[PATCH v8 0/6] mm/memory-failure: add panic option for unrecoverable pages

[PATCH v8 0/6] mm/memory-failure: add panic option for unrecoverable pages

[Breno Leitao]

A multi-bit ECC error on a kernel-owned page that the memory failure
handler cannot recover is currently swallowed: PG_hwpoison is set, the
event is logged, and the kernel keeps running.  The corrupted memory
remains accessible to the kernel and either drives silent data
corruption or surfaces seconds-to-minutes later as an apparently
unrelated crash.  In a large fleet that delayed, unattributable crash
turns into significant engineering effort to root-cause; in a kdump
configuration, by the time the crash happens the original error
context (faulting PFN, MCE/GHES record, page state) is long gone.

This series adds an opt-in sysctl,
vm.panic_on_unrecoverable_memory_failure, that converts an
unrecoverable kernel-page hwpoison event into an immediate panic with
a clean dmesg/vmcore that still contains the original failure
context.  The default is disabled so existing workloads see no
change.

There is a selftest that test different cases, and I tested it using
the following variants:

  ┌─────────┬──────────┬───────────────────────────────────────────────────────────┐
  │ Variant │   PFN    │                          Result                           │
  ├─────────┼──────────┼───────────────────────────────────────────────────────────┤
  │ rodata  │ 0x2600   │ Panic with "Memory failure: 0x2600: unrecoverable page"   │
  ├─────────┼──────────┼───────────────────────────────────────────────────────────┤
  │ slab    │ 0x100032 │ Panic with "Memory failure: 0x100032: unrecoverable page" │
  ├─────────┼──────────┼───────────────────────────────────────────────────────────┤
  │ pgtable │ 0x100000 │ Panic with "Memory failure: 0x100000: unrecoverable page" │
  └─────────┴──────────┴───────────────────────────────────────────────────────────┘

Each one shows the same call trace, exactly the path the series builds:

  hard_offline_page_store
    → memory_failure
      → action_result
        → panic("Memory failure: %#lx: unrecoverable page")

Signed-off-by: Breno Leitao <leitao@debian.org>
---
Changes in v8:
- Commit message rewording (David)
- Add HWPoisonKernelOwned() helper (Lance)
- Removed patch "mm/memory-failure: short-circuit PG_reserved before get_hwpoison_page()"
- Broaden the selftest (Lance)
- Link to v7: https://patch.msgid.link/20260513-ecc_panic-v7-0-be2e578e61da@debian.org

Changes in v7:
- Move the PG_reserved / unhandlable-kernel-page classification into
  get_any_page() and surface it via -ENOTRECOVERABLE, per David
  Hildenbrand's and Lance Yang's review of v6.  This drops the
  is_reserved snapshot in memory_failure() and the mf_get_page_status
  enum / out-parameter introduced in v6.
- Restructure the post-call branch in memory_failure() as a switch
  over the get_hwpoison_page() return code (David).
- Drop the "reserved" qualifier from the MF_MSG_KERNEL label and the
  matching tracepoint string; the enum now covers both PG_reserved
  pages and other unhandlable kernel pages.
- Squash the former patches 1/4 ("MF_MSG_KERNEL for reserved pages")
  and 2/4 ("classify get_any_page() failures by reason") into a
  single classification patch; the series is now 3 patches.
- Simplify panic_on_unrecoverable_mf() to a single return statement
  (David).
- Link to v6: https://patch.msgid.link/20260511-ecc_panic-v6-0-183012ba7d4b@debian.org

Changes in v6:
- Dropped the selftest given the value was not clear
- Get the status of the failure from get_any_page()
- Small nits from different people/AIs.
- Link to v5: https://patch.msgid.link/20260424-ecc_panic-v5-0-a35f4b50425c@debian.org

Changes in v5:
- Add vm.panic_on_unrecoverable_memory_failure sysctl to panic on
  unrecoverable kernel page hwpoison events (reserved pages, refcount-0
  non-buddy pages, unknown state), with a recheck to avoid racing with
  concurrent buddy allocations. (Miaohe)
- Distinguish reserved pages as MF_MSG_KERNEL in memory_failure(),
  document the new sysctl in Documentation/admin-guide/sysctl/vm.rst,
  and add a selftest verifying SIGBUS recovery on userspace pages still
  works when the sysctl is enabled. (Miaohe)
- Added a selftest
- Link to v4:
  https://patch.msgid.link/20260415-ecc_panic-v4-0-2d0277f8f601@debian.org

Changes in v4:
- Drop CONFIG_BOOTPARAM_MEMORY_FAILURE_PANIC kernel configuration option.
- Split the reserved page classification (MF_MSG_KERNEL) into its own
  patch, separate from the panic mechanism.
- Document why the buddy allocator TOCTOU race (between
  get_hwpoison_page() and is_free_buddy_page()) cannot cause false
  positives: PG_hwpoison is set beforehand and check_new_page() in the
  page allocator rejects hwpoisoned pages.
- Document the narrow LRU isolation race window for MF_MSG_UNKNOWN and
  its mitigation via identify_page_state()'s two-pass design.
- Explicitly document why MF_MSG_GET_HWPOISON is excluded from the
  panic conditions (shared path with transient races and non-reserved
  kernel memory).
- Link to v3: https://patch.msgid.link/20260413-ecc_panic-v3-0-1dcbb2f12bc4@debian.org

