bcachefs: Factor out acc_u64s()

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/btree_gc.c

@@ -482,29 +482,35 @@ static void bch2_gc_free(struct bch_fs *c)
 	c->usage[1] = NULL;
 }
 
-static void fs_usage_reset(struct bch_fs_usage *fs_usage)
+/*
+ * Accumulate percpu counters onto one cpu's copy - only valid when access
+ * against any percpu counter is guarded against
+ */
+static u64 *acc_percpu_u64s(u64 __percpu *p, unsigned nr)
 {
-	unsigned offset = offsetof(typeof(*fs_usage), s.gc_start);
+	u64 *ret;
+	int cpu;
 
-	memset((void *) fs_usage + offset, 0,
-	       sizeof(*fs_usage) - offset);
-}
+	preempt_disable();
+	ret = this_cpu_ptr(p);
+	preempt_enable();
 
-static void fs_usage_cpy(struct bch_fs_usage *dst,
-			 struct bch_fs_usage *src)
-{
-	unsigned offset = offsetof(typeof(*dst), s.gc_start);
+	for_each_possible_cpu(cpu) {
+		u64 *i = per_cpu_ptr(p, cpu);
 
-	memcpy((void *) dst + offset,
-	       (void *) src + offset,
-	       sizeof(*dst) - offset);
+		if (i != ret) {
+			acc_u64s(ret, i, nr);
+			memset(i, 0, nr * sizeof(u64));
+		}
+	}
+
+	return ret;
 }
 
 static void bch2_gc_done_nocheck(struct bch_fs *c)
 {
 	struct bch_dev *ca;
 	unsigned i;
-	int cpu;
 
 	{
 		struct genradix_iter dst_iter = genradix_iter_init(&c->stripes[0], 0);
@@ -534,42 +540,39 @@ static void bch2_gc_done_nocheck(struct bch_fs *c)
 	};
 
 	for_each_member_device(ca, c, i) {
-		struct bch_dev_usage *p;
+		unsigned nr = sizeof(struct bch_dev_usage) / sizeof(u64);
+		struct bch_dev_usage *dst = (void *)
+			acc_percpu_u64s((void *) ca->usage[0], nr);
+		struct bch_dev_usage *src = (void *)
+			acc_percpu_u64s((void *) ca->usage[1], nr);
 
-		for_each_possible_cpu(cpu) {
-			p = per_cpu_ptr(ca->usage[0], cpu);
-			memset(p, 0, sizeof(*p));
-		}
-
-		preempt_disable();
-		*this_cpu_ptr(ca->usage[0]) = __bch2_dev_usage_read(ca, 1);
-		preempt_enable();
+		*dst = *src;
 	}
 
 	{
-		struct bch_fs_usage src = __bch2_fs_usage_read(c, 1);
-
-		for_each_possible_cpu(cpu)
-			fs_usage_reset(per_cpu_ptr(c->usage[0], cpu));
+		unsigned nr = sizeof(struct bch_fs_usage) / sizeof(u64);
+		struct bch_fs_usage *dst = (void *)
+			acc_percpu_u64s((void *) c->usage[0], nr);
+		struct bch_fs_usage *src = (void *)
+			acc_percpu_u64s((void *) c->usage[1], nr);
+		unsigned offset = offsetof(typeof(*dst), s.gc_start);
 
-		preempt_disable();
-		fs_usage_cpy(this_cpu_ptr(c->usage[0]), &src);
-		preempt_enable();
+		memcpy((void *) dst + offset,
+		       (void *) src + offset,
+		       sizeof(*dst) - offset);
 	}
-
 }
 
 static void bch2_gc_done(struct bch_fs *c, bool initial)
 {
 	struct bch_dev *ca;
 	unsigned i;
-	int cpu;
 
