IB/hfi1: Add 16B RC/UC support

Add 16B bypass packet support for RC/UC traffic types.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Don Hiatt <don.hiatt@intel.com>
Signed-off-by: Dasaratharaman Chandramouli <dasaratharaman.chandramouli@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

drivers/infiniband/hw/hfi1/hfi.h

@@ -459,6 +459,8 @@ static inline u16 hfi1_16B_get_entropy(struct hfi1_16b_header *hdr)
 	return (u16)(hdr->lrh[3] & OPA_16B_ENTROPY_MASK);
 }
 
+#define OPA_16B_MAKE_QW(low_dw, high_dw) (((u64)(high_dw) << 32) | (low_dw))
+
 /*
  * BTH
  */
@@ -1538,7 +1540,7 @@ static inline u32 egress_cycles(u32 len, u32 rate)
 }
 
 void set_link_ipg(struct hfi1_pportdata *ppd);
-void process_becn(struct hfi1_pportdata *ppd, u8 sl,  u16 rlid, u32 lqpn,
+void process_becn(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn,
 		  u32 rqpn, u8 svc_type);
 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
 		u32 pkey, u32 slid, u32 dlid, u8 sc5,

drivers/infiniband/hw/hfi1/rc.c

@@ -100,8 +100,12 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp,
 	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK))
 		goto bail;
 
-	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
-	hwords = 5;
+	if (priv->hdr_type == HFI1_PKT_TYPE_9B)
+		/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+		hwords = 5;
+	else
+		/* header size in 32-bit words 16B LRH+BTH = (16+12)/4. */
+		hwords = 7;
 
 	switch (qp->s_ack_state) {
 	case OP(RDMA_READ_RESPONSE_LAST):
@@ -258,8 +262,7 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
 	struct ib_other_headers *ohdr;
 	struct rvt_sge_state *ss;
 	struct rvt_swqe *wqe;
-	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
-	u32 hwords = 5;
+	u32 hwords;
 	u32 len;
 	u32 bth0 = 0;
 	u32 bth2;
@@ -273,9 +276,23 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
 	if (IS_ERR(ps->s_txreq))
 		goto bail_no_tx;
 
-	ohdr = &ps->s_txreq->phdr.hdr.ibh.u.oth;
-	if (rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)
-		ohdr = &ps->s_txreq->phdr.hdr.ibh.u.l.oth;
+	ps->s_txreq->phdr.hdr.hdr_type = priv->hdr_type;
+	if (priv->hdr_type == HFI1_PKT_TYPE_9B) {
+		/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+		hwords = 5;
+		if (rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)
+			ohdr = &ps->s_txreq->phdr.hdr.ibh.u.l.oth;
+		else
+			ohdr = &ps->s_txreq->phdr.hdr.ibh.u.oth;
+	} else {
+		/* header size in 32-bit words 16B LRH+BTH = (16+12)/4. */
+		hwords = 7;
+		if ((rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH) &&
+		    (hfi1_check_mcast(rdma_ah_get_dlid(&qp->remote_ah_attr))))
+			ohdr = &ps->s_txreq->phdr.hdr.opah.u.l.oth;
+		else
+			ohdr = &ps->s_txreq->phdr.hdr.opah.u.oth;
+	}
 
 	/* Sending responses has higher priority over sending requests. */
 	if ((qp->s_flags & RVT_S_RESP_PENDING) &&
@@ -703,6 +720,154 @@ bail_no_tx:
 	return 0;
 }
 
