Re: [PATCH] Add nid sanity on alloc_pages_node

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 13 Jul 2007 01:29:06 -0700 Paul Jackson <pj@sgi.com> wrote:

> > I'm scratching my head over that min_t in __first_node(), too.   I don't think
> > it's possible for find_first_bit(..., N) to return anything >N _anyway_.  And if
> > it does, we want to know about it.
> > 
> > <looks at Paul>
> 
> I'm not sure I've got this right, but looks like that min_t went in after
> Zwane Mwaikambo, then <zwane@fsmlabs.com>, whom I am presuming is the same
> person as now at <zwane@arm.linux.org.uk>, found a problem with the i386
> find_next_bit implementation returning > N when merging i386 cpu hotplug.

Ah, Zwane was involved - say no more ;)

> See the thread:
> 
>   http://lkml.org/lkml/2004/7/31/102
>   [PATCH][2.6] first/next_cpu returns values > NR_CPUS
> 
> I apparently lobbied at the time to mandate that find_first_bit(..., N)
> return exactly N on failure to find a set bit, but gave up, after some
> confusions on my part.

iirc, find_first_bit(..., N) _must_ return N on nothing-found.  It'd be
untidy to return some randomly-larger number.

I wonder which was the culpable architecture?  Oh, i386.

Note how the i386 implementation's documentation carefully avoids describing
the return value.  I don't think _any_ of our find_foo_bit()
implementations have return-value docs, and here we see the result.

Sigh.  What crap.  I guess we leave it as-is.