From 671bf2b8b22f6c737fb6927b1cb7308ba50900ad Mon Sep 17 00:00:00 2001
From: Navdeep Parhar <np@FreeBSD.org>
Date: Tue, 5 Jul 2016 01:29:24 +0000
Subject: [PATCH] cxgbe(4): Changes to the CPL-handler registration mechanism
 and code related to "shared" CPLs.

a) Combine t4_set_tcb_field and t4_set_tcb_field_rpl into a single
function.  Allow callers to direct the response to any iq.  Tidy up
set_ulp_mode_iscsi while there to use names from t4_tcb.h instead of
magic constants.

b) Remove all CPL handler tables from struct adapter.  This reduces its
size by around 2KB.  All handlers are now registered at MOD_LOAD instead
of attach or some kind of initialization/activation.  The registration
functions do not need an adapter parameter any more.

c) Add per-iq handlers to deal with CPLs whose destination cannot be
determined solely from the opcode.  There are 2 such CPLs in use right
now: SET_TCB_RPL and L2T_WRITE_RPL.  The base driver continues to send
filter and L2T_WRITEs over the mgmtq and solicits the reply on fwq.
t4_tom (including the DDP code) now uses the port's ctrlq to send
L2T_WRITEs and SET_TCB_FIELDs and solicits the reply on an ofld_rxq.
fwq and ofld_rxq have different handlers that know what kind of tid to
expect in the reply.  Update t4_write_l2e and callers to to support any
wrq/iq combination.

Approved by:	re@ (kib@)
Sponsored by:	Chelsio Communications
---
 sys/dev/cxgbe/adapter.h           |  29 +++--
 sys/dev/cxgbe/cxgbei/cxgbei.c     |  28 ++---
 sys/dev/cxgbe/cxgbei/icl_cxgbei.c |  13 ++-
 sys/dev/cxgbe/iw_cxgbe/cm.c       |  30 ++---
 sys/dev/cxgbe/iw_cxgbe/device.c   |   1 -
 sys/dev/cxgbe/iw_cxgbe/iw_cxgbe.h |   3 -
 sys/dev/cxgbe/t4_l2t.c            |  22 ++--
 sys/dev/cxgbe/t4_l2t.h            |   4 +-
 sys/dev/cxgbe/t4_main.c           | 179 ++++++++----------------------
 sys/dev/cxgbe/t4_sge.c            | 132 +++++++++++++++++++---
 sys/dev/cxgbe/tom/t4_connect.c    |   6 +-
 sys/dev/cxgbe/tom/t4_cpl_io.c     |  71 +++++-------
 sys/dev/cxgbe/tom/t4_ddp.c        |  26 +++--
 sys/dev/cxgbe/tom/t4_listen.c     |  10 +-
 sys/dev/cxgbe/tom/t4_tom.c        |  25 +++--
 sys/dev/cxgbe/tom/t4_tom.h        |  15 ++-
 sys/dev/cxgbe/tom/t4_tom_l2t.c    |  45 ++------
 sys/dev/cxgbe/tom/t4_tom_l2t.h    |   4 +-
 18 files changed, 303 insertions(+), 340 deletions(-)

diff --git a/sys/dev/cxgbe/adapter.h b/sys/dev/cxgbe/adapter.h
index 34cb38fd3633..d31124581efa 100644
--- a/sys/dev/cxgbe/adapter.h
+++ b/sys/dev/cxgbe/adapter.h
@@ -372,6 +372,13 @@ enum {
 	NM_BUSY	= 2,
 };
 
+struct sge_iq;
+struct rss_header;
+typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *,
+    struct mbuf *);
+typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *);
+typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *);
+
 /*
  * Ingress Queue: T4 is producer, driver is consumer.
  */
@@ -379,6 +386,8 @@ struct sge_iq {
 	uint32_t flags;
 	volatile int state;
 	struct adapter *adapter;
+	cpl_handler_t set_tcb_rpl;
+	cpl_handler_t l2t_write_rpl;
 	struct iq_desc  *desc;	/* KVA of descriptor ring */
 	int8_t   intr_pktc_idx;	/* packet count threshold index */
 	uint8_t  gen;		/* generation bit */
@@ -739,12 +748,6 @@ struct sge {
 	struct hw_buf_info hw_buf_info[SGE_FLBUF_SIZES];
 };
 
-struct rss_header;
-typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *,
-    struct mbuf *);
-typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *);
-typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *);
-
 struct adapter {
 	SLIST_ENTRY(adapter) link;
 	device_t dev;
@@ -783,6 +786,7 @@ struct adapter {
 
 	struct sge sge;
 	int lro_timeout;
+	int sc_do_rxcopy;
 
 	struct taskqueue *tq[MAX_NCHAN];	/* General purpose taskqueues */
 	struct port_info *port[MAX_NPORTS];
@@ -842,15 +846,9 @@ struct adapter {
 
 	struct memwin memwin[NUM_MEMWIN];	/* memory windows */
 
-	an_handler_t an_handler __aligned(CACHE_LINE_SIZE);
-	fw_msg_handler_t fw_msg_handler[7];	/* NUM_FW6_TYPES */
-	cpl_handler_t cpl_handler[0xef];	/* NUM_CPL_CMDS */
-
 	const char *last_op;
 	const void *last_op_thr;
 	int last_op_flags;
-
-	int sc_do_rxcopy;
 };
 
 #define ADAPTER_LOCK(sc)		mtx_lock(&(sc)->sc_lock)
@@ -1080,9 +1078,6 @@ int t4_os_pci_restore_state(struct adapter *);
 void t4_os_portmod_changed(const struct adapter *, int);
 void t4_os_link_changed(struct adapter *, int, int, int);
 void t4_iterate(void (*)(struct adapter *, void *), void *);
-int t4_register_cpl_handler(struct adapter *, int, cpl_handler_t);
-int t4_register_an_handler(struct adapter *, an_handler_t);
-int t4_register_fw_msg_handler(struct adapter *, int, fw_msg_handler_t);
 int t4_filter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
 int begin_synchronized_op(struct adapter *, struct vi_info *, int, char *);
 void doom_vi(struct adapter *, struct vi_info *);
@@ -1107,7 +1102,6 @@ void t4_nm_intr(void *);
 void t4_sge_modload(void);
 void t4_sge_modunload(void);
 uint64_t t4_sge_extfree_refs(void);
-void t4_init_sge_cpl_handlers(struct adapter *);
 void t4_tweak_chip_settings(struct adapter *);
 int t4_read_chip_settings(struct adapter *);
 int t4_create_dma_tag(struct adapter *);
@@ -1129,6 +1123,9 @@ int parse_pkt(struct mbuf **);
 void *start_wrq_wr(struct sge_wrq *, int, struct wrq_cookie *);
 void commit_wrq_wr(struct sge_wrq *, void *, struct wrq_cookie *);
 int tnl_cong(struct port_info *, int);
+int t4_register_an_handler(an_handler_t);
+int t4_register_fw_msg_handler(int, fw_msg_handler_t);
+int t4_register_cpl_handler(int, cpl_handler_t);
 
 /* t4_tracer.c */
 struct t4_tracer;
diff --git a/sys/dev/cxgbe/cxgbei/cxgbei.c b/sys/dev/cxgbe/cxgbei/cxgbei.c
index 775e3a2c9474..face22d02fc1 100644
--- a/sys/dev/cxgbe/cxgbei/cxgbei.c
+++ b/sys/dev/cxgbe/cxgbei/cxgbei.c
@@ -745,24 +745,6 @@ do_rx_iscsi_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 	return (0);
 }
 