Changes in v3:
- Rename is_unrecoverable_memory_failure() to panic_on_unrecoverable_mf()
  as suggested by maintainer.
- Add CONFIG_BOOTPARAM_MEMORY_FAILURE_PANIC kernel configuration option,
  similar to CONFIG_BOOTPARAM_HARDLOCKUP_PANIC.
- Add documentation for the sysctl and CONFIG option.
- Add code comments documenting the panic condition design rationale and
  how the retry mechanism mitigates false positives from buddy allocator
  races.
- Link to v2: https://patch.msgid.link/20260331-ecc_panic-v2-0-9e40d0f64f7a@debian.org

Changes in v2:
- Panic on MF_MSG_KERNEL, MF_MSG_KERNEL_HIGH_ORDER and MF_MSG_UNKNOWN
  instead of MF_MSG_GET_HWPOISON.
- Report MF_MSG_KERNEL for reserved pages when get_hwpoison_page() fails
  instead of MF_MSG_GET_HWPOISON.
- Link to v1: https://patch.msgid.link/20260323-ecc_panic-v1-0-72a1921726c5@debian.org

To: Miaohe Lin <linmiaohe@huawei.com>
To: Naoya Horiguchi <nao.horiguchi@gmail.com>
To: Andrew Morton <akpm@linux-foundation.org>
To: Steven Rostedt <rostedt@goodmis.org>
To: Masami Hiramatsu <mhiramat@kernel.org>
To: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
To: Jonathan Corbet <corbet@lwn.net>
To: Shuah Khan <skhan@linuxfoundation.org>
To: David Hildenbrand <david@kernel.org>
To: Lorenzo Stoakes <ljs@kernel.org>
To: "Liam R. Howlett" <liam@infradead.org>
To: Vlastimil Babka <vbabka@kernel.org>
To: Mike Rapoport <rppt@kernel.org>
To: Suren Baghdasaryan <surenb@google.com>
To: Michal Hocko <mhocko@suse.com>
To: Shuah Khan <shuah@kernel.org>
Cc: linux-mm@kvack.org
Cc: linux-kernel@vger.kernel.org
Cc: linux-trace-kernel@vger.kernel.org
Cc: linux-doc@vger.kernel.org
Cc: linux-kselftest@vger.kernel.org

---
Breno Leitao (6):
      mm/memory-failure: drop dead error_states[] entry for reserved pages
      mm/memory-failure: surface unhandlable kernel pages as -ENOTRECOVERABLE
      mm/memory-failure: report MF_MSG_KERNEL for unrecoverable kernel pages
      mm/memory-failure: add panic option for unrecoverable pages
      Documentation: document panic_on_unrecoverable_memory_failure sysctl
      selftests/mm: add hwpoison-panic destructive test

 Documentation/admin-guide/sysctl/vm.rst      |  85 ++++++++++++
 mm/memory-failure.c                          |  96 ++++++++++---
 tools/testing/selftests/mm/Makefile          |   1 +
 tools/testing/selftests/mm/hwpoison-panic.sh | 193 +++++++++++++++++++++++++++
 4 files changed, 357 insertions(+), 18 deletions(-)
---
base-commit: e7e28506af98ce4e1059e5ec59334b335c00a246
change-id: 20260323-ecc_panic-4e473b83087c

Best regards,
--  
Breno Leitao <leitao@debian.org>

[PATCH v8 1/6] mm/memory-failure: drop dead error_states[] entry for reserved pages

[Breno Leitao]

The first entry of error_states[],

	{ reserved,	reserved,	MF_MSG_KERNEL,	me_kernel },

is unreachable.  identify_page_state() has two callers, and neither
one can dispatch a PG_reserved page to me_kernel():

  * memory_failure() reaches identify_page_state() only after
    get_hwpoison_page() returned 1.  get_any_page() reaches that
    return only via __get_hwpoison_page(), which only takes a
    refcount when the page is HWPoisonHandlable().
    HWPoisonHandlable() is an allowlist for LRU, free-buddy, and
    (for soft-offline) movable_ops pages -- PG_reserved pages do
    not satisfy any of these, so they fail with -EBUSY/-EIO long
    before identify_page_state() runs.

  * try_memory_failure_hugetlb() reaches identify_page_state() only
    via the MF_HUGETLB_IN_USED branch, where the page is necessarily
    a hugetlb folio.  hugetlb folios don't carry PG_reserved at that
    point: hugetlb_folio_init_vmemmap() calls __folio_clear_reserved()
    during init, so the reserved entry would not match even if it
    were still present.

me_kernel() never executes and the entry exists only to be matched
against by code that cannot see it.

Drop the entry, the me_kernel() helper, and the now-unused
"reserved" macro.  Leave the MF_MSG_KERNEL enum value in place: it
remains part of the tracepoint and pr_err() string tables, and
follow-on work to classify unrecoverable kernel pages can reuse it
without churning the user-visible enum.

No functional change.

