Some page allocation paths that call post_alloc_hook() but skip
kmsan_alloc_page(), leaving stale KMSAN shadow on allocated pages.
Although there is no reproducer for this issue, this patch should
be able to fix it by explicitly calling kmsan_alloc_page()
after they successfully get new pages.

Reported-by: syzbot+2aee6839a252e612ce34@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=2aee6839a252e612ce34

Signed-off-by: Ke Zhao <ke.zhao.kernel@gmail.com>
---
Changes in v2:
- Use correct variable in alloc_contig_frozen_range_noprof() 
- Remove trace_mm_page_alloc() in alloc_contig_frozen_range_noprof()
  since we does not trace it in the above branch, suggested by Vlastimil Babka
- Link to v1: https://lore.kernel.org/r/20260330-fix-kmsan-v1-1-e9c672a4b9eb@gmail.com
---
 mm/page_alloc.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 2d4b6f1a554e..e08678b9e9cd 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -5189,6 +5189,10 @@ unsigned long alloc_pages_bulk_noprof(gfp_t gfp, int preferred_nid,
 
 		prep_new_page(page, 0, gfp, 0);
 		set_page_refcounted(page);
+
+		trace_mm_page_alloc(page, 0, gfp, ac.migratetype);
+		kmsan_alloc_page(page, 0, gfp);
+
 		page_array[nr_populated++] = page;
 	}
 
@@ -6911,6 +6915,12 @@ static void split_free_frozen_pages(struct list_head *list, gfp_t gfp_mask)
 			int i;
 
 			post_alloc_hook(page, order, gfp_mask);
+			/*
+			 * Initialize KMSAN state right after post_alloc_hook().
+			 * This prepares the pages for subsequent outer callers
+			 * that might free sub-pages after the split.
+			 */
+			kmsan_alloc_page(page, order, gfp_mask);
 			if (!order)
 				continue;
 
@@ -7117,6 +7127,8 @@ int alloc_contig_frozen_range_noprof(unsigned long start, unsigned long end,
 
 		check_new_pages(head, order);
 		prep_new_page(head, order, gfp_mask, 0);
+
+		kmsan_alloc_page(head, order, gfp_mask);
 	} else {
 		ret = -EINVAL;
 		WARN(true, "PFN range: requested [%lu, %lu), allocated [%lu, %lu)\n",

---
base-commit: bbeb83d3182abe0d245318e274e8531e5dd7a948
change-id: 20260325-fix-kmsan-e291f752a949

Best regards,
-- 
Ke Zhao <ke.zhao.kernel@gmail.com>

On Mon, 13 Apr 2026 10:12:40 +0800 Ke Zhao <ke.zhao.kernel@gmail.com> wrote:

> Some page allocation paths that call post_alloc_hook() but skip
> kmsan_alloc_page(), leaving stale KMSAN shadow on allocated pages.
> Although there is no reproducer for this issue, this patch should
> be able to fix it by explicitly calling kmsan_alloc_page()
> after they successfully get new pages.
> 

Please refresh, retest and resend this.

When doing so, please cc the people who commented on the v1 patch
(Vlastimil, Usama).

Please also cc the KMSAN developers, as listed in the MAINTAINERS file.

Thanks.