-static void
-t4_register_cpl_handler_with_tom(struct adapter *sc)
-{
-
-	t4_register_cpl_handler(sc, CPL_ISCSI_HDR, do_rx_iscsi_hdr);
-	t4_register_cpl_handler(sc, CPL_ISCSI_DATA, do_rx_iscsi_data);
-	t4_register_cpl_handler(sc, CPL_RX_ISCSI_DDP, do_rx_iscsi_ddp);
-}
-
-static void
-t4_unregister_cpl_handler_with_tom(struct adapter *sc)
-{
-
-	t4_register_cpl_handler(sc, CPL_ISCSI_HDR, NULL);
-	t4_register_cpl_handler(sc, CPL_ISCSI_DATA, NULL);
-	t4_register_cpl_handler(sc, CPL_RX_ISCSI_DDP, NULL);
-}
-
 /* initiator */
 void
 cxgbei_conn_task_reserve_itt(void *conn, void **prv,
@@ -835,7 +817,6 @@ cxgbei_activate(struct adapter *sc)
 		return (rc);
 	}
 
-	t4_register_cpl_handler_with_tom(sc);
 	sc->iscsi_ulp_softc = ci;
 
 	return (0);
@@ -849,7 +830,6 @@ cxgbei_deactivate(struct adapter *sc)
 
 	if (sc->iscsi_ulp_softc != NULL) {
 		cxgbei_ddp_cleanup(sc->iscsi_ulp_softc);
-		t4_unregister_cpl_handler_with_tom(sc);
 		free(sc->iscsi_ulp_softc, M_CXGBE);
 		sc->iscsi_ulp_softc = NULL;
 	}
@@ -1062,6 +1042,10 @@ cxgbei_mod_load(void)
 {
 	int rc;
 
+	t4_register_cpl_handler(CPL_ISCSI_HDR, do_rx_iscsi_hdr);
+	t4_register_cpl_handler(CPL_ISCSI_DATA, do_rx_iscsi_data);
+	t4_register_cpl_handler(CPL_RX_ISCSI_DDP, do_rx_iscsi_ddp);
+
 	rc = start_worker_threads();
 	if (rc != 0)
 		return (rc);
@@ -1088,6 +1072,10 @@ cxgbei_mod_unload(void)
 
 	stop_worker_threads();
 
+	t4_register_cpl_handler(CPL_ISCSI_HDR, NULL);
+	t4_register_cpl_handler(CPL_ISCSI_DATA, NULL);
+	t4_register_cpl_handler(CPL_RX_ISCSI_DDP, NULL);
+
 	return (0);
 }
 #endif
diff --git a/sys/dev/cxgbe/cxgbei/icl_cxgbei.c b/sys/dev/cxgbe/cxgbei/icl_cxgbei.c
index d0e7f399195a..8f29452196fe 100644
--- a/sys/dev/cxgbe/cxgbei/icl_cxgbei.c
+++ b/sys/dev/cxgbe/cxgbei/icl_cxgbei.c
@@ -71,6 +71,7 @@ __FBSDID("$FreeBSD$");
 #include <icl_conn_if.h>
 
 #include "common/common.h"
+#include "common/t4_tcb.h"
 #include "tom/t4_tom.h"
 #include "cxgbei.h"
 
@@ -584,19 +585,19 @@ send_iscsi_flowc_wr(struct adapter *sc, struct toepcb *toep, int maxlen)
 static void
 set_ulp_mode_iscsi(struct adapter *sc, struct toepcb *toep, int hcrc, int dcrc)
 {
-	uint64_t val = 0;
+	uint64_t val = ULP_MODE_ISCSI;
 
 	if (hcrc)
-		val |= ULP_CRC_HEADER;
+		val |= ULP_CRC_HEADER << 4;
 	if (dcrc)
-		val |= ULP_CRC_DATA;
-	val <<= 4;
-	val |= ULP_MODE_ISCSI;
+		val |= ULP_CRC_DATA << 4;
 
 	CTR4(KTR_CXGBE, "%s: tid %u, ULP_MODE_ISCSI, CRC hdr=%d data=%d",
 	    __func__, toep->tid, hcrc, dcrc);
 
-	t4_set_tcb_field(sc, toep, 1, 0, 0xfff, val);
+	t4_set_tcb_field(sc, toep->ctrlq, toep->tid, W_TCB_ULP_TYPE,
+	    V_TCB_ULP_TYPE(M_TCB_ULP_TYPE) | V_TCB_ULP_RAW(M_TCB_ULP_RAW), val,
+	    0, 0, toep->ofld_rxq->iq.abs_id);
 }
 
 /*
diff --git a/sys/dev/cxgbe/iw_cxgbe/cm.c b/sys/dev/cxgbe/iw_cxgbe/cm.c
index 1f6cc3741fff..bea5aa01b850 100644
--- a/sys/dev/cxgbe/iw_cxgbe/cm.c
+++ b/sys/dev/cxgbe/iw_cxgbe/cm.c
@@ -2463,28 +2463,14 @@ static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbu
 	return 0;
 }
 
-	void
-c4iw_cm_init_cpl(struct adapter *sc)
-{
-
-	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, terminate);
-	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, fw6_wr_rpl);
-	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, fw6_cqe_handler);
-	t4_register_an_handler(sc, c4iw_ev_handler);
-}
-
-	void
-c4iw_cm_term_cpl(struct adapter *sc)
-{
-
-	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, NULL);
-	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, NULL);
-	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, NULL);
-}
-
 int __init c4iw_cm_init(void)
 {
 
+	t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate);
+	t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl);
+	t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler);
+	t4_register_an_handler(c4iw_ev_handler);
+
 	TAILQ_INIT(&req_list);
 	spin_lock_init(&req_lock);
 	INIT_LIST_HEAD(&timeout_list);
@@ -2496,7 +2482,6 @@ int __init c4iw_cm_init(void)
 	if (!c4iw_taskq)
 		return -ENOMEM;
 
-
 	return 0;
 }
 
@@ -2506,5 +2491,10 @@ void __exit c4iw_cm_term(void)
 	WARN_ON(!list_empty(&timeout_list));
 	flush_workqueue(c4iw_taskq);
 	destroy_workqueue(c4iw_taskq);
+
+	t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL);
+	t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL);
+	t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL);
+	t4_register_an_handler(NULL);
 }
 #endif
diff --git a/sys/dev/cxgbe/iw_cxgbe/device.c b/sys/dev/cxgbe/iw_cxgbe/device.c
index ea04190628ae..310b99bdc89f 100644
--- a/sys/dev/cxgbe/iw_cxgbe/device.c
+++ b/sys/dev/cxgbe/iw_cxgbe/device.c
@@ -227,7 +227,6 @@ c4iw_activate(struct adapter *sc)
 	}
 
 	sc->iwarp_softc = iwsc;
-	c4iw_cm_init_cpl(sc);
 
 	rc = -c4iw_register_device(iwsc);
 	if (rc) {
diff --git a/sys/dev/cxgbe/iw_cxgbe/iw_cxgbe.h b/sys/dev/cxgbe/iw_cxgbe/iw_cxgbe.h
index 0aa2568ee907..faeb5f4bace8 100644
--- a/sys/dev/cxgbe/iw_cxgbe/iw_cxgbe.h
+++ b/sys/dev/cxgbe/iw_cxgbe/iw_cxgbe.h
@@ -936,9 +936,6 @@ extern int c4iw_max_read_depth;
 #define L1_CACHE_BYTES 32
 #endif
 
-void c4iw_cm_init_cpl(struct adapter *);
-void c4iw_cm_term_cpl(struct adapter *);
-
 void your_reg_device(struct c4iw_dev *dev);
 
