Remove hash_elements_max accounting from DBUF and ARC

Those values require global atomics to get current hash_elements
values in few of the hottest code paths, while in all the years I
never cared about it.  If somebody wants, it should be easy to
get it by periodic sampling, since neither ARC header nor DBUF
counts change so fast that it would be difficult to catch.

For now I've left hash_elements_max kstat for ARC, since it was
used/reported by arc_summary and it would break older versions,
but now it just reports the current value.

Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Alexander Motin <[email protected]>
Sponsored by:	iXsystems, Inc.
Closes openzfs#16759

Author: amotin

cmd/arc_summary

@@ -662,10 +662,7 @@ def section_arc(kstats_dict):
     print()
 
     print('ARC hash breakdown:')
-    prt_i1('Elements max:', f_hits(arc_stats['hash_elements_max']))
-    prt_i2('Elements current:',
-           f_perc(arc_stats['hash_elements'], arc_stats['hash_elements_max']),
-           f_hits(arc_stats['hash_elements']))
+    prt_i1('Elements:', f_hits(arc_stats['hash_elements']))
     prt_i1('Collisions:', f_hits(arc_stats['hash_collisions']))
 
     prt_i1('Chain max:', f_hits(arc_stats['hash_chain_max']))

include/sys/arc_impl.h

@@ -942,6 +942,7 @@ typedef struct arc_sums {
 	wmsum_t arcstat_evict_l2_eligible_mru;
 	wmsum_t arcstat_evict_l2_ineligible;
 	wmsum_t arcstat_evict_l2_skip;
+	wmsum_t arcstat_hash_elements;
 	wmsum_t arcstat_hash_collisions;
 	wmsum_t arcstat_hash_chains;
 	aggsum_t arcstat_size;

module/zfs/arc.c

@@ -1074,12 +1074,9 @@ buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp)
 		ARCSTAT_BUMP(arcstat_hash_collisions);
 		if (i == 1)
 			ARCSTAT_BUMP(arcstat_hash_chains);
-
 		ARCSTAT_MAX(arcstat_hash_chain_max, i);
 	}
-	uint64_t he = atomic_inc_64_nv(
-	    &arc_stats.arcstat_hash_elements.value.ui64);
-	ARCSTAT_MAX(arcstat_hash_elements_max, he);
+	ARCSTAT_BUMP(arcstat_hash_elements);
 
 	return (NULL);
 }
@@ -1103,8 +1100,7 @@ buf_hash_remove(arc_buf_hdr_t *hdr)
 	arc_hdr_clear_flags(hdr, ARC_FLAG_IN_HASH_TABLE);
 
 	/* collect some hash table performance data */
-	atomic_dec_64(&arc_stats.arcstat_hash_elements.value.ui64);
-
+	ARCSTAT_BUMPDOWN(arcstat_hash_elements);
 	if (buf_hash_table.ht_table[idx] &&
 	    buf_hash_table.ht_table[idx]->b_hash_next == NULL)
 		ARCSTAT_BUMPDOWN(arcstat_hash_chains);
@@ -7008,6 +7004,9 @@ arc_kstat_update(kstat_t *ksp, int rw)
 	    wmsum_value(&arc_sums.arcstat_evict_l2_ineligible);
 	as->arcstat_evict_l2_skip.value.ui64 =
 	    wmsum_value(&arc_sums.arcstat_evict_l2_skip);
+	as->arcstat_hash_elements.value.ui64 =
+	    as->arcstat_hash_elements_max.value.ui64 =
+	    wmsum_value(&arc_sums.arcstat_hash_elements);
 	as->arcstat_hash_collisions.value.ui64 =
 	    wmsum_value(&arc_sums.arcstat_hash_collisions);
 	as->arcstat_hash_chains.value.ui64 =
@@ -7432,6 +7431,7 @@ arc_state_init(void)
 	wmsum_init(&arc_sums.arcstat_evict_l2_eligible_mru, 0);
 	wmsum_init(&arc_sums.arcstat_evict_l2_ineligible, 0);
 	wmsum_init(&arc_sums.arcstat_evict_l2_skip, 0);
+	wmsum_init(&arc_sums.arcstat_hash_elements, 0);
 	wmsum_init(&arc_sums.arcstat_hash_collisions, 0);
 	wmsum_init(&arc_sums.arcstat_hash_chains, 0);
 	aggsum_init(&arc_sums.arcstat_size, 0);
@@ -7590,6 +7590,7 @@ arc_state_fini(void)
 	wmsum_fini(&arc_sums.arcstat_evict_l2_eligible_mru);
 	wmsum_fini(&arc_sums.arcstat_evict_l2_ineligible);
 	wmsum_fini(&arc_sums.arcstat_evict_l2_skip);
+	wmsum_fini(&arc_sums.arcstat_hash_elements);
 	wmsum_fini(&arc_sums.arcstat_hash_collisions);
 	wmsum_fini(&arc_sums.arcstat_hash_chains);
 	aggsum_fini(&arc_sums.arcstat_size);

