summaryrefslogtreecommitdiffstats
path: root/arch/x86/mm/numa.c
diff options
context:
space:
mode:
authorYinghai Lu <yinghai@kernel.org>2011-05-02 17:24:49 +0200
committerTejun Heo <tj@kernel.org>2011-05-02 17:24:49 +0200
commite5a10c1bd12a5d71bbb6406c1b0dbbc9d8958397 (patch)
tree1b8ee2a5cddd890e2058f167155f396ae9d69f40 /arch/x86/mm/numa.c
parenta56bca80db8903bb557b9ac38da68dc5b98ea672 (diff)
downloadop-kernel-dev-e5a10c1bd12a5d71bbb6406c1b0dbbc9d8958397.zip
op-kernel-dev-e5a10c1bd12a5d71bbb6406c1b0dbbc9d8958397.tar.gz
x86, NUMA: Trim numa meminfo with max_pfn in a separate loop
During testing 32bit numa unifying code from tj, found one system with more than 64g fails to use numa. It turns out we do not trim numa meminfo correctly against max_pfn in case start address of a node is higher than 64GiB. Bug fix made it to tip tree. This patch moves the checking and trimming to a separate loop. So we don't need to compare low/high in following merge loops. It makes the code more readable. Also it makes the node merge printouts less strange. On a 512GiB numa system with 32bit, before: > NUMA: Node 0 [0,a0000) + [100000,80000000) -> [0,80000000) > NUMA: Node 0 [0,80000000) + [100000000,1080000000) -> [0,1000000000) after: > NUMA: Node 0 [0,a0000) + [100000,80000000) -> [0,80000000) > NUMA: Node 0 [0,80000000) + [100000000,1000000000) -> [0,1000000000) Signed-off-by: Yinghai Lu <yinghai@kernel.org> [Updated patch description and comment slightly.] Signed-off-by: Tejun Heo <tj@kernel.org>
Diffstat (limited to 'arch/x86/mm/numa.c')
-rw-r--r--arch/x86/mm/numa.c15
1 files changed, 10 insertions, 5 deletions
diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 9a0ed31..f5510d8 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -270,6 +270,7 @@ int __init numa_cleanup_meminfo(struct numa_meminfo *mi)
const u64 high = PFN_PHYS(max_pfn);
int i, j, k;
+ /* first, trim all entries */
for (i = 0; i < mi->nr_blks; i++) {
struct numa_memblk *bi = &mi->blk[i];
@@ -278,10 +279,13 @@ int __init numa_cleanup_meminfo(struct numa_meminfo *mi)
bi->end = min(bi->end, high);
/* and there's no empty block */
- if (bi->start >= bi->end) {
+ if (bi->start >= bi->end)
numa_remove_memblk_from(i--, mi);
- continue;
- }
+ }
+
+ /* merge neighboring / overlapping entries */
+ for (i = 0; i < mi->nr_blks; i++) {
+ struct numa_memblk *bi = &mi->blk[i];
for (j = i + 1; j < mi->nr_blks; j++) {
struct numa_memblk *bj = &mi->blk[j];
@@ -311,8 +315,8 @@ int __init numa_cleanup_meminfo(struct numa_meminfo *mi)
*/
if (bi->nid != bj->nid)
continue;
- start = max(min(bi->start, bj->start), low);
- end = min(max(bi->end, bj->end), high);
+ start = min(bi->start, bj->start);
+ end = max(bi->end, bj->end);
for (k = 0; k < mi->nr_blks; k++) {
struct numa_memblk *bk = &mi->blk[k];
@@ -332,6 +336,7 @@ int __init numa_cleanup_meminfo(struct numa_meminfo *mi)
}
}
+ /* clear unused ones */
for (i = mi->nr_blks; i < ARRAY_SIZE(mi->blk); i++) {
mi->blk[i].start = mi->blk[i].end = 0;
mi->blk[i].nid = NUMA_NO_NODE;
OpenPOWER on IntegriCloud