 #define SGE_CTRLQ_NUM	0
diff --git a/sys/dev/cxgbe/t4_l2t.c b/sys/dev/cxgbe/t4_l2t.c
index 7c4cedefb280..b4b1ee4d312d 100644
--- a/sys/dev/cxgbe/t4_l2t.c
+++ b/sys/dev/cxgbe/t4_l2t.c
@@ -111,27 +111,34 @@ t4_alloc_l2e(struct l2t_data *d)
  * The write may be synchronous or asynchronous.
  */
 int
-t4_write_l2e(struct adapter *sc, struct l2t_entry *e, int sync)
+t4_write_l2e(struct l2t_entry *e, int sync)
 {
+	struct sge_wrq *wrq;
+	struct adapter *sc;
 	struct wrq_cookie cookie;
 	struct cpl_l2t_write_req *req;
-	int idx = e->idx + sc->vres.l2t.start;
+	int idx;
 
 	mtx_assert(&e->lock, MA_OWNED);
+	MPASS(e->wrq != NULL);
 
-	req = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*req), 16), &cookie);
+	wrq = e->wrq;
+	sc = wrq->adapter;
+
+	req = start_wrq_wr(wrq, howmany(sizeof(*req), 16), &cookie);
 	if (req == NULL)
 		return (ENOMEM);
 
+	idx = e->idx + sc->vres.l2t.start;
 	INIT_TP_WR(req, 0);
 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, idx |
-	    V_SYNC_WR(sync) | V_TID_QID(sc->sge.fwq.abs_id)));
+	    V_SYNC_WR(sync) | V_TID_QID(e->iqid)));
 	req->params = htons(V_L2T_W_PORT(e->lport) | V_L2T_W_NOREPLY(!sync));
 	req->l2t_idx = htons(idx);
 	req->vlan = htons(e->vlan);
 	memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
 
-	commit_wrq_wr(&sc->sge.mgmtq, req, &cookie);
+	commit_wrq_wr(wrq, req, &cookie);
 
 	if (sync && e->state != L2T_STATE_SWITCHING)
 		e->state = L2T_STATE_SYNC_WRITE;
@@ -173,9 +180,11 @@ t4_l2t_set_switching(struct adapter *sc, struct l2t_entry *e, uint16_t vlan,
 
 	e->vlan = vlan;
 	e->lport = port;
+	e->wrq = &sc->sge.mgmtq;
+	e->iqid = sc->sge.fwq.abs_id;
 	memcpy(e->dmac, eth_addr, ETHER_ADDR_LEN);
 	mtx_lock(&e->lock);
-	rc = t4_write_l2e(sc, e, 0);
+	rc = t4_write_l2e(e, 0);
 	mtx_unlock(&e->lock);
 	return (rc);
 }
@@ -211,7 +220,6 @@ t4_init_l2t(struct adapter *sc, int flags)
 	}
 
 	sc->l2t = d;
-	t4_register_cpl_handler(sc, CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
 
 	return (0);
 }
diff --git a/sys/dev/cxgbe/t4_l2t.h b/sys/dev/cxgbe/t4_l2t.h
index c60eef1367b2..2d861dc9a2fd 100644
--- a/sys/dev/cxgbe/t4_l2t.h
+++ b/sys/dev/cxgbe/t4_l2t.h
@@ -61,6 +61,8 @@ struct l2t_entry {
 	uint16_t state;			/* entry state */
 	uint16_t idx;			/* entry index */
 	uint32_t addr[4];		/* next hop IP or IPv6 address */
+	uint32_t iqid;			/* iqid for reply to write_l2e */
+	struct sge_wrq *wrq;		/* queue to use for write_l2e */
 	struct ifnet *ifp;		/* outgoing interface */
 	uint16_t smt_idx;		/* SMT index */
 	uint16_t vlan;			/* VLAN TCI (id: 0-11, prio: 13-15) */
@@ -90,7 +92,7 @@ struct l2t_entry *t4_alloc_l2e(struct l2t_data *);
 struct l2t_entry *t4_l2t_alloc_switching(struct l2t_data *);
 int t4_l2t_set_switching(struct adapter *, struct l2t_entry *, uint16_t,
     uint8_t, uint8_t *);
-int t4_write_l2e(struct adapter *, struct l2t_entry *, int);
+int t4_write_l2e(struct l2t_entry *, int);
 int do_l2t_write_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
 
 static inline void
diff --git a/sys/dev/cxgbe/t4_main.c b/sys/dev/cxgbe/t4_main.c
index eeb118078969..247175d660c3 100644
--- a/sys/dev/cxgbe/t4_main.c
+++ b/sys/dev/cxgbe/t4_main.c
@@ -458,10 +458,6 @@ static void vi_refresh_stats(struct adapter *, struct vi_info *);
 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
 static void cxgbe_tick(void *);
 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
-static int cpl_not_handled(struct sge_iq *, const struct rss_header *,
-    struct mbuf *);
-static int an_not_handled(struct sge_iq *, const struct rsp_ctrl *);
-static int fw_msg_not_handled(struct adapter *, const __be64 *);
 static void t4_sysctls(struct adapter *);
 static void cxgbe_sysctls(struct port_info *);
 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
@@ -525,6 +521,8 @@ static int del_filter(struct adapter *, struct t4_filter *);
 static void clear_filter(struct filter_entry *);
 static int set_filter_wr(struct adapter *, int);
 static int del_filter_wr(struct adapter *, int);
+static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
+    struct mbuf *);
 static int get_sge_context(struct adapter *, struct t4_sge_context *);
 static int load_fw(struct adapter *, struct t4_data *);
 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
@@ -589,11 +587,6 @@ struct {
 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
 #endif
-
-/* No easy way to include t4_msg.h before adapter.h so we check this way */
-CTASSERT(nitems(((struct adapter *)0)->cpl_handler) == NUM_CPL_CMDS);
-CTASSERT(nitems(((struct adapter *)0)->fw_msg_handler) == NUM_FW6_TYPES);
-
 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
 
 static int
@@ -739,15 +732,6 @@ t4_attach(device_t dev)
 	sc->mbox = sc->pf;
 
 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
-	sc->an_handler = an_not_handled;
-	for (i = 0; i < nitems(sc->cpl_handler); i++)
-		sc->cpl_handler[i] = cpl_not_handled;
-	for (i = 0; i < nitems(sc->fw_msg_handler); i++)
-		sc->fw_msg_handler[i] = fw_msg_not_handled;
-	t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl);
-	t4_register_cpl_handler(sc, CPL_TRACE_PKT, t4_trace_pkt);
-	t4_register_cpl_handler(sc, CPL_T5_TRACE_PKT, t5_trace_pkt);
-	t4_init_sge_cpl_handlers(sc);
 
 	/* Prepare the adapter for operation. */
 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
@@ -4500,98 +4484,6 @@ cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
 	VLAN_SETCOOKIE(vlan, ifp);
 }
 
