prov/opx: 0 byte write_data Error

lsavers · j-xiong · commit 110cdbe99d06 · 2025-10-09T13:49:12.000-07:00
0 byte write with CQ data was not posting because there is no data to send after the RTS/CTS exchange. Updated to post the CQ data when receiving the RTS and the sender completes when receiving the CTS.

Other minor changes:
Tail was not set properly in fi_opx_ep_rx_process_header()
Added the endpoint pointer to the flight recorder dump
Added opx_debug_ep_list_dump to dump the endpoint list

Signed-off-by: Lindsay Reiser &lt;lindsay.reiser@cornelisnetworks.com&gt;
diff --git a/prov/opx/include/rdma/opx/fi_opx_endpoint.h b/prov/opx/include/rdma/opx/fi_opx_endpoint.h
@@ -2007,7 +2007,11 @@ void fi_opx_ep_rx_process_header_rma_rts(struct fi_opx_ep *opx_ep, const union o
 		fprintf(stderr, "%s:%s():%d\n", __FILE__, __func__, __LINE__);
 		abort();
 	}
-	slist_insert_tail((struct slist_entry *) context, rx->cq_pending_ptr);
+	if (context->len) {
+		slist_insert_tail((struct slist_entry *) context, rx->cq_pending_ptr);
+	} else { /* length is 0, there will be no RZV data sent, so post now */
+		slist_insert_tail((struct slist_entry *) context, rx->cq_completed_ptr);
+	}
 
 	OPX_TRACER_TRACE(OPX_TRACER_END_SUCCESS, "RECV-RMA-RTS");
 	FI_DBG_TRACE(fi_opx_global.prov, FI_LOG_EP_DATA,
@@ -2831,11 +2835,13 @@ static inline void fi_opx_ep_rx_process_header(struct fid_ep *ep, const union op
 			if (prev) {
 				prev->next = context->next;
 			} else {
+				assert(opx_ep->rx->queue[kind].mq.head == (struct slist_entry *) context);
 				opx_ep->rx->queue[kind].mq.head = (struct slist_entry *) context->next;
 			}
 
 			if (context->next == NULL) {
-				opx_ep->rx->queue[kind].mq.tail = NULL;
+				assert(opx_ep->rx->queue[kind].mq.tail == (struct slist_entry *) context);
+				opx_ep->rx->queue[kind].mq.tail = (struct slist_entry *) prev;
 			}
 
 			context->next = NULL;
diff --git a/prov/opx/include/rdma/opx/fi_opx_flight_recorder.h b/prov/opx/include/rdma/opx/fi_opx_flight_recorder.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021,2024 Cornelis Networks.
+ * Copyright (C) 2021,2025 Cornelis Networks.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -139,31 +139,34 @@ static inline void flight_recorder_dump_packet_payload(struct flight_recorder_en
 	}
 }
 
-static inline void flight_recorder_dump(struct flight_recorder *fr)
+static inline void flight_recorder_dump(struct fi_opx_ep *opx_ep, struct flight_recorder *fr)
 {
 	const unsigned	count = fr->count;
 	struct timespec current_time;
 	fi_opx_timer_now(fr->now, fr->timer);
 	if (!fr->timer->cycle_timer.use_cycle_timer) {
 		current_time = fr->now->tp;
 		fprintf(stderr,
+			"#FLIGHT_RECORDER t%d |EP: %p\n"
 			"#FLIGHT_RECORDER t%d |Last Dump Time: %ld.%ld\n"
 			"#FLIGHT_RECORDER t%d |Current   Time: %ld.%ld\n"
 			"#FLIGHT_RECORDER t%d |Entry Count   : %u\n",
-			fr->tid, fr->last_dump.tv_sec, fr->last_dump.tv_nsec, fr->tid, current_time.tv_sec,
-			current_time.tv_nsec, fr->tid, count);
+			fr->tid, opx_ep, fr->tid, fr->last_dump.tv_sec, fr->last_dump.tv_nsec, fr->tid,
+			current_time.tv_sec, current_time.tv_nsec, fr->tid, count);
 		fr->last_dump = current_time;
 		if (count == 0) {
 			fr->last_dump = current_time;
 			return;
 		}
 	} else {
 		fprintf(stderr,
+			"#FLIGHT_RECORDER t%d |EP: %p\n"
 			"#FLIGHT_RECORDER t%d |Last Dump Time: %0.9lf\n"
 			"#FLIGHT_RECORDER t%d |Current   Time: %0.9lf\n"
 			"#FLIGHT_RECORDER t%d |Entry Count   : %u\n",
-			fr->tid, fr->last_dump_cycles * fr->timer->cycle_timer.picos_per_cycle / 1e12, fr->tid,
-			fr->now->cycle_timer.cycles * fr->timer->cycle_timer.picos_per_cycle / 1e12, fr->tid, count);
+			fr->tid, opx_ep, fr->tid, fr->last_dump_cycles * fr->timer->cycle_timer.picos_per_cycle / 1e12,
+			fr->tid, fr->now->cycle_timer.cycles * fr->timer->cycle_timer.picos_per_cycle / 1e12, fr->tid,
+			count);
 		fr->last_dump_cycles = fr->now->cycle_timer.cycles;
 		if (count == 0) {
 			fr->last_dump_cycles = fr->now->cycle_timer.cycles;
@@ -199,33 +202,33 @@ static inline void flight_recorder_dump(struct flight_recorder *fr)
 	fr->count = 0;
 }
 
