[PATCH 6/8] lockdep: Free unused reservation slabs to buddy allocator at late boot

[Jim Cromie <jim.cromie@gmail.com>]

During early boot, lockdep reserves a generous slab pool (e.g. 64 slabs =
4 MB) from memblock to guarantee uninterrupted initialization. Once
the kernel reaches late_initcall, the boot locking storm is complete,
and lockdep's steady-state working set is known.

Add lockdep_post_init_trim() as a late_initcall_sync handler. Calculate
required runtime headroom (default 100% headroom over boot usage, with
a floor of 32 slabs, or custom lockdep_slabs=N / lockdep_headroom=M%),
and return all excess slabs to the buddy page allocator via
free_reserved_page().

Also register a reboot notifier to log total lifetime slab consumption
and remaining headroom on clean system shutdown.

Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
---
 kernel/locking/lockdep.c      | 71 +++++++++++++++++++++++++++++++++++++++++++
 kernel/locking/lockdep_proc.c | 23 +++++++-------
 2 files changed, 83 insertions(+), 11 deletions(-)

diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
index b2dc7619a5e3..3fb6c07cea36 100644
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -6992,6 +6992,77 @@ void __init lockdep_init(void)
 	       sizeof(((struct task_struct *)NULL)->held_locks));
 }
 
+static int lockdep_shutdown_notify(struct notifier_block *nb,
+				   unsigned long code, void *unused)
+{
+	unsigned int used = lockdep_slabs_used - lockdep_nr_free_slabs;
+	unsigned int total = lockdep_nr_slabs;
+	unsigned int free_slabs = total > used ? total - used : 0;
+	unsigned int headroom_pct = used ? (free_slabs * 100) / used : 0;
+
+	pr_info("lockdep: shutdown summary : %u/%u slabs (%u kB/%u kB, %u%% headroom left), %lu classes, %lu chains, %u hlocks\n",
+		used, total,
+		(used * LOCKDEP_SLAB_SIZE) / 1024,
+		(total * LOCKDEP_SLAB_SIZE) / 1024,
+		headroom_pct,
+		nr_lock_classes, lock_chain_count(), chain_hlocks_used());
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block lockdep_reboot_nb = {
+	.notifier_call = lockdep_shutdown_notify,
+};
+
+static int __init lockdep_post_init_trim(void)
+{
+	unsigned int used = lockdep_slabs_used - lockdep_nr_free_slabs;
+	unsigned int initial_slabs = lockdep_nr_slabs;
+	unsigned int target_slabs, kept_headroom_slabs;
+	unsigned int freed_slabs = 0;
+	unsigned int i;
+
+	if (!lockdep_nr_slabs)
+		return 0;
+
+	/* Compute headroom requirement (default 100% or custom percentage) with 32-slab floor */
+	kept_headroom_slabs = DIV_ROUND_UP(used * requested_lockdep_headroom_pct, 100);
+	if (kept_headroom_slabs < 32)
+		kept_headroom_slabs = 32;
+
+	target_slabs = used + kept_headroom_slabs;
+
+	/* Satisfy both: target_slabs >= requested_lockdep_slabs AND headroom >= M% */
+	if (requested_lockdep_slabs > target_slabs)
+		target_slabs = requested_lockdep_slabs;
+
+	target_slabs = clamp_t(unsigned int, target_slabs, used, lockdep_nr_slabs);
+
+	/* Release excess slabs to buddy allocator */
+	if (target_slabs < lockdep_nr_slabs) {
+		for (i = target_slabs; i < lockdep_nr_slabs; i++) {
+			struct page *page = virt_to_page(lockdep_slabs[i]);
+			unsigned long p;
+
+			for (p = 0; p < (LOCKDEP_SLAB_SIZE >> PAGE_SHIFT); p++)
+				free_reserved_page(page + p);
+
+			lockdep_slabs[i] = NULL;
+		}
+		freed_slabs = lockdep_nr_slabs - target_slabs;
+		lockdep_nr_slabs = target_slabs;
+	}
+
+	pr_info("lockdep: boot complete : %u/%u slabs used, %u kept (%u%% headroom), %u returned to buddy (%u kB freed)\n",
+		used, initial_slabs, lockdep_nr_slabs,
+		lockdep_nr_slabs > used ? ((lockdep_nr_slabs - used) * 100) / used : 0,
+		freed_slabs, (freed_slabs * LOCKDEP_SLAB_SIZE) / 1024);
+
+	register_reboot_notifier(&lockdep_reboot_nb);
+	return 0;
+}
+late_initcall_sync(lockdep_post_init_trim);
+
 static void
 print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
 		     const void *mem_to, struct held_lock *hlock)
diff --git a/kernel/locking/lockdep_proc.c b/kernel/locking/lockdep_proc.c
index 1916db9aa46b..95b76047918d 100644
--- a/kernel/locking/lockdep_proc.c
+++ b/kernel/locking/lockdep_proc.c
@@ -32,16 +32,17 @@
  * bitmap and max_lock_class_idx.
  */
 #define iterate_lock_classes(idx, class)				\
-	for (idx = 0, class = lock_classes; idx <= max_lock_class_idx;	\
-	     idx++, class++)
+	for (idx = 0, class = idx_to_lock_class(0);			\
+	     idx <= max_lock_class_idx;					\
+	     idx++, class = idx_to_lock_class(idx))
 
 static void *l_next(struct seq_file *m, void *v, loff_t *pos)
 {
-	struct lock_class *class = v;
+	unsigned long idx = ++*pos;
 
-	++class;
-	*pos = class - lock_classes;
-	return (*pos > max_lock_class_idx) ? NULL : class;
+	if (idx > max_lock_class_idx)
+		return NULL;
+	return idx_to_lock_class(idx);
 }
 
 static void *l_start(struct seq_file *m, loff_t *pos)
@@ -50,7 +51,7 @@ static void *l_start(struct seq_file *m, loff_t *pos)
 
 	if (idx > max_lock_class_idx)
 		return NULL;
-	return lock_classes + idx;
+	return idx_to_lock_class(idx);
 }
 
 static void l_stop(struct seq_file *m, void *v)
@@ -59,7 +60,7 @@ static void l_stop(struct seq_file *m, void *v)
 
 static void print_name(struct seq_file *m, struct lock_class *class)
 {
-	char str[KSYM_NAME_LEN];
+	char str[128];
 	const char *name = class->name;
 
 	if (!name) {
@@ -79,9 +80,9 @@ static int l_show(struct seq_file *m, void *v)
 	struct lock_class *class = v;
 	struct lock_list *entry;
 	char usage[LOCK_USAGE_CHARS];
-	int idx = class - lock_classes;
+	int idx = class->class_idx;
 
-	if (v == lock_classes)
+	if (idx == 0)
 		seq_printf(m, "all lock classes:\n");
 
 	if (!test_bit(idx, lock_classes_in_use))
@@ -133,7 +134,7 @@ static void *lc_start(struct seq_file *m, loff_t *pos)
 	if (*pos == 0)
 		return SEQ_START_TOKEN;
 
-	return lock_chains + (*pos - 1);
+	return idx_to_lock_chain(*pos - 1);
 }
 
 static void *lc_next(struct seq_file *m, void *v, loff_t *pos)

-- 
2.55.0