bcachefs: Better inlining in core write path

Provide inline versions of some allocation functions
 - bch2_alloc_sectors_done_inlined()
 - bch2_alloc_sectors_append_ptrs_inlined()

and use them in the core IO path.

Also, inline bch2_extent_update_i_size_sectors() and
bch2_bkey_append_ptr().

In the core write path, function call overhead matters - every function
call is a jump to a new location and a potential cache miss.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/alloc_foreground.c

@@ -1244,34 +1244,11 @@ struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob)
 	};
 }
 
-/*
- * Append pointers to the space we just allocated to @k, and mark @sectors space
- * as allocated out of @ob
- */
 void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp,
 				    struct bkey_i *k, unsigned sectors,
 				    bool cached)
 {
-	struct open_bucket *ob;
-	unsigned i;
-
-	BUG_ON(sectors > wp->sectors_free);
-	wp->sectors_free	-= sectors;
-	wp->sectors_allocated	+= sectors;
-
-	open_bucket_for_each(c, &wp->ptrs, ob, i) {
-		struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
-		struct bch_extent_ptr ptr = bch2_ob_ptr(c, ob);
-
-		ptr.cached = cached ||
-			(!ca->mi.durability &&
-			 wp->data_type == BCH_DATA_user);
-
-		bch2_bkey_append_ptr(k, ptr);
-
-		BUG_ON(sectors > ob->sectors_free);
-		ob->sectors_free -= sectors;
-	}
+	bch2_alloc_sectors_append_ptrs_inlined(c, wp, k, sectors, cached);
 }
 
 /*
@@ -1280,17 +1257,7 @@ void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp,
  */
 void bch2_alloc_sectors_done(struct bch_fs *c, struct write_point *wp)
 {
-	struct open_buckets ptrs = { .nr = 0 }, keep = { .nr = 0 };
-	struct open_bucket *ob;
-	unsigned i;
-
-	open_bucket_for_each(c, &wp->ptrs, ob, i)
-		ob_push(c, !ob->sectors_free ? &ptrs : &keep, ob);
-	wp->ptrs = keep;
-
-	mutex_unlock(&wp->lock);
-
-	bch2_open_buckets_put(c, &ptrs);
+	bch2_alloc_sectors_done_inlined(c, wp);
 }
 
 static inline void writepoint_init(struct write_point *wp,

fs/bcachefs/alloc_foreground.h

@@ -4,6 +4,8 @@
 
 #include "bcachefs.h"
 #include "alloc_types.h"
+#include "extents.h"
+#include "super.h"
 
 #include <linux/hash.h>
 
@@ -81,6 +83,21 @@ static inline void bch2_open_buckets_put(struct bch_fs *c,
 	ptrs->nr = 0;
 }
 
+static inline void bch2_alloc_sectors_done_inlined(struct bch_fs *c, struct write_point *wp)
+{
+	struct open_buckets ptrs = { .nr = 0 }, keep = { .nr = 0 };
+	struct open_bucket *ob;
+	unsigned i;
+
+	open_bucket_for_each(c, &wp->ptrs, ob, i)
+		ob_push(c, !ob->sectors_free ? &ptrs : &keep, ob);
+	wp->ptrs = keep;
+
+	mutex_unlock(&wp->lock);
+
+	bch2_open_buckets_put(c, &ptrs);
+}
+
 static inline void bch2_open_bucket_get(struct bch_fs *c,
 					struct write_point *wp,
 					struct open_buckets *ptrs)
@@ -149,6 +166,38 @@ int bch2_alloc_sectors_start_trans(struct btree_trans *,
 				   struct write_point **);
 
 struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *, struct open_bucket *);
+
+/*
+ * Append pointers to the space we just allocated to @k, and mark @sectors space
+ * as allocated out of @ob
+ */
+static inline void
+bch2_alloc_sectors_append_ptrs_inlined(struct bch_fs *c, struct write_point *wp,
+				       struct bkey_i *k, unsigned sectors,
+				       bool cached)
+{
+	struct open_bucket *ob;
+	unsigned i;
+
+	BUG_ON(sectors > wp->sectors_free);
+	wp->sectors_free	-= sectors;
+	wp->sectors_allocated	+= sectors;
+
+	open_bucket_for_each(c, &wp->ptrs, ob, i) {
+		struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
+		struct bch_extent_ptr ptr = bch2_ob_ptr(c, ob);
+
+		ptr.cached = cached ||
+			(!ca->mi.durability &&
+			 wp->data_type == BCH_DATA_user);
+
+		bch2_bkey_append_ptr(k, ptr);
+
+		BUG_ON(sectors > ob->sectors_free);
+		ob->sectors_free -= sectors;
+	}
+}
+
 void bch2_alloc_sectors_append_ptrs(struct bch_fs *, struct write_point *,
 				    struct bkey_i *, unsigned, bool);
 void bch2_alloc_sectors_done(struct bch_fs *, struct write_point *);

fs/bcachefs/extents.c

@@ -698,29 +698,6 @@ void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry
 	k->k.u64s -= extent_entry_u64s(entry);
 }
 