-#define FLIGHT_RECORDER_STRING(fr, event_id, format, ...)                                                              \
+#define FLIGHT_RECORDER_STRING(ep, fr, event_id, format, ...)                                                          \
 	{                                                                                                              \
 		struct flight_recorder_entry *next =                                                                   \
 			flight_recorder_init_next_entry((fr), (event_id), FR_ENTRY_TYPE_STRING);                       \
 		int actual_len = snprintf((char *) next->data, FLIGHT_RECORDER_ENTRY_DATA_LEN, format, ##__VA_ARGS__); \
 		int end_of_string	  = MIN(actual_len, FLIGHT_RECORDER_ENTRY_DATA_LEN - 1);                       \
 		next->data[end_of_string] = 0;                                                                         \
 		if ((fr)->count + 1 == FLIGHT_RECORDER_ENTRY_COUNT)                                                    \
-			flight_recorder_dump((fr));                                                                    \
+			flight_recorder_dump(ep, fr);                                                                  \
 	}
 
-#define FLIGHT_RECORDER_PACKET_HDR(fr, event_id, packet_hdr)                                                       \
+#define FLIGHT_RECORDER_PACKET_HDR(ep, fr, event_id, packet_hdr)                                                   \
 	{                                                                                                          \
 		struct flight_recorder_entry *next =                                                               \
 			flight_recorder_init_next_entry((fr), (event_id), FR_ENTRY_TYPE_PACKET_HDR);               \
 		memcpy((void *) next->data, (void *) &(packet_hdr), sizeof(struct fi_opx_hfi1_stl_packet_hdr_9B)); \
 		if ((fr)->count + 1 == FLIGHT_RECORDER_ENTRY_COUNT)                                                \
-			flight_recorder_dump((fr));                                                                \
+			flight_recorder_dump(ep, fr);                                                              \
 	}
 
-#define FLIGHT_RECORDER_PACKET(fr, event_id, packet)                                                       \
+#define FLIGHT_RECORDER_PACKET(ep, fr, event_id, packet)                                                   \
 	{                                                                                                  \
 		struct flight_recorder_entry *next =                                                       \
 			flight_recorder_init_next_entry((fr), (event_id), FR_ENTRY_TYPE_PACKET);           \
 		memcpy((void *) next->data, (void *) &(packet), sizeof(union fi_opx_hfi1_packet_payload)); \
 		if ((fr)->count + 1 == FLIGHT_RECORDER_ENTRY_COUNT)                                        \
-			flight_recorder_dump((fr));                                                        \
+			flight_recorder_dump(ep, fr);                                                      \
 	}
 
 #define FLIGHT_RECORDER_INIT(fr)          \
@@ -235,9 +238,9 @@ static inline void flight_recorder_dump(struct flight_recorder *fr)
 
 #else /* !FLIGHT_RECORDER_ENABLE */
 
-#define FLIGHT_RECORDER_STRING(fr, event_id, format, ...)
-#define FLIGHT_RECORDER_PACKET_HDR(fr, event_id, packet_hdr)
-#define FLIGHT_RECORDER_PACKET(fr, event_id, packet)
+#define FLIGHT_RECORDER_STRING(ep, fr, event_id, format, ...)
+#define FLIGHT_RECORDER_PACKET_HDR(ep, fr, event_id, packet_hdr)
+#define FLIGHT_RECORDER_PACKET(ep, fr, event_id, packet)
 
 #endif /* #ifdef FLIGHT_RECORDER_ENABLE */
 
diff --git a/prov/opx/include/rdma/opx/fi_opx_hfi1_progress.h b/prov/opx/include/rdma/opx/fi_opx_hfi1_progress.h
@@ -376,7 +376,7 @@ void fi_opx_hfi1_handle_packet(struct fi_opx_ep *opx_ep, uint64_t *p_rhf_seq, ui
 			*p_last_egrbfr_index = egrbfr_index;
 		}
 
-		FLIGHT_RECORDER_PACKET_HDR(opx_ep->fr, FR_EVENT_HFI1_POLL_ONCE, hdr);
+		FLIGHT_RECORDER_PACKET_HDR(opx_ep, opx_ep->fr, FR_EVENT_HFI1_POLL_ONCE, hdr);
 	}
 
