Index: sys/dev/cxgbe/adapter.h
===================================================================
--- sys/dev/cxgbe/adapter.h
+++ sys/dev/cxgbe/adapter.h
@@ -316,10 +316,6 @@
 	u_int tx_parse_error;
 	int fcs_reg;
 	uint64_t fcs_base;
-	u_long	tx_toe_tls_records;
-	u_long	tx_toe_tls_octets;
-	u_long	rx_toe_tls_records;
-	u_long	rx_toe_tls_octets;
 
 	struct callout tick;
 };
@@ -654,6 +650,8 @@
 struct sge_ofld_rxq {
 	struct sge_iq iq;	/* MUST be first */
 	struct sge_fl fl;	/* MUST follow iq */
+	u_long	rx_toe_tls_records;
+	u_long	rx_toe_tls_octets;
 } __aligned(CACHE_LINE_SIZE);
 
 static inline struct sge_ofld_rxq *
@@ -715,6 +713,8 @@
 /* ofld_txq: SGE egress queue + miscellaneous items */
 struct sge_ofld_txq {
 	struct sge_wrq wrq;
+	counter_u64_t tx_toe_tls_records;
+	counter_u64_t tx_toe_tls_octets;
 } __aligned(CACHE_LINE_SIZE);
 
 #define INVALID_NM_RXQ_CNTXT_ID ((uint16_t)(-1))
Index: sys/dev/cxgbe/t4_main.c
===================================================================
--- sys/dev/cxgbe/t4_main.c
+++ sys/dev/cxgbe/t4_main.c
@@ -7217,19 +7217,6 @@
 
 #undef T4_REGSTAT
 #undef T4_PORTSTAT
-
-	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_toe_tls_records",
-	    CTLFLAG_RD, &pi->tx_toe_tls_records,
-	    "# of TOE TLS records transmitted");
-	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_toe_tls_octets",
-	    CTLFLAG_RD, &pi->tx_toe_tls_octets,
-	    "# of payload octets in transmitted TOE TLS records");
-	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_toe_tls_records",
-	    CTLFLAG_RD, &pi->rx_toe_tls_records,
-	    "# of TOE TLS records received");
-	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_toe_tls_octets",
-	    CTLFLAG_RD, &pi->rx_toe_tls_octets,
-	    "# of payload octets in received TOE TLS records");
 }
 
 static int
@@ -10764,6 +10751,8 @@
 			for_each_ofld_txq(vi, i, ofld_txq) {
 				ofld_txq->wrq.tx_wrs_direct = 0;
 				ofld_txq->wrq.tx_wrs_copied = 0;
+				counter_u64_zero(ofld_txq->tx_toe_tls_records);
+				counter_u64_zero(ofld_txq->tx_toe_tls_octets);
 			}
 #endif
 #ifdef TCP_OFFLOAD
@@ -10771,6 +10760,8 @@
 				ofld_rxq->fl.cl_allocated = 0;
 				ofld_rxq->fl.cl_recycled = 0;
 				ofld_rxq->fl.cl_fast_recycled = 0;
+				ofld_rxq->rx_toe_tls_records = 0;
+				ofld_rxq->rx_toe_tls_octets = 0;
 			}
 #endif
 
