not reflect the current situation.
> > > If SFI is not selected, Intel MID can't select APB_TIMER otherwise this
> > > happens:
> > > 
> > > arch/x86/built-in.o: In function `apbt_set_mapping':
> > > arch/x86/kernel/apb_timer.c:91: undefined reference to `sfi_get_mtmr'
> > > arch/x86/kernel/apb_timer.c:109: undefined reference to `sfi_free_mtmr'
> > > arch/x86/kernel/apb_timer.c:112: undefined reference to `sfi_get_mtmr'
> > > arch/x86/kernel/apb_timer.c:91: undefined reference to `sfi_get_mtmr'
> > > arch/x86/kernel/apb_timer.c:109: undefined reference to `sfi_free_mtmr'
> > > arch/x86/kernel/apb_timer.c:112: undefined reference to `sfi_get_mtmr'
> > > 
> > > This patch solves such problem by adding SFI as dependence of
> > > X86_INTEL_MID. As soon as we make Intel MID completely independent of
> > > SFI this patch is safe to be reverted.
> > > 
> > > Signed-off-by: David Cohen <david.a.cohen@linux.intel.com>
> > > ---
> > > 
> > > This patch is intended to be applied on kernel 3.13-rc release.
> > > 
> > > As a matter of curiosity, CONFIG_APB_TIMER does depend on CONFIG_SFI. But looks
> > > like CONFIG_X86_INTEL_MID is able to bypass it.
> > > Maybe we need to check Kconfig itself too?
> > > 
> > >  arch/x86/Kconfig | 1 +
> > >  1 file changed, 1 insertion(+)
> > > 
> > > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > > index 00fa3b7d2abe..5802476d4e91 100644
> > > --- a/arch/x86/Kconfig
> > > +++ b/arch/x86/Kconfig
> > > @@ -447,6 +447,7 @@ config X86_INTEL_MID
> > >  	depends on PCI
> > >  	depends on PCI_GOANY
> > >  	depends on X86_IO_APIC
> > > +	depends on SFI
> > >  	select DW_APB_TIMER
> > >  	select APB_TIMER
> > >  	select INTEL_SCU_IPC
> > 
> > Can SFI be selected instead? That's what CONFIG_X86_MDFLD does, and 
> > it's a conceptual extension of 'select DW_APB_TIMER'.
> 
> Right. It makes sense indeed to move SFI line from X86_MDFLD to
> X86_INTEL_MID.
> But it remembers me I'll have to rework this patch, since X86_MDFLD
> doesn't exist in my -next branch anymore:
> https://patchwork.kernel.org/patch/3179421/

It looks like if we apply the patch above using the attached config
we're able to bypass CONFIG_SFI dependence on CONFIG_APB_TIMER (i.e.
select APB_TIMER without SFI even though APB_TIMER explicitly depends on
SFI) breaking compilation (ARCH=i386 is a must). But the same doesn't
happen without the patch.

So we do not need this fix for 3.13-rc. I will just send a new version
of the patch mentioned on link above selecting CONFIG_SFI explicitly
from X86_INTEL_MID.

Br, David

> 
> Br, David Cohen
> 
> > 
> > Thanks,
> > 
> > 	Ingo