Suggested-by: David Hildenbrand <david@kernel.org>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Reviewed-by: Lance Yang <lance.yang@linux.dev>
Acked-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Breno Leitao <leitao@debian.org>
---
 mm/memory-failure.c | 14 --------------
 1 file changed, 14 deletions(-)

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 51508a55c405..f4d3e6e20e13 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -980,17 +980,6 @@ static bool has_extra_refcount(struct page_state *ps, struct page *p,
 	return false;
 }
 
-/*
- * Error hit kernel page.
- * Do nothing, try to be lucky and not touch this instead. For a few cases we
- * could be more sophisticated.
- */
-static int me_kernel(struct page_state *ps, struct page *p)
-{
-	unlock_page(p);
-	return MF_IGNORED;
-}
-
 /*
  * Page in unknown state. Do nothing.
  * This is a catch-all in case we fail to make sense of the page state.
@@ -1199,10 +1188,8 @@ static int me_huge_page(struct page_state *ps, struct page *p)
 #define mlock		(1UL << PG_mlocked)
 #define lru		(1UL << PG_lru)
 #define head		(1UL << PG_head)
-#define reserved	(1UL << PG_reserved)
 
 static struct page_state error_states[] = {
-	{ reserved,	reserved,	MF_MSG_KERNEL,	me_kernel },
 	/*
 	 * free pages are specially detected outside this table:
 	 * PG_buddy pages only make a small fraction of all free pages.
@@ -1234,7 +1221,6 @@ static struct page_state error_states[] = {
 #undef mlock
 #undef lru
 #undef head
-#undef reserved
 
 static void update_per_node_mf_stats(unsigned long pfn,
 				     enum mf_result result)

-- 
2.54.0

[PATCH v8 2/6] mm/memory-failure: surface unhandlable kernel pages as -ENOTRECOVERABLE

[Breno Leitao]

get_any_page() collapses every HWPoisonHandlable() rejection into a
single -EIO via the __get_hwpoison_page() -> -EBUSY -> shake_page()
-> retry path.  That is correct for the transient case (a userspace
folio briefly off LRU during migration or compaction, which a later
shake can drag back), but wrong for stable kernel-owned pages: slab,
page-table, large-kmalloc and PG_reserved pages will never become
HWPoisonHandlable(), so the retry loop is wasted work and the final
-EIO loses the "this is structurally unrecoverable" information.
memory_failure() then maps -EIO into MF_MSG_GET_HWPOISON, which the
panic-on-unrecoverable sysctl deliberately does not act on.

Introduce HWPoisonKernelOwned(), a small predicate that positively
identifies pages the hwpoison handler cannot recover from:

  HWPoisonKernelOwned(p, flags) :=
      !(MF_SOFT_OFFLINE && page_has_movable_ops(p)) &&
      (PageReserved(p) || PageSlab(p) ||
       PageTable(p)    || PageLargeKmalloc(p))

The MF_SOFT_OFFLINE / page_has_movable_ops() opt-out mirrors the
same exception in HWPoisonHandlable(): soft-offline is allowed to
migrate movable_ops pages even though they are not on the LRU, and
we must not pre-empt that with an unrecoverable verdict.

The list is intentionally not exhaustive.  vmalloc and kernel-stack
pages, for example, do not carry a page_type bit and would need a
different oracle; they keep going through the existing retry path
unchanged.  This is the smallest set we can identify with certainty
by page type.

Wire the helper into the top of get_any_page() to short-circuit
those pages before the retry loop runs.  On a hit, drop the caller's
MF_COUNT_INCREASED reference (if any) and return -ENOTRECOVERABLE
straight away.  Pages outside the helper's positive list still take
the existing retry path and return -EIO, leaving operator-visible
behaviour for those cases unchanged.

Extend the unhandlable-page pr_err() to fire for either errno and
update the get_hwpoison_page() kerneldoc to document the new return.

memory_failure() still folds every negative return into
MF_MSG_GET_HWPOISON via its existing "else if (res < 0)" branch, so
this patch on its own only changes the errno that soft_offline_page()
can propagate to its callers.  A follow-up wires -ENOTRECOVERABLE
through memory_failure() and reports MF_MSG_KERNEL for the
unrecoverable cases, which is what the
panic_on_unrecoverable_memory_failure sysctl observes.

Suggested-by: David Hildenbrand <david@kernel.org>
Suggested-by: Lance Yang <lance.yang@linux.dev>
Signed-off-by: Breno Leitao <leitao@debian.org>
---
 mm/memory-failure.c | 42 ++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 40 insertions(+), 2 deletions(-)

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index f4d3e6e20e13..8f63bdfeff8f 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -1325,6 +1325,28 @@ static inline bool HWPoisonHandlable(struct page *page, unsigned long flags)
 	return PageLRU(page) || is_free_buddy_page(page);
 }
 
+/*
+ * Positive identification of pages the hwpoison handler cannot recover.
+ * These page types are owned by kernel internals (no userspace mapping
+ * to unmap, no file mapping to invalidate, no migration target), so the
+ * shake_page() / retry loop in get_any_page() can never turn them into
+ * something HWPoisonHandlable() will accept.  Short-circuit them to
+ * -ENOTRECOVERABLE so callers can panic on operator request instead of
+ * spinning through retries that exit as a transient-looking -EIO.
+ *
+ * The MF_SOFT_OFFLINE / page_has_movable_ops() opt-out mirrors
+ * HWPoisonHandlable(): soft-offline is allowed to migrate movable_ops
+ * pages even though they are not on the LRU.
+ */
+static inline bool HWPoisonKernelOwned(struct page *page, unsigned long flags)
+{
+	if ((flags & MF_SOFT_OFFLINE) && page_has_movable_ops(page))
+		return false;
+
+	return PageReserved(page) || PageSlab(page) ||
+	       PageTable(page) || PageLargeKmalloc(page);
+}
+
 static int __get_hwpoison_page(struct page *page, unsigned long flags)
 {
 	struct folio *folio = page_folio(page);
@@ -1371,6 +1393,19 @@ static int get_any_page(struct page *p, unsigned long flags)
 	if (flags & MF_COUNT_INCREASED)
 		count_increased = true;
 
