Re: [PATCH 1/2] perf/x86: Disable non generic regs for software/probe events

[Liang, Kan]

On 5/25/2019 4:48 AM, Ingo Molnar wrote:
> * kan.liang@xxxxxxxxxxxxxxx <kan.liang@xxxxxxxxxxxxxxx> wrote:
>
> > @@ -57,6 +57,11 @@ static unsigned int pt_regs_offset[PERF_REG_X86_MAX] = {
> >   #endif
> >   };
> >   
> > +u64 non_generic_regs_mask(void)
> > +{
> > +	return (~((1ULL << PERF_REG_X86_XMM0) - 1));
> > +}
> > +
> >   u64 perf_reg_value(struct pt_regs *regs, int idx)
> >   {
> >   	struct x86_perf_regs *perf_regs;
> > diff --git a/include/linux/perf_regs.h b/include/linux/perf_regs.h
> > index 4767474..c1c3454 100644
> > --- a/include/linux/perf_regs.h
> > +++ b/include/linux/perf_regs.h
> > @@ -9,6 +9,8 @@ struct perf_regs {
> >   	struct pt_regs	*regs;
> >   };
> >   
> > +u64 non_generic_regs_mask(void);
>
> This is a *constant* value, why is it in a separate function, not an
> inline?
>
> Or rather, since it's obviously a constant, name it in such a way.
> (PERF_REG_X86_NON_GENERIC_MASK or so. >
> To the generic code define it as 0 if arch headers haven't overriden it.

I will name it PERF_REG_NON_GENERIC_MASK in generic code.

Perf tool also defined a similar macro. I think I will define 
PERF_REG_NON_GENERIC_MASK in X86 uapi header. So both kernel and user 
space can use it.

I will send out V2 to address all comments.

Thanks,
Kan

> > +u64 __weak non_generic_regs_mask(void)
> > +{
> > +	return 0;
> > +}
> > +
> > +static inline bool has_non_generic_regs(struct perf_event *event)
> > +{
> > +	u64 mask = non_generic_regs_mask();
> > +
> > +	return ((event->attr.sample_regs_user & mask) ||
> > +		(event->attr.sample_regs_intr & mask));
>
> 'return' is not a function ...
>
> > +	/* only support generic regs */
> > +	if (has_non_generic_regs(event))
> > +		return -EOPNOTSUPP;
>
> In human readable comments please use complete sentences with no
> unnecessary abbreviations, i.e. "Only support generic registers".
>
> Thanks,
>
> 	Ingo