Skip to contentSkip to conversation

mm/hugetlb: fix hugetlb cgroup rsvd charge/uncharge mismatch

PATCH Applied Deepanshu Kartikey 3 participants 4 messages Mar 28 – Apr 26

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm/hugetlb: fix hugetlb cgroup rsvd charge/uncharge mismatch

In alloc_hugetlb_folio(), a single h_cg pointer is used for both the rsvd and non-rsvd hugetlb cgroup charges. When map_chg is set, hugetlb_cgroup_charge_cgroup_rsvd() stores the charged cgroup in h_cg, but the immediately following hugetlb_cgroup_charge_cgroup() overwrites h_cg with the non-rsvd cgroup pointer.

As a result, hugetlb_cgroup_commit_charge_rsvd() stores the wrong (non-rsvd) cgroup pointer into the folio's rsvd slot.

When the folio is later freed, free_huge_folio() unconditionally calls both hugetlb_cgroup_uncharge_folio() and hugetlb_cgroup_uncharge_folio_rsvd(). The rsvd uncharge reads back the wrong cgroup from the folio and decrements a counter that was never charged for that cgroup, causing a page_counter underflow:

page_counter underflow: -512 nr_pages=512

  WARNING: mm/page_counter.c:61 at page_counter_cancel

Fix this by introducing a separate h_cg_rsvd pointer exclusively for the rsvd charge path, keeping the rsvd and non-rsvd charges fully independent through their charge, commit, and error uncharge paths.

Fixes 08cf9faf7558 ("hugetlb_cgroup: support noreserve mappings")
Reported-by [email protected]
---
 mm/hugetlb.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
mm/hugetlb.c +4 -3
index 327eaa4074d3..5be36a888e70 100644
@@ -2915,6 +2915,7 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 	map_chg_state map_chg;
 	int ret, idx;
 	struct hugetlb_cgroup *h_cg = NULL;
+	struct hugetlb_cgroup *h_cg_rsvd = NULL;
 	gfp_t gfp = htlb_alloc_mask(h) | __GFP_RETRY_MAYFAIL;
 
 	idx = hstate_index(h);
@@ -2965,7 +2966,7 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 	 */
 	if (map_chg) {
 		ret = hugetlb_cgroup_charge_cgroup_rsvd(
-			idx, pages_per_huge_page(h), &h_cg);
+			idx, pages_per_huge_page(h), &h_cg_rsvd);
 		if (ret)
 			goto out_subpool_put;
 	}
@@ -3007,7 +3008,7 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 	 */
 	if (map_chg) {
 		hugetlb_cgroup_commit_charge_rsvd(idx, pages_per_huge_page(h),
-						  h_cg, folio);
+						  h_cg_rsvd, folio);
 	}
 
 	spin_unlock_irq(&hugetlb_lock);
@@ -3059,7 +3060,7 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 out_uncharge_cgroup_reservation:
 	if (map_chg)
 		hugetlb_cgroup_uncharge_cgroup_rsvd(idx, pages_per_huge_page(h),
-						    h_cg);
+						    h_cg_rsvd);
 out_subpool_put:
 	/*
 	 * put page to subpool iff the quota of subpool's rsv_hpages is used
Signature · 1 line
2.43.0
maintainer
#1 lore raw
Quoted 15 lines · On Sat, 28 Mar 2026 12:25:34 +0530 Deepanshu Kartikey <[email protected]>…

In alloc_hugetlb_folio(), a single h_cg pointer is used for both the rsvd and non-rsvd hugetlb cgroup charges. When map_chg is set, hugetlb_cgroup_charge_cgroup_rsvd() stores the charged cgroup in h_cg, but the immediately following hugetlb_cgroup_charge_cgroup() overwrites h_cg with the non-rsvd cgroup pointer.

As a result, hugetlb_cgroup_commit_charge_rsvd() stores the wrong (non-rsvd) cgroup pointer into the folio's rsvd slot.

