bcachefs: Use try_cmpxchg() family of functions instead of cmpxchg()

Use try_cmpxchg() family of functions instead of
cmpxchg (*ptr, old, new) == old. x86 CMPXCHG instruction returns
success in ZF flag, so this change saves a compare after cmpxchg
(and related move instruction in front of cmpxchg).

Also, try_cmpxchg() implicitly assigns old *ptr value to "old" when
cmpxchg fails. There is no need to re-read the value in the loop.

No functional change intended.

Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Brian Foster <bfoster@redhat.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/btree_cache.c

@@ -602,8 +602,8 @@ int bch2_btree_cache_cannibalize_lock(struct btree_trans *trans, struct closure
 	struct btree_cache *bc = &c->btree_cache;
 	struct task_struct *old;
 
-	old = cmpxchg(&bc->alloc_lock, NULL, current);
+	old = NULL;
-	if (old == NULL || old == current)
+	if (try_cmpxchg(&bc->alloc_lock, &old, current) || old == current)
 		goto success;
 
 	if (!cl) {
@@ -614,8 +614,8 @@ int bch2_btree_cache_cannibalize_lock(struct btree_trans *trans, struct closure
 	closure_wait(&bc->alloc_wait, cl);
 
 	/* Try again, after adding ourselves to waitlist */
-	old = cmpxchg(&bc->alloc_lock, NULL, current);
+	old = NULL;
-	if (old == NULL || old == current) {
+	if (try_cmpxchg(&bc->alloc_lock, &old, current) || old == current) {
 		/* We raced */
 		closure_wake_up(&bc->alloc_wait);
 		goto success;

fs/bcachefs/btree_io.c

@@ -1796,15 +1796,16 @@ int bch2_btree_root_read(struct bch_fs *c, enum btree_id id,
 static void bch2_btree_complete_write(struct bch_fs *c, struct btree *b,
 				      struct btree_write *w)
 {
-	unsigned long old, new, v = READ_ONCE(b->will_make_reachable);
+	unsigned long old, new;
 
+	old = READ_ONCE(b->will_make_reachable);
 	do {
-		old = new = v;
+		new = old;
 		if (!(old & 1))
 			break;
 
 		new &= ~1UL;
-	} while ((v = cmpxchg(&b->will_make_reachable, old, new)) != old);
+	} while (!try_cmpxchg(&b->will_make_reachable, &old, new));
 
 	if (old & 1)
 		closure_put(&((struct btree_update *) new)->cl);
@@ -1815,14 +1816,14 @@ static void bch2_btree_complete_write(struct bch_fs *c, struct btree *b,
 static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
 {
 	struct btree_write *w = btree_prev_write(b);
-	unsigned long old, new, v;
+	unsigned long old, new;
 	unsigned type = 0;
 
 	bch2_btree_complete_write(c, b, w);
 
-	v = READ_ONCE(b->flags);
+	old = READ_ONCE(b->flags);
 	do {
-		old = new = v;
+		new = old;
 
 		if ((old & (1U << BTREE_NODE_dirty)) &&
 		    (old & (1U << BTREE_NODE_need_write)) &&
@@ -1842,7 +1843,7 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
 			new &= ~(1U << BTREE_NODE_write_in_flight);
 			new &= ~(1U << BTREE_NODE_write_in_flight_inner);
 		}
-	} while ((v = cmpxchg(&b->flags, old, new)) != old);
+	} while (!try_cmpxchg(&b->flags, &old, new));
 
 	if (new & (1U << BTREE_NODE_write_in_flight))
 		__bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED|type);
@@ -2014,8 +2015,9 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags)
 	 * dirty bit requires a write lock, we can't race with other threads
 	 * redirtying it:
 	 */
+	old = READ_ONCE(b->flags);
 	do {
-		old = new = READ_ONCE(b->flags);
+		new = old;
 
 		if (!(old & (1 << BTREE_NODE_dirty)))
 			return;
@@ -2046,7 +2048,7 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags)
 		new |=  (1 << BTREE_NODE_write_in_flight_inner);
 		new |=  (1 << BTREE_NODE_just_written);
 		new ^=  (1 << BTREE_NODE_write_idx);
-	} while (cmpxchg_acquire(&b->flags, old, new) != old);
+	} while (!try_cmpxchg_acquire(&b->flags, &old, new));
 
 	if (new & (1U << BTREE_NODE_need_write))
 		return;

fs/bcachefs/btree_trans_commit.c

@@ -228,14 +228,14 @@ static int __btree_node_flush(struct journal *j, struct journal_entry_pin *pin,
 	struct btree_write *w = container_of(pin, struct btree_write, journal);
 	struct btree *b = container_of(w, struct btree, writes[i]);
 	struct btree_trans *trans = bch2_trans_get(c);
-	unsigned long old, new, v;
+	unsigned long old, new;
 	unsigned idx = w - b->writes;
 
 	btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
-	v = READ_ONCE(b->flags);
 