 #define copy_field(_f, _msg, ...)					\
-	if (dst._f != src._f) {						\
-		bch_err(c, _msg ": got %llu, should be %llu, fixing"\
-			, ##__VA_ARGS__, dst._f, src._f);		\
-		dst._f = src._f;					\
+	if (dst->_f != src->_f) {					\
+		bch_err(c, _msg ": got %llu, should be %llu, fixing"	\
+			, ##__VA_ARGS__, dst->_f, src->_f);		\
+		dst->_f = src->_f;					\
 	}
 #define copy_stripe_field(_f, _msg, ...)				\
 	if (dst->_f != src->_f) {					\
@@ -650,9 +653,11 @@ static void bch2_gc_done(struct bch_fs *c, bool initial)
 	};
 
 	for_each_member_device(ca, c, i) {
-		struct bch_dev_usage dst = __bch2_dev_usage_read(ca, 0);
-		struct bch_dev_usage src = __bch2_dev_usage_read(ca, 1);
-		struct bch_dev_usage *p;
+		unsigned nr = sizeof(struct bch_dev_usage) / sizeof(u64);
+		struct bch_dev_usage *dst = (void *)
+			acc_percpu_u64s((void *) ca->usage[0], nr);
+		struct bch_dev_usage *src = (void *)
+			acc_percpu_u64s((void *) ca->usage[1], nr);
 		unsigned b;
 
 		for (b = 0; b < BCH_DATA_NR; b++)
@@ -666,21 +671,14 @@ static void bch2_gc_done(struct bch_fs *c, bool initial)
 				       "sectors[%s]", bch2_data_types[b]);
 		copy_dev_field(sectors_fragmented,
 			       "sectors_fragmented");
-
-		for_each_possible_cpu(cpu) {
-			p = per_cpu_ptr(ca->usage[0], cpu);
-			memset(p, 0, sizeof(*p));
-		}
-
-		preempt_disable();
-		p = this_cpu_ptr(ca->usage[0]);
-		*p = dst;
-		preempt_enable();
 	}
 
 	{
-		struct bch_fs_usage dst = __bch2_fs_usage_read(c, 0);
-		struct bch_fs_usage src = __bch2_fs_usage_read(c, 1);
+		unsigned nr = sizeof(struct bch_fs_usage) / sizeof(u64);
+		struct bch_fs_usage *dst = (void *)
+			acc_percpu_u64s((void *) c->usage[0], nr);
+		struct bch_fs_usage *src = (void *)
+			acc_percpu_u64s((void *) c->usage[1], nr);
 		unsigned r, b;
 
 		copy_fs_field(s.hidden,		"hidden");
@@ -703,13 +701,6 @@ static void bch2_gc_done(struct bch_fs *c, bool initial)
 		for (b = 0; b < BCH_DATA_NR; b++)
 			copy_fs_field(buckets[b],
 				      "buckets[%s]", bch2_data_types[b]);
-
-		for_each_possible_cpu(cpu)
-			fs_usage_reset(per_cpu_ptr(c->usage[0], cpu));
-
-		preempt_disable();
-		fs_usage_cpy(this_cpu_ptr(c->usage[0]), &dst);
-		preempt_enable();
 	}
 out:
 	percpu_up_write(&c->mark_lock);

fs/bcachefs/buckets.c

@@ -78,77 +78,6 @@
 
 static inline u64 __bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage);
 
