Re: [PATCH hw_breakpoint] percpu: add __percpu sparse annotations to hw_breakpoint

[Frederic Weisbecker <fweisbec@gmail.com>]

On Thu, Feb 18, 2010 at 09:49:08AM +0900, Tejun Heo wrote:
> Hello,
> 
> On 02/18/2010 01:39 AM, Frederic Weisbecker wrote:
> > On Wed, Feb 17, 2010 at 10:50:50AM +0900, Tejun Heo wrote:
> > Yeah, looks good, I'm queuing it.
> > Just few comments below, for nano-considerations.
> >>  	cpu_events = alloc_percpu(typeof(*cpu_events));
> >>  	if (!cpu_events)
> >> -		return ERR_PTR(-ENOMEM);
> >> +		return (void __percpu __force *)ERR_PTR(-ENOMEM);
> > 
> > Is this pattern common enough that we can think about a ERR_CPU_PTR ?
> 
> I thought about that but there aren't too many yet, so I just added
> the ugly castings.  It would be cool if sparse can be taught that
> ERR_PTR() returns universal pseudo pointer.



Yeah, it would be nice to just have a universal address space
that is compatible with all others. It's sad to see such
uglification to make a secondary tool happy.

 
> >>  	sample_hbp = register_wide_hw_breakpoint(&attr, sample_hbp_handler);
> >> -	if (IS_ERR(sample_hbp)) {
> >> -		ret = PTR_ERR(sample_hbp);
> >> +	if (IS_ERR((void __force *)sample_hbp)) {
> >> +		ret = PTR_ERR((void __force *)sample_hbp);
> > 
> > Same comments here, although I wouldn't like much a CPU_PTR_ERR or
> > IS_ERR_CPU.... CPP is just so poor in magic for that.
> > 
> > I must confess I miss a bit the old per_cpu prefix that guarded the implicit
> > separate namespace.
> 
> Yeap, I agree that the prefix had its advantages.  It's just that it
> can't scale to the new situation where static and dynamic percpu
> variables behave uniformly.


Well, I miss a bit of per cpu internals so I won't argue further :)

Thanks.