-static int
-cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
-{
-
-#ifdef INVARIANTS
-	panic("%s: opcode 0x%02x on iq %p with payload %p",
-	    __func__, rss->opcode, iq, m);
-#else
-	log(LOG_ERR, "%s: opcode 0x%02x on iq %p with payload %p\n",
-	    __func__, rss->opcode, iq, m);
-	m_freem(m);
-#endif
-	return (EDOOFUS);
-}
-
-int
-t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
-{
-	uintptr_t *loc, new;
-
-	if (opcode >= nitems(sc->cpl_handler))
-		return (EINVAL);
-
-	new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
-	loc = (uintptr_t *) &sc->cpl_handler[opcode];
-	atomic_store_rel_ptr(loc, new);
-
-	return (0);
-}
-
-static int
-an_not_handled(struct sge_iq *iq, const struct rsp_ctrl *ctrl)
-{
-
-#ifdef INVARIANTS
-	panic("%s: async notification on iq %p (ctrl %p)", __func__, iq, ctrl);
-#else
-	log(LOG_ERR, "%s: async notification on iq %p (ctrl %p)\n",
-	    __func__, iq, ctrl);
-#endif
-	return (EDOOFUS);
-}
-
-int
-t4_register_an_handler(struct adapter *sc, an_handler_t h)
-{
-	uintptr_t *loc, new;
-
-	new = h ? (uintptr_t)h : (uintptr_t)an_not_handled;
-	loc = (uintptr_t *) &sc->an_handler;
-	atomic_store_rel_ptr(loc, new);
-
-	return (0);
-}
-
-static int
-fw_msg_not_handled(struct adapter *sc, const __be64 *rpl)
-{
-	const struct cpl_fw6_msg *cpl =
-	    __containerof(rpl, struct cpl_fw6_msg, data[0]);
-
-#ifdef INVARIANTS
-	panic("%s: fw_msg type %d", __func__, cpl->type);
-#else
-	log(LOG_ERR, "%s: fw_msg type %d\n", __func__, cpl->type);
-#endif
-	return (EDOOFUS);
-}
-
-int
-t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
-{
-	uintptr_t *loc, new;
-
-	if (type >= nitems(sc->fw_msg_handler))
-		return (EINVAL);
-
-	/*
-	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
-	 * handler dispatch table.  Reject any attempt to install a handler for
-	 * this subtype.
-	 */
-	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
-		return (EINVAL);
-
-	new = h ? (uintptr_t)h : (uintptr_t)fw_msg_not_handled;
-	loc = (uintptr_t *) &sc->fw_msg_handler[type];
-	atomic_store_rel_ptr(loc, new);
-
-	return (0);
-}
-
 /*
  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
  */
@@ -8262,38 +8154,53 @@ t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 
 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
 	    rss->opcode));
+	MPASS(iq == &sc->sge.fwq);
+	MPASS(is_ftid(sc, idx));
 
-	if (is_ftid(sc, idx)) {
+	idx -= sc->tids.ftid_base;
+	f = &sc->tids.ftid_tab[idx];
+	rc = G_COOKIE(rpl->cookie);
 
-		idx -= sc->tids.ftid_base;
-		f = &sc->tids.ftid_tab[idx];
-		rc = G_COOKIE(rpl->cookie);
-
-		mtx_lock(&sc->tids.ftid_lock);
-		if (rc == FW_FILTER_WR_FLT_ADDED) {
-			KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
-			    __func__, idx));
-			f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
-			f->pending = 0;  /* asynchronous setup completed */
-			f->valid = 1;
-		} else {
-			if (rc != FW_FILTER_WR_FLT_DELETED) {
-				/* Add or delete failed, display an error */
-				log(LOG_ERR,
-				    "filter %u setup failed with error %u\n",
-				    idx, rc);
-			}
-
-			clear_filter(f);
-			sc->tids.ftids_in_use--;
+	mtx_lock(&sc->tids.ftid_lock);
+	if (rc == FW_FILTER_WR_FLT_ADDED) {
+		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
+		    __func__, idx));
+		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
+		f->pending = 0;  /* asynchronous setup completed */
+		f->valid = 1;
+	} else {
+		if (rc != FW_FILTER_WR_FLT_DELETED) {
+			/* Add or delete failed, display an error */
+			log(LOG_ERR,
+			    "filter %u setup failed with error %u\n",
+			    idx, rc);
 		}
-		wakeup(&sc->tids.ftid_tab);
-		mtx_unlock(&sc->tids.ftid_lock);
+
+		clear_filter(f);
+		sc->tids.ftids_in_use--;
 	}
+	wakeup(&sc->tids.ftid_tab);
+	mtx_unlock(&sc->tids.ftid_lock);
 
 	return (0);
 }
 
+static int
+set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
+{
+
+	MPASS(iq->set_tcb_rpl != NULL);
+	return (iq->set_tcb_rpl(iq, rss, m));
+}
+
+static int
+l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
+{
+
+	MPASS(iq->l2t_write_rpl != NULL);
+	return (iq->l2t_write_rpl(iq, rss, m));
+}
+
 static int
 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
 {
@@ -9469,6 +9376,10 @@ mod_event(module_t mod, int cmd, void *arg)
 		sx_xlock(&mlu);
 		if (loaded++ == 0) {
 			t4_sge_modload();
+			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
+			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
+			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
+			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
 			sx_init(&t4_list_lock, "T4/T5 adapters");
 			SLIST_INIT(&t4_list);
 #ifdef TCP_OFFLOAD
diff --git a/sys/dev/cxgbe/t4_sge.c b/sys/dev/cxgbe/t4_sge.c
index 4cf3a5352098..6bf8d2297582 100644
--- a/sys/dev/cxgbe/t4_sge.c
+++ b/sys/dev/cxgbe/t4_sge.c
@@ -68,6 +68,7 @@ __FBSDID("$FreeBSD$");
 #include "common/t4_regs.h"
 #include "common/t4_regs_values.h"
 #include "common/t4_msg.h"
+#include "t4_l2t.h"
 #include "t4_mp_ring.h"
 
 #ifdef T4_PKT_TIMESTAMP
@@ -253,12 +254,110 @@ static int sysctl_tc(SYSCTL_HANDLER_ARGS);
 static counter_u64_t extfree_refs;
 static counter_u64_t extfree_rels;
 
+an_handler_t t4_an_handler;
+fw_msg_handler_t t4_fw_msg_handler[NUM_FW6_TYPES];
+cpl_handler_t t4_cpl_handler[NUM_CPL_CMDS];
+
+
+static int
+an_not_handled(struct sge_iq *iq, const struct rsp_ctrl *ctrl)
+{
+
+#ifdef INVARIANTS
+	panic("%s: async notification on iq %p (ctrl %p)", __func__, iq, ctrl);
+#else
+	log(LOG_ERR, "%s: async notification on iq %p (ctrl %p)\n",
+	    __func__, iq, ctrl);
+#endif
+	return (EDOOFUS);
+}
+
+int
+t4_register_an_handler(an_handler_t h)
+{
+	uintptr_t *loc, new;
+
+	new = h ? (uintptr_t)h : (uintptr_t)an_not_handled;
+	loc = (uintptr_t *) &t4_an_handler;
+	atomic_store_rel_ptr(loc, new);
+
+	return (0);
+}
+
+static int
+fw_msg_not_handled(struct adapter *sc, const __be64 *rpl)
+{
+	const struct cpl_fw6_msg *cpl =
+	    __containerof(rpl, struct cpl_fw6_msg, data[0]);
+
+#ifdef INVARIANTS
+	panic("%s: fw_msg type %d", __func__, cpl->type);
+#else
+	log(LOG_ERR, "%s: fw_msg type %d\n", __func__, cpl->type);
+#endif
+	return (EDOOFUS);
+}
+
+int
+t4_register_fw_msg_handler(int type, fw_msg_handler_t h)
+{
+	uintptr_t *loc, new;
+
+	if (type >= nitems(t4_fw_msg_handler))
+		return (EINVAL);
+
+	/*
+	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
+	 * handler dispatch table.  Reject any attempt to install a handler for
+	 * this subtype.
+	 */
+	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
+		return (EINVAL);
+
+	new = h ? (uintptr_t)h : (uintptr_t)fw_msg_not_handled;
+	loc = (uintptr_t *) &t4_fw_msg_handler[type];
+	atomic_store_rel_ptr(loc, new);
+
+	return (0);
+}
+
+static int
+cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
+{
+
+#ifdef INVARIANTS
+	panic("%s: opcode 0x%02x on iq %p with payload %p",
+	    __func__, rss->opcode, iq, m);
+#else
+	log(LOG_ERR, "%s: opcode 0x%02x on iq %p with payload %p\n",
+	    __func__, rss->opcode, iq, m);
+	m_freem(m);
+#endif
+	return (EDOOFUS);
+}
+
+int
+t4_register_cpl_handler(int opcode, cpl_handler_t h)
+{
+	uintptr_t *loc, new;
+
+	if (opcode >= nitems(t4_cpl_handler))
+		return (EINVAL);
+
+	new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
+	loc = (uintptr_t *) &t4_cpl_handler[opcode];
+	atomic_store_rel_ptr(loc, new);
+
+	return (0);
+}
+
 /*
  * Called on MOD_LOAD.  Validates and calculates the SGE tunables.
  */
 void
 t4_sge_modload(void)
 {
+	int i;
 
