On Wed, Feb 17, 2021 at 02:30:43PM +0100, Michal Hocko wrote:

On Wed 17-02-21 11:08:15, Oscar Salvador wrote:
I do not think fallback to a different zone is ok. If yes then this
really requires a very good reasoning. alloc_contig_range is an
optimistic allocation interface at best and it shouldn't break carefully
node aware preallocation done by administrator.

Yeah, previous version (RFC) was more careful with that.
I somehow thought that it might be ok to fallback to other nodes in case
we failed to allocate on the preferred nid.

I will get RFC handling back wrt. allocation once I gather more feedback.

quoted

Note that gigantic hugetlb pages are fenced off since there is a cyclic
dependency between them and alloc_contig_range.

Why do we need/want to do all this in the first place?

When trying to allocate a memory chunk with alloc_contig_range, it will fail
if it ever sees a Hugetlb page because isolate_migratepages_range() does not
"recognize" those for migration (or to put it different, does not know about
them).

Given that HugeTLB pages can be migrated or dissolved if free, it makes sense
to enable isolate_migratepages_range() to recognize HugeTLB pages in order
to handle them, as it currently does with LRU and Movable pages.

-- 
Oscar Salvador
SUSE L3