+static inline void hfi1_make_bth_aeth(struct rvt_qp *qp,
+				      struct ib_other_headers *ohdr,
+				      u32 bth0, u32 bth1)
+{
+	if (qp->r_nak_state)
+		ohdr->u.aeth = cpu_to_be32((qp->r_msn & IB_MSN_MASK) |
+					    (qp->r_nak_state <<
+					     IB_AETH_CREDIT_SHIFT));
+	else
+		ohdr->u.aeth = rvt_compute_aeth(qp);
+
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	ohdr->bth[1] = cpu_to_be32(bth1 | qp->remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(mask_psn(qp->r_ack_psn));
+}
+
+static inline void hfi1_queue_rc_ack(struct rvt_qp *qp, bool is_fecn)
+{
+	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK))
+		goto unlock;
+	this_cpu_inc(*ibp->rvp.rc_qacks);
+	qp->s_flags |= RVT_S_ACK_PENDING | RVT_S_RESP_PENDING;
+	qp->s_nak_state = qp->r_nak_state;
+	qp->s_ack_psn = qp->r_ack_psn;
+	if (is_fecn)
+		qp->s_flags |= RVT_S_ECN;
+
+	/* Schedule the send tasklet. */
+	hfi1_schedule_send(qp);
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}
+
+static inline void hfi1_make_rc_ack_9B(struct rvt_qp *qp,
+				       struct hfi1_opa_header *opa_hdr,
+				       u8 sc5, bool is_fecn,
+				       u64 *pbc_flags, u32 *hwords,
+				       u32 *nwords)
+{
+	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
+	struct ib_header *hdr = &opa_hdr->ibh;
+	struct ib_other_headers *ohdr;
+	u16 lrh0 = HFI1_LRH_BTH;
+	u16 pkey;
+	u32 bth0, bth1;
+
+	opa_hdr->hdr_type = HFI1_PKT_TYPE_9B;
+	ohdr = &hdr->u.oth;
+	/* header size in 32-bit words LRH+BTH+AETH = (8+12+4)/4 */
+	*hwords = 6;
+
+	if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)) {
+		*hwords += hfi1_make_grh(ibp, &hdr->u.l.grh,
+					 rdma_ah_read_grh(&qp->remote_ah_attr),
+					 *hwords - 2, SIZE_OF_CRC);
+		ohdr = &hdr->u.l.oth;
+		lrh0 = HFI1_LRH_GRH;
+	}
+	/* set PBC_DC_INFO bit (aka SC[4]) in pbc_flags */
+	*pbc_flags |= ((!!(sc5 & 0x10)) << PBC_DC_INFO_SHIFT);
+
+	/* read pkey_index w/o lock (its atomic) */
+	pkey = hfi1_get_pkey(ibp, qp->s_pkey_index);
+
+	lrh0 |= (sc5 & IB_SC_MASK) << IB_SC_SHIFT |
+		(rdma_ah_get_sl(&qp->remote_ah_attr) & IB_SL_MASK) <<
+			IB_SL_SHIFT;
+
+	hfi1_make_ib_hdr(hdr, lrh0, *hwords + SIZE_OF_CRC,
+			 opa_get_lid(rdma_ah_get_dlid(&qp->remote_ah_attr), 9B),
+			 ppd->lid | rdma_ah_get_path_bits(&qp->remote_ah_attr));
+
+	bth0 = pkey | (OP(ACKNOWLEDGE) << 24);
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth0 |= IB_BTH_MIG_REQ;
+	bth1 = (!!is_fecn) << IB_BECN_SHIFT;
+	hfi1_make_bth_aeth(qp, ohdr, bth0, bth1);
+}
+
+static inline void hfi1_make_rc_ack_16B(struct rvt_qp *qp,
+					struct hfi1_opa_header *opa_hdr,
+					u8 sc5, bool is_fecn,
+					u64 *pbc_flags, u32 *hwords,
+					u32 *nwords)
+{
+	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
+	struct hfi1_16b_header *hdr = &opa_hdr->opah;
+	struct ib_other_headers *ohdr;
+	u32 bth0, bth1;
+	u16 len, pkey;
+	u8 becn = !!is_fecn;
+	u8 l4 = OPA_16B_L4_IB_LOCAL;
+	u8 extra_bytes;
+
+	opa_hdr->hdr_type = HFI1_PKT_TYPE_16B;
+	ohdr = &hdr->u.oth;
+	/* header size in 32-bit words 16B LRH+BTH+AETH = (16+12+4)/4 */
+	*hwords = 8;
+	extra_bytes = hfi1_get_16b_padding(*hwords << 2, 0);
+	*nwords = SIZE_OF_CRC + ((extra_bytes + SIZE_OF_LT) >> 2);
+
+	if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH) &&
+	    hfi1_check_mcast(rdma_ah_get_dlid(&qp->remote_ah_attr))) {
+		*hwords += hfi1_make_grh(ibp, &hdr->u.l.grh,
+					 rdma_ah_read_grh(&qp->remote_ah_attr),
+					 *hwords - 4, *nwords);
+		ohdr = &hdr->u.l.oth;
+		l4 = OPA_16B_L4_IB_GLOBAL;
+	}
+	*pbc_flags |= PBC_PACKET_BYPASS | PBC_INSERT_BYPASS_ICRC;
+
+	/* read pkey_index w/o lock (its atomic) */
+	pkey = hfi1_get_pkey(ibp, qp->s_pkey_index);
+
+	/* Convert dwords to flits */
+	len = (*hwords + *nwords) >> 1;
+
+	hfi1_make_16b_hdr(hdr,
+			  ppd->lid | rdma_ah_get_path_bits(&qp->remote_ah_attr),
+			  opa_get_lid(rdma_ah_get_dlid(&qp->remote_ah_attr),
+				      16B),
+			  len, pkey, becn, 0, l4, sc5);
+
+	bth0 = pkey | (OP(ACKNOWLEDGE) << 24);
+	bth0 |= extra_bytes << 20;
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth1 = OPA_BTH_MIG_REQ;
+	hfi1_make_bth_aeth(qp, ohdr, bth0, bth1);
+}
+
+typedef void (*hfi1_make_rc_ack)(struct rvt_qp *qp,
+				 struct hfi1_opa_header *opa_hdr,
+				 u8 sc5, bool is_fecn,
+				 u64 *pbc_flags, u32 *hwords,
+				 u32 *nwords);
+
+/* We support only two types - 9B and 16B for now */
+static const hfi1_make_rc_ack hfi1_make_rc_ack_tbl[2] = {
+	[HFI1_PKT_TYPE_9B] = &hfi1_make_rc_ack_9B,
+	[HFI1_PKT_TYPE_16B] = &hfi1_make_rc_ack_16B
+};
+
 /**
  * hfi1_send_rc_ack - Construct an ACK packet and send it
  * @qp: a pointer to the QP
@@ -711,87 +876,48 @@ bail_no_tx:
  * Note that RDMA reads and atomics are handled in the
  * send side QP state and send engine.
  */
-void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp,
-		      int is_fecn)
+void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd,
+		      struct rvt_qp *qp, bool is_fecn)
 {
 	struct hfi1_ibport *ibp = rcd_to_iport(rcd);
+	struct hfi1_qp_priv *priv = qp->priv;
 	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
+	u8 sc5 = ibp->sl_to_sc[rdma_ah_get_sl(&qp->remote_ah_attr)];
 	u64 pbc, pbc_flags = 0;
-	u16 lrh0;
-	u16 sc5;
-	u32 bth0;
-	u32 hwords;
-	u32 vl, plen;
-	struct send_context *sc;
+	u32 hwords = 0;
+	u32 nwords = 0;
+	u32 plen;
 	struct pio_buf *pbuf;
-	struct hfi1_opa_header opah;
-	struct ib_header *hdr;
-	struct ib_other_headers *ohdr;
-	unsigned long flags;
+	struct hfi1_opa_header opa_hdr;
 