+	/*
+	 * Page types we know are kernel-owned and cannot be recovered.
+	 * Short-circuit before the shake_page() / retry loop, which
+	 * cannot turn any of these into something HWPoisonHandlable().
+	 * Drop the caller's reference if MF_COUNT_INCREASED took one.
+	 */
+	if (HWPoisonKernelOwned(p, flags)) {
+		if (count_increased)
+			put_page(p);
+		ret = -ENOTRECOVERABLE;
+		goto out;
+	}
+
 try_again:
 	if (!count_increased) {
 		ret = __get_hwpoison_page(p, flags);
@@ -1418,7 +1453,7 @@ static int get_any_page(struct page *p, unsigned long flags)
 		ret = -EIO;
 	}
 out:
-	if (ret == -EIO)
+	if (ret == -EIO || ret == -ENOTRECOVERABLE)
 		pr_err("%#lx: unhandlable page.\n", page_to_pfn(p));
 
 	return ret;
@@ -1475,7 +1510,10 @@ static int __get_unpoison_page(struct page *page)
  *         -EIO for pages on which we can not handle memory errors,
  *         -EBUSY when get_hwpoison_page() has raced with page lifecycle
  *         operations like allocation and free,
- *         -EHWPOISON when the page is hwpoisoned and taken off from buddy.
+ *         -EHWPOISON when the page is hwpoisoned and taken off from buddy,
+ *         -ENOTRECOVERABLE for kernel-owned pages identified by
+ *         HWPoisonKernelOwned() (PG_reserved, slab,
+ *         page-table, large-kmalloc) that the handler cannot recover.
  */
 static int get_hwpoison_page(struct page *p, unsigned long flags)
 {

-- 
2.54.0

[PATCH v8 3/6] mm/memory-failure: report MF_MSG_KERNEL for unrecoverable kernel pages

[Breno Leitao]

The previous patch teaches get_any_page() to return -ENOTRECOVERABLE
for stable unhandlable kernel pages (PG_reserved, slab, page tables,
large-kmalloc).  memory_failure() still folds every negative return
into MF_MSG_GET_HWPOISON, so callers that want to react to the
unrecoverable cases (a panic option, smarter logging) cannot tell
them apart from transient page-allocator races.

Turn the post-call branch into a switch over the get_hwpoison_page()
return code: map -ENOTRECOVERABLE to MF_MSG_KERNEL and any other
negative return to MF_MSG_GET_HWPOISON.  case 0 keeps the existing
free-buddy / kernel-high-order handling and case 1 falls through to
the rest of memory_failure() unchanged.

The MF_MSG_KERNEL label and tracepoint string are kept as
"reserved kernel page" to avoid breaking userspace tools that match
on those literals; the enum value still adequately tags the failure
even though it now also covers slab, page tables and large-kmalloc
pages.

Suggested-by: David Hildenbrand <david@kernel.org>
Signed-off-by: Breno Leitao <leitao@debian.org>
---
 mm/memory-failure.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 8f63bdfeff8f..14c0a958638c 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -2426,7 +2426,8 @@ int memory_failure(unsigned long pfn, int flags)
 	 * that may make page_ref_freeze()/page_ref_unfreeze() mismatch.
 	 */
 	res = get_hwpoison_page(p, flags);
-	if (!res) {
+	switch (res) {
+	case 0:
 		if (is_free_buddy_page(p)) {
 			if (take_page_off_buddy(p)) {
 				page_ref_inc(p);
@@ -2445,7 +2446,19 @@ int memory_failure(unsigned long pfn, int flags)
 			res = action_result(pfn, MF_MSG_KERNEL_HIGH_ORDER, MF_IGNORED);
 		}
 		goto unlock_mutex;
-	} else if (res < 0) {
+	case 1:
+		/* Got a refcount on a handlable page. */
+		break;
+	case -ENOTRECOVERABLE:
+		/*
+		 * Stable unhandlable kernel-owned page (PG_reserved,
+		 * slab, page tables, large-kmalloc).
+		 * No recovery possible.
+		 */
+		res = action_result(pfn, MF_MSG_KERNEL, MF_IGNORED);
+		goto unlock_mutex;
+	default:
+		/* Transient lifecycle race with the page allocator. */
 		res = action_result(pfn, MF_MSG_GET_HWPOISON, MF_IGNORED);
 		goto unlock_mutex;
 	}

-- 
2.54.0

[PATCH v8 4/6] mm/memory-failure: add panic option for unrecoverable pages

[Breno Leitao]

Add a sysctl panic_on_unrecoverable_memory_failure (disabled by
default) that triggers a kernel panic when memory_failure()
encounters pages that cannot be recovered.  This provides a clean
crash with useful debug information rather than allowing silent
data corruption or a delayed crash at an unrelated code path.

Panic eligibility is intentionally narrow: only MF_MSG_KERNEL with
result == MF_IGNORED panics.  After the previous patch, MF_MSG_KERNEL
covers PG_reserved pages and the kernel-owned pages promoted from
get_hwpoison_page() via -ENOTRECOVERABLE (slab, page tables,
large-kmalloc).