When the folio is later freed, free_huge_folio() unconditionally calls both hugetlb_cgroup_uncharge_folio() and hugetlb_cgroup_uncharge_folio_rsvd(). The rsvd uncharge reads back the wrong cgroup from the folio and decrements a counter that was never charged for that cgroup, causing a page_counter underflow:

Quoted 7 lines

page_counter underflow: -512 nr_pages=512

  WARNING: mm/page_counter.c:61 at page_counter_cancel

Fix this by introducing a separate h_cg_rsvd pointer exclusively for the rsvd charge path, keeping the rsvd and non-rsvd charges fully independent through their charge, commit, and error uncharge paths.

Thanks.

Quoted 1 line
Fixes 08cf9faf7558 ("hugetlb_cgroup: support noreserve mappings")

Merged in 2020!

Could reviewers please give consideration to whether we should backport this?

Quoted 3 lines

This doesn't seem super-urgent so for now I'll park it in my pile to revisit after 7.1-rc1.

maintainer
#2 lore raw
Quoted 25 lines · On Mon, 30 Mar 2026 13:15:25 -0700 Andrew Morton <[email protected]>…
Quoted 15 lines · On Sat, 28 Mar 2026 12:25:34 +0530 Deepanshu Kartikey <[email protected]>…

In alloc_hugetlb_folio(), a single h_cg pointer is used for both the rsvd and non-rsvd hugetlb cgroup charges. When map_chg is set, hugetlb_cgroup_charge_cgroup_rsvd() stores the charged cgroup in h_cg, but the immediately following hugetlb_cgroup_charge_cgroup() overwrites h_cg with the non-rsvd cgroup pointer.

As a result, hugetlb_cgroup_commit_charge_rsvd() stores the wrong (non-rsvd) cgroup pointer into the folio's rsvd slot.

When the folio is later freed, free_huge_folio() unconditionally calls both hugetlb_cgroup_uncharge_folio() and hugetlb_cgroup_uncharge_folio_rsvd(). The rsvd uncharge reads back the wrong cgroup from the folio and decrements a counter that was never charged for that cgroup, causing a page_counter underflow:

Quoted 7 lines

page_counter underflow: -512 nr_pages=512

  WARNING: mm/page_counter.c:61 at page_counter_cancel

Fix this by introducing a separate h_cg_rsvd pointer exclusively for the rsvd charge path, keeping the rsvd and non-rsvd charges fully independent through their charge, commit, and error uncharge paths.

Quoted 9 lines

Thanks.

Quoted 1 line
Fixes 08cf9faf7558 ("hugetlb_cgroup: support noreserve mappings")

Merged in 2020!

Could reviewers please give consideration to whether we should backport this?

OK, then ;)

I'll queue this up and shall add the cc:stable - that underflow warning needs to be addressed.

I'll add a needs-review note-to-self.

From: Deepanshu Kartikey <[email protected]>
Subject: mm/hugetlb: fix hugetlb cgroup rsvd charge/uncharge mismatch
Date: Sat, 28 Mar 2026 12:25:34 +0530

In alloc_hugetlb_folio(), a single h_cg pointer is used for both the rsvd and non-rsvd hugetlb cgroup charges. When map_chg is set, hugetlb_cgroup_charge_cgroup_rsvd() stores the charged cgroup in h_cg, but the immediately following hugetlb_cgroup_charge_cgroup() overwrites h_cg with the non-rsvd cgroup pointer.

As a result, hugetlb_cgroup_commit_charge_rsvd() stores the wrong (non-rsvd) cgroup pointer into the folio's rsvd slot.

When the folio is later freed, free_huge_folio() unconditionally calls both hugetlb_cgroup_uncharge_folio() and hugetlb_cgroup_uncharge_folio_rsvd(). The rsvd uncharge reads back the wrong cgroup from the folio and decrements a counter that was never charged for that cgroup, causing a page_counter underflow:

page_counter underflow: -512 nr_pages=512

  WARNING: mm/page_counter.c:61 at page_counter_cancel

Fix this by introducing a separate h_cg_rsvd pointer exclusively for the rsvd charge path, keeping the rsvd and non-rsvd charges fully independent through their charge, commit, and error uncharge paths.

