perfmgr: Add xmit_wait support

index 0304d9c..93e57a6 100644
@@ -75,6 +75,7 @@ extern "C" {
 #define OSM_PERFMGR_DEFAULT_SWEEP_TIME_S 180
 #define OSM_PERFMGR_DEFAULT_DUMP_FILE "opensm_port_counters.log"
 #define OSM_PERFMGR_DEFAULT_MAX_OUTSTANDING_QUERIES 500
+#define OSM_PERFMGR_DEFAULT_XMIT_WAIT_THRESHOLD 0x0000FFFF
 
 /****s* OpenSM: PerfMgr/osm_perfmgr_state_t */
 typedef enum {

@@ -150,6 +151,8 @@ typedef struct osm_perfmgr {
 	int16_t local_port;
 	int rm_nodes;
 	boolean_t query_cpi;
+	boolean_t xmit_wait_log;
+	uint32_t xmit_wait_threshold;
 } osm_perfmgr_t;
 /*
 * FIELDS

index ed14f76..25d9c58 100644
@@ -90,6 +90,7 @@ typedef struct {
 	uint64_t link_integrity;
 	uint64_t buffer_overrun;
 	uint64_t vl15_dropped;
+	uint64_t xmit_wait;
 	time_t time;
 } perfmgr_db_err_reading_t;
 

@@ -205,7 +206,8 @@ void perfmgr_db_print_by_guid(perfmgr_db_t * db, uint64_t guid, FILE *fp,
  */
 
 void perfmgr_db_fill_err_read(ib_port_counters_t * wire_read,
-			      perfmgr_db_err_reading_t * reading);
+			      perfmgr_db_err_reading_t * reading,
+			      boolean_t xmit_wait_sup);
 void perfmgr_db_fill_data_cnt_read_pc(ib_port_counters_t * wire_read,
 				      perfmgr_db_data_cnt_reading_t * reading);
 void perfmgr_db_fill_data_cnt_read_pce(ib_port_counters_ext_t * wire_read,

index 37e0fb3..a4ce2ed 100644
@@ -363,6 +363,8 @@ typedef struct osm_subn_opt {
 	int perfmgr_rm_nodes;
 	boolean_t perfmgr_log_errors;
 	boolean_t perfmgr_query_cpi;
+	boolean_t perfmgr_xmit_wait_log;
+	uint32_t perfmgr_xmit_wait_threshold;
 #endif				/* ENABLE_OSM_PERF_MGR */
 	char *event_plugin_name;
 	char *event_plugin_options;

170 lines not on this page · raw message

index 92e1a39..5b476bd 100644
@@ -341,6 +341,10 @@ debug_dump_err_reading(perfmgr_db_t * db, uint64_t guid, uint8_t port_num,
 		   "vld %" PRIu64 " <-- %" PRIu64 " (%" PRIu64 ")\n",
 		   cur->vl15_dropped, port->err_previous.vl15_dropped,
 		   port->err_total.vl15_dropped);
+	osm_log_v2(log, OSM_LOG_DEBUG, FILE_ID,
+		   "xw %" PRIu64 " <-- %" PRIu64 " (%" PRIu64 ")\n",
+		   cur->xmit_wait, port->err_previous.xmit_wait,
+		   port->err_total.xmit_wait);
 }
 
 /**********************************************************************

@@ -355,6 +359,7 @@ perfmgr_db_add_err_reading(perfmgr_db_t * db, uint64_t guid, uint8_t port,
 	perfmgr_db_err_reading_t *previous = NULL;
 	perfmgr_db_err_t rc = PERFMGR_EVENT_DB_SUCCESS;
 	osm_epi_pe_event_t epi_pe_data;
+	uint64_t xmit_wait_diff;
 
 	cl_plock_excl_acquire(&db->lock);
 	node = get(db, guid);

@@ -410,6 +415,9 @@ perfmgr_db_add_err_reading(perfmgr_db_t * db, uint64_t guid, uint8_t port,
 	epi_pe_data.vl15_dropped =
 	    (reading->vl15_dropped - previous->vl15_dropped);
 	p_port->err_total.vl15_dropped += epi_pe_data.vl15_dropped;
+	xmit_wait_diff =
+	    (reading->xmit_wait - previous->xmit_wait);
+	p_port->err_total.xmit_wait += xmit_wait_diff;
 
 	p_port->err_previous = *reading;
 