 	/* clear the defer count */
 	qp->r_adefered = 0;
 
 	/* Don't send ACK or NAK if a RDMA read or atomic is pending. */
-	if (qp->s_flags & RVT_S_RESP_PENDING)
-		goto queue_ack;
+	if (qp->s_flags & RVT_S_RESP_PENDING) {
+		hfi1_queue_rc_ack(qp, is_fecn);
+		return;
+	}
 
 	/* Ensure s_rdma_ack_cnt changes are committed */
 	smp_read_barrier_depends();
-	if (qp->s_rdma_ack_cnt)
-		goto queue_ack;
-
-	/* Construct the header */
-	opah.hdr_type = 0;
-	hdr = &opah.ibh;
-
-	/* header size in 32-bit words LRH+BTH+AETH = (8+12+4)/4 */
-	hwords = 6;
-	if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)) {
-		hwords += hfi1_make_grh(ibp, &hdr->u.l.grh,
-					rdma_ah_read_grh(&qp->remote_ah_attr),
-					hwords, 0);
-		ohdr = &hdr->u.l.oth;
-		lrh0 = HFI1_LRH_GRH;
-	} else {
-		ohdr = &hdr->u.oth;
-		lrh0 = HFI1_LRH_BTH;
+	if (qp->s_rdma_ack_cnt) {
+		hfi1_queue_rc_ack(qp, is_fecn);
+		return;
 	}
-	/* read pkey_index w/o lock (its atomic) */
-	bth0 = hfi1_get_pkey(ibp, qp->s_pkey_index) | (OP(ACKNOWLEDGE) << 24);
-	if (qp->s_mig_state == IB_MIG_MIGRATED)
-		bth0 |= IB_BTH_MIG_REQ;
-	if (qp->r_nak_state)
-		ohdr->u.aeth = cpu_to_be32((qp->r_msn & IB_MSN_MASK) |
-					    (qp->r_nak_state <<
-					     IB_AETH_CREDIT_SHIFT));
-	else
-		ohdr->u.aeth = rvt_compute_aeth(qp);
-	sc5 = ibp->sl_to_sc[rdma_ah_get_sl(&qp->remote_ah_attr)];
-	/* set PBC_DC_INFO bit (aka SC[4]) in pbc_flags */
-	pbc_flags |= (ib_is_sc5(sc5) << PBC_DC_INFO_SHIFT);
-	lrh0 |= (sc5 & 0xf) << 12 | (rdma_ah_get_sl(&qp->remote_ah_attr)
-				     & 0xf) << 4;
-	hdr->lrh[0] = cpu_to_be16(lrh0);
-	hdr->lrh[1] = cpu_to_be16(rdma_ah_get_dlid(&qp->remote_ah_attr));
-	hdr->lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
-	hdr->lrh[3] = cpu_to_be16(ppd->lid |
-				  rdma_ah_get_path_bits(&qp->remote_ah_attr));
-	ohdr->bth[0] = cpu_to_be32(bth0);
-	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
-	ohdr->bth[1] |= cpu_to_be32((!!is_fecn) << IB_BECN_SHIFT);
-	ohdr->bth[2] = cpu_to_be32(mask_psn(qp->r_ack_psn));
 
 	/* Don't try to send ACKs if the link isn't ACTIVE */
 	if (driver_lstate(ppd) != IB_PORT_ACTIVE)
 		return;
 
-	sc = rcd->sc;
-	plen = 2 /* PBC */ + hwords;
-	vl = sc_to_vlt(ppd->dd, sc5);
-	pbc = create_pbc(ppd, pbc_flags, qp->srate_mbps, vl, plen);
+	/* Make the appropriate header */
+	hfi1_make_rc_ack_tbl[priv->hdr_type](qp, &opa_hdr, sc5, is_fecn,
+					     &pbc_flags, &hwords, &nwords);
 
