Merge HFI1 updates into k.o/for-next

Based on rdma.git for-rc for dependencies.

From Dennis Dalessandro:

====================

Here are some code improvement patches and fixes for less serious bugs to
TID RDMA than we sent for RC.

====================

* HFI1 updates:
  IB/hfi1: Implement CCA for TID RDMA protocol
  IB/hfi1: Remove WARN_ON when freeing expected receive groups
  IB/hfi1: Unify the software PSN check for TID RDMA READ/WRITE
  IB/hfi1: Add a function to read next expected psn from hardware flow
  IB/hfi1: Delay the release of destination mr for TID RDMA WRITE DATA
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

drivers/infiniband/hw/hfi1/chip.c

@@ -13232,7 +13232,7 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 	int total_contexts;
 	int ret;
 	unsigned ngroups;
-	int qos_rmt_count;
+	int rmt_count;
 	int user_rmt_reduced;
 	u32 n_usr_ctxts;
 	u32 send_contexts = chip_send_contexts(dd);
@@ -13294,10 +13294,23 @@ static int set_up_context_variables(struct hfi1_devdata *dd)
 		n_usr_ctxts = rcv_contexts - total_contexts;
 	}
 
-	/* each user context requires an entry in the RMT */
-	qos_rmt_count = qos_rmt_entries(dd, NULL, NULL);
-	if (qos_rmt_count + n_usr_ctxts > NUM_MAP_ENTRIES) {
-		user_rmt_reduced = NUM_MAP_ENTRIES - qos_rmt_count;
+	/*
+	 * The RMT entries are currently allocated as shown below:
+	 * 1. QOS (0 to 128 entries);
+	 * 2. FECN (num_kernel_context - 1 + num_user_contexts +
+	 *    num_vnic_contexts);
+	 * 3. VNIC (num_vnic_contexts).
+	 * It should be noted that FECN oversubscribe num_vnic_contexts
+	 * entries of RMT because both VNIC and PSM could allocate any receive
+	 * context between dd->first_dyn_alloc_text and dd->num_rcv_contexts,
+	 * and PSM FECN must reserve an RMT entry for each possible PSM receive
+	 * context.
+	 */
+	rmt_count = qos_rmt_entries(dd, NULL, NULL) + (num_vnic_contexts * 2);
+	if (HFI1_CAP_IS_KSET(TID_RDMA))
+		rmt_count += num_kernel_contexts - 1;
+	if (rmt_count + n_usr_ctxts > NUM_MAP_ENTRIES) {
+		user_rmt_reduced = NUM_MAP_ENTRIES - rmt_count;
 		dd_dev_err(dd,
 			   "RMT size is reducing the number of user receive contexts from %u to %d\n",
 			   n_usr_ctxts,
@@ -14278,35 +14291,43 @@ static void init_qos(struct hfi1_devdata *dd, struct rsm_map_table *rmt)
 	init_qpmap_table(dd, FIRST_KERNEL_KCTXT, dd->n_krcv_queues - 1);
 }
 
-static void init_user_fecn_handling(struct hfi1_devdata *dd,
-				    struct rsm_map_table *rmt)
+static void init_fecn_handling(struct hfi1_devdata *dd,
+			       struct rsm_map_table *rmt)
 {
 	struct rsm_rule_data rrd;
 	u64 reg;
-	int i, idx, regoff, regidx;
+	int i, idx, regoff, regidx, start;
 	u8 offset;
+	u32 total_cnt;
+
+	if (HFI1_CAP_IS_KSET(TID_RDMA))
+		/* Exclude context 0 */
+		start = 1;
+	else
+		start = dd->first_dyn_alloc_ctxt;
+
+	total_cnt = dd->num_rcv_contexts - start;
 
 	/* there needs to be enough room in the map table */
-	if (rmt->used + dd->num_user_contexts >= NUM_MAP_ENTRIES) {
-		dd_dev_err(dd, "User FECN handling disabled - too many user contexts allocated\n");
+	if (rmt->used + total_cnt >= NUM_MAP_ENTRIES) {
+		dd_dev_err(dd, "FECN handling disabled - too many contexts allocated\n");
 		return;
 	}
 