All other action types are excluded:

- MF_MSG_GET_HWPOISON and MF_MSG_KERNEL_HIGH_ORDER can be reached by
  transient refcount races with the page allocator (an in-flight buddy
  allocation has refcount 0 and is no longer on the buddy free list,
  briefly), and panicking on them would risk killing the box for what
  is actually a recoverable userspace page.

- MF_MSG_UNKNOWN means identify_page_state() could not classify the
  page; that is precisely the wrong basis for a panic decision.

Signed-off-by: Breno Leitao <leitao@debian.org>
---
 mm/memory-failure.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 14c0a958638c..dcd53dbc6aec 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -74,6 +74,8 @@ static int sysctl_memory_failure_recovery __read_mostly = 1;
 
 static int sysctl_enable_soft_offline __read_mostly = 1;
 
+static int sysctl_panic_on_unrecoverable_mf __read_mostly;
+
 atomic_long_t num_poisoned_pages __read_mostly = ATOMIC_LONG_INIT(0);
 
 static bool hw_memory_failure __read_mostly = false;
@@ -155,6 +157,15 @@ static const struct ctl_table memory_failure_table[] = {
 		.proc_handler	= proc_dointvec_minmax,
 		.extra1		= SYSCTL_ZERO,
 		.extra2		= SYSCTL_ONE,
+	},
+	{
+		.procname	= "panic_on_unrecoverable_memory_failure",
+		.data		= &sysctl_panic_on_unrecoverable_mf,
+		.maxlen		= sizeof(sysctl_panic_on_unrecoverable_mf),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
 	}
 };
 
@@ -1255,6 +1266,15 @@ static void update_per_node_mf_stats(unsigned long pfn,
 	++mf_stats->total;
 }
 
+static bool panic_on_unrecoverable_mf(enum mf_action_page_type type,
+				      enum mf_result result)
+{
+	if (!sysctl_panic_on_unrecoverable_mf || result != MF_IGNORED)
+		return false;
+
+	return type == MF_MSG_KERNEL;
+}
+
 /*
  * "Dirty/Clean" indication is not 100% accurate due to the possibility of
  * setting PG_dirty outside page lock. See also comment above set_page_dirty().
@@ -1272,6 +1292,9 @@ static int action_result(unsigned long pfn, enum mf_action_page_type type,
 	pr_err("%#lx: recovery action for %s: %s\n",
 		pfn, action_page_types[type], action_name[result]);
 
+	if (panic_on_unrecoverable_mf(type, result))
+		panic("Memory failure: %#lx: unrecoverable page", pfn);
+
 	return (result == MF_RECOVERED || result == MF_DELAYED) ? 0 : -EBUSY;
 }
 

-- 
2.54.0

[PATCH v8 5/6] Documentation: document panic_on_unrecoverable_memory_failure sysctl

[Breno Leitao]

Add documentation for the new vm.panic_on_unrecoverable_memory_failure
sysctl, describing which failures trigger a panic (kernel-owned pages
the handler cannot recover) and which are intentionally left out
(transient allocator races and unclassified pages).

Signed-off-by: Breno Leitao <leitao@debian.org>
---
 Documentation/admin-guide/sysctl/vm.rst | 85 +++++++++++++++++++++++++++++++++
 1 file changed, 85 insertions(+)

diff --git a/Documentation/admin-guide/sysctl/vm.rst b/Documentation/admin-guide/sysctl/vm.rst
index 97e12359775c..f71d87039904 100644
--- a/Documentation/admin-guide/sysctl/vm.rst
+++ b/Documentation/admin-guide/sysctl/vm.rst
@@ -67,6 +67,7 @@ Currently, these files are in /proc/sys/vm:
 - page-cluster
 - page_lock_unfairness
 - panic_on_oom
+- panic_on_unrecoverable_memory_failure
 - percpu_pagelist_high_fraction
 - stat_interval
 - stat_refresh
@@ -925,6 +926,90 @@ panic_on_oom=2+kdump gives you very strong tool to investigate
 why oom happens. You can get snapshot.
 
