diff --git a/sys/dev/cxgbe/adapter.h b/sys/dev/cxgbe/adapter.h
--- a/sys/dev/cxgbe/adapter.h
+++ b/sys/dev/cxgbe/adapter.h
@@ -1404,7 +1404,6 @@
 void t4_register_cpl_handler(int, cpl_handler_t);
 void t4_register_shared_cpl_handler(int, cpl_handler_t, int);
 #ifdef RATELIMIT
-int ethofld_transmit(struct ifnet *, struct mbuf *);
 void send_etid_flush_wr(struct cxgbe_rate_tag *);
 #endif
 
diff --git a/sys/dev/cxgbe/t4_main.c b/sys/dev/cxgbe/t4_main.c
--- a/sys/dev/cxgbe/t4_main.c
+++ b/sys/dev/cxgbe/t4_main.c
@@ -3062,16 +3062,16 @@
 
 	rc = parse_pkt(&m, vi->flags & TX_USES_VM_WR);
 	if (__predict_false(rc != 0)) {
+		if (__predict_true(rc == EINPROGRESS)) {
+			/* queued by parse_pkt */
+			MPASS(m != NULL);
+			return (0);
+		}
+
 		MPASS(m == NULL);			/* was freed already */
 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
 		return (rc);
 	}
-#ifdef RATELIMIT
-	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
-		if (m->m_pkthdr.snd_tag->sw->type == IF_SND_TAG_TYPE_RATE_LIMIT)
-			return (ethofld_transmit(ifp, m));
-	}
-#endif
 
 	/* Select a txq. */
 	sc = vi->adapter;
diff --git a/sys/dev/cxgbe/t4_sge.c b/sys/dev/cxgbe/t4_sge.c
--- a/sys/dev/cxgbe/t4_sge.c
+++ b/sys/dev/cxgbe/t4_sge.c
@@ -345,6 +345,7 @@
 #endif
 static int ethofld_fw4_ack(struct sge_iq *, const struct rss_header *,
     struct mbuf *);
+static int ethofld_transmit(struct ifnet *, struct mbuf *);
 #endif
 
 static counter_u64_t extfree_refs;
@@ -2960,6 +2961,10 @@
 		set_mbuf_eo_nsegs(m0, nsegs);
 		set_mbuf_eo_len16(m0,
 		    txpkt_eo_len16(nsegs, immhdrs, needs_tso(m0)));
+		rc = ethofld_transmit(mst->ifp, m0);
+		if (rc != 0)
+			goto fail;
+		return (EINPROGRESS);
 	}
 #endif
 #endif
@@ -6847,7 +6852,7 @@
 	}
 }
 
-int
+static int
 ethofld_transmit(struct ifnet *ifp, struct mbuf *m0)
 {
 	struct cxgbe_rate_tag *cst;
@@ -6903,8 +6908,6 @@
 
 done:
 	mtx_unlock(&cst->lock);
-	if (__predict_false(rc != 0))
-		m_freem(m0);
 	return (rc);
 }