Re: [PATCH 00/12] kmap_atomic cleanup for 3.6

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Mon, 2012-06-25 at 20:43 +0000, Arnd Bergmann wrote:
> On Monday 25 June 2012, Peter Zijlstra wrote:
> > 
> > On Mon, 2012-06-25 at 15:18 +0000, Arnd Bergmann wrote:
> > > > Different arch has different values for KM_TYPE_NR, I am not sure if
> > > > unifying them to a fixed value could fit all? 
> > 
> > No you can't. Some arch's have arch specific KM_TYPE thingies, like FRV.
> 
> Ah, right. Is it only FRV or are there any others?

FRV is the only one I can remember, but like always, just check all
archs.

> > > > For safety, I kept their original values.
> > > 
> > > My fear is that it will make it harder to clean that code up for
> > > real, when there is no longer an indication about where the number
> > > comes from.
> > 
> > Agreed, I'd much prefer it if we'd come up with a sane way to compute
> > the max value before doing away with these enums.
> > 
> > Sadly I haven't been able to come up with a sane way short of whole
> > program analysis.
> 
> How about putting that constant into asm/highmem.h then, and adding a
> default like
> 
> #ifndef KM_TYPE_NR
> #define KM_TYPE_NR 8
> #endif
> 
> in linux/highmem.h? Then FRV and anything else that needs it can override
> the value and the other ones don't need to bother.

At least put in a hand-wavy argument supporting whatever one number
that's being put in. That way a reader at least as some incling as to
where it comes from and what needs checking if it turns out its wrong.