 
+panic_on_unrecoverable_memory_failure
+======================================
+
+When a hardware memory error (e.g. multi-bit ECC) hits a kernel page
+that cannot be recovered by the memory failure handler, the default
+behaviour is to ignore the error and continue operation.  This is
+dangerous because the corrupted data remains accessible to the kernel,
+risking silent data corruption or a delayed crash when the poisoned
+memory is next accessed.
+
+When enabled, this sysctl triggers a panic on memory failure events
+hitting kernel-owned pages that the handler cannot recover:
+``PageReserved`` (firmware reservations, kernel image, vDSO, zero
+page, and similar memblock-reserved regions), ``PageSlab``,
+``PageTable``, and ``PageLargeKmalloc``.  These are owned by the
+kernel and the memory failure handler cannot reliably evict their
+contents.
+
+For soft offline (``madvise(MADV_SOFT_OFFLINE)``,
+``/sys/devices/system/memory/soft_offline_page``), pages owned by
+``movable_ops`` are exempted, since soft offline is allowed to
+migrate them even though they are not on the LRU.
+
+Other unrecoverable kernel-owned populations (vmalloc allocations,
+kernel stack pages, ...) are not currently covered because the
+handler has no page-type signal that distinguishes them from a
+userspace folio temporarily off the LRU during migration or
+compaction.  Such pages still go through the standard
+MF_MSG_GET_HWPOISON path: ``PG_hwpoison`` is set on them and a
+delayed crash on the next access remains possible.  Coverage may
+grow as the handler gains stronger kernel-ownership signals.
+
+Recoverable failure paths are also intentionally left out: in-flight
+buddy allocations and other transient races with the page allocator
+can reach the same diagnostic, and panicking on them would risk
+killing the box for a page destined for userspace where the standard
+SIGBUS recovery path applies.  Pages whose state could not be
+classified at all are not covered either, since an unknown state is
+not a sound basis for a panic decision.
+
+For many environments it is preferable to panic immediately with a clean
+crash dump that captures the original error context, rather than to
+continue and face a random crash later whose cause is difficult to
+diagnose.
+
+Use cases
+---------
+
+This option is most useful in environments where unattributed crashes
+are expensive to debug or where data integrity must take precedence
+over availability:
+
+* Large fleets, where multi-bit ECC errors on kernel pages are observed
+  regularly and post-mortem analysis of an unrelated downstream crash
+  (often seconds to minutes after the original error) consumes
+  significant engineering effort.
+
+* Systems configured with kdump, where panicking at the moment of the
+  hardware error produces a vmcore that still contains the faulting
+  address, the affected page state, and the originating MCE/GHES
+  record — context that is typically lost by the time a delayed crash
+  occurs.
+
+* High-availability clusters that rely on fast, deterministic node
+  failure for failover, and prefer an immediate panic over silent data
+  corruption propagating to replicas or persistent storage.
+
+* Kernel and platform developers reproducing hwpoison issues with
+  tools such as ``mce-inject`` or error-injection debugfs interfaces,
+  where panicking on the unrecoverable path makes regressions
+  immediately visible instead of surfacing as later, unrelated
+  failures.
+
+= =====================================================================
+0 Try to continue operation (default).
+1 Panic immediately.  If the ``panic`` sysctl is also non-zero then the
+  machine will be rebooted.
+= =====================================================================
+
+Example::
+
+     echo 1 > /proc/sys/vm/panic_on_unrecoverable_memory_failure
+
+
 percpu_pagelist_high_fraction
 =============================
 

-- 
2.54.0

[PATCH v8 6/6] selftests/mm: add hwpoison-panic destructive test

[Breno Leitao]

Add a destructive selftest that verifies
vm.panic_on_unrecoverable_memory_failure actually panics when a
hwpoison error hits a kernel-owned page.

Three "kinds" of kernel-owned page can be targeted, selectable via
the script's first positional argument (default: rodata):

  rodata  - a PG_reserved page in the kernel rodata range, sourced
            from the "Kernel rodata" sub-resource of "System RAM" in
            /proc/iomem.  That entry is reported on every major
            architecture and guarantees the chosen PFN is backed by
            struct page (an online System RAM range, not a firmware
            hole), is PG_reserved, and is read-only -- so even if
            the panic fails to fire for some reason, the resulting
            PG_hwpoison marker on rodata does not corrupt writable
            kernel state.

  slab    - a slab page found by walking /proc/kpageflags for the
            first PFN with KPF_SLAB set (and KPF_HWPOISON / KPF_NOPAGE
            / KPF_COMPOUND_TAIL clear).  Exercises the get_any_page()
            path on a non PG_reserved kernel-owned page and so
            catches regressions where get_any_page() collapses
            kernel-owned pages into a transient -EIO instead of
            -ENOTRECOVERABLE.

  pgtable - same as slab, but the PFN is selected via KPF_PGTABLE.

PageLargeKmalloc, the fourth page type matched by
HWPoisonKernelOwned(), is intentionally not covered: it is a
PAGE_TYPE_OPS flag with no /proc/kpageflags bit, so selecting such
a PFN from userspace is not feasible.  The slab and pgtable
variants already exercise the same get_any_page() positive-check
branch.

The script enables the sysctl and writes the selected physical
address to /sys/devices/system/memory/hard_offline_page.  A
successful run crashes the kernel with

  Memory failure: <pfn>: unrecoverable page

A return from the inject means the panic did not fire and the test
fails.  Test outcome is therefore observed externally (serial
console, kdump) rather than from the script's own exit code.

The script is intentionally NOT wired into run_vmtests.sh: every
successful run panics the kernel, which is incompatible with the
sequential "run each category in the same VM" model that
run_vmtests.sh assumes.  It is also not registered as a TEST_PROGS /
ksft_* wrapper so a default kselftest run does not opt itself into
a panic.  The script is meant to be executed manually inside a
disposable VM (e.g. virtme-ng), one variant per VM boot, and
requires RUN_DESTRUCTIVE=1 in the environment as a safety net.

