Index: sys/kern/subr_smr.c
===================================================================
--- sys/kern/subr_smr.c
+++ sys/kern/subr_smr.c
@@ -35,6 +35,8 @@
 #include <sys/proc.h>
 #include <sys/smp.h>
 #include <sys/smr.h>
+#include <sys/sysctl.h>
+#include <sys/counter.h>
 
 #include <vm/uma.h>
 
@@ -162,6 +164,17 @@
 #define	SMR_SEQ_MAX_ADVANCE	SMR_SEQ_MAX_DELTA / 2
 #endif
 
+static SYSCTL_NODE(_debug, OID_AUTO, smr, CTLFLAG_RW, NULL, "SMR Stats");
+static counter_u64_t advance;
+SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, advance, CTLFLAG_RD, &advance, "");
+static counter_u64_t advance_wait;
+SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, advance_wait, CTLFLAG_RD, &advance_wait, "");
+static counter_u64_t poll_fast;
+SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, poll_fast, CTLFLAG_RD, &poll_fast, "");
+static counter_u64_t poll_scan;
+SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, poll_scan, CTLFLAG_RD, &poll_scan, "");
+
+
 /*
  * Advance the write sequence and return the new value for use as the
  * wait goal.  This guarantees that any changes made by the calling
@@ -197,14 +210,17 @@
 	 */
 	s = zpcpu_get(smr)->c_shared;
 	goal = atomic_fetchadd_int(&s->s_wr_seq, SMR_SEQ_INCR) + SMR_SEQ_INCR;
+	counter_u64_add(advance, 1);
 
 	/*
 	 * Force a synchronization here if the goal is getting too
 	 * far ahead of the read sequence number.  This keeps the
 	 * wrap detecting arithmetic working in pathological cases.
 	 */
-	if (goal - atomic_load_int(&s->s_rd_seq) >= SMR_SEQ_MAX_DELTA)
+	if (goal - atomic_load_int(&s->s_rd_seq) >= SMR_SEQ_MAX_DELTA) {
+		counter_u64_add(advance_wait, 1);
 		smr_wait(smr, goal - SMR_SEQ_MAX_ADVANCE);
+	}
 
 	return (goal);
 }
@@ -296,14 +312,17 @@
 	 * A valid goal must be greater than s_rd_seq or we have not verified
 	 * that it has been observed and must fall through to polling.
 	 */
-	if (SMR_SEQ_GEQ(s_rd_seq, goal) || SMR_SEQ_LT(s_wr_seq, goal))
+	if (SMR_SEQ_GEQ(s_rd_seq, goal) || SMR_SEQ_LT(s_wr_seq, goal)) {
+		counter_u64_add(poll_fast, 1);
 		goto out;
+	}
 
 	/*
 	 * Loop until all cores have observed the goal sequence or have
 	 * gone inactive.  Keep track of the oldest sequence currently
 	 * active as rd_seq.
 	 */
+	counter_u64_add(poll_scan, 1);
 	rd_seq = s_wr_seq;
 	CPU_FOREACH(i) {
 		c = zpcpu_get_cpu(smr, i);
@@ -364,7 +383,7 @@
 	s_rd_seq = atomic_load_int(&s->s_rd_seq);
 	do {
 		if (SMR_SEQ_LEQ(rd_seq, s_rd_seq))
-			break;
+			goto out;
 	} while (atomic_fcmpset_int(&s->s_rd_seq, &s_rd_seq, rd_seq) == 0);
 
 out:
@@ -424,3 +443,14 @@
 	smr_zone = uma_zcreate("SMR CPU", sizeof(struct smr),
 	    NULL, NULL, NULL, NULL, (CACHE_LINE_SIZE * 2) - 1, UMA_ZONE_PCPU);
 }
+
+static void
+smr_init_counters(void *unused)
+{
+
+	advance = counter_u64_alloc(M_WAITOK);
+	advance_wait = counter_u64_alloc(M_WAITOK);
+	poll_fast = counter_u64_alloc(M_WAITOK);
+	poll_scan = counter_u64_alloc(M_WAITOK);
+}
+SYSINIT(smr_counters, SI_SUB_CPU, SI_ORDER_ANY, smr_init_counters, NULL);