-	pbuf = sc_buffer_alloc(sc, plen, NULL, NULL);
+	plen = 2 /* PBC */ + hwords + nwords;
+	pbc = create_pbc(ppd, pbc_flags, qp->srate_mbps,
+			 sc_to_vlt(ppd->dd, sc5), plen);
+	pbuf = sc_buffer_alloc(rcd->sc, plen, NULL, NULL);
 	if (!pbuf) {
 		/*
 		 * We have no room to send at the moment.  Pass
@@ -799,32 +925,18 @@ void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp,
 		 * so that when enough buffer space becomes available,
 		 * the ACK is sent ahead of other outgoing packets.
 		 */
-		goto queue_ack;
+		hfi1_queue_rc_ack(qp, is_fecn);
+		return;
 	}
-
 	trace_ack_output_ibhdr(dd_from_ibdev(qp->ibqp.device),
-			       &opah, ib_is_sc5(sc5));
+			       &opa_hdr, ib_is_sc5(sc5));
 
 	/* write the pbc and data */
-	ppd->dd->pio_inline_send(ppd->dd, pbuf, pbc, hdr, hwords);
-
+	ppd->dd->pio_inline_send(ppd->dd, pbuf, pbc,
+				 (priv->hdr_type == HFI1_PKT_TYPE_9B ?
+				 (void *)&opa_hdr.ibh :
+				 (void *)&opa_hdr.opah), hwords);
 	return;
-
-queue_ack:
-	spin_lock_irqsave(&qp->s_lock, flags);
-	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK))
-		goto unlock;
-	this_cpu_inc(*ibp->rvp.rc_qacks);
-	qp->s_flags |= RVT_S_ACK_PENDING | RVT_S_RESP_PENDING;
-	qp->s_nak_state = qp->r_nak_state;
-	qp->s_ack_psn = qp->r_ack_psn;
-	if (is_fecn)
-		qp->s_flags |= RVT_S_ECN;
-
-	/* Schedule the send engine. */
-	hfi1_schedule_send(qp);
-unlock:
-	spin_unlock_irqrestore(&qp->s_lock, flags);
 }
 
 /**
@@ -992,8 +1104,10 @@ static void reset_sending_psn(struct rvt_qp *qp, u32 psn)
 void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah)
 {
 	struct ib_other_headers *ohdr;
-	struct ib_header *hdr = &opah->ibh;
+	struct hfi1_qp_priv *priv = qp->priv;
 	struct rvt_swqe *wqe;
+	struct ib_header *hdr = NULL;
+	struct hfi1_16b_header *hdr_16b = NULL;
 	u32 opcode;
 	u32 psn;
 
@@ -1002,10 +1116,22 @@ void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah)
 		return;
 
 	/* Find out where the BTH is */
-	if (ib_get_lnh(hdr) == HFI1_LRH_BTH)
-		ohdr = &hdr->u.oth;
-	else
-		ohdr = &hdr->u.l.oth;
+	if (priv->hdr_type == HFI1_PKT_TYPE_9B) {
+		hdr = &opah->ibh;
+		if (ib_get_lnh(hdr) == HFI1_LRH_BTH)
+			ohdr = &hdr->u.oth;
+		else
+			ohdr = &hdr->u.l.oth;
+	} else {
+		u8 l4;
+
+		hdr_16b = &opah->opah;
+		l4  = hfi1_16B_get_l4(hdr_16b);
+		if (l4 == OPA_16B_L4_IB_LOCAL)
+			ohdr = &hdr_16b->u.oth;
+		else
+			ohdr = &hdr_16b->u.l.oth;
+	}
 
 	opcode = ib_bth_get_opcode(ohdr);
 	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
@@ -1405,36 +1531,34 @@ static void rdma_seq_err(struct rvt_qp *qp, struct hfi1_ibport *ibp, u32 psn,
 
 /**
  * rc_rcv_resp - process an incoming RC response packet
- * @ibp: the port this packet came in on
- * @ohdr: the other headers for this packet
- * @data: the packet data
- * @tlen: the packet length
- * @qp: the QP for this packet
- * @opcode: the opcode for this packet
- * @psn: the packet sequence number for this packet
- * @hdrsize: the header length
- * @pmtu: the path MTU
+ * @packet: data packet information
  *
  * This is called from hfi1_rc_rcv() to process an incoming RC response
  * packet for the given QP.
  * Called at interrupt level.
  */
-static void rc_rcv_resp(struct hfi1_ibport *ibp,
-			struct ib_other_headers *ohdr,
-			void *data, u32 tlen, struct rvt_qp *qp,
-			u32 opcode, u32 psn, u32 hdrsize, u32 pmtu,
-			struct hfi1_ctxtdata *rcd)
+static void rc_rcv_resp(struct hfi1_packet *packet)
 {
+	struct hfi1_ctxtdata *rcd = packet->rcd;
+	void *data = packet->payload;
+	u32 tlen = packet->tlen;
+	struct rvt_qp *qp = packet->qp;
+	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct ib_other_headers *ohdr = packet->ohdr;
 	struct rvt_swqe *wqe;
 	enum ib_wc_status status;
 	unsigned long flags;
 	int diff;
-	u32 pad;
-	u32 aeth;
 	u64 val;
+	u32 aeth;
+	u32 psn = ib_bth_get_psn(packet->ohdr);
+	u32 pmtu = qp->pmtu;
+	u16 hdrsize = packet->hlen;
+	u8 opcode = packet->opcode;
+	u8 pad = packet->pad;
+	u8 extra_bytes = pad + packet->extra_byte + (SIZE_OF_CRC << 2);
 
 	spin_lock_irqsave(&qp->s_lock, flags);
-
 	trace_hfi1_ack(qp, psn);
 
