On Wed, 22 Jul 2015, Vlastimil Babka wrote:
> > alloc_pages_exact_node(), as you said, connotates that the allocation will
> > take place on that node or will fail. So why not go beyond this patch and
> > actually make alloc_pages_exact_node() set __GFP_THISNODE and then call
> > into a new alloc_pages_prefer_node(), which would be the current
> > alloc_pages_exact_node() implementation, and then fix up the callers?
>
> OK, but then we have alloc_pages_node(), alloc_pages_prefer_node() and
> alloc_pages_exact_node(). Isn't that a bit too much? The first two
> differ only in tiny bit:
>
> static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
> unsigned int order)
> {
> /* Unknown node is current node */
> if (nid < 0)
> nid = numa_node_id();
>
> return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
> }
>
> static inline struct page *alloc_pages_prefer_node(int nid, gfp_t gfp_mask,
> unsigned int order)
> {
> VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
>
> return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
> }
>
Eek, yeah, that does look bad. I'm not even sure the
if (nid < 0)
nid = numa_node_id();
is correct; I think this should be comparing to NUMA_NO_NODE rather than
all negative numbers, otherwise we silently ignore overflow and nobody
ever knows.
> So _prefer_node is just a tiny optimization over the other one. It
> should be maybe called __alloc_pages_node() then? This would perhaps
> discourage users outside of mm/arch code (where it may matter). The
> savings of a skipped branch is likely dubious anyway... It would be also
> nice if alloc_pages_node() could use __alloc_pages_node() internally, but
> I'm not sure if all callers are safe wrt the
> VM_BUG_ON(!node_online(nid)) part.
>
I'm not sure how large you want to make your patch :) In a perfect world
I would think that we wouldn't have an alloc_pages_prefer_node() at all
and everything would be converted to alloc_pages_node() which would do
if (nid == NUMA_NO_NODE)
nid = numa_mem_id();
VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
and then alloc_pages_exact_node() would do
return alloc_pages_node(nid, gfp_mask | __GFP_THISNODE, order);
and existing alloc_pages_exact_node() callers fixed up depending on
whether they set the bit or not.
The only possible downside would be existing users of
alloc_pages_node() that are calling it with an offline node. Since it's a
VM_BUG_ON() that would catch that, I think it should be changed to a
VM_WARN_ON() and eventually fixed up because it's nonsensical.
VM_BUG_ON() here should be avoided.
Or just go with a single alloc_pages_node() and rename __GFP_THISNODE to
__GFP_EXACT_NODE which may accomplish the same thing :)
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>