On Thu 23-08-18 16:00:53, Oscar Salvador wrote:

On Thu, Aug 23, 2018 at 03:25:26PM +0200, Michal Hocko wrote:

quoted

On Thu 23-08-18 21:07:32, Wei Yang wrote:

quoted

And it is known, __highest_present_section_nr is a more strict boundary
than NR_MEM_SECTIONS.

This patch uses a __highest_present_section_nr to check a valid pfn.

But why is this an improvement? Sure when you loop over all sections
than __highest_present_section_nr makes a lot of sense. But all the
updated function perform a trivial comparision.

I think it makes some sense.
NR_MEM_SECTIONS can be a big number, but we might not be using
all sections, so __highest_present_section_nr ends up being a much lower
value.

And how exactly does it help to check for the smaller vs. a larger number?
Both are O(1) operations AFAICS. __highest_present_section_nr makes
perfect sense when we iterate over all sections or similar operations
where it smaller number of iterations really makes sense.

I am not saying the patch is wrong but I just do not see this being an
improvement. You have to export an internal symbol to achieve something
that is hardly an optimization.
-- 
Michal Hocko
SUSE Labs