 	/* Ignore invalid responses. */
@@ -1500,7 +1624,7 @@ static void rc_rcv_resp(struct hfi1_ibport *ibp,
 		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
 			goto ack_op_err;
 read_middle:
-		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+		if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
 			goto ack_len_err;
 		if (unlikely(pmtu >= qp->s_rdma_read_len))
 			goto ack_len_err;
@@ -1532,13 +1656,11 @@ read_middle:
 		aeth = be32_to_cpu(ohdr->u.aeth);
 		if (!do_rc_ack(qp, aeth, psn, opcode, 0, rcd))
 			goto ack_done;
-		/* Get the number of bytes the message was padded by. */
-		pad = ib_bth_get_pad(ohdr);
 		/*
 		 * Check that the data size is >= 0 && <= pmtu.
 		 * Remember to account for ICRC (4).
 		 */
-		if (unlikely(tlen < (hdrsize + pad + 4)))
+		if (unlikely(tlen < (hdrsize + extra_bytes)))
 			goto ack_len_err;
 		/*
 		 * If this is a response to a resent RDMA read, we
@@ -1556,16 +1678,14 @@ read_middle:
 			goto ack_seq_err;
 		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
 			goto ack_op_err;
-		/* Get the number of bytes the message was padded by. */
-		pad = ib_bth_get_pad(ohdr);
 		/*
 		 * Check that the data size is >= 1 && <= pmtu.
 		 * Remember to account for ICRC (4).
 		 */
-		if (unlikely(tlen <= (hdrsize + pad + 4)))
+		if (unlikely(tlen <= (hdrsize + extra_bytes)))
 			goto ack_len_err;
 read_last:
-		tlen -= hdrsize + pad + 4;
+		tlen -= hdrsize + extra_bytes;
 		if (unlikely(tlen != qp->s_rdma_read_len))
 			goto ack_len_err;
 		aeth = be32_to_cpu(ohdr->u.aeth);
@@ -1850,7 +1970,7 @@ static void log_cca_event(struct hfi1_pportdata *ppd, u8 sl, u32 rlid,
 	spin_unlock_irqrestore(&ppd->cc_log_lock, flags);
 }
 
-void process_becn(struct hfi1_pportdata *ppd, u8 sl, u16 rlid, u32 lqpn,
+void process_becn(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn,
 		  u32 rqpn, u8 svc_type)
 {
 	struct cca_timer *cca_timer;
@@ -1907,12 +2027,7 @@ void process_becn(struct hfi1_pportdata *ppd, u8 sl, u16 rlid, u32 lqpn,
 
 /**
  * hfi1_rc_rcv - process an incoming RC packet
- * @rcd: the context pointer
- * @hdr: the header of this packet
- * @rcv_flags: flags relevant to rcv processing
- * @data: the packet data
- * @tlen: the packet length
- * @qp: the QP for this packet
+ * @packet: data packet information
  *
  * This is called from qp_rcv() to process an incoming RC packet
  * for the given QP.
@@ -1926,10 +2041,10 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 	struct rvt_qp *qp = packet->qp;
 	struct hfi1_ibport *ibp = rcd_to_iport(rcd);
 	struct ib_other_headers *ohdr = packet->ohdr;
-	u32 bth0;
+	u32 bth0 = be32_to_cpu(ohdr->bth[0]);
 	u32 opcode = packet->opcode;
 	u32 hdrsize = packet->hlen;
-	u32 psn;
+	u32 psn = ib_bth_get_psn(packet->ohdr);
 	u32 pad = packet->pad;
 	struct ib_wc wc;
 	u32 pmtu = qp->pmtu;
@@ -1940,15 +2055,14 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 	bool is_fecn = false;
 	bool copy_last = false;
 	u32 rkey;
+	u8 extra_bytes = pad + packet->extra_byte + (SIZE_OF_CRC << 2);
 
 	lockdep_assert_held(&qp->r_lock);
 
-	bth0 = be32_to_cpu(ohdr->bth[0]);
 	if (hfi1_ruc_check_hdr(ibp, packet))
 		return;
 
 	is_fecn = process_ecn(qp, packet, false);
-	psn = ib_bth_get_psn(ohdr);
 
