prov/efa: Make num_runt_bytes_in_flight per domain

This counter is used to prevent using runting read when the outstanding send bytes is reaching a limit that makes EFA NIC busy. In that case runting read would be less performant than a read. Signed-off-by: Shi Jin <sjina@amazon.com>
ofiwg · Apr 2, 2024 · a1d8aeb · a1d8aeb
1 parent f8bdd09
commit a1d8aeb
Show file tree

Hide file tree

Showing 12 changed files with 139 additions and 147 deletions.
diff --git a/prov/efa/src/efa_domain.c b/prov/efa/src/efa_domain.c
@@ -160,6 +160,7 @@ static int efa_domain_init_rdm(struct efa_domain *efa_domain, struct fi_info *in
 	efa_domain->rdm_cq_size = MAX(info->rx_attr->size + info->tx_attr->size,
 				  efa_env.cq_size);
 	efa_domain->num_read_msg_in_flight = 0;
+	efa_domain->num_runt_bytes_in_flight = 0;
 	return 0;
 }
 

diff --git a/prov/efa/src/efa_domain.h b/prov/efa/src/efa_domain.h
@@ -30,7 +30,15 @@ struct efa_domain {
 	size_t			rdm_cq_size;
 	struct dlist_entry	list_entry; /* linked to g_efa_domain_list */
 	struct ofi_genlock	srx_lock; /* shared among peer providers */
+	/**
+	 * @brief number of messages that are using read based protocol
+	 */
 	uint64_t		num_read_msg_in_flight;
+	/**
+	 * @brief number of bytes that has been sent as part of runting protocols
+	 * @details this value is capped by efa_env.efa_runt_size
+	 */
+	int64_t 		num_runt_bytes_in_flight;
 };
 
 extern struct dlist_entry g_efa_domain_list;

diff --git a/prov/efa/src/rdm/efa_rdm_ep.h b/prov/efa/src/rdm/efa_rdm_ep.h
@@ -271,6 +271,10 @@ ssize_t efa_rdm_ep_post_queued_pkts(struct efa_rdm_ep *ep,
 
 size_t efa_rdm_ep_get_memory_alignment(struct efa_rdm_ep *ep, enum fi_hmem_iface iface);
 
+size_t efa_rdm_ep_get_runt_size(struct efa_rdm_ep *ep, struct efa_rdm_ope *ope);
+
+int efa_rdm_ep_select_readbase_rtm(struct efa_rdm_ep *ep, struct efa_rdm_ope *ope);
+
 static inline
 struct efa_domain *efa_rdm_ep_domain(struct efa_rdm_ep *ep)
 {

diff --git a/prov/efa/src/rdm/efa_rdm_ep_utils.c b/prov/efa/src/rdm/efa_rdm_ep_utils.c
@@ -679,3 +679,59 @@ size_t efa_rdm_ep_get_memory_alignment(struct efa_rdm_ep *ep, enum fi_hmem_iface
 	return memory_alignment;
 }
 
+/**
+ * @brief Get the runt size for a given ep and ope
+ *
+ * @param ep efa rdm ep
+ * @param ope efa rdm ope
+ * @return size_t the number of bytes that can be runt
+ */
+size_t efa_rdm_ep_get_runt_size(struct efa_rdm_ep *ep, struct efa_rdm_ope *ope)
+{
+	struct efa_domain *domain = efa_rdm_ep_domain(ep);
+	struct efa_hmem_info *hmem_info;
+	size_t runt_size;
+	size_t memory_alignment;
+	int iface;
+
+	hmem_info = domain->hmem_info;
+	iface = ope->desc[0] ? ((struct efa_mr*) ope->desc[0])->peer.iface : FI_HMEM_SYSTEM;
+
+	if (domain->hmem_info[iface].runt_size < domain->num_runt_bytes_in_flight)
+		return 0;
+
+	runt_size = MIN(hmem_info[iface].runt_size - domain->num_runt_bytes_in_flight, ope->total_len);
+	memory_alignment = efa_rdm_ep_get_memory_alignment(ep, iface);
+	/*
+	 * runt size must be aligned because:
+	 * 1. For LL128 protocol, the size to be copied on the receiver side must be 128-multiple,
+	 * 128 is the alignment in this case.
+	 * 2. For non-LL128 protocol, using aligned runt size has optimal performance for data copy.
+	 * Note the returned value can be 0. In that case we will not use runting read protocol.
+	 */
+	return (runt_size & ~(memory_alignment - 1));
+}
+
+/**
+ * @brief Determine which Read based protocol to use
+ *
+ * @param[in] ep		efa rdm ep
+ * @param[in] efa_rdm_ope	efa rdm ope
+ * @return The read-based protocol to use based on inputs.
+ */
+int efa_rdm_ep_select_readbase_rtm(struct efa_rdm_ep *ep, struct efa_rdm_ope *ope)
+{
+	int op = ope->op;
+
+	assert(op == ofi_op_tagged || op == ofi_op_msg);
+
+	if (efa_rdm_ep_domain(ep)->num_read_msg_in_flight == 0 &&
+	    efa_rdm_ep_get_runt_size(ep, ope) > 0 &&
+	    !(ope->fi_flags & FI_DELIVERY_COMPLETE)) {
+		return (op == ofi_op_tagged) ? EFA_RDM_RUNTREAD_TAGRTM_PKT
+					     : EFA_RDM_RUNTREAD_MSGRTM_PKT;
+	} else {
+		return (op == ofi_op_tagged) ? EFA_RDM_LONGREAD_TAGRTM_PKT
+					     : EFA_RDM_LONGREAD_MSGRTM_PKT;
+	}
+}
diff --git a/prov/efa/src/rdm/efa_rdm_msg.c b/prov/efa/src/rdm/efa_rdm_msg.c
@@ -86,7 +86,7 @@ int efa_rdm_msg_select_rtm(struct efa_rdm_ep *efa_rdm_ep, struct efa_rdm_ope *tx
 
