From: Kunwu Chan <kunwu.chan@gmail.com>
To: paulmck@kernel.org, corbet@lwn.net, mingo@redhat.com,
frederic@kernel.org, neeraj.upadhyay@kernel.org,
josh@joshtriplett.org, urezki@gmail.com, dave@stgolabs.net,
lianux.mm@gmail.com
Cc: stern@rowland.harvard.edu, parri.andrea@gmail.com,
will@kernel.org, peterz@infradead.org, boqun@kernel.org,
npiggin@gmail.com, dhowells@redhat.com, j.alglave@ucl.ac.uk,
luc.maranget@inria.fr, akiyks@gmail.com, dlustig@nvidia.com,
joelagnelf@nvidia.com, skhan@linuxfoundation.org,
rdunlap@infradead.org, longman@redhat.com, rostedt@goodmis.org,
mathieu.desnoyers@efficios.com, jiangshanlai@gmail.com,
qiang.zhang@linux.dev, kunwu.chan@gmail.com,
brads@mainlining.org, linux-kernel@vger.kernel.org,
linux-arch@vger.kernel.org, lkmm@lists.linux.dev,
linux-doc@vger.kernel.org, rcu@vger.kernel.org,
linux-kselftest@vger.kernel.org
Subject: [PATCH RFC v2 03/15] hazptr: scan all per-CPU slots before overflow lists
Date: Sat, 3 Oct 2026 01:08:35 +0800 [thread overview]
Message-ID: <20261002170847.3653663-4-kunwu.chan@gmail.com> (raw)
In-Reply-To: <20261002170847.3653663-1-kunwu.chan@gmail.com>
Boqun Feng pointed out that a promote racing a scan can cause the
scan to miss a hazard pointer: hazptr_promote_to_backup_slot()
chains the backup slot into an overflow list before clearing the
per-CPU slot, so the scan must observe the old location before the
new one. The scan currently visits each CPU's overflow lists right
after that CPU's per-CPU slots, so a promote that moves a hazard
pointer from a per-CPU slot of a later-scanned CPU into an overflow
list of an earlier-scanned CPU can be missed.
The direct two-phase scan has a related race because the overflow
list selection follows the wildcard phase: a promote racing the
wildcard flip can move a backup slot to the list not covered by the
second phase.
Fix this by scanning all per-CPU slots before any overflow-list
slot, in both the shared-scan walk and the direct fallback scan.
Give the overflow lists a flip phase of their own, independent of
the wildcard, so each list is scanned while it is the non-live one:
new backup slots are chained to the other list, preserving forward
progress.
The same race was also addressed by Mathieu Desnoyers.
Fixes: 0b8114b25f17 ("hazptr: Implement two-phase wildcard scan")
Reported-by: Boqun Feng <boqun@kernel.org>
Link: https://lore.kernel.org/all/20260925195958.4766-1-mathieu.desnoyers@efficios.com/
Signed-off-by: Kunwu Chan <kunwu.chan@gmail.com>
---
kernel/hazptr.c | 88 +++++++++++++++++++++++++++++++++++--------------
1 file changed, 64 insertions(+), 24 deletions(-)
diff --git a/kernel/hazptr.c b/kernel/hazptr.c
index 900ba35de2cb..799784f698ed 100644
--- a/kernel/hazptr.c
+++ b/kernel/hazptr.c
@@ -23,13 +23,20 @@
* The current hazard pointer wildcard. Flips between 1UL and 2UL to guarantee
* hazptr_synchronize forward progress even with a steady stream of readers.
* This wildcard value is used by acquire to temporarily tag the per-CPU slots.
- * This also affects the overflow list selection: the current list used by
- * readers is array[(unsigned long) hazptr_wildcard - 1].
*/
-static DEFINE_MUTEX(hazptr_wildcard_lock); /* Protect the wildcard flip. */
+static DEFINE_MUTEX(hazptr_phase_lock);
+/* Protect the wildcard and overflow-list phase flips. */
void *hazptr_wildcard = (void *) 1UL;
EXPORT_SYMBOL_GPL(hazptr_wildcard);
+/*
+ * The current overflow list phase. Independent from the wildcard so that the
+ * overflow list scan can be placed after the per-CPU slot scan while still
+ * scanning the non-live list: the current list used by readers is
+ * array[hazptr_overflow_list_phase].
+ */
+static unsigned int hazptr_overflow_list_phase;
+
struct hazptr_overflow_list {
raw_spinlock_t lock; /* Lock protecting overflow list and list generation. */
struct hlist_head head; /* Overflow list head. */
@@ -59,6 +66,12 @@ void *flip_wildcard(void *wildcard)
return ((unsigned long) wildcard == 1UL) ? (void *) 2UL : (void *) 1UL;
}
+static
+unsigned int flip_list_phase(unsigned int phase)
+{
+ return 1 - phase;
+}
+
static
bool is_wildcard(void *addr)
{
@@ -184,15 +197,12 @@ void hazptr_synchronize_cpu_slots(int cpu, void *addr, void *scan_wildcard)
}
static
-void hazptr_scan_period(void *addr, void *scan_wildcard)
+void hazptr_scan_cpu_slots_period(void *addr, void *scan_wildcard)
{
- unsigned int scan_idx = (unsigned long) scan_wildcard - 1;
int cpu;
/* Scan all CPUs slots. */
for_each_possible_cpu(cpu) {
- struct hazptr_overflow_list_flip *overflow_list_flip = per_cpu_ptr(&percpu_overflow_list_flip, cpu);
-
/*
* Scan CPU slots.