 	/*
 	 * Process responses (ACKs) before anything else.  Note that the
@@ -1958,8 +2072,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 	 */
 	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
 	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
-		rc_rcv_resp(ibp, ohdr, data, tlen, qp, opcode, psn,
-			    hdrsize, pmtu, rcd);
+		rc_rcv_resp(packet);
 		if (is_fecn)
 			goto send_ack;
 		return;
@@ -2026,7 +2139,12 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 	case OP(RDMA_WRITE_MIDDLE):
 send_middle:
 		/* Check for invalid length PMTU or posted rwqe len. */
-		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+		/*
+		 * There will be no padding for 9B packet but 16B packets
+		 * will come in with some padding since we always add
+		 * CRC and LT bytes which will need to be flit aligned
+		 */
+		if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
 			goto nack_inv;
 		qp->r_rcv_len += pmtu;
 		if (unlikely(qp->r_rcv_len > qp->r_len))
@@ -2080,10 +2198,10 @@ no_immediate_data:
 send_last:
 		/* Check for invalid length. */
 		/* LAST len should be >= 1 */
-		if (unlikely(tlen < (hdrsize + pad + 4)))
+		if (unlikely(tlen < (hdrsize + extra_bytes)))
 			goto nack_inv;
-		/* Don't count the CRC. */
-		tlen -= (hdrsize + pad + 4);
+		/* Don't count the CRC(and padding and LT byte for 16B). */
+		tlen -= (hdrsize + extra_bytes);
 		wc.byte_len = tlen + qp->r_rcv_len;
 		if (unlikely(wc.byte_len > qp->r_len))
 			goto nack_inv;

drivers/infiniband/hw/hfi1/ruc.c

@@ -735,73 +735,186 @@ static inline void build_ahg(struct rvt_qp *qp, u32 npsn)
 	}
 }
 
