[PATCH] mm/mm_init: avoid overflow in percentage-based core sizing

[PATCH] mm/mm_init: avoid overflow in percentage-based core sizing

[Avi Weiss]

The kernelcore= and movablecore= percentage calculations multiply the
total number of pages before dividing. On 32-bit kernels, the
intermediate unsigned long product can overflow, producing incorrect
memory zone boundaries. With 2 GiB of RAM and kernelcore=100%, a
32-bit HIGHMEM kernel incorrectly creates ZONE_MOVABLE.

Use mult_frac() to calculate both percentages without overflowing
the intermediate product.

Fixes: a5c6d6509342 ("mm, page_alloc: extend kernelcore and movablecore for percent")
Signed-off-by: Avi Weiss <thnkslprpt@gmail.com>
---
 mm/mm_init.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/mm/mm_init.c b/mm/mm_init.c
index 2a0e782a151d..27a50ecd5c20 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -7,6 +7,7 @@
  *
  */
 #include <linux/kernel.h>
+#include <linux/math.h>
 #include <linux/init.h>
 #include <linux/kobject.h>
 #include <linux/export.h>
@@ -448,11 +449,9 @@ static void __init find_zone_movable_pfns_for_nodes(void)
 	 * amount of necessary memory.
 	 */
 	if (required_kernelcore_percent)
-		required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
-				       10000UL;
+		required_kernelcore = mult_frac(totalpages, required_kernelcore_percent, 100);
 	if (required_movablecore_percent)
-		required_movablecore = (totalpages * 100 * required_movablecore_percent) /
-					10000UL;
+		required_movablecore = mult_frac(totalpages, required_movablecore_percent, 100);
 
 	/*
 	 * If movablecore= was specified, calculate what size of

base-commit: 36aad21d50d1af3f6954bb0c988257a908202477
-- 
2.43.0

Re: [PATCH] mm/mm_init: avoid overflow in percentage-based core sizing

[Mike Rapoport]

From: Mike Rapoport (Microsoft) <rppt@kernel.org>

On Fri, 06 Nov 2026 09:23:55 +0200, Avi Weiss wrote:
> The kernelcore= and movablecore= percentage calculations multiply the
> total number of pages before dividing. On 32-bit kernels, the
> intermediate unsigned long product can overflow, producing incorrect
> memory zone boundaries. With 2 GiB of RAM and kernelcore=100%, a
> 32-bit HIGHMEM kernel incorrectly creates ZONE_MOVABLE.
> 
> Use mult_frac() to calculate both percentages without overflowing
> the intermediate product.
> 
> [...]

Applied to for-next branch of memblock.git tree, thanks!

[1/1] mm/mm_init: avoid overflow in percentage-based core sizing
      commit: 36557d74a757a4bb71bb1492449680177687f5a1

tree: https://git.kernel.org/pub/scm/linux/kernel/git/mm/memblock
branch: for-next

--
Sincerely yours,
Mike.