 	if (fl_pktshift < 0 || fl_pktshift > 7) {
 		printf("Invalid hw.cxgbe.fl_pktshift value (%d),"
@@ -291,6 +390,18 @@ t4_sge_modload(void)
 	extfree_rels = counter_u64_alloc(M_WAITOK);
 	counter_u64_zero(extfree_refs);
 	counter_u64_zero(extfree_rels);
+
+	t4_an_handler = an_not_handled;
+	for (i = 0; i < nitems(t4_fw_msg_handler); i++)
+		t4_fw_msg_handler[i] = fw_msg_not_handled;
+	for (i = 0; i < nitems(t4_cpl_handler); i++)
+		t4_cpl_handler[i] = cpl_not_handled;
+
+	t4_register_cpl_handler(CPL_FW4_MSG, handle_fw_msg);
+	t4_register_cpl_handler(CPL_FW6_MSG, handle_fw_msg);
+	t4_register_cpl_handler(CPL_SGE_EGR_UPDATE, handle_sge_egr_update);
+	t4_register_cpl_handler(CPL_RX_PKT, t4_eth_rx);
+	t4_register_fw_msg_handler(FW6_TYPE_CMD_RPL, t4_handle_fw_rpl);
 }
 
 void
@@ -312,17 +423,6 @@ t4_sge_extfree_refs(void)
 	return (refs - rels);
 }
 
-void
-t4_init_sge_cpl_handlers(struct adapter *sc)
-{
-
-	t4_register_cpl_handler(sc, CPL_FW4_MSG, handle_fw_msg);
-	t4_register_cpl_handler(sc, CPL_FW6_MSG, handle_fw_msg);
-	t4_register_cpl_handler(sc, CPL_SGE_EGR_UPDATE, handle_sge_egr_update);
-	t4_register_cpl_handler(sc, CPL_RX_PKT, t4_eth_rx);
-	t4_register_fw_msg_handler(sc, FW6_TYPE_CMD_RPL, t4_handle_fw_rpl);
-}
-
 static inline void
 setup_pad_and_pack_boundaries(struct adapter *sc)
 {
@@ -1316,7 +1416,7 @@ service_iq(struct sge_iq *iq, int budget)
 				KASSERT(d->rss.opcode < NUM_CPL_CMDS,
 				    ("%s: bad opcode %02x.", __func__,
 				    d->rss.opcode));
-				sc->cpl_handler[d->rss.opcode](iq, &d->rss, m0);
+				t4_cpl_handler[d->rss.opcode](iq, &d->rss, m0);
 				break;
 
 			case X_RSPD_TYPE_INTR:
@@ -1338,7 +1438,7 @@ service_iq(struct sge_iq *iq, int budget)
 				 * iWARP async notification.
 				 */
 				if (lq >= 1024) {
-                                        sc->an_handler(iq, &d->rsp);
+                                        t4_an_handler(iq, &d->rsp);
                                         break;
                                 }
 
@@ -2789,6 +2889,8 @@ alloc_fwq(struct adapter *sc)
 	init_iq(fwq, sc, 0, 0, FW_IQ_QSIZE);
 	fwq->flags |= IQ_INTR;	/* always */
 	intr_idx = sc->intr_count > 1 ? 1 : 0;
+	fwq->set_tcb_rpl = t4_filter_rpl;
+	fwq->l2t_write_rpl = do_l2t_write_rpl;
 	rc = alloc_iq_fl(&sc->port[0]->vi[0], fwq, NULL, intr_idx, -1);
 	if (rc != 0) {
 		device_printf(sc->dev,
@@ -4674,10 +4776,10 @@ handle_fw_msg(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 		const struct rss_header *rss2;
 
 		rss2 = (const struct rss_header *)&cpl->data[0];
-		return (sc->cpl_handler[rss2->opcode](iq, rss2, m));
+		return (t4_cpl_handler[rss2->opcode](iq, rss2, m));
 	}
 
-	return (sc->fw_msg_handler[cpl->type](sc, &cpl->data[0]));
+	return (t4_fw_msg_handler[cpl->type](sc, &cpl->data[0]));
 }
 
 static int
diff --git a/sys/dev/cxgbe/tom/t4_connect.c b/sys/dev/cxgbe/tom/t4_connect.c
index c386a83fea09..f550723cb210 100644
--- a/sys/dev/cxgbe/tom/t4_connect.c
+++ b/sys/dev/cxgbe/tom/t4_connect.c
@@ -261,11 +261,11 @@ calc_opt2a(struct socket *so, struct toepcb *toep)
 }
 
 void
-t4_init_connect_cpl_handlers(struct adapter *sc)
+t4_init_connect_cpl_handlers(void)
 {
 
-	t4_register_cpl_handler(sc, CPL_ACT_ESTABLISH, do_act_establish);
-	t4_register_cpl_handler(sc, CPL_ACT_OPEN_RPL, do_act_open_rpl);
+	t4_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish);
+	t4_register_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl);
 }
 
 #define DONT_OFFLOAD_ACTIVE_OPEN(x)	do { \
diff --git a/sys/dev/cxgbe/tom/t4_cpl_io.c b/sys/dev/cxgbe/tom/t4_cpl_io.c
index 387ee5148c13..f7ef499f4488 100644
--- a/sys/dev/cxgbe/tom/t4_cpl_io.c
+++ b/sys/dev/cxgbe/tom/t4_cpl_io.c
@@ -1679,7 +1679,7 @@ do_fw4_ack(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 	return (0);
 }
 
-static int
+int
 do_set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 {
 	struct adapter *sc = iq->adapter;
@@ -1693,9 +1693,7 @@ do_set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 	KASSERT(opcode == CPL_SET_TCB_RPL,
 	    ("%s: unexpected opcode 0x%x", __func__, opcode));
 	KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
-
-	if (is_ftid(sc, tid))
-		return (t4_filter_rpl(iq, rss, m)); /* TCB is a filter */
+	MPASS(iq != &sc->sge.fwq);
 
 	toep = lookup_tid(sc, tid);
 	if (toep->ulp_mode == ULP_MODE_TCPDDP) {
@@ -1720,47 +1718,26 @@ do_set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 }
 
 void
-t4_set_tcb_field(struct adapter *sc, struct toepcb *toep, int ctrl,
-    uint16_t word, uint64_t mask, uint64_t val)
+t4_set_tcb_field(struct adapter *sc, struct sge_wrq *wrq, int tid,
+    uint16_t word, uint64_t mask, uint64_t val, int reply, int cookie, int iqid)
 {
 	struct wrqe *wr;
 	struct cpl_set_tcb_field *req;
 