-void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
-			  u32 bth0, u32 bth2, int middle,
-			  struct hfi1_pkt_state *ps)
+static inline void hfi1_make_ruc_bth(struct rvt_qp *qp,
+				     struct ib_other_headers *ohdr,
+				     u32 bth0, u32 bth1, u32 bth2)
+{
+	bth1 |= qp->remote_qpn;
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	ohdr->bth[1] = cpu_to_be32(bth1);
+	ohdr->bth[2] = cpu_to_be32(bth2);
+}
+
+static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
+					    struct ib_other_headers *ohdr,
+					    u32 bth0, u32 bth2, int middle,
+					    struct hfi1_pkt_state *ps)
+{
+	struct hfi1_qp_priv *priv = qp->priv;
+	struct hfi1_ibport *ibp = ps->ibp;
+	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
+	u32 bth1 = 0;
+	u32 slid;
+	u16 pkey = hfi1_get_pkey(ibp, qp->s_pkey_index);
+	u8 l4 = OPA_16B_L4_IB_LOCAL;
+	u8 extra_bytes = hfi1_get_16b_padding((qp->s_hdrwords << 2),
+				   ps->s_txreq->s_cur_size);
+	u32 nwords = SIZE_OF_CRC + ((ps->s_txreq->s_cur_size +
+				 extra_bytes + SIZE_OF_LT) >> 2);
+	u8 becn = 0;
+
+	if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH) &&
+	    hfi1_check_mcast(rdma_ah_get_dlid(&qp->remote_ah_attr))) {
+		struct ib_grh *grh;
+		struct ib_global_route *grd =
+			rdma_ah_retrieve_grh(&qp->remote_ah_attr);
+		int hdrwords;
+
+		/*
+		 * Ensure OPA GIDs are transformed to IB gids
+		 * before creating the GRH.
+		 */
+		if (grd->sgid_index == OPA_GID_INDEX)
+			grd->sgid_index = 0;
+		grh = &ps->s_txreq->phdr.hdr.opah.u.l.grh;
+		l4 = OPA_16B_L4_IB_GLOBAL;
+		hdrwords = qp->s_hdrwords - 4;
+		qp->s_hdrwords += hfi1_make_grh(ibp, grh, grd,
+						hdrwords, nwords);
+		middle = 0;
+	}
+
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth1 |= OPA_BTH_MIG_REQ;
+	else
+		middle = 0;
+
+	if (middle)
+		build_ahg(qp, bth2);
+	else
+		qp->s_flags &= ~RVT_S_AHG_VALID;
+
+	bth0 |= pkey;
+	bth0 |= extra_bytes << 20;
+	if (qp->s_flags & RVT_S_ECN) {
+		qp->s_flags &= ~RVT_S_ECN;
+		/* we recently received a FECN, so return a BECN */
+		becn = 1;
+	}
+	hfi1_make_ruc_bth(qp, ohdr, bth0, bth1, bth2);
+
+	if (!ppd->lid)
+		slid = be32_to_cpu(OPA_LID_PERMISSIVE);
+	else
+		slid = ppd->lid |
+			(rdma_ah_get_path_bits(&qp->remote_ah_attr) &
+			((1 << ppd->lmc) - 1));
+
+	hfi1_make_16b_hdr(&ps->s_txreq->phdr.hdr.opah,
+			  slid,
+			  opa_get_lid(rdma_ah_get_dlid(&qp->remote_ah_attr),
+				      16B),
+			  (qp->s_hdrwords + nwords) >> 1,
+			  pkey, becn, 0, l4, priv->s_sc);
+}
+
+static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp,
+					   struct ib_other_headers *ohdr,
+					   u32 bth0, u32 bth2, int middle,
+					   struct hfi1_pkt_state *ps)
 {
 	struct hfi1_qp_priv *priv = qp->priv;
 	struct hfi1_ibport *ibp = ps->ibp;
-	u16 lrh0;
-	u32 nwords;
-	u32 extra_bytes;
-	u32 bth1;
-
-	/* Construct the header. */
-	extra_bytes = -ps->s_txreq->s_cur_size & 3;
-	nwords = (ps->s_txreq->s_cur_size + extra_bytes) >> 2;
-	lrh0 = HFI1_LRH_BTH;
+	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
+	u32 bth1 = 0;
+	u16 pkey = hfi1_get_pkey(ibp, qp->s_pkey_index);
+	u16 lrh0 = HFI1_LRH_BTH;
+	u16 slid;
+	u8 extra_bytes = -ps->s_txreq->s_cur_size & 3;
+	u32 nwords = SIZE_OF_CRC + ((ps->s_txreq->s_cur_size +
+					 extra_bytes) >> 2);
+
 	if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)) {
-		qp->s_hdrwords +=
-			hfi1_make_grh(ibp,
-				      &ps->s_txreq->phdr.hdr.ibh.u.l.grh,
-				      &qp->remote_ah_attr.grh,
-				      qp->s_hdrwords, nwords);
+		struct ib_grh *grh = &ps->s_txreq->phdr.hdr.ibh.u.l.grh;
+		int hdrwords = qp->s_hdrwords - 2;
+
 		lrh0 = HFI1_LRH_GRH;
+		qp->s_hdrwords +=
+			hfi1_make_grh(ibp, grh,
+				      rdma_ah_read_grh(&qp->remote_ah_attr),
+				      hdrwords, nwords);
 		middle = 0;
 	}
 	lrh0 |= (priv->s_sc & 0xf) << 12 |
 		(rdma_ah_get_sl(&qp->remote_ah_attr) & 0xf) << 4;
-	/*
-	 * reset s_ahg/AHG fields
-	 *
-	 * This insures that the ahgentry/ahgcount
-	 * are at a non-AHG default to protect
-	 * build_verbs_tx_desc() from using
-	 * an include ahgidx.
-	 *
-	 * build_ahg() will modify as appropriate
-	 * to use the AHG feature.
-	 */
-	priv->s_ahg->tx_flags = 0;
-	priv->s_ahg->ahgcount = 0;
-	priv->s_ahg->ahgidx = 0;
+
 	if (qp->s_mig_state == IB_MIG_MIGRATED)
 		bth0 |= IB_BTH_MIG_REQ;
 	else
 		middle = 0;
+
 	if (middle)
 		build_ahg(qp, bth2);
 	else
 		qp->s_flags &= ~RVT_S_AHG_VALID;
-	ps->s_txreq->phdr.hdr.ibh.lrh[0] = cpu_to_be16(lrh0);
-	ps->s_txreq->phdr.hdr.ibh.lrh[1] =
-		cpu_to_be16(rdma_ah_get_dlid(&qp->remote_ah_attr));
-	ps->s_txreq->phdr.hdr.ibh.lrh[2] =
-		cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
-	ps->s_txreq->phdr.hdr.ibh.lrh[3] =
-		cpu_to_be16(ppd_from_ibp(ibp)->lid |
-		rdma_ah_get_path_bits(&qp->remote_ah_attr));
-	bth0 |= hfi1_get_pkey(ibp, qp->s_pkey_index);
+
+	bth0 |= pkey;
 	bth0 |= extra_bytes << 20;
-	ohdr->bth[0] = cpu_to_be32(bth0);
-	bth1 = qp->remote_qpn;
 	if (qp->s_flags & RVT_S_ECN) {
 		qp->s_flags &= ~RVT_S_ECN;
 		/* we recently received a FECN, so return a BECN */
 		bth1 |= (IB_BECN_MASK << IB_BECN_SHIFT);
 	}
-	ohdr->bth[1] = cpu_to_be32(bth1);
-	ohdr->bth[2] = cpu_to_be32(bth2);
+	hfi1_make_ruc_bth(qp, ohdr, bth0, bth1, bth2);
+
+	if (!ppd->lid)
+		slid = be16_to_cpu(IB_LID_PERMISSIVE);
+	else
+		slid = ppd->lid |
+			(rdma_ah_get_path_bits(&qp->remote_ah_attr) &
+			((1 << ppd->lmc) - 1));
+	hfi1_make_ib_hdr(&ps->s_txreq->phdr.hdr.ibh,
+			 lrh0,
+			 qp->s_hdrwords + nwords,
+			 opa_get_lid(rdma_ah_get_dlid(&qp->remote_ah_attr), 9B),
+			 ppd_from_ibp(ibp)->lid |
+				rdma_ah_get_path_bits(&qp->remote_ah_attr));
+}
+
+typedef void (*hfi1_make_ruc_hdr)(struct rvt_qp *qp,
+				  struct ib_other_headers *ohdr,
+				  u32 bth0, u32 bth2, int middle,
+				  struct hfi1_pkt_state *ps);
+
+/* We support only two types - 9B and 16B for now */
+static const hfi1_make_ruc_hdr hfi1_ruc_header_tbl[2] = {
+	[HFI1_PKT_TYPE_9B] = &hfi1_make_ruc_header_9B,
+	[HFI1_PKT_TYPE_16B] = &hfi1_make_ruc_header_16B
+};
+
+void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
+			  u32 bth0, u32 bth2, int middle,
+			  struct hfi1_pkt_state *ps)
+{
+	struct hfi1_qp_priv *priv = qp->priv;
+
+	/*
+	 * reset s_ahg/AHG fields
+	 *
+	 * This insures that the ahgentry/ahgcount
+	 * are at a non-AHG default to protect
+	 * build_verbs_tx_desc() from using
+	 * an include ahgidx.
+	 *
+	 * build_ahg() will modify as appropriate
+	 * to use the AHG feature.
+	 */
+	priv->s_ahg->tx_flags = 0;
+	priv->s_ahg->ahgcount = 0;
+	priv->s_ahg->ahgidx = 0;
+
+	/* Make the appropriate header */
+	hfi1_ruc_header_tbl[priv->hdr_type](qp, ohdr, bth0, bth2, middle, ps);
 }
 