 	eager_rtm_max_data_size = efa_rdm_txe_max_req_data_capacity(efa_rdm_ep, txe, eager_rtm);
 
-	readbase_rtm = efa_rdm_peer_select_readbase_rtm(txe->peer, efa_rdm_ep, txe);
+	readbase_rtm = efa_rdm_ep_select_readbase_rtm(efa_rdm_ep, txe);
 
 	if (txe->total_len >= hmem_info[iface].min_read_msg_size &&
 		efa_rdm_interop_rdma_read(efa_rdm_ep, txe->peer) &&

diff --git a/prov/efa/src/rdm/efa_rdm_ope.c b/prov/efa/src/rdm/efa_rdm_ope.c
@@ -336,7 +336,7 @@ void efa_rdm_txe_set_runt_size(struct efa_rdm_ep *ep, struct efa_rdm_ope *txe)
 	peer = efa_rdm_ep_get_peer(ep, txe->addr);
 
 	assert(peer);
-	txe->bytes_runt = efa_rdm_peer_get_runt_size(peer, ep, txe);
+	txe->bytes_runt = efa_rdm_ep_get_runt_size(ep, txe);
 
 	assert(txe->bytes_runt);
 }

diff --git a/prov/efa/src/rdm/efa_rdm_peer.c b/prov/efa/src/rdm/efa_rdm_peer.c
@@ -24,7 +24,6 @@ void efa_rdm_peer_construct(struct efa_rdm_peer *peer, struct efa_rdm_ep *ep, st
 	peer->efa_fiaddr = conn->fi_addr;
 	peer->is_self = efa_is_same_addr(&ep->base_ep.src_addr, conn->ep_addr);
 	peer->host_id = peer->is_self ? ep->host_id : 0;	/* Peer host id is exchanged via handshake */
-	peer->num_runt_bytes_in_flight = 0;
 	ofi_recvwin_buf_alloc(&peer->robuf, efa_env.recvwin_size);
 	dlist_init(&peer->outstanding_tx_pkts);
 	dlist_init(&peer->txe_list);
@@ -223,62 +222,3 @@ void efa_rdm_peer_proc_pending_items_in_robuf(struct efa_rdm_peer *peer, struct
 	return;
 }
 
