Aslan found a bug in the per-node hugetlb_cma area size calculation:
the total remaining size should cap the per-node area size instead
of be the minimal possible allocation.
Without the fix:
[ 0.004136] hugetlb_cma: reserve 2048 MiB, up to 1024 MiB per node
[ 0.004138] cma: Reserved 2048 MiB at 0x0000000180000000
[ 0.004139] hugetlb_cma: reserved 2048 MiB on node 0
With the fix:
[ 0.006780] hugetlb_cma: reserve 2048 MiB, up to 1024 MiB per node
[ 0.006786] cma: Reserved 1024 MiB at 0x00000001c0000000
[ 0.006787] hugetlb_cma: reserved 1024 MiB on node 0
[ 0.006788] cma: Reserved 1024 MiB at 0x00000003c0000000
[ 0.006789] hugetlb_cma: reserved 1024 MiB on node 1
Reported-by: Aslan Barkirov <aslan@xxxxxx>
Signed-off-by: Roman Gushchin <guro@xxxxxx>
---
mm/hugetlb.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 456f4c010fea..44b47c2b6fab 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -5590,7 +5590,7 @@ void __init hugetlb_cma_reserve(int order)
min_pfn = min_low_pfn;
max_pfn = max_low_pfn;
#endif
- size = max(per_node, hugetlb_cma_size - reserved);
+ size = min(per_node, hugetlb_cma_size - reserved);
size = round_up(size, PAGE_SIZE << order);
if (size > ((max_pfn - min_pfn) << PAGE_SHIFT) / 2) {
--
2.25.1
