Re: "mm: use early_pfn_to_nid in page_ext_init" broken on some configurations?

[Michal Hocko <mhocko@kernel.org>]

On Fri 30-06-17 16:18:47, Michal Hocko wrote:
> fe53ca54270a ("mm: use early_pfn_to_nid in page_ext_init") seem
> to silently depend on CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID resp.
> CONFIG_HAVE_MEMBLOCK_NODE_MAP. early_pfn_to_nid is returning zero with
> !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
> I am not sure how widely is this used but such a code is tricky. I see
> how catching early allocations during defered initialization might be
> useful but a subtly broken code sounds like a problem to me.  So is
> fe53ca54270a worth this or we should revert it?

I've dug little bit further. It seems that only s390 and ia64 select
HAVE_ARCH_EARLY_PFN_TO_NID. Much more architectures enabled
HAVE_MEMBLOCK_NODE_MAP though but still
alpha, arc, arm, avr32, blackfin, c6x, cris, frv, h8300, hexagon,
Kconfig, m32r, m68k, mn10300, nios2, openrisc, parisc, tile, um,
unicore32, xtensa do not. I can only see alpha having NUMA and even that
is marked BROKEN. So it seems that this is not a real problem after all.

Still subtle, so I guess we want to have the following. What do you
think?

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 16532fa0bb64..894697c1e6f5 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -1055,6 +1055,7 @@ static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
 	!defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
 {
+	BUILD_BUG_ON(!IS_ENABLED(CONFIG_NUMA));
 	return 0;
 }
 #endif
-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>