-void bch2_bkey_append_ptr(struct bkey_i *k,
-			  struct bch_extent_ptr ptr)
-{
-	EBUG_ON(bch2_bkey_has_device(bkey_i_to_s_c(k), ptr.dev));
-
-	switch (k->k.type) {
-	case KEY_TYPE_btree_ptr:
-	case KEY_TYPE_btree_ptr_v2:
-	case KEY_TYPE_extent:
-		EBUG_ON(bkey_val_u64s(&k->k) >= BKEY_EXTENT_VAL_U64s_MAX);
-
-		ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
-
-		memcpy((void *) &k->v + bkey_val_bytes(&k->k),
-		       &ptr,
-		       sizeof(ptr));
-		k->u64s++;
-		break;
-	default:
-		BUG();
-	}
-}
-
 static inline void __extent_entry_insert(struct bkey_i *k,
 					 union bch_extent_entry *dst,
 					 union bch_extent_entry *new)

fs/bcachefs/extents.h

@@ -581,8 +581,35 @@ unsigned bch2_bkey_sectors_compressed(struct bkey_s_c);
 unsigned bch2_bkey_replicas(struct bch_fs *, struct bkey_s_c);
 unsigned bch2_bkey_durability(struct bch_fs *, struct bkey_s_c);
 
+void bch2_bkey_drop_device(struct bkey_s, unsigned);
+void bch2_bkey_drop_device_noerror(struct bkey_s, unsigned);
+const struct bch_extent_ptr *bch2_bkey_has_device(struct bkey_s_c, unsigned);
+bool bch2_bkey_has_target(struct bch_fs *, struct bkey_s_c, unsigned);
+
 void bch2_bkey_extent_entry_drop(struct bkey_i *, union bch_extent_entry *);
-void bch2_bkey_append_ptr(struct bkey_i *, struct bch_extent_ptr);
+
+static inline void bch2_bkey_append_ptr(struct bkey_i *k, struct bch_extent_ptr ptr)
+{
+	EBUG_ON(bch2_bkey_has_device(bkey_i_to_s_c(k), ptr.dev));
+
+	switch (k->k.type) {
+	case KEY_TYPE_btree_ptr:
+	case KEY_TYPE_btree_ptr_v2:
+	case KEY_TYPE_extent:
+		EBUG_ON(bkey_val_u64s(&k->k) >= BKEY_EXTENT_VAL_U64s_MAX);
+
+		ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
+
+		memcpy((void *) &k->v + bkey_val_bytes(&k->k),
+		       &ptr,
+		       sizeof(ptr));
+		k->u64s++;
+		break;
+	default:
+		BUG();
+	}
+}
+
 void bch2_extent_ptr_decoded_append(struct bkey_i *,
 				    struct extent_ptr_decoded *);
 union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s,
@@ -605,11 +632,6 @@ do {									\
 	}								\
 } while (0)
 
-void bch2_bkey_drop_device(struct bkey_s, unsigned);
-void bch2_bkey_drop_device_noerror(struct bkey_s, unsigned);
-const struct bch_extent_ptr *bch2_bkey_has_device(struct bkey_s_c, unsigned);
-bool bch2_bkey_has_target(struct bch_fs *, struct bkey_s_c, unsigned);
-
 bool bch2_bkey_matches_ptr(struct bch_fs *, struct bkey_s_c,
 			   struct bch_extent_ptr, u64);
 bool bch2_extents_match(struct bkey_s_c, struct bkey_s_c);

fs/bcachefs/io.c

@@ -832,7 +832,7 @@ static void init_append_extent(struct bch_write_op *op,
 	    crc.nonce)
 		bch2_extent_crc_append(&e->k_i, crc);
 
-	bch2_alloc_sectors_append_ptrs(op->c, wp, &e->k_i, crc.compressed_size,
+	bch2_alloc_sectors_append_ptrs_inlined(op->c, wp, &e->k_i, crc.compressed_size,
 				       op->flags & BCH_WRITE_CACHED);
 
 	bch2_keylist_push(&op->insert_keys);
@@ -1275,7 +1275,7 @@ static void __bch2_write(struct bch_write_op *op)
 		bch2_open_bucket_get(c, wp, &op->open_buckets);
 		ret = bch2_write_extent(op, wp, &bio);
 
-		bch2_alloc_sectors_done(c, wp);
+		bch2_alloc_sectors_done_inlined(c, wp);
 err:
 		if (ret <= 0) {
 			if (!(op->flags & BCH_WRITE_SYNC)) {