+	old = READ_ONCE(b->flags);
 	do {
-		old = new = v;
+		new = old;
 
 		if (!(old & (1 << BTREE_NODE_dirty)) ||
 		    !!(old & (1 << BTREE_NODE_write_idx)) != idx ||
@@ -245,7 +245,7 @@ static int __btree_node_flush(struct journal *j, struct journal_entry_pin *pin,
 		new &= ~BTREE_WRITE_TYPE_MASK;
 		new |= BTREE_WRITE_journal_reclaim;
 		new |= 1 << BTREE_NODE_need_write;
-	} while ((v = cmpxchg(&b->flags, old, new)) != old);
+	} while (!try_cmpxchg(&b->flags, &old, new));
 
 	btree_node_write_if_need(c, b, SIX_LOCK_read);
 	six_unlock_read(&b->c.lock);

fs/bcachefs/btree_update_interior.c

@@ -1356,7 +1356,7 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
 	struct bch_fs *c = as->c;
 	struct bkey_packed *k;
 	struct printbuf buf = PRINTBUF;
-	unsigned long old, new, v;
+	unsigned long old, new;
 
 	BUG_ON(insert->k.type == KEY_TYPE_btree_ptr_v2 &&
 	       !btree_ptr_sectors_written(insert));
@@ -1395,14 +1395,14 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
 	bch2_btree_bset_insert_key(trans, path, b, node_iter, insert);
 	set_btree_node_dirty_acct(c, b);
 
-	v = READ_ONCE(b->flags);
+	old = READ_ONCE(b->flags);
 	do {
-		old = new = v;
+		new = old;
 
 		new &= ~BTREE_WRITE_TYPE_MASK;
 		new |= BTREE_WRITE_interior;
 		new |= 1 << BTREE_NODE_need_write;
-	} while ((v = cmpxchg(&b->flags, old, new)) != old);
+	} while (!try_cmpxchg(&b->flags, &old, new));
 
 	printbuf_exit(&buf);
 }

fs/bcachefs/buckets.c

@@ -916,13 +916,13 @@ void bch2_trans_account_disk_usage_change(struct btree_trans *trans)
 	 */
 	s64 should_not_have_added = added - (s64) disk_res_sectors;
 	if (unlikely(should_not_have_added > 0)) {
-		u64 old, new, v = atomic64_read(&c->sectors_available);
+		u64 old, new;
 
+		old = atomic64_read(&c->sectors_available);
 		do {
-			old = v;
 			new = max_t(s64, 0, old - should_not_have_added);
-		} while ((v = atomic64_cmpxchg(&c->sectors_available,
+		} while (!atomic64_try_cmpxchg(&c->sectors_available,
-					       old, new)) != old);
+					       &old, new));
 
 		added -= should_not_have_added;
 		warn = true;
@@ -1523,7 +1523,7 @@ int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
 			      u64 sectors, int flags)
 {
 	struct bch_fs_pcpu *pcpu;
-	u64 old, v, get;
+	u64 old, get;
 	s64 sectors_available;
 	int ret;
 
@@ -1534,17 +1534,16 @@ int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
 	if (sectors <= pcpu->sectors_available)
 		goto out;
 
-	v = atomic64_read(&c->sectors_available);
+	old = atomic64_read(&c->sectors_available);
 	do {
-		old = v;
 		get = min((u64) sectors + SECTORS_CACHE, old);
 
 		if (get < sectors) {
 			preempt_enable();
 			goto recalculate;
 		}
-	} while ((v = atomic64_cmpxchg(&c->sectors_available,
+	} while (!atomic64_try_cmpxchg(&c->sectors_available,
-				       old, old - get)) != old);
+				       &old, old - get));
 
 	pcpu->sectors_available		+= get;
 

fs/bcachefs/buckets.h

@@ -432,13 +432,13 @@ static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reserv
 #ifdef __KERNEL__
 	u64 old, new;
 
+	old = this_cpu_read(c->pcpu->sectors_available);
 	do {
-		old = this_cpu_read(c->pcpu->sectors_available);
 		if (sectors > old)
 			return __bch2_disk_reservation_add(c, res, sectors, flags);
 
 		new = old - sectors;
-	} while (this_cpu_cmpxchg(c->pcpu->sectors_available, old, new) != old);
+	} while (!this_cpu_try_cmpxchg(c->pcpu->sectors_available, &old, new));
 
 	this_cpu_add(*c->online_reserved, sectors);
 	res->sectors			+= sectors;

fs/bcachefs/io_write.c

@@ -69,11 +69,10 @@ void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw)
 	u64 io_latency = time_after64(now, submit_time)
 		? now - submit_time
 		: 0;
-	u64 old, new, v = atomic64_read(latency);
+	u64 old, new;
 
+	old = atomic64_read(latency);
 	do {
-		old = v;
-
 		/*
 		 * If the io latency was reasonably close to the current
 		 * latency, skip doing the update and atomic operation - most of
@@ -84,7 +83,7 @@ void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw)
 			break;
 
 		new = ewma_add(old, io_latency, 5);
-	} while ((v = atomic64_cmpxchg(latency, old, new)) != old);
+	} while (!atomic64_try_cmpxchg(latency, &old, new));
 
 	bch2_congested_acct(ca, io_latency, now, rw);
 

fs/bcachefs/journal.c

@@ -230,7 +230,6 @@ static void __journal_entry_close(struct journal *j, unsigned closed_val, bool t
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_buf *buf = journal_cur_buf(j);
 	union journal_res_state old, new;
-	u64 v = atomic64_read(&j->reservations.counter);
 	unsigned sectors;
 