 	/*
 	 * RSM will extract the destination context as an index into the
 	 * map table.  The destination contexts are a sequential block
-	 * in the range first_dyn_alloc_ctxt...num_rcv_contexts-1 (inclusive).
+	 * in the range start...num_rcv_contexts-1 (inclusive).
 	 * Map entries are accessed as offset + extracted value.  Adjust
 	 * the added offset so this sequence can be placed anywhere in
 	 * the table - as long as the entries themselves do not wrap.
 	 * There are only enough bits in offset for the table size, so
 	 * start with that to allow for a "negative" offset.
 	 */
-	offset = (u8)(NUM_MAP_ENTRIES + (int)rmt->used -
-						(int)dd->first_dyn_alloc_ctxt);
+	offset = (u8)(NUM_MAP_ENTRIES + rmt->used - start);
 
-	for (i = dd->first_dyn_alloc_ctxt, idx = rmt->used;
-				i < dd->num_rcv_contexts; i++, idx++) {
+	for (i = start, idx = rmt->used; i < dd->num_rcv_contexts;
+	     i++, idx++) {
 		/* replace with identity mapping */
 		regoff = (idx % 8) * 8;
 		regidx = idx / 8;
@@ -14341,7 +14362,7 @@ static void init_user_fecn_handling(struct hfi1_devdata *dd,
 	/* add rule 1 */
 	add_rsm_rule(dd, RSM_INS_FECN, &rrd);
 
-	rmt->used += dd->num_user_contexts;
+	rmt->used += total_cnt;
 }
 
 /* Initialize RSM for VNIC */
@@ -14428,7 +14449,7 @@ static void init_rxe(struct hfi1_devdata *dd)
 	rmt = alloc_rsm_map_table(dd);
 	/* set up QOS, including the QPN map table */
 	init_qos(dd, rmt);
-	init_user_fecn_handling(dd, rmt);
+	init_fecn_handling(dd, rmt);
 	complete_rsm_map_table(dd, rmt);
 	/* record number of used rsm map entries for vnic */
 	dd->vnic.rmt_start = rmt->used;

drivers/infiniband/hw/hfi1/driver.c

@@ -514,7 +514,9 @@ bool hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
 	 */
 	do_cnp = prescan ||
 		(opcode >= IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST &&
-		 opcode <= IB_OPCODE_RC_ATOMIC_ACKNOWLEDGE);
+		 opcode <= IB_OPCODE_RC_ATOMIC_ACKNOWLEDGE) ||
+		opcode == TID_OP(READ_RESP) ||
+		opcode == TID_OP(ACK);
 
 	/* Call appropriate CNP handler */
 	if (!ignore_fecn && do_cnp && fecn)

drivers/infiniband/hw/hfi1/exp_rcv.c

@@ -112,9 +112,6 @@ int hfi1_alloc_ctxt_rcv_groups(struct hfi1_ctxtdata *rcd)
  */
 void hfi1_free_ctxt_rcv_groups(struct hfi1_ctxtdata *rcd)
 {
-	WARN_ON(!EXP_TID_SET_EMPTY(rcd->tid_full_list));
-	WARN_ON(!EXP_TID_SET_EMPTY(rcd->tid_used_list));
-
 	kfree(rcd->groups);
 	rcd->groups = NULL;
 	hfi1_exp_tid_group_init(rcd);

drivers/infiniband/hw/hfi1/qp.c

@@ -900,7 +900,9 @@ void notify_error_qp(struct rvt_qp *qp)
 		if (!list_empty(&priv->s_iowait.list) &&
 		    !(qp->s_flags & RVT_S_BUSY) &&
 		    !(priv->s_flags & RVT_S_BUSY)) {
-			qp->s_flags &= ~RVT_S_ANY_WAIT_IO;
+			qp->s_flags &= ~HFI1_S_ANY_WAIT_IO;
+			iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_IB);
+			iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
 			list_del_init(&priv->s_iowait.list);
 			priv->s_iowait.lock = NULL;
 			rvt_put_qp(qp);