-#ifdef DEBUG_BUCKETS
-
-#define lg_local_lock	lg_global_lock
-#define lg_local_unlock	lg_global_unlock
-
-static void bch2_fs_stats_verify(struct bch_fs *c)
-{
-	struct bch_fs_usage stats =_bch2_fs_usage_read(c);
-	unsigned i, j;
-
-	for (i = 0; i < ARRAY_SIZE(stats.replicas); i++) {
-		for (j = 0; j < ARRAY_SIZE(stats.replicas[i].data); j++)
-			if ((s64) stats.replicas[i].data[j] < 0)
-				panic("replicas %u %s sectors underflow: %lli\n",
-				      i + 1, bch_data_types[j],
-				      stats.replicas[i].data[j]);
-
-		if ((s64) stats.replicas[i].persistent_reserved < 0)
-			panic("replicas %u reserved underflow: %lli\n",
-			      i + 1, stats.replicas[i].persistent_reserved);
-	}
-
-	for (j = 0; j < ARRAY_SIZE(stats.buckets); j++)
-		if ((s64) stats.replicas[i].data_buckets[j] < 0)
-			panic("%s buckets underflow: %lli\n",
-			      bch_data_types[j],
-			      stats.buckets[j]);
-
-	if ((s64) stats.s.online_reserved < 0)
-		panic("sectors_online_reserved underflow: %lli\n",
-		      stats.s.online_reserved);
-}
-
-static void bch2_dev_stats_verify(struct bch_dev *ca)
-{
-	struct bch_dev_usage stats =
-		__bch2_dev_usage_read(ca);
-	u64 n = ca->mi.nbuckets - ca->mi.first_bucket;
-	unsigned i;
-
-	for (i = 0; i < ARRAY_SIZE(stats.buckets); i++)
-		BUG_ON(stats.buckets[i]		> n);
-	BUG_ON(stats.buckets_alloc		> n);
-	BUG_ON(stats.buckets_unavailable	> n);
-}
-
-static void bch2_disk_reservations_verify(struct bch_fs *c, int flags)
-{
-	if (!(flags & BCH_DISK_RESERVATION_NOFAIL)) {
-		u64 used = __bch2_fs_sectors_used(c);
-		u64 cached = 0;
-		u64 avail = atomic64_read(&c->sectors_available);
-		int cpu;
-
-		for_each_possible_cpu(cpu)
-			cached += per_cpu_ptr(c->usage_percpu, cpu)->available_cache;
-
-		if (used + avail + cached > c->capacity)
-			panic("used %llu avail %llu cached %llu capacity %llu\n",
-			      used, avail, cached, c->capacity);
-	}
-}
-
-#else
-
-static void bch2_fs_stats_verify(struct bch_fs *c) {}
-static void bch2_dev_stats_verify(struct bch_dev *ca) {}
-static void bch2_disk_reservations_verify(struct bch_fs *c, int flags) {}
-
-#endif
-
 /*
  * Clear journal_seq_valid for buckets for which it's not needed, to prevent
  * wraparound:
@@ -186,43 +115,23 @@ void bch2_bucket_seq_cleanup(struct bch_fs *c)
 	}
 }
 
-#define bch2_usage_add(_acc, _stats)					\
-do {									\
-	typeof(_acc) _a = (_acc), _s = (_stats);			\
-	unsigned i;							\
-									\
-	for (i = 0; i < sizeof(*_a) / sizeof(u64); i++)			\
-		((u64 *) (_a))[i] += ((u64 *) (_s))[i];			\
-} while (0)
-
 #define bch2_usage_read_raw(_stats)					\
 ({									\
 	typeof(*this_cpu_ptr(_stats)) _acc;				\
-	int cpu;							\
 									\
 	memset(&_acc, 0, sizeof(_acc));					\
-									\
-	for_each_possible_cpu(cpu)					\
-		bch2_usage_add(&_acc, per_cpu_ptr((_stats), cpu));	\
+	acc_u64s_percpu((u64 *) &_acc,					\
+			(u64 __percpu *) _stats,			\
+			sizeof(_acc) / sizeof(u64));			\
 									\
 	_acc;								\
 })
 
-struct bch_dev_usage __bch2_dev_usage_read(struct bch_dev *ca, bool gc)
-{
-	return bch2_usage_read_raw(ca->usage[gc]);
-}
-
 struct bch_dev_usage bch2_dev_usage_read(struct bch_fs *c, struct bch_dev *ca)
 {
 	return bch2_usage_read_raw(ca->usage[0]);
 }
 