* Forward progress against recurring wildcards is guaranteed
@@ -205,12 +215,22 @@ void hazptr_scan_period(void *addr, void *scan_wildcard)
* to acquire that same hazard pointer value.
*/
hazptr_synchronize_cpu_slots(cpu, addr, scan_wildcard);
+ }
+}
+
+static
+void hazptr_scan_overflow_list_period(void *addr, unsigned int scan_idx)
+{
+ int cpu;
+
+ /*
+ * Scan backup slots in percpu overflow lists.
+ * Forward progress is guaranteed by scanning one list
+ * while new elements are added into the other list.
+ */
+ for_each_possible_cpu(cpu) {
+ struct hazptr_overflow_list_flip *overflow_list_flip = per_cpu_ptr(&percpu_overflow_list_flip, cpu);
- /*
- * Scan backup slots in percpu overflow lists.
- * Forward progress is guaranteed by scanning one list
- * while new elements are added into the other list.
- */
hazptr_synchronize_overflow_list(&overflow_list_flip->array[scan_idx], addr);
}
}
@@ -285,10 +305,13 @@ static struct hazptr_scan_state hazptr_scan;
* Walk all slots and return true if @watch is present. If @bloom
* is non-NULL, record observed non-wildcard addresses in it.
*
- * Per-CPU slots are examined before overflow-list slots on each CPU
- * to preserve the acquisition ordering required by the promote path:
- * synchronize must observe the per-CPU slot release before the
- * overflow-list entry can be missed.
+ * All per-CPU slots are examined before any overflow-list slot: a
+ * promote (hazptr_detach() or a context switch) moves a hazard
+ * pointer from a per-CPU slot to an overflow list by chaining the
+ * backup slot before clearing the per-CPU slot, see
+ * hazptr_promote_to_backup_slot(). Therefore the scan must observe
+ * the old location before the new location for every slot/list
+ * pair, including pairs on different CPUs.
*/
static bool hazptr_scan_walk(void *watch, struct hazptr_bloom *bloom)
{
@@ -297,9 +320,9 @@ static bool hazptr_scan_walk(void *watch, struct hazptr_bloom *bloom)
if (bloom)
hazptr_bloom_reset(bloom);
+ /* Scan all per-CPU slots before any overflow-list slot. */
for_each_possible_cpu(cpu) {
struct hazptr_percpu_slots *percpu_slots = per_cpu_ptr(&hazptr_percpu_slots, cpu);
- struct hazptr_overflow_list_flip *overflow_list_flip = per_cpu_ptr(&percpu_overflow_list_flip, cpu);
unsigned int idx;
for (idx = 0; idx < NR_HAZPTR_PERCPU_SLOTS; idx++) {
@@ -313,6 +336,10 @@ static bool hazptr_scan_walk(void *watch, struct hazptr_bloom *bloom)
if (bloom && v && !is_wildcard(v))
hazptr_bloom_add(bloom, v);
}
+ }
+ for_each_possible_cpu(cpu) {
+ struct hazptr_overflow_list_flip *overflow_list_flip = per_cpu_ptr(&percpu_overflow_list_flip, cpu);
+
for (int i = 0; i < 2; i++) {
struct hazptr_overflow_list *list = &overflow_list_flip->array[i];
struct hazptr_backup_slot *b;
@@ -359,14 +386,14 @@ static void hazptr_scan_do_cycle(void)
struct hazptr_waiter *w, *n;
LIST_HEAD(done);
- mutex_lock(&hazptr_wildcard_lock);
+ mutex_lock(&hazptr_phase_lock);
mutex_lock(&hazptr_scan.lock);
list_splice_tail_init(&hazptr_scan.pending, &hazptr_scan.scanning);
mutex_unlock(&hazptr_scan.lock);
if (list_empty(&hazptr_scan.scanning)) {
- mutex_unlock(&hazptr_wildcard_lock);
+ mutex_unlock(&hazptr_phase_lock);
return;
}
@@ -391,7 +418,7 @@ static void hazptr_scan_do_cycle(void)
list_move(&w->node, &done);
}
- mutex_unlock(&hazptr_wildcard_lock);
+ mutex_unlock(&hazptr_phase_lock);
list_for_each_entry_safe(w, n, &done, node) {
list_del_init(&w->node);
@@ -463,6 +490,7 @@ static void hazptr_synchronize_queued(void *addr)
void hazptr_synchronize(void *addr)
{
void *scan_wildcard;
+ unsigned int scan_list_phase;
/*
* Busy-wait should only be done from preemptible context.