drivers/infiniband/hw/hfi1/rc.c

@@ -140,10 +140,7 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp,
 	case OP(RDMA_READ_RESPONSE_LAST):
 	case OP(RDMA_READ_RESPONSE_ONLY):
 		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		/* FALLTHROUGH */
 	case OP(ATOMIC_ACKNOWLEDGE):
 		/*
@@ -343,7 +340,8 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp,
 			break;
 
 		e->sent = 1;
-		qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+		/* Do not free e->rdma_sge until all data are received */
+		qp->s_ack_state = OP(ATOMIC_ACKNOWLEDGE);
 		break;
 
 	case TID_OP(READ_RESP):
@@ -2643,10 +2641,7 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data,
 		len = be32_to_cpu(reth->length);
 		if (unlikely(offset + len != e->rdma_sge.sge_length))
 			goto unlock_done;
-		if (e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		if (len != 0) {
 			u32 rkey = be32_to_cpu(reth->rkey);
 			u64 vaddr = get_ib_reth_vaddr(reth);
@@ -3088,10 +3083,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 			update_ack_queue(qp, next);
 		}
 		e = &qp->s_ack_queue[qp->r_head_ack_queue];
-		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		reth = &ohdr->u.rc.reth;
 		len = be32_to_cpu(reth->length);
 		if (len) {
@@ -3166,10 +3158,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 			update_ack_queue(qp, next);
 		}
 		e = &qp->s_ack_queue[qp->r_head_ack_queue];
-		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
-			rvt_put_mr(e->rdma_sge.mr);
-			e->rdma_sge.mr = NULL;
-		}
+		release_rdma_sge_mr(e);
 		/* Process OPFN special virtual address */
 		if (opfn) {
 			opfn_conn_response(qp, e, ateth);

drivers/infiniband/hw/hfi1/rc.h

@@ -41,6 +41,14 @@ static inline u32 restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe,
 	return rvt_restart_sge(ss, wqe, len);
 }
 
+static inline void release_rdma_sge_mr(struct rvt_ack_entry *e)
+{
+	if (e->rdma_sge.mr) {
+		rvt_put_mr(e->rdma_sge.mr);
+		e->rdma_sge.mr = NULL;
+	}
+}
+
 struct rvt_ack_entry *find_prev_entry(struct rvt_qp *qp, u32 psn, u8 *prev,
 				      u8 *prev_ack, bool *scheduled);
 int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, u64 val,

drivers/infiniband/hw/hfi1/tid_rdma.h

@@ -76,10 +76,8 @@ struct tid_rdma_qp_params {
 struct tid_flow_state {
 	u32 generation;
 	u32 psn;
-	u32 r_next_psn;      /* next PSN to be received (in TID space) */
 	u8 index;
 	u8 last_index;
-	u8 flags;
 };
 
 enum tid_rdma_req_state {

drivers/infiniband/hw/hfi1/trace_tid.h

@@ -53,7 +53,7 @@ u16 hfi1_trace_get_tid_idx(u32 ent);
 			    "tid_r_comp %u pending_tid_r_segs %u " \
 			    "s_flags 0x%x ps_flags 0x%x iow_flags 0x%lx " \
 			    "s_state 0x%x hw_flow_index %u generation 0x%x " \
-			    "fpsn 0x%x flow_flags 0x%x"
+			    "fpsn 0x%x"
 