-	wr = alloc_wrqe(sizeof(*req), ctrl ? toep->ctrlq : toep->ofld_txq);
+	MPASS((cookie & ~M_COOKIE) == 0);
+	MPASS((iqid & ~M_QUEUENO) == 0);
+
+	wr = alloc_wrqe(sizeof(*req), wrq);
 	if (wr == NULL) {
 		/* XXX */
 		panic("%s: allocation failure.", __func__);
 	}
 	req = wrtod(wr);
 
-	INIT_TP_WR_MIT_CPL(req, CPL_SET_TCB_FIELD, toep->tid);
-	req->reply_ctrl = htobe16(V_NO_REPLY(1) |
-	    V_QUEUENO(toep->ofld_rxq->iq.abs_id));
-	req->word_cookie = htobe16(V_WORD(word) | V_COOKIE(0));
-	req->mask = htobe64(mask);
-	req->val = htobe64(val);
-
-	t4_wrq_tx(sc, wr);
-}
-
-void
-t4_set_tcb_field_rpl(struct adapter *sc, struct toepcb *toep, int ctrl,
-    uint16_t word, uint64_t mask, uint64_t val, uint8_t cookie)
-{
-	struct wrqe *wr;
-	struct cpl_set_tcb_field *req;
-
-	KASSERT((cookie & ~M_COOKIE) == 0, ("%s: invalid cookie %#x", __func__,
-	    cookie));
-	wr = alloc_wrqe(sizeof(*req), ctrl ? toep->ctrlq : toep->ofld_txq);
-	if (wr == NULL) {
-		/* XXX */
-		panic("%s: allocation failure.", __func__);
-	}
-	req = wrtod(wr);
-
-	INIT_TP_WR_MIT_CPL(req, CPL_SET_TCB_FIELD, toep->tid);
-	req->reply_ctrl = htobe16(V_QUEUENO(toep->ofld_rxq->iq.abs_id));
+	INIT_TP_WR_MIT_CPL(req, CPL_SET_TCB_FIELD, tid);
+	req->reply_ctrl = htobe16(V_QUEUENO(iqid));
+	if (reply == 0)
+		req->reply_ctrl |= htobe16(F_NO_REPLY);
 	req->word_cookie = htobe16(V_WORD(word) | V_COOKIE(cookie));
 	req->mask = htobe64(mask);
 	req->val = htobe64(val);
@@ -1769,22 +1746,26 @@ t4_set_tcb_field_rpl(struct adapter *sc, struct toepcb *toep, int ctrl,
 }
 
 void
-t4_init_cpl_io_handlers(struct adapter *sc)
+t4_init_cpl_io_handlers(void)
 {
 
-	t4_register_cpl_handler(sc, CPL_PEER_CLOSE, do_peer_close);
-	t4_register_cpl_handler(sc, CPL_CLOSE_CON_RPL, do_close_con_rpl);
-	t4_register_cpl_handler(sc, CPL_ABORT_REQ_RSS, do_abort_req);
-	t4_register_cpl_handler(sc, CPL_ABORT_RPL_RSS, do_abort_rpl);
-	t4_register_cpl_handler(sc, CPL_RX_DATA, do_rx_data);
-	t4_register_cpl_handler(sc, CPL_FW4_ACK, do_fw4_ack);
-	t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, do_set_tcb_rpl);
+	t4_register_cpl_handler(CPL_PEER_CLOSE, do_peer_close);
+	t4_register_cpl_handler(CPL_CLOSE_CON_RPL, do_close_con_rpl);
+	t4_register_cpl_handler(CPL_ABORT_REQ_RSS, do_abort_req);
+	t4_register_cpl_handler(CPL_ABORT_RPL_RSS, do_abort_rpl);
+	t4_register_cpl_handler(CPL_RX_DATA, do_rx_data);
+	t4_register_cpl_handler(CPL_FW4_ACK, do_fw4_ack);
 }
 
 void
-t4_uninit_cpl_io_handlers(struct adapter *sc)
+t4_uninit_cpl_io_handlers(void)
 {
 
-	t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl);
+	t4_register_cpl_handler(CPL_PEER_CLOSE, do_peer_close);
+	t4_register_cpl_handler(CPL_CLOSE_CON_RPL, do_close_con_rpl);
+	t4_register_cpl_handler(CPL_ABORT_REQ_RSS, do_abort_req);
+	t4_register_cpl_handler(CPL_ABORT_RPL_RSS, do_abort_rpl);
+	t4_register_cpl_handler(CPL_RX_DATA, do_rx_data);
+	t4_register_cpl_handler(CPL_FW4_ACK, do_fw4_ack);
 }
 #endif
diff --git a/sys/dev/cxgbe/tom/t4_ddp.c b/sys/dev/cxgbe/tom/t4_ddp.c
index bcc8d1d8f528..ff73c8bb3cad 100644
--- a/sys/dev/cxgbe/tom/t4_ddp.c
+++ b/sys/dev/cxgbe/tom/t4_ddp.c
@@ -786,6 +786,8 @@ handle_ddp_close(struct toepcb *toep, struct tcpcb *tp, __be32 rcv_nxt)
 	 F_DDP_INVALID_TAG | F_DDP_COLOR_ERR | F_DDP_TID_MISMATCH |\
 	 F_DDP_INVALID_PPOD | F_DDP_HDRCRC_ERR | F_DDP_DATACRC_ERR)
 
+extern cpl_handler_t t4_cpl_handler[];
+
 static int
 do_rx_data_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 {
@@ -807,7 +809,7 @@ do_rx_data_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
 	}
 
 	if (toep->ulp_mode == ULP_MODE_ISCSI) {
-		sc->cpl_handler[CPL_RX_ISCSI_DDP](iq, rss, m);
+		t4_cpl_handler[CPL_RX_ISCSI_DDP](iq, rss, m);
 		return (0);
 	}
 
@@ -848,13 +850,14 @@ enable_ddp(struct adapter *sc, struct toepcb *toep)
 
 	DDP_ASSERT_LOCKED(toep);
 	toep->ddp_flags |= DDP_SC_REQ;
-	t4_set_tcb_field(sc, toep, 1, W_TCB_RX_DDP_FLAGS,
+	t4_set_tcb_field(sc, toep->ctrlq, toep->tid, W_TCB_RX_DDP_FLAGS,
 	    V_TF_DDP_OFF(1) | V_TF_DDP_INDICATE_OUT(1) |
 	    V_TF_DDP_BUF0_INDICATE(1) | V_TF_DDP_BUF1_INDICATE(1) |
 	    V_TF_DDP_BUF0_VALID(1) | V_TF_DDP_BUF1_VALID(1),
-	    V_TF_DDP_BUF0_INDICATE(1) | V_TF_DDP_BUF1_INDICATE(1));
-	t4_set_tcb_field(sc, toep, 1, W_TCB_T_FLAGS,
-	    V_TF_RCV_COALESCE_ENABLE(1), 0);
+	    V_TF_DDP_BUF0_INDICATE(1) | V_TF_DDP_BUF1_INDICATE(1), 0, 0,
+	    toep->ofld_rxq->iq.abs_id);
+	t4_set_tcb_field(sc, toep->ctrlq, toep->tid, W_TCB_T_FLAGS,
+	    V_TF_RCV_COALESCE_ENABLE(1), 0, 0, 0, toep->ofld_rxq->iq.abs_id);
 }
 
