Re: [PATCH 1/4] x86/cpu: Add and use new CPUID region helper

[Huang, Kai]

On Tue, 2024-04-02 at 10:13 -0700, Dave Hansen wrote:
> On 3/25/24 05:24, Huang, Kai wrote:
> > 
> > Nit:
> > 
> > > +
> > > +/* Returns true if the leaf exists and @value was populated */
> > 
> > 						 ^ is ?
> 
> It's a subtle difference, but I think it's better as I wrote it.
> Returning true happens *after* the value _was_ populated.
> 
> > > +static inline bool get_cpuid_region_leaf(u32 leaf, enum cpuid_regs_idx reg,
> > > +					 u32 *value)
> > > +{
> > > +	u16 region = leaf >> 16;
> > > +	u32 regs[4];
> > > +
> > > +	if (cpuid_region_max_leaf(region) < leaf)
> > > +		return false;
> > > +
> > > +	cpuid(leaf, &regs[CPUID_EAX], &regs[CPUID_EBX],
> > > +	            &regs[CPUID_ECX], &regs[CPUID_EDX]);
> > > +
> > > +	*value = regs[reg];
> > > +
> > > +	return true;
> > > +}
> > 
> > I found despite the get_cpuid_region_leaf() returns true/false, the return value
> > is never used in this series.  Instead, this series uses below pattern:
> > 
> > 	u32 data = 0; 	/* explicit initialization */
> > 
> > 	get_cpuid_region_leaf(leaf, ..., &data);
> > 
> > Which kinda implies the 'data' won't be touched if the requested leaf isn't
> > supported I suppose?
> > 
> > Since the return value is never used, should we consider just making this
> > function void?
> 
> I certainly considered it.
> 
> But I do think that get_cpuid_region_leaf() looks a lot more obviously
> correct and useful when it explicitly returns what it did, even if the
> existing callers don't take advantage of it.
> 
> I suspect it generates the same code either way.

Agreed:

Reviewed-by: Kai Huang <kai.huang@xxxxxxxxx>