Signed-off-by: Breno Leitao <leitao@debian.org>
---
 tools/testing/selftests/mm/Makefile          |   1 +
 tools/testing/selftests/mm/hwpoison-panic.sh | 193 +++++++++++++++++++++++++++
 2 files changed, 194 insertions(+)

diff --git a/tools/testing/selftests/mm/Makefile b/tools/testing/selftests/mm/Makefile
index e6df968f0971..170e376c97b4 100644
--- a/tools/testing/selftests/mm/Makefile
+++ b/tools/testing/selftests/mm/Makefile
@@ -181,6 +181,7 @@ TEST_FILES += charge_reserved_hugetlb.sh
 TEST_FILES += hugetlb_reparenting_test.sh
 TEST_FILES += test_page_frag.sh
 TEST_FILES += run_vmtests.sh
+TEST_FILES += hwpoison-panic.sh
 
 # required by charge_reserved_hugetlb.sh
 TEST_FILES += write_hugetlb_memory.sh
diff --git a/tools/testing/selftests/mm/hwpoison-panic.sh b/tools/testing/selftests/mm/hwpoison-panic.sh
new file mode 100755
index 000000000000..43fc379f8761
--- /dev/null
+++ b/tools/testing/selftests/mm/hwpoison-panic.sh
@@ -0,0 +1,193 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+#
+# Verify vm.panic_on_unrecoverable_memory_failure by injecting a hwpoison
+# error on a kernel-owned page and confirming the kernel panics.
+#
+# Three "kinds" of kernel-owned page can be targeted, selectable via the
+# first positional argument (default: rodata):
+#
+#   rodata  - a PG_reserved page in the kernel rodata range
+#             (sourced from /proc/iomem "Kernel rodata").  Exercises
+#             memory_failure() -> get_any_page() on a PageReserved page.
+#
+#   slab    - a slab page found via /proc/kpageflags (KPF_SLAB).
+#             Exercises memory_failure() -> get_any_page() on a non
+#             PG_reserved kernel-owned page.  This path is what catches
+#             regressions where get_any_page() collapses kernel-owned
+#             pages into a transient -EIO instead of -ENOTRECOVERABLE.
+#
+#   pgtable - a page-table page found via /proc/kpageflags (KPF_PGTABLE).
+#             Same path as slab, different page type.
+#
+# This test is DESTRUCTIVE: a successful run crashes the kernel.  It is
+# meant to be executed inside a disposable VM (e.g. virtme-ng) with a
+# serial console captured by the harness.  It is skipped unless the
+# caller opts in via RUN_DESTRUCTIVE=1.
+#
+# Test passes externally: the kernel must panic with
+#   "Memory failure: <pfn>: unrecoverable page"
+# A return from the inject means the panic did not fire and the test
+# fails.
+#
+# Author: Breno Leitao <leitao@debian.org>
+
+set -u
+
+ksft_skip=4
+sysctl_path=/proc/sys/vm/panic_on_unrecoverable_memory_failure
+inject_path=/sys/devices/system/memory/hard_offline_page
+kpageflags_path=/proc/kpageflags
+
+# /proc/kpageflags bit positions (see include/uapi/linux/kernel-page-flags.h)
+KPF_SLAB=7
+KPF_COMPOUND_TAIL=16
+KPF_HWPOISON=19
+KPF_NOPAGE=20
+KPF_PGTABLE=26
+
+kind=${1:-rodata}
+
+ksft_print() { echo "# $*"; }
+ksft_exit_skip() { ksft_print "$*"; exit "$ksft_skip"; }
+ksft_exit_fail() { echo "not ok 1 $*"; exit 1; }
+
+if [ "$(id -u)" -ne 0 ]; then
+	ksft_exit_skip "must run as root"
+fi
+
+if [ ! -w "$sysctl_path" ]; then
+	ksft_exit_skip "$sysctl_path not present (kernel without the sysctl?)"
+fi
+
+if [ ! -w "$inject_path" ]; then
+	ksft_exit_skip "$inject_path not present (no MEMORY_HOTPLUG?)"
+fi
+
+if [ "${RUN_DESTRUCTIVE:-0}" != "1" ]; then
+	ksft_exit_skip "destructive test; re-run with RUN_DESTRUCTIVE=1 inside a disposable VM"
+fi
+
+# Pick a PFN inside the kernel image rodata region of /proc/iomem.
+# This is preferred over a top-level "Reserved" entry because top-level
+# Reserved ranges are often firmware holes that have no backing struct
+# page; pfn_to_online_page() returns NULL on those and memory_failure()
+# bails out with -ENXIO before reaching the panic path.
+#
+# "Kernel rodata" is reported as a sub-resource of "System RAM" on every
+# major architecture, which guarantees:
+#   - the PFN is backed by struct page (within an online memory range);
+#   - PG_reserved is set on the page (kernel image area);
+#   - the memory is read-only, so setting PG_hwpoison on it does not
+#     corrupt writable kernel state if the panic somehow does not fire.
+#
+# /proc/iomem entries look like (indented for sub-resources):
+#     "  02500000-02ffffff : Kernel rodata"
+pick_rodata_phys_addr() {
+	awk -v pagesize="$(getconf PAGE_SIZE)" '
+	/: Kernel rodata[[:space:]]*$/ {
+		sub(/^[[:space:]]+/, "")
+		n = split($0, a, /[- ]/)
+		start = strtonum("0x" a[1])
+		end   = strtonum("0x" a[2])
+		if (end <= start)
+			next
+		# Page-align upward and emit the first byte of that page.
+		pfn = int((start + pagesize - 1) / pagesize)
+		printf "0x%x\n", pfn * pagesize
+		exit 0
+	}
+	' /proc/iomem
+}
+
+# Walk /proc/kpageflags and return the phys addr of the first PFN that
+# has bit $1 set, with KPF_HWPOISON, KPF_NOPAGE and KPF_COMPOUND_TAIL
+# all clear (so we attack a real, non-tail, not-already-poisoned page).
+#
+# We skip the first 16 MiB of PFNs to step past low-memory special
+# ranges (BIOS/EFI/ACPI/etc.) that often are PG_reserved and would not
+# exhibit the slab/pgtable type we are looking for.
+pick_kpageflags_phys_addr() {
+	local want_bit=$1
+	local pagesize skip_pfn
+
+	[ -r "$kpageflags_path" ] || return
+
+	pagesize=$(getconf PAGE_SIZE)
+	skip_pfn=$(((16 * 1024 * 1024) / pagesize))
+
+	od -An -tx8 -v -w8 -j "$((skip_pfn * 8))" "$kpageflags_path" 2>/dev/null | \
+	awk -v want_bit="$want_bit" \
+	    -v hwp_bit="$KPF_HWPOISON" \
+	    -v nopage_bit="$KPF_NOPAGE" \
+	    -v tail_bit="$KPF_COMPOUND_TAIL" \
+	    -v base_pfn="$skip_pfn" \
+	    -v pagesize="$pagesize" '
+	# Test whether bit "b" is set in the 16-hex-digit value "hex".
+	# Done with substring + per-digit lookup so we never rely on awk
+	# bitwise operators (mawk lacks them) or 64-bit FP precision.
+	function bit_set(hex, b,    di, bi, c, v) {
+		di = int(b / 4)
+		bi = b - di * 4
+		c = substr(hex, length(hex) - di, 1)
+		v = strtonum("0x" c)
+		if (bi == 0) return (v % 2) == 1
+		if (bi == 1) return int(v / 2) % 2 == 1
+		if (bi == 2) return int(v / 4) % 2 == 1
+		return int(v / 8) % 2 == 1
+	}
+	{
+		gsub(/^[[:space:]]+/, "")
+		h = $1
+		if (bit_set(h, want_bit) &&
+		    !bit_set(h, hwp_bit) &&
+		    !bit_set(h, nopage_bit) &&
+		    !bit_set(h, tail_bit)) {
+			pfn = base_pfn + NR - 1
+			printf "0x%x\n", pfn * pagesize
+			exit 0
+		}
+	}
+	'
+}
+
+case "$kind" in
+rodata)
+	phys_addr=$(pick_rodata_phys_addr)
+	missing_msg='no "Kernel rodata" entry in /proc/iomem'
+	;;
+slab)
+	phys_addr=$(pick_kpageflags_phys_addr "$KPF_SLAB")
+	missing_msg="no usable slab PFN found in $kpageflags_path"
+	;;
+pgtable)
+	phys_addr=$(pick_kpageflags_phys_addr "$KPF_PGTABLE")
+	missing_msg="no usable page-table PFN found in $kpageflags_path"
+	;;
+*)
+	ksft_exit_fail "unknown kind '$kind' (expected: rodata|slab|pgtable)"
+	;;
+esac
+
+if [ -z "$phys_addr" ]; then
+	ksft_exit_skip "$missing_msg"
+fi
+
+ksft_print "enabling $sysctl_path"
+prior=$(cat "$sysctl_path")
+echo 1 > "$sysctl_path" || ksft_exit_fail "failed to enable sysctl"
+
+ksft_print "injecting hwpoison at phys 0x$(printf '%x' "$phys_addr") (kind=$kind)"
+ksft_print "expecting kernel panic: 'Memory failure: <pfn>: unrecoverable page'"
+
+# If this returns, the kernel did not panic → test failed.  Restore the
+# sysctl before reporting so the system is left as we found it.
+if echo "$phys_addr" > "$inject_path"; then
+	echo "$prior" > "$sysctl_path"
+	ksft_exit_fail "inject returned without panic; sysctl ineffective"
+fi
+
+# Write failed (e.g. -EINVAL on offlining a non-online region): also a
+# failure for this test, since we expected the panic path.
+echo "$prior" > "$sysctl_path"
+ksft_exit_fail "inject failed before reaching the panic path"

