Re: [PATCH v4 04/17] x86/split_lock: Align x86_capability to unsigned long to avoid split locked access

[Peter Zijlstra]

On Mon, Mar 04, 2019 at 11:15:12AM -0800, Fenghua Yu wrote:
> On Mon, Mar 04, 2019 at 10:52:19AM -0800, Dave Hansen wrote:
> > On 3/1/19 6:44 PM, Fenghua Yu wrote:
> > > diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
> > > index 33051436c864..eb8ae701ef65 100644
> > > --- a/arch/x86/include/asm/processor.h
> > > +++ b/arch/x86/include/asm/processor.h
> > > @@ -93,7 +93,9 @@ struct cpuinfo_x86 {
> > >  	__u32			extended_cpuid_level;
> > >  	/* Maximum supported CPUID level, -1=no CPUID: */
> > >  	int			cpuid_level;
> > > -	__u32			x86_capability[NCAPINTS + NBUGINTS];
> > > +	/* Unsigned long alignment to avoid split lock in atomic bitmap ops */
> > > +	__u32			x86_capability[NCAPINTS + NBUGINTS]
> > > +				__aligned(sizeof(unsigned long));
> > 
> > I think this also warrants a comment in the changelog about the
> > alignment of 'struct cpuinfo_x86'.
> 
> How about add "Depending on the starting address where GCC generates
> for data of struct cpuinfo_x86, x86_capability[] may or may not align to
> unsigned long...."?

Composite types inherit the strictest alignment of their members.
Without that alignment hints would be utterly useless.