 	*p_rhf_seq   = OPX_RHF_SEQ_INCREMENT(rhf_seq, hfi1_type);
diff --git a/prov/opx/include/rdma/opx/opx_debug.h b/prov/opx/include/rdma/opx/opx_debug.h
@@ -45,6 +45,7 @@ struct opx_debug_ep_entry {
 
 void opx_debug_ep_list_append(void *opx_ep);
 void opx_debug_ep_list_free(void *opx_ep);
+void opx_debug_ep_list_dump();
 void opx_debug_install_handler();
 
 #endif
diff --git a/prov/opx/src/fi_opx_hfi1.c b/prov/opx/src/fi_opx_hfi1.c
@@ -4397,6 +4397,15 @@ fi_opx_hfi1_rx_rzv_cts(struct fi_opx_ep *opx_ep, const union opx_hfi1_packet_hdr
 		       const enum ofi_reliability_kind reliability, const uint32_t u32_extended_rx,
 		       const enum opx_hfi1_type hfi1_type)
 {
+	if (dput_opcode == FI_OPX_HFI_DPUT_OPCODE_PUT_CQ) {
+		struct fi_opx_completion_counter *cc = ((struct fi_opx_rma_request *) rma_request_vaddr)->cc;
+		if (cc->byte_counter == 0) {
+			OPX_BUF_FREE((struct fi_opx_rma_request *) rma_request_vaddr);
+			cc->hit_zero(cc);
+			return NULL;
+		}
+	}
+
 	union fi_opx_hfi1_deferred_work *work	= ofi_buf_alloc(opx_ep->tx->work_pending_pool);
 	struct fi_opx_hfi1_dput_params	*params = &work->dput;
 
diff --git a/prov/opx/src/opx_debug.c b/prov/opx/src/opx_debug.c
@@ -424,13 +424,18 @@ void opx_debug_ep_list_free(void *opx_ep)
 	}
 }
 
-static void opx_debug_signal_handler(int signum, siginfo_t *info, void *ucontext)
+void opx_debug_ep_list_dump()
 {
 	struct opx_debug_ep_entry *entry = (struct opx_debug_ep_entry *) ep_list.head;
 	while (entry) {
 		opx_debug_dump_endpoint((struct fi_opx_ep *) entry->ep);
 		entry = entry->next;
 	}
+}
+
+static void opx_debug_signal_handler(int signum, siginfo_t *info, void *ucontext)
+{
+	opx_debug_ep_list_dump();
 
 	if (prev_sig_handler && prev_sig_handler != SIG_DFL && prev_sig_handler != SIG_IGN) {
 		prev_sig_handler(signum);

Original file line number	Diff line number	Diff line change
`@@ -376,7 +376,7 @@ void fi_opx_hfi1_handle_packet(struct fi_opx_ep opx_ep, uint64_t p_rhf_seq, ui`
`376`	`376`	`*p_last_egrbfr_index = egrbfr_index;`
`377`	`377`	`}`
`378`	`378`
`379`		`- FLIGHT_RECORDER_PACKET_HDR(opx_ep->fr, FR_EVENT_HFI1_POLL_ONCE, hdr);`
	`379`	`+ FLIGHT_RECORDER_PACKET_HDR(opx_ep, opx_ep->fr, FR_EVENT_HFI1_POLL_ONCE, hdr);`
`380`	`380`	`}`
`381`	`381`
`382`	`382`	`*p_rhf_seq = OPX_RHF_SEQ_INCREMENT(rhf_seq, hfi1_type);`
Original file line number	Diff line number	Diff line change
`@@ -424,13 +424,18 @@ void opx_debug_ep_list_free(void *opx_ep)`
`424`	`424`	`}`
`425`	`425`	`}`
`426`	`426`
`427`		`-static void opx_debug_signal_handler(int signum, siginfo_t info, void ucontext)`
	`427`	`+void opx_debug_ep_list_dump()`
`428`	`428`	`{`
`429`	`429`	`struct opx_debug_ep_entry entry = (struct opx_debug_ep_entry ) ep_list.head;`
`430`	`430`	`while (entry) {`
`431`	`431`	`opx_debug_dump_endpoint((struct fi_opx_ep *) entry->ep);`
`432`	`432`	`entry = entry->next;`
`433`	`433`	`}`
	`434`	`+}`
	`435`	`+`
	`436`	`+static void opx_debug_signal_handler(int signum, siginfo_t info, void ucontext)`
	`437`	`+{`
	`438`	`+ opx_debug_ep_list_dump();`
`434`	`439`
`435`	`440`	`if (prev_sig_handler && prev_sig_handler != SIG_DFL && prev_sig_handler != SIG_IGN) {`
`436`	`441`	`prev_sig_handler(signum);`