@@ -662,7 +670,7 @@ static void dump_node_mr(db_node_t * node, FILE * fp)
 		"%s\t%s\t"
 		"%s\t%s\t%s\t%s\t%s\t%s\t%s\t"
 		"%s\t%s\t%s\t%s\t%s\t%s\t%s\t"
-		"%s\t%s\t%s\t%s\n",
+		"%s\t%s\t%s\t%s\t%s\n",
 		"symbol_err_cnt",
 		"link_err_recover",
 		"link_downed",

@@ -675,6 +683,7 @@ static void dump_node_mr(db_node_t * node, FILE * fp)
 		"link_int_err",
 		"buf_overrun_err",
 		"vl15_dropped",
+		"xmit_wait",
 		"xmit_data",
 		"rcv_data",
 		"xmit_pkts",

@@ -704,7 +713,7 @@ static void dump_node_mr(db_node_t * node, FILE * fp)
 			"%" PRIu64 "\t%" PRIu64 "\t%" PRIu64 "\t" "%" PRIu64
 			"\t%" PRIu64 "\t%" PRIu64 "\t" "%" PRIu64 "\t%" PRIu64
 			"\t%" PRIu64 "\t%" PRIu64 "\t" "%" PRIu64 "\t%" PRIu64
-			"\t%" PRIu64 "\t%" PRIu64 "\n", node->node_name,
+			"\t%" PRIu64 "\t%" PRIu64 "\t%" PRIu64 "\n", node->node_name,
 			node->node_guid,
 			node->active ? "TRUE" : "FALSE",
 			i, last_reset, last_err_update, last_data_update,

@@ -720,6 +729,7 @@ static void dump_node_mr(db_node_t * node, FILE * fp)
 			node->ports[i].err_total.link_integrity,
 			node->ports[i].err_total.buffer_overrun,
 			node->ports[i].err_total.vl15_dropped,
+			node->ports[i].err_total.xmit_wait,
 			node->ports[i].dc_total.xmit_data,
 			node->ports[i].dc_total.rcv_data,
 			node->ports[i].dc_total.xmit_pkts,

@@ -827,7 +837,8 @@ static void dump_node_hr(db_node_t * node, FILE * fp, char *port, int err_only)
 		    && err->rcv_constraint_err == 0
 		    && err->link_integrity == 0
 		    && err->buffer_overrun == 0
-		    && err->vl15_dropped == 0)
+		    && err->vl15_dropped == 0
+		    && err->xmit_wait == 0)
 			continue;
 
 		fprintf(fp, "\"%s\" 0x%" PRIx64 " active %s port %d\n"

@@ -873,6 +884,9 @@ static void dump_node_hr(db_node_t * node, FILE * fp, char *port, int err_only)
 		if (!err_only || err->vl15_dropped != 0)
 			fprintf(fp, "     vl15_dropped         : %" PRIu64 "\n",
 				err->vl15_dropped);
+		if (!err_only || err->xmit_wait != 0)
+			fprintf(fp, "     xmit_wait            : %" PRIu64 "\n",
+				err->xmit_wait);
 
 		if (err_only)
 			continue;

@@ -1022,7 +1036,8 @@ perfmgr_db_dump(perfmgr_db_t * db, char *file, perfmgr_db_dump_t dump_type)
  **********************************************************************/
 void
 perfmgr_db_fill_err_read(ib_port_counters_t * wire_read,
-			 perfmgr_db_err_reading_t * reading)
+			 perfmgr_db_err_reading_t * reading,
+			 boolean_t xmit_wait_sup)
 {
 	reading->symbol_err_cnt = cl_ntoh16(wire_read->symbol_err_cnt);
 	reading->link_err_recover = wire_read->link_err_recover;

