On Mon, 24 Feb 2020, Michal Hocko wrote:

Hmm, nasty. Is there any reason why kmalloc_node behaves differently
from the page allocator?

The page allocator will do the same thing if you pass GFP_THISNODE and
insist on allocating memory from a node that does not exist.

quoted

quoted

A short summary. kmalloc_node blows up when trying to allocate from a
memory less node.

Use kmalloc instead? And set a memory allocation policy?

The current code (memcg_expand_one_shrinker_map resp. memcg_alloc_shrinker_maps)
already use kvmalloc. Kirill's patch wanted to make those data structure
on the respective node and kvmalloc_node sounded like the right thing to
do. It comes as a surprise that the kernel simply blows up on a memory
less node rather than falling back to a close node gracefully. I suspect
this already happens when the target node is out of memory, right?

No. If the target node is out of memory then direct reclaim is going to be
invovked.

How would a memory allocation policy help in this case btw.?

It would allow fallback to other nodes.