From: wen.yang@linux.dev
To: Gabriele Monaco <gmonaco@redhat.com>
Cc: Nam Cao <namcao@linutronix.de>,
linux-trace-kernel@vger.kernel.org, linux-kernel@vger.kernel.org,
Wen Yang <wen.yang@linux.dev>
Subject: [PATCH v8 04/14] rv: Disarm HA timers on the storage free path
Date: Fri, 6 Nov 2026 01:49:04 +0800 [thread overview]
Message-ID: <0ab3632933008e028855e158b4b978881a2715dc.1793900349.git.wen.yang@linux.dev> (raw)
In-Reply-To: <cover.1793900349.git.wen.yang@linux.dev>
From: Wen Yang <wen.yang@linux.dev>
da_destroy_storage() cancels the HA timer through da_monitor_reset_hook()
and only then unlinks the storage with hash_del_rcu(). A handler on
another CPU that looked the storage up before the unlink can still call
ha_start_timer_*() after the cancel. The timer then stays queued on a
storage that kfree_rcu() frees a grace period later, and fires on freed
memory, reporting a violation for an object that no longer exists.
A monitor cannot close this from its own handlers: any cancel it issues
precedes the unlink. The first point no reader can race with is the
RCU callback that frees the storage.
Add da_monitor_free_hook(), called right before a per-object storage is
freed, from the RCU callback and from da_monitor_destroy(). HA monitors
implement it by calling ha_cancel_timer_sync() if da_monitoring() is true.
Replace kfree_rcu() with call_rcu() so there is a callback to hook.
Cancelling synchronously from the RCU callback is safe: HA hrtimers are
always armed in HARD mode, so hrtimer_cancel() only spins, and on
PREEMPT_RT, where timer_delete_sync() may sleep, RCU callbacks run in
the rcuc kthread.
Storages whose monitor never started are skipped via da_monitoring().
Fixes: 4a24127bd6cb ("rv: Add support for per-object monitors in DA/HA")
Suggested-by: Gabriele Monaco <gmonaco@redhat.com>
Signed-off-by: Wen Yang <wen.yang@linux.dev>
---
include/rv/da_monitor.h | 27 ++++++++++++++++++++++++++-
include/rv/ha_monitor.h | 8 ++++++++
2 files changed, 34 insertions(+), 1 deletion(-)
diff --git a/include/rv/da_monitor.h b/include/rv/da_monitor.h
index e3cf85c9ce55..c7f9733fb8a4 100644
--- a/include/rv/da_monitor.h
+++ b/include/rv/da_monitor.h
@@ -67,6 +67,17 @@ static struct rv_monitor rv_this;
#define da_monitor_sync_hook()
#endif
+/*
+ * Called on the last reference to a storage, right before it is freed.
+ * Monitors embedding a timer disarm it here: an earlier point races with a
+ * reader that resolved the storage before hash_del_rcu(), and
+ * tracepoint_synchronize_unregister() only covers tracepoint probes.
+ * Default: no-op.
+ */
+#ifndef da_monitor_free_hook
+#define da_monitor_free_hook(da_mon)
+#endif
+
/*
* Type for the target id, default to int but can be overridden.
* A long type can work as hash table key (PER_OBJ) but will be downgraded to
@@ -543,6 +554,15 @@ static inline monitor_target da_get_target_by_id(da_id_type id)
return mon_storage->target;
}
+static void da_kfree_cb(struct rcu_head *head)
+{
+ struct da_monitor_storage *ms =
+ container_of(head, struct da_monitor_storage, rcu);
+
+ da_monitor_free_hook(&ms->rv.da_mon);
+ kfree(ms);
+}
+
/*
* da_destroy_storage - destroy the per-object storage
*
@@ -564,7 +584,11 @@ static inline void da_destroy_storage(da_id_type id)
return;
da_monitor_reset_hook(&mon_storage->rv.da_mon);
hash_del_rcu(&mon_storage->node);
- kfree_rcu(mon_storage, rcu);
+ /*
+ * A reader from before hash_del_rcu() may still arm; the free
+ * callback runs past it, so the hook disarms there.
+ */
+ call_rcu(&mon_storage->rcu, da_kfree_cb);
}
static void __da_monitor_reset_all(void (*reset)(struct da_monitor *))
@@ -607,6 +631,7 @@ static inline void da_monitor_destroy(void)
* pending, we can safely assume no concurrent user.