-- 
2.54.0

Re: [PATCH v8 0/6] mm/memory-failure: add panic option for unrecoverable pages

[Andrew Morton]

On Wed, 27 May 2026 07:06:13 -0700 Breno Leitao <leitao@debian.org> wrote:

> A multi-bit ECC error on a kernel-owned page that the memory failure
> handler cannot recover is currently swallowed: PG_hwpoison is set, the
> event is logged, and the kernel keeps running.  The corrupted memory
> remains accessible to the kernel and either drives silent data
> corruption or surfaces seconds-to-minutes later as an apparently
> unrelated crash.  In a large fleet that delayed, unattributable crash
> turns into significant engineering effort to root-cause; in a kdump
> configuration, by the time the crash happens the original error
> context (faulting PFN, MCE/GHES record, page state) is long gone.
> 
> This series adds an opt-in sysctl,
> vm.panic_on_unrecoverable_memory_failure, that converts an
> unrecoverable kernel-page hwpoison event into an immediate panic with
> a clean dmesg/vmcore that still contains the original failure
> context.  The default is disabled so existing workloads see no
> change.

Thanks.  That does seem useful.

I'll pass at this time, due to -rc5 and not-very-reviewed.

AI review said a few things.  It claims to have found one pre-existing
issue.

	https://sashiko.dev/#/patchset/20260527-ecc_panic-v8-0-9ea0cfa16bb0@debian.org