Update bcachefs sources to c241a5bf54ed bcachefs: nopromote sub counters

.bcachefs_revision

@@ -1 +1 @@
-8efd93eb2d0f977564b0db7d9e54369ae870c437
+c241a5bf54ed4aeb29d029d8f1dae1dd592cdda4

libbcachefs/btree_io.c

@@ -1336,6 +1336,10 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 		k = bkey_p_next(k);
 	}
 
+	for (k = i->start; k != vstruct_last(i);) {
+		BUG_ON(!k->u64s);
+	}
+
 	bch2_bset_build_aux_tree(b, b->set, false);
 
 	set_needs_whiteout(btree_bset_first(b), true);

libbcachefs/io_read.c

@@ -39,38 +39,73 @@ MODULE_PARM_DESC(read_corrupt_ratio, "");
 
 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
 
+static inline u32 bch2_dev_congested_read(struct bch_dev *ca, u64 now)
+{
+	s64 congested = atomic_read(&ca->congested);
+	u64 last = READ_ONCE(ca->congested_last);
+	if (time_after64(now, last))
+		congested -= (now - last) >> 12;
+
+	return clamp(congested, 0LL, CONGESTED_MAX);
+}
+
 static bool bch2_target_congested(struct bch_fs *c, u16 target)
 {
 	const struct bch_devs_mask *devs;
 	unsigned d, nr = 0, total = 0;
-	u64 now = local_clock(), last;
+	u64 now = local_clock();
-	s64 congested;
-	struct bch_dev *ca;
-
-	if (!target)
-		return false;
 
 	guard(rcu)();
 	devs = bch2_target_to_mask(c, target) ?:
 		&c->rw_devs[BCH_DATA_user];
 
 	for_each_set_bit(d, devs->d, BCH_SB_MEMBERS_MAX) {
-		ca = rcu_dereference(c->devs[d]);
+		struct bch_dev *ca = rcu_dereference(c->devs[d]);
 		if (!ca)
 			continue;
 
-		congested = atomic_read(&ca->congested);
+		total += bch2_dev_congested_read(ca, now);
-		last = READ_ONCE(ca->congested_last);
-		if (time_after64(now, last))
-			congested -= (now - last) >> 12;
-
-		total += max(congested, 0LL);
 		nr++;
 	}
 
 	return get_random_u32_below(nr * CONGESTED_MAX) < total;
 }
 
+void bch2_dev_congested_to_text(struct printbuf *out, struct bch_dev *ca)
+{
+	printbuf_tabstop_push(out, 32);
+
+	prt_printf(out, "current:\t%u%%\n",
+		   bch2_dev_congested_read(ca, local_clock()) *
+		   100 / CONGESTED_MAX);
+
+	prt_printf(out, "raw:\t%i/%u\n", atomic_read(&ca->congested), CONGESTED_MAX);
+
+	prt_printf(out, "last io over threshold:\t");
+	bch2_pr_time_units(out, local_clock() - ca->congested_last);
+	prt_newline(out);
+
+	prt_printf(out, "read latency threshold:\t");
+	bch2_pr_time_units(out,
+			   ca->io_latency[READ].quantiles.entries[QUANTILE_IDX(1)].m * 2);
+	prt_newline(out);
+
+	prt_printf(out, "median read latency:\t");
+	bch2_pr_time_units(out,
+			   ca->io_latency[READ].quantiles.entries[QUANTILE_IDX(7)].m);
+	prt_newline(out);
+
+	prt_printf(out, "write latency threshold:\t");
+	bch2_pr_time_units(out,
+			   ca->io_latency[WRITE].quantiles.entries[QUANTILE_IDX(1)].m * 3);
+	prt_newline(out);
+
+	prt_printf(out, "median write latency:\t");
+	bch2_pr_time_units(out,
+			   ca->io_latency[WRITE].quantiles.entries[QUANTILE_IDX(7)].m);
+	prt_newline(out);
+}
+
 #else
 
 static bool bch2_target_congested(struct bch_fs *c, u16 target)
@@ -130,22 +165,32 @@ static inline int should_promote(struct bch_fs *c, struct bkey_s_c k,
 	if (!have_io_error(failed)) {
 		BUG_ON(!opts.promote_target);
 
