Re: [PATCH] total_free_shortage() using zone_free_shortage()

[Linus Torvalds]

[ Hey, I can have long discussions by myself. I don't need you guys to
  answer me at all.

  This must be what senility feels like. Linus "doddering fool" Torvalds ]

On Tue, 7 Aug 2001, Linus Torvalds wrote:
>
> We should really document the ranges clearly. Right now (with these
> changes), the ranges would be (with users in parenthesis):
>
>  free (free + inactive_clean):
>
> 	low water mark: zone->pages_high	(__alloc_pages, zone_free_shortage)
> 	high water mark: zone->pages_high*2	(zone_free_plenty)

That's subtly wrong, __alloc_pages really has "zone->pages_low", not
"pages_high"  as the low water mark on when to start kswapd.

It does use "pages_high" too - but it is used as a "prefer this zone
because it isn't close to empty" marker. In short, it's really more of a
local high water mark.

So the free_shortage() logic really should be:
 - we want zone to have "pages_high" on _average_, but it's a shortage
   only if any zone is under the "pages_low" thing.

Which is actually exactly what you get if "total_free_shortage()" first
tests the global sums against "freepages.high", and then checks individual
zones with "zone_free_shortage()" against zone->pages_low. So I think I
happened to get that one right originally by pure luck.

		Linus