 #define TID_REQ_PRN "[%s] qpn 0x%x newreq %u opcode 0x%x psn 0x%x lpsn 0x%x " \
 		    "cur_seg %u comp_seg %u ack_seg %u alloc_seg %u " \
@@ -71,7 +71,7 @@ u16 hfi1_trace_get_tid_idx(u32 ent);
 			    "pending_tid_w_segs %u sync_pt %s " \
 			    "ps_nak_psn 0x%x ps_nak_state 0x%x " \
 			    "prnr_nak_state 0x%x hw_flow_index %u generation "\
-			    "0x%x fpsn 0x%x flow_flags 0x%x resync %s" \
+			    "0x%x fpsn 0x%x resync %s" \
 			    "r_next_psn_kdeth 0x%x"
 
 #define TID_WRITE_SENDER_PRN "[%s] qpn 0x%x newreq %u s_tid_cur %u " \
@@ -973,7 +973,6 @@ DECLARE_EVENT_CLASS(/* tid_read_sender */
 		__field(u32, hw_flow_index)
 		__field(u32, generation)
 		__field(u32, fpsn)
-		__field(u32, flow_flags)
 	),
 	TP_fast_assign(/* assign */
 		struct hfi1_qp_priv *priv = qp->priv;
@@ -991,7 +990,6 @@ DECLARE_EVENT_CLASS(/* tid_read_sender */
 		__entry->hw_flow_index = priv->flow_state.index;
 		__entry->generation = priv->flow_state.generation;
 		__entry->fpsn = priv->flow_state.psn;
-		__entry->flow_flags = priv->flow_state.flags;
 	),
 	TP_printk(/* print */
 		TID_READ_SENDER_PRN,
@@ -1007,8 +1005,7 @@ DECLARE_EVENT_CLASS(/* tid_read_sender */
 		__entry->s_state,
 		__entry->hw_flow_index,
 		__entry->generation,
-		__entry->fpsn,
-		__entry->flow_flags
+		__entry->fpsn
 	)
 );
 
@@ -1338,7 +1335,6 @@ DECLARE_EVENT_CLASS(/* tid_write_sp */
 		__field(u32, hw_flow_index)
 		__field(u32, generation)
 		__field(u32, fpsn)
-		__field(u32, flow_flags)
 		__field(bool, resync)
 		__field(u32, r_next_psn_kdeth)
 	),
@@ -1360,7 +1356,6 @@ DECLARE_EVENT_CLASS(/* tid_write_sp */
 		__entry->hw_flow_index = priv->flow_state.index;
 		__entry->generation = priv->flow_state.generation;
 		__entry->fpsn = priv->flow_state.psn;
-		__entry->flow_flags = priv->flow_state.flags;
 		__entry->resync = priv->resync;
 		__entry->r_next_psn_kdeth = priv->r_next_psn_kdeth;
 	),
@@ -1381,7 +1376,6 @@ DECLARE_EVENT_CLASS(/* tid_write_sp */
 		__entry->hw_flow_index,
 		__entry->generation,
 		__entry->fpsn,
-		__entry->flow_flags,
 		__entry->resync ? "yes" : "no",
 		__entry->r_next_psn_kdeth
 	)

drivers/infiniband/hw/mlx5/odp.c

@@ -585,7 +585,7 @@ static int pagefault_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr,
 	struct ib_umem_odp *odp_mr = to_ib_umem_odp(mr->umem);
 	bool downgrade = flags & MLX5_PF_FLAGS_DOWNGRADE;
 	bool prefetch = flags & MLX5_PF_FLAGS_PREFETCH;
-	u64 access_mask = ODP_READ_ALLOWED_BIT;
+	u64 access_mask;
 	u64 start_idx, page_mask;
 	struct ib_umem_odp *odp;
 	size_t size;
@@ -607,6 +607,7 @@ static int pagefault_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr,
 	page_shift = mr->umem->page_shift;
 	page_mask = ~(BIT(page_shift) - 1);
 	start_idx = (io_virt - (mr->mmkey.iova & page_mask)) >> page_shift;
+	access_mask = ODP_READ_ALLOWED_BIT;
 
 	if (prefetch && !downgrade && !mr->umem->writable) {
 		/* prefetch with write-access must