 	BUG_ON(closed_val != JOURNAL_ENTRY_CLOSED_VAL &&
@@ -238,15 +237,16 @@ static void __journal_entry_close(struct journal *j, unsigned closed_val, bool t
 
 	lockdep_assert_held(&j->lock);
 
+	old.v = atomic64_read(&j->reservations.counter);
 	do {
-		old.v = new.v = v;
+		new.v = old.v;
 		new.cur_entry_offset = closed_val;
 
 		if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL ||
 		    old.cur_entry_offset == new.cur_entry_offset)
 			return;
-	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
+	} while (!atomic64_try_cmpxchg(&j->reservations.counter,
-				       old.v, new.v)) != old.v);
+				       &old.v, new.v));
 
 	if (!__journal_entry_is_open(old))
 		return;
@@ -353,7 +353,6 @@ static int journal_entry_open(struct journal *j)
 		((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 	union journal_res_state old, new;
 	int u64s;
-	u64 v;
 
 	lockdep_assert_held(&j->lock);
 	BUG_ON(journal_entry_is_open(j));
@@ -432,9 +431,9 @@ static int journal_entry_open(struct journal *j)
 	 */
 	j->cur_entry_u64s = u64s;
 
-	v = atomic64_read(&j->reservations.counter);
+	old.v = atomic64_read(&j->reservations.counter);
 	do {
-		old.v = new.v = v;
+		new.v = old.v;
 
 		BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 
@@ -446,8 +445,8 @@ static int journal_entry_open(struct journal *j)
 
 		/* Handle any already added entries */
 		new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
-	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
+	} while (!atomic64_try_cmpxchg(&j->reservations.counter,
-				       old.v, new.v)) != old.v);
+				       &old.v, new.v));
 
 	if (nr_unwritten_journal_entries(j) == 1)
 		mod_delayed_work(j->wq,

fs/bcachefs/journal.h

@@ -327,10 +327,10 @@ static inline int journal_res_get_fast(struct journal *j,
 				       unsigned flags)
 {
 	union journal_res_state old, new;
-	u64 v = atomic64_read(&j->reservations.counter);
 
+	old.v = atomic64_read(&j->reservations.counter);
 	do {
-		old.v = new.v = v;
+		new.v = old.v;
 
 		/*
 		 * Check if there is still room in the current journal
@@ -356,8 +356,8 @@ static inline int journal_res_get_fast(struct journal *j,
 
 		if (flags & JOURNAL_RES_GET_CHECK)
 			return 1;
-	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
+	} while (!atomic64_try_cmpxchg(&j->reservations.counter,
-				       old.v, new.v)) != old.v);
+				       &old.v, new.v));
 
 	res->ref	= true;
 	res->idx	= old.idx;

fs/bcachefs/journal_io.c

@@ -1588,7 +1588,7 @@ static CLOSURE_CALLBACK(journal_write_done)
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct bch_replicas_padded replicas;
 	union journal_res_state old, new;
-	u64 v, seq = le64_to_cpu(w->data->seq);
+	u64 seq = le64_to_cpu(w->data->seq);
 	int err = 0;
 
 	bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
@@ -1647,14 +1647,15 @@ static CLOSURE_CALLBACK(journal_write_done)
 		if (j->watermark != BCH_WATERMARK_stripe)
 			journal_reclaim_kick(&c->journal);
 
-		v = atomic64_read(&j->reservations.counter);
+		old.v = atomic64_read(&j->reservations.counter);
 		do {
-			old.v = new.v = v;
+			new.v = old.v;
 			BUG_ON(journal_state_count(new, new.unwritten_idx));
 			BUG_ON(new.unwritten_idx != (seq & JOURNAL_BUF_MASK));
 
 			new.unwritten_idx++;
-		} while ((v = atomic64_cmpxchg(&j->reservations.counter, old.v, new.v)) != old.v);
+		} while (!atomic64_try_cmpxchg(&j->reservations.counter,
+					       &old.v, new.v));
 
 		closure_wake_up(&w->wait);
 		completed = true;

fs/bcachefs/two_state_shared_lock.h

@@ -36,15 +36,14 @@ static inline void bch2_two_state_unlock(two_state_lock_t *lock, int s)
 static inline bool bch2_two_state_trylock(two_state_lock_t *lock, int s)
 {
 	long i = s ? 1 : -1;
-	long v = atomic_long_read(&lock->v), old;
+	long old;
 
+	old = atomic_long_read(&lock->v);
 	do {
-		old = v;
+		if (i > 0 ? old < 0 : old > 0)
-
-		if (i > 0 ? v < 0 : v > 0)
 			return false;
-	} while ((v = atomic_long_cmpxchg_acquire(&lock->v,
+	} while (!atomic_long_try_cmpxchg_acquire(&lock->v, &old, old + i));
-					old, old + i)) != old);
+
 	return true;
 }