sparsemem: ensure we initialise the node mapping for SPARSEMEM_STATIC
authorAndy Whitcroft <apw@shadowen.org>
Wed, 22 Aug 2007 21:01:03 +0000 (14:01 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 23 Aug 2007 02:52:44 +0000 (19:52 -0700)
commit85770ffe4f0cdd4396b17f14762adc25a571a348
tree23f69f720e53cee67632b04c1754c3a7a64dab72
parentdf068464169a84a6a66c05d140f43a46d5eb6176
sparsemem: ensure we initialise the node mapping for SPARSEMEM_STATIC

Booting SPARSEMEM on NUMA systems trips a BUG in page_alloc.c:

Initializing HighMem for node 0 (00038000:00100000)
Initializing HighMem for node 1 (00100000:001ffe00)
------------[ cut here ]------------
kernel BUG at /home/apw/git/linux-2.6/mm/page_alloc.c:456!
[...]

This occurs because the section to node id mapping is not being
setup correctly during init under SPARSEMEM_STATIC, leading to an
attempt to free pages from all nodes into the zones on node 0.

When the zone_table[] was removed in the following commit, a new
section to node mapping table was introduced:

    commit 89689ae7f95995723fbcd5c116c47933a3bb8b13
    [PATCH] Get rid of zone_table[]

That conversion inadvertantly only initialised the node mapping in
SPARSEMEM_EXTREME.  Ensure we initialise the node mapping in
SPARSEMEM_STATIC.

[akpm@linux-foundation.org: make the stubs static inline]
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/sparse.c