Fixes 08cf9faf7558 ("hugetlb_cgroup: support noreserve mappings")
Reported-by [email protected]
Cc David Hildenbrand <[email protected]>
Cc Muchun Song <[email protected]>
Cc Oscar Salvador <[email protected]>
Cc Mina Almasry <[email protected]>

---

 mm/hugetlb.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
mm/hugetlb.c renamed from mm/hugetlb.c~mm-hugetlb-fix-hugetlb-cgroup-rsvd-charge-uncharge-mismatch +4 -3
@@ -2879,6 +2879,7 @@ struct folio *alloc_hugetlb_folio(struct
 	map_chg_state map_chg;
 	int ret, idx;
 	struct hugetlb_cgroup *h_cg = NULL;
+	struct hugetlb_cgroup *h_cg_rsvd = NULL;
 	gfp_t gfp = htlb_alloc_mask(h) | __GFP_RETRY_MAYFAIL;
 
 	idx = hstate_index(h);
@@ -2929,7 +2930,7 @@ struct folio *alloc_hugetlb_folio(struct
 	 */
 	if (map_chg) {
 		ret = hugetlb_cgroup_charge_cgroup_rsvd(
-			idx, pages_per_huge_page(h), &h_cg);
+			idx, pages_per_huge_page(h), &h_cg_rsvd);
 		if (ret)
 			goto out_subpool_put;
 	}
@@ -2971,7 +2972,7 @@ struct folio *alloc_hugetlb_folio(struct
 	 */
 	if (map_chg) {
 		hugetlb_cgroup_commit_charge_rsvd(idx, pages_per_huge_page(h),
-						  h_cg, folio);
+						  h_cg_rsvd, folio);
 	}
 
 	spin_unlock_irq(&hugetlb_lock);
@@ -3023,7 +3024,7 @@ out_uncharge_cgroup:
 out_uncharge_cgroup_reservation:
 	if (map_chg)
 		hugetlb_cgroup_uncharge_cgroup_rsvd(idx, pages_per_huge_page(h),
-						    h_cg);
+						    h_cg_rsvd);
 out_subpool_put:
 	/*
 	 * put page to subpool iff the quota of subpool's rsv_hpages is used

_

maintainer Reviewed-by on v1 1/1
#3 lore raw
In reply to Deepanshu Kartikey
Quoted 27 lines · Deepanshu Kartikey

In alloc_hugetlb_folio(), a single h_cg pointer is used for both the rsvd and non-rsvd hugetlb cgroup charges. When map_chg is set, hugetlb_cgroup_charge_cgroup_rsvd() stores the charged cgroup in h_cg, but the immediately following hugetlb_cgroup_charge_cgroup() overwrites h_cg with the non-rsvd cgroup pointer.

As a result, hugetlb_cgroup_commit_charge_rsvd() stores the wrong (non-rsvd) cgroup pointer into the folio's rsvd slot.

When the folio is later freed, free_huge_folio() unconditionally calls both hugetlb_cgroup_uncharge_folio() and hugetlb_cgroup_uncharge_folio_rsvd(). The rsvd uncharge reads back the wrong cgroup from the folio and decrements a counter that was never charged for that cgroup, causing a page_counter underflow:

page_counter underflow: -512 nr_pages=512

 WARNING: mm/page_counter.c:61 at page_counter_cancel

Fix this by introducing a separate h_cg_rsvd pointer exclusively for the rsvd charge path, keeping the rsvd and non-rsvd charges fully independent through their charge, commit, and error uncharge paths.

Quoted 4 lines
Fixes 08cf9faf7558 ("hugetlb_cgroup: support noreserve mappings")
Reported-by [email protected]

Thanks.