-/**
- * @brief Get the runt size for a given peer and ope
- *
- * @param peer rdm peer
- * @param ep efa rdm ep
- * @param ope efa rdm ope
- * @return size_t the number of bytes that can be runt
- */
-size_t efa_rdm_peer_get_runt_size(struct efa_rdm_peer *peer,
-				  struct efa_rdm_ep *ep, struct efa_rdm_ope *ope)
-{
-	struct efa_hmem_info *hmem_info;
-	size_t runt_size;
-	size_t memory_alignment;
-	int iface;
-
-	hmem_info = efa_rdm_ep_domain(ep)->hmem_info;
-	iface = ope->desc[0] ? ((struct efa_mr*) ope->desc[0])->peer.iface : FI_HMEM_SYSTEM;
-
-	if (hmem_info[iface].runt_size < peer->num_runt_bytes_in_flight)
-		return 0;
-
-	runt_size = MIN(hmem_info[iface].runt_size - peer->num_runt_bytes_in_flight, ope->total_len);
-	memory_alignment = efa_rdm_ep_get_memory_alignment(ep, iface);
-	/*
-	 * runt size must be aligned because:
-	 * 1. For LL128 protocol, the size to be copied on the receiver side must be 128-multiple,
-	 * 128 is the alignment in this case.
-	 * 2. For non-LL128 protocol, using aligned runt size has optimal performance for data copy.
-	 * Note the returned value can be 0. In that case we will not use runting read protocol.
-	 */
-	return (runt_size & ~(memory_alignment - 1));
-}
-
-/**
- * @brief Determine which Read based protocol to use for a given peer
- *
- * @param[in] peer		rdm peer
- * @param[in] ep		efa rdm ep
- * @param[in] efa_rdm_ope	efa rdm ope
- * @return The read-based protocol to use based on inputs.
- */
-int efa_rdm_peer_select_readbase_rtm(struct efa_rdm_peer *peer,
-				     struct efa_rdm_ep *ep, struct efa_rdm_ope *ope)
-{
-	int op = ope->op;
-
-	assert(op == ofi_op_tagged || op == ofi_op_msg);
-
-	if (efa_rdm_ep_domain(ep)->num_read_msg_in_flight == 0 &&
-	    efa_rdm_peer_get_runt_size(peer, ep, ope) > 0 &&
-	    !(ope->fi_flags & FI_DELIVERY_COMPLETE)) {
-		return (op == ofi_op_tagged) ? EFA_RDM_RUNTREAD_TAGRTM_PKT
-					     : EFA_RDM_RUNTREAD_MSGRTM_PKT;
-	} else {
-		return (op == ofi_op_tagged) ? EFA_RDM_LONGREAD_TAGRTM_PKT
-					     : EFA_RDM_LONGREAD_MSGRTM_PKT;
-	}
-}
diff --git a/prov/efa/src/rdm/efa_rdm_peer.h b/prov/efa/src/rdm/efa_rdm_peer.h
@@ -52,12 +52,6 @@ struct efa_rdm_peer {
 	struct dlist_entry rx_unexp_tagged_list; /**< a list of unexpected tagged rxe for this peer */
 	struct dlist_entry txe_list; /**< a list of txe related to this peer */
 	struct dlist_entry rxe_list; /**< a list of rxe relased to this peer */
-
-	/**
-	 * @brief number of bytes that has been sent as part of runting protocols
-	 * @details this value is capped by efa_env.efa_runt_size
-	 */
-	int64_t num_runt_bytes_in_flight;
 };
 
 /**
@@ -248,8 +242,4 @@ int efa_rdm_peer_reorder_msg(struct efa_rdm_peer *peer, struct efa_rdm_ep *ep, s
 
 void efa_rdm_peer_proc_pending_items_in_robuf(struct efa_rdm_peer *peer, struct efa_rdm_ep *ep);
 
-size_t efa_rdm_peer_get_runt_size(struct efa_rdm_peer *peer, struct efa_rdm_ep *ep, struct efa_rdm_ope *ope);
-
-int efa_rdm_peer_select_readbase_rtm(struct efa_rdm_peer *peer, struct efa_rdm_ep *ep, struct efa_rdm_ope *ope);
-
 #endif /* EFA_RDM_PEER_H */
diff --git a/prov/efa/src/rdm/efa_rdm_pke_rtm.c b/prov/efa/src/rdm/efa_rdm_pke_rtm.c
@@ -1349,7 +1349,7 @@ void efa_rdm_pke_handle_runtread_rtm_sent(struct efa_rdm_pke *pkt_entry)
 
 	txe = pkt_entry->ope;
 	txe->bytes_sent += pkt_data_size;
-	peer->num_runt_bytes_in_flight += pkt_data_size;
+	efa_rdm_ep_domain(ep)->num_runt_bytes_in_flight += pkt_data_size;
 
 	if (efa_rdm_pke_get_runtread_rtm_base_hdr(pkt_entry)->seg_offset == 0 &&
 	    txe->total_len > txe->bytes_runt)
@@ -1368,18 +1368,15 @@ void efa_rdm_pke_handle_runtread_rtm_send_completion(struct efa_rdm_pke *pkt_ent
 {
 	struct efa_rdm_ep *ep;
 	struct efa_rdm_ope *txe;
-	struct efa_rdm_peer *peer;
 	size_t pkt_data_size;
 
 	ep = pkt_entry->ep;
 	txe = pkt_entry->ope;
 	pkt_data_size = pkt_entry->payload_size;
 	txe->bytes_acked += pkt_data_size;
 
-	peer = efa_rdm_ep_get_peer(ep, pkt_entry->addr);
-	assert(peer);
-	assert(peer->num_runt_bytes_in_flight >= pkt_data_size);
-	peer->num_runt_bytes_in_flight -= pkt_data_size;
+	assert(efa_rdm_ep_domain(ep)->num_runt_bytes_in_flight >= pkt_data_size);
+	efa_rdm_ep_domain(ep)->num_runt_bytes_in_flight -= pkt_data_size;
 	if (txe->total_len == txe->bytes_acked)
 		efa_rdm_ope_handle_send_completed(txe);
 }