-struct bch_fs_usage __bch2_fs_usage_read(struct bch_fs *c, bool gc)
-{
-	return bch2_usage_read_raw(c->usage[gc]);
-}
-
 struct bch_fs_usage bch2_fs_usage_read(struct bch_fs *c)
 {
 	return bch2_usage_read_raw(c->usage[0]);
@@ -326,12 +235,16 @@ void bch2_fs_usage_apply(struct bch_fs *c,
 	}
 
 	preempt_disable();
-	bch2_usage_add(this_cpu_ptr(c->usage[0]), fs_usage);
-
-	if (gc_visited(c, gc_pos))
-		bch2_usage_add(this_cpu_ptr(c->usage[1]), fs_usage);
-
-	bch2_fs_stats_verify(c);
+	acc_u64s((u64 *) this_cpu_ptr(c->usage[0]),
+		 (u64 *) fs_usage,
+		 sizeof(*fs_usage) / sizeof(u64));
+
+	if (gc_visited(c, gc_pos)) {
+		BUG_ON(!c->usage[1]);
+		acc_u64s((u64 *) this_cpu_ptr(c->usage[1]),
+			 (u64 *) fs_usage,
+			 sizeof(*fs_usage) / sizeof(u64));
+	}
 	preempt_enable();
 
 	memset(fs_usage, 0, sizeof(*fs_usage));
@@ -392,8 +305,6 @@ static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
 
 	if (!is_available_bucket(old) && is_available_bucket(new))
 		bch2_wake_allocator(ca);
-
-	bch2_dev_stats_verify(ca);
 }
 
 void bch2_dev_usage_from_buckets(struct bch_fs *c, struct bch_dev *ca)
@@ -1011,8 +922,6 @@ void __bch2_disk_reservation_put(struct bch_fs *c, struct disk_reservation *res)
 {
 	percpu_down_read(&c->mark_lock);
 	this_cpu_sub(c->usage[0]->s.online_reserved, res->sectors);
-
-	bch2_fs_stats_verify(c);
 	percpu_up_read(&c->mark_lock);
 
 	res->sectors = 0;
@@ -1055,8 +964,6 @@ int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
 	this_cpu_add(c->usage[0]->s.online_reserved, sectors);
 	res->sectors			+= sectors;
 
-	bch2_disk_reservations_verify(c, flags);
-	bch2_fs_stats_verify(c);
 	preempt_enable();
 	percpu_up_read(&c->mark_lock);
 	return 0;
@@ -1089,14 +996,11 @@ int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
 		this_cpu_add(c->usage[0]->s.online_reserved, sectors);
 		res->sectors			+= sectors;
 		ret = 0;
-
-		bch2_disk_reservations_verify(c, flags);
 	} else {
 		atomic64_set(&c->sectors_available, sectors_available);
 		ret = -ENOSPC;
 	}
 
-	bch2_fs_stats_verify(c);
 	percpu_up_write(&c->mark_lock);
 
 	if (!(flags & BCH_DISK_RESERVATION_GC_LOCK_HELD))

fs/bcachefs/buckets.h

@@ -180,7 +180,6 @@ static inline bool bucket_needs_journal_commit(struct bucket_mark m,
 
 /* Device usage: */
 
-struct bch_dev_usage __bch2_dev_usage_read(struct bch_dev *, bool);
 struct bch_dev_usage bch2_dev_usage_read(struct bch_fs *, struct bch_dev *);
 
 static inline u64 __dev_buckets_available(struct bch_dev *ca,
@@ -219,7 +218,6 @@ static inline u64 dev_buckets_free(struct bch_fs *c, struct bch_dev *ca)
 
 /* Filesystem usage: */
 
-struct bch_fs_usage __bch2_fs_usage_read(struct bch_fs *, bool);
 struct bch_fs_usage bch2_fs_usage_read(struct bch_fs *);
 
 u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage);

fs/bcachefs/util.h

@@ -700,4 +700,21 @@ do {									\
 	}								\
 } while (0)
 
+static inline void acc_u64s(u64 *acc, const u64 *src, unsigned nr)
+{
+	unsigned i;
+
+	for (i = 0; i < nr; i++)
+		acc[i] += src[i];
+}
+
+static inline void acc_u64s_percpu(u64 *acc, const u64 __percpu *src,
+				   unsigned nr)
+{
+	int cpu;
+
+	for_each_possible_cpu(cpu)
+		acc_u64s(acc, per_cpu_ptr(src, cpu), nr);
+}
+
 #endif /* _BCACHEFS_UTIL_H */