 static int
@@ -1070,9 +1073,6 @@ t4_init_ddp(struct adapter *sc, struct tom_data *td)
 	td->ppod_start = sc->vres.ddp.start;
 	td->ppod_arena = vmem_create("DDP page pods", sc->vres.ddp.start,
 	    sc->vres.ddp.size, 1, 32, M_FIRSTFIT | M_NOWAIT);
-
-	t4_register_cpl_handler(sc, CPL_RX_DATA_DDP, do_rx_data_ddp);
-	t4_register_cpl_handler(sc, CPL_RX_DDP_COMPLETE, do_rx_ddp_complete);
 }
 
 void
@@ -1683,8 +1683,10 @@ t4_aio_cancel_active(struct kaiocb *job)
 			 */
 			valid_flag = i == 0 ? V_TF_DDP_BUF0_VALID(1) :
 			    V_TF_DDP_BUF1_VALID(1);
-			t4_set_tcb_field_rpl(sc, toep, 1, W_TCB_RX_DDP_FLAGS,
-			    valid_flag, 0, i + DDP_BUF0_INVALIDATED);
+			t4_set_tcb_field(sc, toep->ctrlq, toep->tid,
+			    W_TCB_RX_DDP_FLAGS, valid_flag, 0, 1,
+			    i + DDP_BUF0_INVALIDATED,
+			    toep->ofld_rxq->iq.abs_id);
 			toep->db[i].cancel_pending = 1;
 			CTR2(KTR_CXGBE, "%s: request %p marked pending",
 			    __func__, job);
@@ -1756,6 +1758,8 @@ int
 t4_ddp_mod_load(void)
 {
 
+	t4_register_cpl_handler(CPL_RX_DATA_DDP, do_rx_data_ddp);
+	t4_register_cpl_handler(CPL_RX_DDP_COMPLETE, do_rx_ddp_complete);
 	TAILQ_INIT(&ddp_orphan_pagesets);
 	mtx_init(&ddp_orphan_pagesets_lock, "ddp orphans", NULL, MTX_DEF);
 	TASK_INIT(&ddp_orphan_task, 0, ddp_free_orphan_pagesets, NULL);
@@ -1769,5 +1773,7 @@ t4_ddp_mod_unload(void)
 	taskqueue_drain(taskqueue_thread, &ddp_orphan_task);
 	MPASS(TAILQ_EMPTY(&ddp_orphan_pagesets));
 	mtx_destroy(&ddp_orphan_pagesets_lock);
+	t4_register_cpl_handler(CPL_RX_DATA_DDP, NULL);
+	t4_register_cpl_handler(CPL_RX_DDP_COMPLETE, NULL);
 }
 #endif
diff --git a/sys/dev/cxgbe/tom/t4_listen.c b/sys/dev/cxgbe/tom/t4_listen.c
index 22f734a26e2e..4a1a0629897b 100644
--- a/sys/dev/cxgbe/tom/t4_listen.c
+++ b/sys/dev/cxgbe/tom/t4_listen.c
@@ -1589,12 +1589,12 @@ do_pass_establish(struct sge_iq *iq, const struct rss_header *rss,
 }
 
 void
-t4_init_listen_cpl_handlers(struct adapter *sc)
+t4_init_listen_cpl_handlers(void)
 {
 
-	t4_register_cpl_handler(sc, CPL_PASS_OPEN_RPL, do_pass_open_rpl);
-	t4_register_cpl_handler(sc, CPL_CLOSE_LISTSRV_RPL, do_close_server_rpl);
-	t4_register_cpl_handler(sc, CPL_PASS_ACCEPT_REQ, do_pass_accept_req);
-	t4_register_cpl_handler(sc, CPL_PASS_ESTABLISH, do_pass_establish);
+	t4_register_cpl_handler(CPL_PASS_OPEN_RPL, do_pass_open_rpl);
+	t4_register_cpl_handler(CPL_CLOSE_LISTSRV_RPL, do_close_server_rpl);
+	t4_register_cpl_handler(CPL_PASS_ACCEPT_REQ, do_pass_accept_req);
+	t4_register_cpl_handler(CPL_PASS_ESTABLISH, do_pass_establish);
 }
 #endif
diff --git a/sys/dev/cxgbe/tom/t4_tom.c b/sys/dev/cxgbe/tom/t4_tom.c
index 452c47e5de96..2a8082e92865 100644
--- a/sys/dev/cxgbe/tom/t4_tom.c
+++ b/sys/dev/cxgbe/tom/t4_tom.c
@@ -381,8 +381,9 @@ t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name)
 
 	switch (name) {
 	case TCP_NODELAY:
-		t4_set_tcb_field(sc, toep, 1, W_TCB_T_FLAGS, V_TF_NAGLE(1),
-		    V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1));
+		t4_set_tcb_field(sc, toep->ctrlq, toep->tid, W_TCB_T_FLAGS,
+		    V_TF_NAGLE(1), V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1),
+		    0, 0, toep->ofld_rxq->iq.abs_id);
 		break;
 	default:
 		break;
@@ -930,8 +931,6 @@ free_tom_data(struct adapter *sc, struct tom_data *td)
 	KASSERT(td->lctx_count == 0,
 	    ("%s: lctx hash table is not empty.", __func__));
 
-	t4_uninit_l2t_cpl_handlers(sc);
-	t4_uninit_cpl_io_handlers(sc);
 	t4_uninit_ddp(sc, td);
 	destroy_clip_table(sc, td);
 
@@ -997,7 +996,8 @@ t4_tom_activate(struct adapter *sc)
 	struct tom_data *td;
 	struct toedev *tod;
 	struct vi_info *vi;
-	int i, rc, v;
+	struct sge_ofld_rxq *ofld_rxq;
+	int i, j, rc, v;
 
 	ASSERT_SYNCHRONIZED_OP(sc);
 
@@ -1031,12 +1031,6 @@ t4_tom_activate(struct adapter *sc)
 	/* CLIP table for IPv6 offload */
 	init_clip_table(sc, td);
 
-	/* CPL handlers */
-	t4_init_connect_cpl_handlers(sc);
-	t4_init_l2t_cpl_handlers(sc);
-	t4_init_listen_cpl_handlers(sc);
-	t4_init_cpl_io_handlers(sc);
-
 	/* toedev ops */
 	tod = &td->tod;
 	init_toedev(tod);
@@ -1059,6 +1053,10 @@ t4_tom_activate(struct adapter *sc)
 	for_each_port(sc, i) {
 		for_each_vi(sc->port[i], v, vi) {
 			TOEDEV(vi->ifp) = &td->tod;
+			for_each_ofld_rxq(vi, j, ofld_rxq) {
+				ofld_rxq->iq.set_tcb_rpl = do_set_tcb_rpl;
+				ofld_rxq->iq.l2t_write_rpl = do_l2t_write_rpl2;
+			}
 		}
 	}
 
@@ -1127,6 +1125,11 @@ t4_tom_mod_load(void)
 	int rc;
 	struct protosw *tcp_protosw, *tcp6_protosw;
 
+	/* CPL handlers */
+	t4_init_connect_cpl_handlers();
+	t4_init_listen_cpl_handlers();
+	t4_init_cpl_io_handlers();
+
 	rc = t4_ddp_mod_load();
 	if (rc != 0)
 		return (rc);
diff --git a/sys/dev/cxgbe/tom/t4_tom.h b/sys/dev/cxgbe/tom/t4_tom.h
index 09238a49b346..f114a3dfb429 100644
--- a/sys/dev/cxgbe/tom/t4_tom.h
+++ b/sys/dev/cxgbe/tom/t4_tom.h
@@ -294,13 +294,13 @@ struct clip_entry *hold_lip(struct tom_data *, struct in6_addr *);
 void release_lip(struct tom_data *, struct clip_entry *);
 
