Minor ARC optimizations

Remove unneeded global, practically constant, state pointer variables
(arc_anon, arc_mru, etc.), replacing them with macros of real state
variables addresses (&ARC_anon, &ARC_mru, etc.). 

Change ARC_EVICT_ALL from -1ULL to UINT64_MAX, not requiring special
handling in inner loop of ARC reclamation.  Respectively change bytes
argument of arc_evict_state() from int64_t to uint64_t.

Reviewed-by: Matthew Ahrens <[email protected]>
Reviewed-by: Mark Maybee <[email protected]>
Reviewed-by: Ryan Moeller <[email protected]>
Signed-off-by: Alexander Motin <[email protected]>
Closes openzfs#12348

include/sys/arc.h

@@ -44,7 +44,7 @@ extern "C" {
  * Used by arc_flush() to inform arc_evict_state() that it should evict
  * all available buffers from the arc state being passed in.
  */
-#define	ARC_EVICT_ALL	-1ULL
+#define	ARC_EVICT_ALL	UINT64_MAX
 
 #define	HDR_SET_LSIZE(hdr, x) do { \
 	ASSERT(IS_P2ALIGNED(x, 1U << SPA_MINBLOCKSHIFT)); \

include/sys/arc_impl.h

@@ -964,6 +964,13 @@ typedef struct arc_evict_waiter {
 #define	arc_c_max	ARCSTAT(arcstat_c_max)	/* max target cache size */
 #define	arc_sys_free	ARCSTAT(arcstat_sys_free) /* target system free bytes */
 
+#define	arc_anon	(&ARC_anon)
+#define	arc_mru		(&ARC_mru)
+#define	arc_mru_ghost	(&ARC_mru_ghost)
+#define	arc_mfu		(&ARC_mfu)
+#define	arc_mfu_ghost	(&ARC_mfu_ghost)
+#define	arc_l2c_only	(&ARC_l2c_only)
+
 extern taskq_t *arc_prune_taskq;
 extern arc_stats_t arc_stats;
 extern arc_sums_t arc_sums;
@@ -974,8 +981,8 @@ extern int arc_no_grow_shift;
 extern int arc_shrink_shift;
 extern kmutex_t arc_prune_mtx;
 extern list_t arc_prune_list;
-extern arc_state_t	*arc_mfu;
-extern arc_state_t	*arc_mru;
+extern arc_state_t	ARC_mfu;
+extern arc_state_t	ARC_mru;
 extern uint_t zfs_arc_pc_percent;
 extern int arc_lotsfree_percent;
 extern unsigned long zfs_arc_min;

module/zfs/arc.c

@@ -648,13 +648,6 @@ arc_sums_t arc_sums;
 	} while (0)
 
 kstat_t			*arc_ksp;
-static arc_state_t	*arc_anon;
-static arc_state_t	*arc_mru_ghost;
-static arc_state_t	*arc_mfu_ghost;
-static arc_state_t	*arc_l2c_only;
-
-arc_state_t	*arc_mru;
-arc_state_t	*arc_mfu;
 
 /*
  * There are several ARC variables that are critical to export as kstats --
@@ -2203,7 +2196,6 @@ arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state)
 		return;
 	}
 
-	ASSERT(!GHOST_STATE(state));
 	if (hdr->b_l1hdr.b_pabd != NULL) {
 		(void) zfs_refcount_add_many(&state->arcs_esize[type],
 		    arc_hdr_size(hdr), hdr);
@@ -2244,7 +2236,6 @@ arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state)
 		return;
 	}
 
-	ASSERT(!GHOST_STATE(state));
 	if (hdr->b_l1hdr.b_pabd != NULL) {
 		(void) zfs_refcount_remove_many(&state->arcs_esize[type],
 		    arc_hdr_size(hdr), hdr);
@@ -4031,23 +4022,21 @@ arc_set_need_free(void)
 
 static uint64_t
 arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker,
-    uint64_t spa, int64_t bytes)
+    uint64_t spa, uint64_t bytes)
 {
 	multilist_sublist_t *mls;
 	uint64_t bytes_evicted = 0, real_evicted = 0;
 	arc_buf_hdr_t *hdr;
 	kmutex_t *hash_lock;
-	int evict_count = 0;
+	int evict_count = zfs_arc_evict_batch_limit;
 
 	ASSERT3P(marker, !=, NULL);
-	IMPLY(bytes < 0, bytes == ARC_EVICT_ALL);
 
 	mls = multilist_sublist_lock(ml, idx);
 
-	for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL;
+	for (hdr = multilist_sublist_prev(mls, marker); likely(hdr != NULL);
 	    hdr = multilist_sublist_prev(mls, marker)) {
-		if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) ||
-		    (evict_count >= zfs_arc_evict_batch_limit))
+		if ((evict_count <= 0) || (bytes_evicted >= bytes))
 			break;
 
 		/*
@@ -4109,7 +4098,7 @@ arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker,
 			 * evict_count in this case.
 			 */
 			if (evicted != 0)
-				evict_count++;
+				evict_count--;
 
 		} else {
 			ARCSTAT_BUMP(arcstat_mutex_miss);
@@ -4170,16 +4159,14 @@ arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker,
  * the given arc state; which is used by arc_flush().
  */
 static uint64_t
-arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes,
+arc_evict_state(arc_state_t *state, uint64_t spa, uint64_t bytes,
     arc_buf_contents_t type)
 {
 	uint64_t total_evicted = 0;
 	multilist_t *ml = &state->arcs_list[type];
 	int num_sublists;
 	arc_buf_hdr_t **markers;
 
-	IMPLY(bytes < 0, bytes == ARC_EVICT_ALL);
-
 	num_sublists = multilist_get_num_sublists(ml);
 
 	/*
@@ -4211,7 +4198,7 @@ arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes,
 	 * While we haven't hit our target number of bytes to evict, or
 	 * we're evicting all available buffers.
 	 */
-	while (total_evicted < bytes || bytes == ARC_EVICT_ALL) {
+	while (total_evicted < bytes) {
 		int sublist_idx = multilist_get_random_index(ml);
 		uint64_t scan_evicted = 0;
 
@@ -4239,9 +4226,7 @@ arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes,
 			uint64_t bytes_remaining;
 			uint64_t bytes_evicted;
 
-			if (bytes == ARC_EVICT_ALL)
-				bytes_remaining = ARC_EVICT_ALL;
-			else if (total_evicted < bytes)
+			if (total_evicted < bytes)
 				bytes_remaining = bytes - total_evicted;
 			else
 				break;
@@ -4336,7 +4321,7 @@ static uint64_t
 arc_evict_impl(arc_state_t *state, uint64_t spa, int64_t bytes,
     arc_buf_contents_t type)
 {
-	int64_t delta;
+	uint64_t delta;
 
 	if (bytes > 0 && zfs_refcount_count(&state->arcs_esize[type]) > 0) {
 		delta = MIN(zfs_refcount_count(&state->arcs_esize[type]),
@@ -7596,13 +7581,6 @@ arc_tuning_update(boolean_t verbose)
 static void
 arc_state_init(void)
 {
-	arc_anon = &ARC_anon;
-	arc_mru = &ARC_mru;
-	arc_mru_ghost = &ARC_mru_ghost;
-	arc_mfu = &ARC_mfu;
-	arc_mfu_ghost = &ARC_mfu_ghost;
-	arc_l2c_only = &ARC_l2c_only;
-
 	multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA],
 	    sizeof (arc_buf_hdr_t),
 	    offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),