 /* when sending, force a reschedule every one of these periods */

drivers/infiniband/hw/hfi1/uc.c

@@ -65,7 +65,7 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
 	struct hfi1_qp_priv *priv = qp->priv;
 	struct ib_other_headers *ohdr;
 	struct rvt_swqe *wqe;
-	u32 hwords = 5;
+	u32 hwords;
 	u32 bth0 = 0;
 	u32 len;
 	u32 pmtu = qp->pmtu;
@@ -93,9 +93,23 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
 		goto done_free_tx;
 	}
 
-	ohdr = &ps->s_txreq->phdr.hdr.ibh.u.oth;
-	if (rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)
-		ohdr = &ps->s_txreq->phdr.hdr.ibh.u.l.oth;
+	ps->s_txreq->phdr.hdr.hdr_type = priv->hdr_type;
+	if (priv->hdr_type == HFI1_PKT_TYPE_9B) {
+		/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+		hwords = 5;
+		if (rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)
+			ohdr = &ps->s_txreq->phdr.hdr.ibh.u.l.oth;
+		else
+			ohdr = &ps->s_txreq->phdr.hdr.ibh.u.oth;
+	} else {
+		/* header size in 32-bit words 16B LRH+BTH = (16+12)/4. */
+		hwords = 7;
+		if ((rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH) &&
+		    (hfi1_check_mcast(rdma_ah_get_dlid(&qp->remote_ah_attr))))
+			ohdr = &ps->s_txreq->phdr.hdr.opah.u.l.oth;
+		else
+			ohdr = &ps->s_txreq->phdr.hdr.opah.u.oth;
+	}
 
 	/* Get the next send request. */
 	wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
@@ -309,6 +323,7 @@ void hfi1_uc_rcv(struct hfi1_packet *packet)
 	u32 pmtu = qp->pmtu;
 	struct ib_reth *reth;
 	int ret;
+	u8 extra_bytes = pad + packet->extra_byte + (SIZE_OF_CRC << 2);
 
 	if (hfi1_ruc_check_hdr(ibp, packet))
 		return;
@@ -408,7 +423,12 @@ send_first:
 		/* FALLTHROUGH */
 	case OP(SEND_MIDDLE):
 		/* Check for invalid length PMTU or posted rwqe len. */
-		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+		/*
+		 * There will be no padding for 9B packet but 16B packets
+		 * will come in with some padding since we always add
+		 * CRC and LT bytes which will need to be flit aligned
+		 */
+		if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
 			goto rewind;
 		qp->r_rcv_len += pmtu;
 		if (unlikely(qp->r_rcv_len > qp->r_len))
@@ -428,10 +448,10 @@ no_immediate_data:
 send_last:
 		/* Check for invalid length. */
 		/* LAST len should be >= 1 */
-		if (unlikely(tlen < (hdrsize + pad + 4)))
+		if (unlikely(tlen < (hdrsize + extra_bytes)))
 			goto rewind;
 		/* Don't count the CRC. */
-		tlen -= (hdrsize + pad + 4);
+		tlen -= (hdrsize + extra_bytes);
 		wc.byte_len = tlen + qp->r_rcv_len;
 		if (unlikely(wc.byte_len > qp->r_len))
 			goto rewind;
@@ -524,7 +544,7 @@ rdma_last_imm:
 		if (unlikely(tlen < (hdrsize + pad + 4)))
 			goto drop;
 		/* Don't count the CRC. */
-		tlen -= (hdrsize + pad + 4);
+		tlen -= (hdrsize + extra_bytes);
 		if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
 			goto drop;
 		if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags)) {
@@ -544,14 +564,12 @@ rdma_last_imm:
 
 	case OP(RDMA_WRITE_LAST):
 rdma_last:
-		/* Get the number of bytes the message was padded by. */
-		pad = ib_bth_get_pad(ohdr);
 		/* Check for invalid length. */
 		/* LAST len should be >= 1 */
 		if (unlikely(tlen < (hdrsize + pad + 4)))
 			goto drop;
 		/* Don't count the CRC. */
-		tlen -= (hdrsize + pad + 4);
+		tlen -= (hdrsize + extra_bytes);
 		if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
 			goto drop;
 		hfi1_copy_sge(&qp->r_sge, data, tlen, true, false);

drivers/infiniband/hw/hfi1/verbs.h

@@ -373,7 +373,8 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread);
 void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
 			enum ib_wc_status status);
 
-void hfi1_send_rc_ack(struct hfi1_ctxtdata *, struct rvt_qp *qp, int is_fecn);
+void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp,
+		      bool is_fecn);
 
 int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);