@@ -1039,6 +1054,10 @@ perfmgr_db_fill_err_read(ib_port_counters_t * wire_read,
 	reading->buffer_overrun =
 	    PC_BUF_OVERRUN(wire_read->link_int_buffer_overrun);
 	reading->vl15_dropped = cl_ntoh16(wire_read->vl15_dropped);
+	if (xmit_wait_sup)
+		reading->xmit_wait = cl_ntoh32(wire_read->xmit_wait);
+	else
+		reading->xmit_wait = 0;
 	reading->time = time(NULL);
 }
 

index a1c7648..f30d852 100644
@@ -854,6 +854,8 @@ static const opt_rec_t opt_tbl[] = {
 	{ "perfmgr_rm_nodes", OPT_OFFSET(perfmgr_rm_nodes), opts_parse_boolean, NULL, 0 },
 	{ "perfmgr_log_errors", OPT_OFFSET(perfmgr_log_errors), opts_parse_boolean, NULL, 0 },
 	{ "perfmgr_query_cpi", OPT_OFFSET(perfmgr_query_cpi), opts_parse_boolean, NULL, 0 },
+	{ "perfmgr_xmit_wait_log", OPT_OFFSET(perfmgr_xmit_wait_log), opts_parse_boolean, NULL, 0 },
+	{ "perfmgr_xmit_wait_threshold", OPT_OFFSET(perfmgr_xmit_wait_threshold), opts_parse_uint32, NULL, 0 },
 #endif				/* ENABLE_OSM_PERF_MGR */
 	{ "event_plugin_name", OPT_OFFSET(event_plugin_name), opts_parse_charp, NULL, 0 },
 	{ "event_plugin_options", OPT_OFFSET(event_plugin_options), opts_parse_charp, NULL, 0 },

@@ -1563,6 +1565,8 @@ void osm_subn_set_default_opt(IN osm_subn_opt_t * p_opt)
 	p_opt->perfmgr_rm_nodes = TRUE;
 	p_opt->perfmgr_log_errors = TRUE;
 	p_opt->perfmgr_query_cpi = FALSE;
+	p_opt->perfmgr_xmit_wait_log = FALSE;
+	p_opt->perfmgr_xmit_wait_threshold = OSM_PERFMGR_DEFAULT_XMIT_WAIT_THRESHOLD;
 #endif				/* ENABLE_OSM_PERF_MGR */
 
 	p_opt->event_plugin_name = NULL;

@@ -2637,7 +2641,12 @@ int osm_subn_output_conf(FILE *out, IN osm_subn_opt_t * p_opts)
 		"# Log error counters to opensm.log\n"
 		"perfmgr_log_errors %s\n\n"
 		"# Query PerfMgrGet(ClassPortInfo) for extended capabilities\n"
-		"perfmgr_query_cpi %s\n\n",
+		"perfmgr_query_cpi %s\n\n"
+		"# Log xmit_wait errors\n"
+		"perfmgr_xmit_wait_log %s\n\n"
+		"# If logging xmit_wait's; set threshold (default %u)\n"
+		"perfmgr_xmit_wait_threshold %u\n\n"
+		,
 		p_opts->perfmgr ? "TRUE" : "FALSE",
 		p_opts->perfmgr_redir ? "TRUE" : "FALSE",
 		p_opts->perfmgr_sweep_time_s,

@@ -2645,7 +2654,10 @@ int osm_subn_output_conf(FILE *out, IN osm_subn_opt_t * p_opts)
 		p_opts->perfmgr_ignore_cas ? "TRUE" : "FALSE",
 		p_opts->perfmgr_rm_nodes ? "TRUE" : "FALSE",
 		p_opts->perfmgr_log_errors ? "TRUE" : "FALSE",
-		p_opts->perfmgr_query_cpi ? "TRUE" : "FALSE");
+		p_opts->perfmgr_query_cpi ? "TRUE" : "FALSE",
+		p_opts->perfmgr_xmit_wait_log ? "TRUE" : "FALSE",
+		OSM_PERFMGR_DEFAULT_XMIT_WAIT_THRESHOLD,
+		p_opts->perfmgr_xmit_wait_threshold);
 
 	fprintf(out,
 		"#\n# Event DB Options\n#\n"