*/
hash_for_each_safe(da_monitor_ht, bkt, tmp, mon_storage, node) {
+ da_monitor_free_hook(&mon_storage->rv.da_mon);
hash_del_rcu(&mon_storage->node);
kfree(mon_storage);
}
diff --git a/include/rv/ha_monitor.h b/include/rv/ha_monitor.h
index ebb7f91e3447..3bf1dd15544c 100644
--- a/include/rv/ha_monitor.h
+++ b/include/rv/ha_monitor.h
@@ -29,6 +29,7 @@ static inline void ha_monitor_reset_env(struct da_monitor *da_mon);
static inline void ha_setup_timer(struct ha_monitor *ha_mon);
static inline bool ha_cancel_timer(struct ha_monitor *ha_mon);
static inline void ha_cancel_timer_sync(struct ha_monitor *ha_mon);
+static inline void ha_monitor_free_env(struct da_monitor *da_mon);
static bool ha_monitor_handle_constraint(struct da_monitor *da_mon,
enum states curr_state,
enum events event,
@@ -38,6 +39,7 @@ static bool ha_monitor_handle_constraint(struct da_monitor *da_mon,
#define da_monitor_init_hook ha_monitor_init_env
#define da_monitor_reset_hook ha_monitor_reset_env
#define da_monitor_sync_hook() synchronize_rcu()
+#define da_monitor_free_hook ha_monitor_free_env
#if !defined(HA_SKIP_AUTO_CLEANUP) && RV_MON_TYPE == RV_MON_PER_TASK
/*
@@ -551,6 +553,12 @@ static inline bool ha_cancel_timer(struct ha_monitor *ha_mon)
static inline void ha_cancel_timer_sync(struct ha_monitor *ha_mon) { }
#endif
+static inline void ha_monitor_free_env(struct da_monitor *da_mon)
+{
+ if (da_monitoring(da_mon))
+ ha_cancel_timer_sync(to_ha_monitor(da_mon));
+}
+
#if IS_ENABLED(CONFIG_RV_MONITORS_KUNIT_TEST)
#ifdef RV_MON_OPS_INIT
#undef RV_MON_OPS_INIT
--
2.25.1
next prev parent reply other threads:[~2026-11-05 17:49 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-05 17:49 [PATCH v8 00/14] rv: Add task latency over budget RV monitor wen.yang
2026-11-05 17:49 ` [PATCH v8 01/14] rv: Force environment reset action on HA monitor start wen.yang
2026-11-05 17:49 ` [PATCH v8 02/14] rv: Fix ha_monitor timer and invariant underflow wen.yang
2026-11-05 17:49 ` [PATCH v8 03/14] rv: Add a KUnit test for the HA timer underflow wen.yang
2026-11-05 17:49 ` wen.yang [this message]
2026-11-05 17:49 ` [PATCH v8 05/14] rv: Add a KUnit test for the storage free path wen.yang
2026-11-05 17:49 ` [PATCH v8 06/14] rv: Make da_monitor_reset_hook, sync_hook and EVENT_NONE_LBL overridable wen.yang
2026-11-05 17:49 ` [PATCH v8 07/14] rv: Introduce DA_MON_ALLOCATION_STRATEGY wen.yang
2026-11-05 17:49 ` [PATCH v8 08/14] rv: Support embedded target types in per-object DA storage wen.yang
2026-11-05 17:49 ` [PATCH v8 09/14] rv: Add generic uprobe infrastructure for RV monitors wen.yang
2026-11-05 17:49 ` [PATCH v8 10/14] rv: Add tlob model DOT file wen.yang
2026-11-05 17:49 ` [PATCH v8 11/14] rv: Add tlob hybrid automaton monitor wen.yang
2026-11-05 17:49 ` [PATCH v8 12/14] rv: Add KUnit tests for the tlob monitor wen.yang
2026-11-05 17:49 ` [PATCH v8 13/14] selftests/verification: Add tlob selftests wen.yang
2026-11-05 17:49 ` [PATCH v8 14/14] selftests/ftrace: Walk up parent dirs to find test.d/functions wen.yang
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=0ab3632933008e028855e158b4b978881a2715dc.1793900349.git.wen.yang@linux.dev \
--to=wen.yang@linux.dev \
--cc=gmonaco@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-trace-kernel@vger.kernel.org \
--cc=namcao@linutronix.de \
/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®