-		if (!(flags & BCH_READ_may_promote))
+		if (!(flags & BCH_READ_may_promote)) {
+			count_event(c, io_read_nopromote_may_not);
 			return bch_err_throw(c, nopromote_may_not);
+		}
 
-		if (bch2_bkey_has_target(c, k, opts.promote_target))
+		if (bch2_bkey_has_target(c, k, opts.promote_target)) {
+			count_event(c, io_read_nopromote_already_promoted);
 			return bch_err_throw(c, nopromote_already_promoted);
+		}
 
-		if (bkey_extent_is_unwritten(k))
+		if (bkey_extent_is_unwritten(k)) {
+			count_event(c, io_read_nopromote_unwritten);
 			return bch_err_throw(c, nopromote_unwritten);
+		}
 
-		if (bch2_target_congested(c, opts.promote_target))
+		if (bch2_target_congested(c, opts.promote_target)) {
+			count_event(c, io_read_nopromote_congested);
 			return bch_err_throw(c, nopromote_congested);
+		}
 	}
 
 	if (rhashtable_lookup_fast(&c->promote_table, &pos,
-				   bch_promote_params))
+				   bch_promote_params)) {
+		count_event(c, io_read_nopromote_in_flight);
 		return bch_err_throw(c, nopromote_in_flight);
+	}
 
 	return 0;
 }

libbcachefs/io_read.h

@@ -7,6 +7,10 @@
 #include "extents_types.h"
 #include "reflink.h"
 
+#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
+void bch2_dev_congested_to_text(struct printbuf *, struct bch_dev *);
+#endif
+
 struct bch_read_bio {
 	struct bch_fs		*c;
 	u64			start_time;

libbcachefs/sb-counters_format.h

@@ -13,6 +13,11 @@ enum counters_flags {
 	x(io_read_hole,					81,	TYPE_SECTORS)	\
 	x(io_read_promote,				30,	TYPE_COUNTER)	\
 	x(io_read_nopromote,				85,	TYPE_COUNTER)	\
+	x(io_read_nopromote_may_not,			86,	TYPE_COUNTER)	\
+	x(io_read_nopromote_already_promoted,		87,	TYPE_COUNTER)	\
+	x(io_read_nopromote_unwritten,			88,	TYPE_COUNTER)	\
+	x(io_read_nopromote_congested,			89,	TYPE_COUNTER)	\
+	x(io_read_nopromote_in_flight,			90,	TYPE_COUNTER)	\
 	x(io_read_bounce,				31,	TYPE_COUNTER)	\
 	x(io_read_split,				33,	TYPE_COUNTER)	\
 	x(io_read_reuse_race,				34,	TYPE_COUNTER)	\

libbcachefs/sysfs.c

@@ -172,7 +172,9 @@ read_attribute(io_latency_read);
 read_attribute(io_latency_write);
 read_attribute(io_latency_stats_read);
 read_attribute(io_latency_stats_write);
+#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
 read_attribute(congested);
+#endif
 
 read_attribute(btree_write_stats);
 
@@ -942,9 +944,10 @@ SHOW(bch2_dev)
 	if (attr == &sysfs_io_latency_stats_write)
 		bch2_time_stats_to_text(out, &ca->io_latency[WRITE].stats);
 
-	sysfs_printf(congested,			"%u%%",
+#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
-		     clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
+	if (attr == &sysfs_congested)
-		     * 100 / CONGESTED_MAX);
+		bch2_dev_congested_to_text(out, ca);
+#endif
 
 	if (attr == &sysfs_alloc_debug)
 		bch2_dev_alloc_debug_to_text(out, ca);
@@ -1015,7 +1018,9 @@ struct attribute *bch2_dev_files[] = {
 	&sysfs_io_latency_write,
 	&sysfs_io_latency_stats_read,
 	&sysfs_io_latency_stats_write,
+#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
 	&sysfs_congested,
+#endif
 
 	&sysfs_read_fua_test,