@@ -486,18 +514,30 @@ void hazptr_synchronize(void *addr)
}
/* Fallback: use the direct scan path. */
- guard(mutex)(&hazptr_wildcard_lock);
+ guard(mutex)(&hazptr_phase_lock);
scan_wildcard = flip_wildcard(hazptr_wildcard);
- hazptr_scan_period(addr, scan_wildcard);
+ /* Scan per-CPU slots. */
+ hazptr_scan_cpu_slots_period(addr, scan_wildcard);
WRITE_ONCE(hazptr_wildcard, scan_wildcard); /* Flip the current wildcard. */
- hazptr_scan_period(addr, flip_wildcard(scan_wildcard));
+ hazptr_scan_cpu_slots_period(addr, flip_wildcard(scan_wildcard));
+
+ /*
+ * Scan overflow lists *after* scanning all per-CPU slots, see
+ * hazptr_promote_to_backup_slot(). Flip the overflow list phase
+ * between the two lists so that each list is scanned while it is
+ * the non-live one.
+ */
+ scan_list_phase = flip_list_phase(hazptr_overflow_list_phase);
+ hazptr_scan_overflow_list_period(addr, scan_list_phase);
+ WRITE_ONCE(hazptr_overflow_list_phase, scan_list_phase);
+ hazptr_scan_overflow_list_period(addr, flip_list_phase(scan_list_phase));
}
EXPORT_SYMBOL_GPL(hazptr_synchronize);
struct hazptr_slot *hazptr_chain_backup_slot(struct hazptr_ctx *ctx)
{
struct hazptr_overflow_list_flip *overflow_list_flip = this_cpu_ptr(&percpu_overflow_list_flip);
- unsigned int list_idx = (unsigned long) READ_ONCE(hazptr_wildcard) - 1;
+ unsigned int list_idx = READ_ONCE(hazptr_overflow_list_phase);
struct hazptr_overflow_list *overflow_list = &overflow_list_flip->array[list_idx];
struct hazptr_slot *slot = &ctx->backup_slot.slot;
--
2.43.0
next prev parent reply other threads:[~2026-10-02 17:09 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-02 17:08 [PATCH RFC v2 00/15] hazptr: batch synchronize operations through a shared scan Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 01/15] hazptr: add shared scan kthread Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 02/15] hazptr: use Bloom filter for shared scan waiters Kunwu Chan
2026-10-02 17:08 ` Kunwu Chan [this message]
2026-10-02 17:08 ` [PATCH RFC v2 04/15] hazptr: add scoped_guard() support Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 05/15] hazptr: add debug option to force the acquire slow path Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 06/15] hazptr: elide redundant first drain pass Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 07/15] Documentation/litmus-tests: add hazptr wildcard-flip escape test Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 08/15] locking/lockdep: use hazptr to wait for dynamic key lookups Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 09/15] rcuscale: add hazptr scale type Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 10/15] hazptr: fix kernel-doc of hazptr_release() Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 11/15] Documentation/litmus-tests: add hazptr acquire-before-scan test Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 12/15] hazptrtorture: add slowpath and lockdep scenarios Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 13/15] hazptrtorture: add READERS4 and READERS0 torture configs Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 14/15] hazptrtorture: add 128- and 256-CPU configs Kunwu Chan
2026-10-02 17:08 ` [PATCH RFC v2 15/15] selftests/rcutorture: add hazptr torture test script Kunwu Chan
2026-10-02 17:12 ` [PATCH RFC v2 00/15] hazptr: batch synchronize operations through a shared scan Bradley Morgan
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20261002170847.3653663-4-kunwu.chan@gmail.com \
--to=kunwu.chan@gmail.com \
--cc=akiyks@gmail.com \
--cc=boqun@kernel.org \
--cc=brads@mainlining.org \
--cc=corbet@lwn.net \
--cc=dave@stgolabs.net \
--cc=dhowells@redhat.com \
--cc=dlustig@nvidia.com \
--cc=frederic@kernel.org \
--cc=j.alglave@ucl.ac.uk \
--cc=jiangshanlai@gmail.com \
--cc=joelagnelf@nvidia.com \
--cc=josh@joshtriplett.org \
--cc=lianux.mm@gmail.com \
--cc=linux-arch@vger.kernel.org \
--cc=linux-doc@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-kselftest@vger.kernel.org \
--cc=lkmm@lists.linux.dev \
--cc=longman@redhat.com \
--cc=luc.maranget@inria.fr \
--cc=mathieu.desnoyers@efficios.com \
--cc=mingo@redhat.com \
--cc=neeraj.upadhyay@kernel.org \
--cc=npiggin@gmail.com \
--cc=parri.andrea@gmail.com \
--cc=paulmck@kernel.org \
--cc=peterz@infradead.org \
--cc=qiang.zhang@linux.dev \
--cc=rcu@vger.kernel.org \
--cc=rdunlap@infradead.org \
--cc=rostedt@goodmis.org \
--cc=skhan@linuxfoundation.org \
--cc=stern@rowland.harvard.edu \
--cc=urezki@gmail.com \
--cc=will@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®