Index: sys/dev/cxgbe/t4_sge.c
===================================================================
--- sys/dev/cxgbe/t4_sge.c
+++ sys/dev/cxgbe/t4_sge.c
@@ -1250,6 +1250,17 @@
 		rc = alloc_wrq(sc, vi, &ofld_txq->wrq, oid2);
 		if (rc != 0)
 			goto done;
+
+		ofld_txq->tx_toe_tls_records = counter_u64_alloc(M_WAITOK);
+		ofld_txq->tx_toe_tls_octets = counter_u64_alloc(M_WAITOK);
+		SYSCTL_ADD_COUNTER_U64(&vi->ctx, SYSCTL_CHILDREN(oid2),
+		    OID_AUTO, "tx_toe_tls_records", CTLFLAG_RD,
+		    &ofld_txq->tx_toe_tls_records,
+		    "# of TOE TLS records transmitted");
+		SYSCTL_ADD_COUNTER_U64(&vi->ctx, SYSCTL_CHILDREN(oid2),
+		    OID_AUTO, "tx_toe_tls_octets", CTLFLAG_RD,
+		    &ofld_txq->tx_toe_tls_octets,
+		    "# of payload octets in transmitted TOE TLS records");
 	}
 #endif
 done:
@@ -1310,6 +1321,8 @@
 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
 	for_each_ofld_txq(vi, i, ofld_txq) {
 		free_wrq(sc, &ofld_txq->wrq);
+		counter_u64_free(ofld_txq->tx_toe_tls_records);
+		counter_u64_free(ofld_txq->tx_toe_tls_octets);
 	}
 #endif
 
@@ -3961,6 +3974,13 @@
 	add_iq_sysctls(&vi->ctx, oid, &ofld_rxq->iq);
 	add_fl_sysctls(pi->adapter, &vi->ctx, oid, &ofld_rxq->fl);
 
+	SYSCTL_ADD_ULONG(&vi->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
+	    "rx_toe_tls_records", CTLFLAG_RD, &ofld_rxq->rx_toe_tls_records,
+	    "# of TOE TLS records received");
+	SYSCTL_ADD_ULONG(&vi->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
+	    "rx_toe_tls_octets", CTLFLAG_RD, &ofld_rxq->rx_toe_tls_octets,
+	    "# of payload octets in received TOE TLS records");
+
 	return (rc);
 }
 
Index: sys/dev/cxgbe/tom/t4_tls.c
===================================================================
--- sys/dev/cxgbe/tom/t4_tls.c
+++ sys/dev/cxgbe/tom/t4_tls.c
@@ -1664,8 +1664,8 @@
 		}
 		toep->txsd_avail--;
 
-		atomic_add_long(&toep->vi->pi->tx_toe_tls_records, 1);
-		atomic_add_long(&toep->vi->pi->tx_toe_tls_octets, plen);
+		counter_u64_add(toep->ofld_txq->tx_toe_tls_records, 1);
+		counter_u64_add(toep->ofld_txq->tx_toe_tls_octets, plen);
 
 		t4_l2t_send(sc, wr, toep->l2te);
 	}
@@ -1966,8 +1966,8 @@
 		}
 		toep->txsd_avail--;
 
-		atomic_add_long(&toep->vi->pi->tx_toe_tls_records, 1);
-		atomic_add_long(&toep->vi->pi->tx_toe_tls_octets, m->m_len);
+		counter_u64_add(toep->ofld_txq->tx_toe_tls_records, 1);
+		counter_u64_add(toep->ofld_txq->tx_toe_tls_octets, m->m_len);
 
 		t4_l2t_send(sc, wr, toep->l2te);
 	}
@@ -2003,7 +2003,7 @@
 	m_adj(m, sizeof(*cpl));
 	len = m->m_pkthdr.len;
 
-	atomic_add_long(&toep->vi->pi->rx_toe_tls_octets, len);
+	toep->ofld_rxq->rx_toe_tls_octets += len;
 
 	KASSERT(len == G_CPL_TLS_DATA_LENGTH(be32toh(cpl->length_pkd)),
 	    ("%s: payload length mismatch", __func__));
@@ -2070,7 +2070,7 @@
 	m_adj(m, sizeof(*cpl));
 	len = m->m_pkthdr.len;
 
-	atomic_add_long(&toep->vi->pi->rx_toe_tls_records, 1);
+	toep->ofld_rxq->rx_toe_tls_records++;
 
 	KASSERT(len == G_CPL_RX_TLS_CMP_LENGTH(be32toh(cpl->pdulength_length)),
 	    ("%s: payload length mismatch", __func__));