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/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 @@ void hfi1_uc_rcv(struct hfi1_packet *packet)
 		/* 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 @@ void hfi1_uc_rcv(struct hfi1_packet *packet)
 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 @@ void hfi1_uc_rcv(struct hfi1_packet *packet)
 		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 @@ void hfi1_uc_rcv(struct hfi1_packet *packet)
 
 	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);