 /* t4_connect.c */
-void t4_init_connect_cpl_handlers(struct adapter *);
+void t4_init_connect_cpl_handlers(void);
 int t4_connect(struct toedev *, struct socket *, struct rtentry *,
     struct sockaddr *);
 void act_open_failure_cleanup(struct adapter *, u_int, u_int);
 
 /* t4_listen.c */
-void t4_init_listen_cpl_handlers(struct adapter *);
+void t4_init_listen_cpl_handlers(void);
 int t4_listen_start(struct toedev *, struct tcpcb *);
 int t4_listen_stop(struct toedev *, struct tcpcb *);
 void t4_syncache_added(struct toedev *, void *);
@@ -313,8 +313,8 @@ int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *,
 void t4_offload_socket(struct toedev *, void *, struct socket *);
 
 /* t4_cpl_io.c */
-void t4_init_cpl_io_handlers(struct adapter *);
-void t4_uninit_cpl_io_handlers(struct adapter *);
+void t4_init_cpl_io_handlers(void);
+void t4_uninit_cpl_io_handlers(void);
 void send_abort_rpl(struct adapter *, struct sge_wrq *, int , int);
 void send_flowc_wr(struct toepcb *, struct flowc_tx_params *);
 void send_reset(struct adapter *, struct toepcb *, uint32_t);
@@ -324,12 +324,11 @@ void t4_rcvd_locked(struct toedev *, struct tcpcb *);
 int t4_tod_output(struct toedev *, struct tcpcb *);
 int t4_send_fin(struct toedev *, struct tcpcb *);
 int t4_send_rst(struct toedev *, struct tcpcb *);
-void t4_set_tcb_field(struct adapter *, struct toepcb *, int, uint16_t,
-    uint64_t, uint64_t);
-void t4_set_tcb_field_rpl(struct adapter *, struct toepcb *, int, uint16_t,
-    uint64_t, uint64_t, uint8_t);
+void t4_set_tcb_field(struct adapter *, struct sge_wrq *, int, uint16_t,
+    uint64_t, uint64_t, int, int, int);
 void t4_push_frames(struct adapter *sc, struct toepcb *toep, int drop);
 void t4_push_pdus(struct adapter *sc, struct toepcb *toep, int drop);
+int do_set_tcb_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
 
 /* t4_ddp.c */
 void t4_init_ddp(struct adapter *, struct tom_data *);
diff --git a/sys/dev/cxgbe/tom/t4_tom_l2t.c b/sys/dev/cxgbe/tom/t4_tom_l2t.c
index 8aadf34a9b5e..f89df02e1000 100644
--- a/sys/dev/cxgbe/tom/t4_tom_l2t.c
+++ b/sys/dev/cxgbe/tom/t4_tom_l2t.c
@@ -219,7 +219,7 @@ update_entry(struct adapter *sc, struct l2t_entry *e, uint8_t *lladdr,
 
 			memcpy(e->dmac, lladdr, ETHER_ADDR_LEN);
 			e->vlan = vtag;
-			t4_write_l2e(sc, e, 1);
+			t4_write_l2e(e, 1);
 		}
 		e->state = L2T_STATE_VALID;
 	}
@@ -309,19 +309,7 @@ t4_l2t_send_slow(struct adapter *sc, struct wrqe *wr, struct l2t_entry *e)
 	return (0);
 }
 
-/*
- * Called when an L2T entry has no more users.  The entry is left in the hash
- * table since it is likely to be reused but we also bump nfree to indicate
- * that the entry can be reallocated for a different neighbor.  We also drop
- * the existing neighbor reference in case the neighbor is going away and is
- * waiting on our reference.
- *
- * Because entries can be reallocated to other neighbors once their ref count
- * drops to 0 we need to take the entry's lock to avoid races with a new
- * incarnation.
- */
-
-static int
+int
 do_l2t_write_rpl2(struct sge_iq *iq, const struct rss_header *rss,
     struct mbuf *m)
 {
@@ -329,11 +317,13 @@ do_l2t_write_rpl2(struct sge_iq *iq, const struct rss_header *rss,
 	const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1);
 	unsigned int tid = GET_TID(rpl);
 	unsigned int idx = tid % L2T_SIZE;
-	int rc;
 
-	rc = do_l2t_write_rpl(iq, rss, m);
-	if (rc != 0)
-		return (rc);
+	if (__predict_false(rpl->status != CPL_ERR_NONE)) {
+		log(LOG_ERR,
+		    "Unexpected L2T_WRITE_RPL (%u) for entry at hw_idx %u\n",
+		    rpl->status, idx);
+		return (EINVAL);
+	}
 
 	if (tid & F_SYNC_WR) {
 		struct l2t_entry *e = &sc->l2t->l2tab[idx - sc->vres.l2t.start];
@@ -349,20 +339,6 @@ do_l2t_write_rpl2(struct sge_iq *iq, const struct rss_header *rss,
 	return (0);
 }
 
-void
-t4_init_l2t_cpl_handlers(struct adapter *sc)
-{
-
-	t4_register_cpl_handler(sc, CPL_L2T_WRITE_RPL, do_l2t_write_rpl2);
-}
-
-void
-t4_uninit_l2t_cpl_handlers(struct adapter *sc)
-{
-
-	t4_register_cpl_handler(sc, CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
-}
-
 /*
  * The TOE wants an L2 table entry that it can use to reach the next hop over
  * the specified port.  Produce such an entry - create one if needed.
@@ -374,7 +350,8 @@ struct l2t_entry *
 t4_l2t_get(struct port_info *pi, struct ifnet *ifp, struct sockaddr *sa)
 {
 	struct l2t_entry *e;
-	struct l2t_data *d = pi->adapter->l2t;
+	struct adapter *sc = pi->adapter;
+	struct l2t_data *d = sc->l2t;
 	u_int hash, smt_idx = pi->port_id;
 
 	KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6,
@@ -409,6 +386,8 @@ t4_l2t_get(struct port_info *pi, struct ifnet *ifp, struct sockaddr *sa)
 		e->smt_idx = smt_idx;
 		e->hash = hash;
 		e->lport = pi->lport;
+		e->wrq = &sc->sge.ctrlq[pi->port_id];
+		e->iqid = sc->sge.ofld_rxq[pi->vi[0].first_ofld_rxq].iq.abs_id;
 		atomic_store_rel_int(&e->refcnt, 1);
 #ifdef VLAN_TAG
 		if (ifp->if_type == IFT_L2VLAN)
diff --git a/sys/dev/cxgbe/tom/t4_tom_l2t.h b/sys/dev/cxgbe/tom/t4_tom_l2t.h
index 3d76735707e3..9a7fb39fbd79 100644
--- a/sys/dev/cxgbe/tom/t4_tom_l2t.h
+++ b/sys/dev/cxgbe/tom/t4_tom_l2t.h
@@ -37,8 +37,8 @@ struct l2t_entry *t4_l2t_get(struct port_info *, struct ifnet *,
     struct sockaddr *);
 void t4_l2_update(struct toedev *, struct ifnet *, struct sockaddr *,
     uint8_t *, uint16_t);
-void t4_init_l2t_cpl_handlers(struct adapter *);
-void t4_uninit_l2t_cpl_handlers(struct adapter *);
+int do_l2t_write_rpl2(struct sge_iq *, const struct rss_header *,
+    struct mbuf *);
 
 static inline int
 t4_l2t_send(struct adapter *sc, struct wrqe *wr, struct l2t_entry *e)