module/zfs/dbuf.c

@@ -89,7 +89,6 @@ typedef struct dbuf_stats {
 	kstat_named_t hash_misses;
 	kstat_named_t hash_collisions;
 	kstat_named_t hash_elements;
-	kstat_named_t hash_elements_max;
 	/*
 	 * Number of sublists containing more than one dbuf in the dbuf
 	 * hash table. Keep track of the longest hash chain.
@@ -134,7 +133,6 @@ dbuf_stats_t dbuf_stats = {
 	{ "hash_misses",			KSTAT_DATA_UINT64 },
 	{ "hash_collisions",			KSTAT_DATA_UINT64 },
 	{ "hash_elements",			KSTAT_DATA_UINT64 },
-	{ "hash_elements_max",			KSTAT_DATA_UINT64 },
 	{ "hash_chains",			KSTAT_DATA_UINT64 },
 	{ "hash_chain_max",			KSTAT_DATA_UINT64 },
 	{ "hash_insert_race",			KSTAT_DATA_UINT64 },
@@ -154,6 +152,7 @@ struct {
 	wmsum_t hash_hits;
 	wmsum_t hash_misses;
 	wmsum_t hash_collisions;
+	wmsum_t hash_elements;
 	wmsum_t hash_chains;
 	wmsum_t hash_insert_race;
 	wmsum_t metadata_cache_count;
@@ -432,8 +431,7 @@ dbuf_hash_insert(dmu_buf_impl_t *db)
 	db->db_hash_next = h->hash_table[idx];
 	h->hash_table[idx] = db;
 	mutex_exit(DBUF_HASH_MUTEX(h, idx));
-	uint64_t he = atomic_inc_64_nv(&dbuf_stats.hash_elements.value.ui64);
-	DBUF_STAT_MAX(hash_elements_max, he);
+	DBUF_STAT_BUMP(hash_elements);
 
 	return (NULL);
 }
@@ -506,7 +504,7 @@ dbuf_hash_remove(dmu_buf_impl_t *db)
 	    h->hash_table[idx]->db_hash_next == NULL)
 		DBUF_STAT_BUMPDOWN(hash_chains);
 	mutex_exit(DBUF_HASH_MUTEX(h, idx));
-	atomic_dec_64(&dbuf_stats.hash_elements.value.ui64);
+	DBUF_STAT_BUMPDOWN(hash_elements);
 }
 
 typedef enum {
@@ -903,6 +901,8 @@ dbuf_kstat_update(kstat_t *ksp, int rw)
 	    wmsum_value(&dbuf_sums.hash_misses);
 	ds->hash_collisions.value.ui64 =
 	    wmsum_value(&dbuf_sums.hash_collisions);
+	ds->hash_elements.value.ui64 =
+	    wmsum_value(&dbuf_sums.hash_elements);
 	ds->hash_chains.value.ui64 =
 	    wmsum_value(&dbuf_sums.hash_chains);
 	ds->hash_insert_race.value.ui64 =
@@ -1004,6 +1004,7 @@ dbuf_init(void)
 	wmsum_init(&dbuf_sums.hash_hits, 0);
 	wmsum_init(&dbuf_sums.hash_misses, 0);
 	wmsum_init(&dbuf_sums.hash_collisions, 0);
+	wmsum_init(&dbuf_sums.hash_elements, 0);
 	wmsum_init(&dbuf_sums.hash_chains, 0);
 	wmsum_init(&dbuf_sums.hash_insert_race, 0);
 	wmsum_init(&dbuf_sums.metadata_cache_count, 0);
@@ -1077,6 +1078,7 @@ dbuf_fini(void)
 	wmsum_fini(&dbuf_sums.hash_hits);
 	wmsum_fini(&dbuf_sums.hash_misses);
 	wmsum_fini(&dbuf_sums.hash_collisions);
+	wmsum_fini(&dbuf_sums.hash_elements);
 	wmsum_fini(&dbuf_sums.hash_chains);
 	wmsum_fini(&dbuf_sums.hash_insert_race);
 	wmsum_fini(&dbuf_sums.metadata_cache_count);