[PATCH 05/12] mm: alloc_contig_range() added

[dave@linux.vnet.ibm.com (Dave Hansen)]

On Thu, 2011-03-31 at 18:26 +0200, Michal Nazarewicz wrote:
> > On Thu, 2011-03-31 at 15:16 +0200, Marek Szyprowski wrote:
> >> +       ret = 0;
> >> +       while (!PageBuddy(pfn_to_page(start & (~0UL << ret))))
> >> +               if (WARN_ON(++ret >= MAX_ORDER))
> >> +                       return -EINVAL;
> 
> On Thu, 31 Mar 2011 18:02:41 +0200, Dave Hansen wrote:
> > Holy cow, that's dense.  Is there really no more straightforward way to
> > do that?
> 
> Which part exactly is dense?  What would be qualify as a more
> straightforward way?

I'm still not 100% sure what it's trying to do.  It looks like it
attempts to check all of "start"'s buddy pages.

unsigned long find_buddy(unsigned long pfn, int buddy)
{
	unsigned long page_idx = pfn & ((1 << MAX_ORDER) - 1); // You had a macro for this I think
	unsigned long buddy_idx = __find_buddy_index(page_idx, order);
	return page_idx + buddy_idx;
}

Is something like this equivalent?

int order;
for (order = 0; order <= MAX_ORDER; order++) {
	unsigned long buddy_pfn = find_buddy(start, order);
	struct page *buddy = pfn_to_page(buddy_pfn);
	if (PageBuddy(buddy)
		break;
	WARN();
	return -EINVAL;
}

I'm wondering also if you can share some code with __rmqueue().

> > In any case, please pull the ++ret bit out of the WARN_ON().  Some
> > people like to do:
> >
> > #define WARN_ON(...) do{}while(0)
> >
> > to save space on some systems.
> 
> I don't think that's the case.  Even if WARN_ON() decides not to print
> a warning, it will still return the value of the argument.  If not,
> a lot of code will brake.

Bah, sorry.  I'm confusing WARN_ON() and WARN().

-- Dave