[PATCH 05/06] opensm/perfmgr: don't clear data counters in PortCounters when ExtendedPortCounters is supported

DORMANTno replies

From: Ira Weiny <hidden>
Date: 2013-02-21 21:33:52
Subsystem: the rest · Maintainer: Linus Torvalds


Some hardware apparently clears the data counters of ExtendedPortCounters when
the PortCounters data counters are cleared.  (Must be using the same hardware
registers.)

Therefore, when ExtendedPortCounters is supported on a port; alter the counter
select of PortCounters to exclude the data counters when clearing.

Signed-off-by: Ira Weiny <redacted>
---
 opensm/osm_perfmgr.c |   88 ++++++++++++++++++++++++++++++++++++++++++--------
 1 files changed, 74 insertions(+), 14 deletions(-)

diff --git a/opensm/osm_perfmgr.c b/opensm/osm_perfmgr.c
index 65886f7..eb0c4f9 100644
--- a/opensm/osm_perfmgr.c
+++ b/opensm/osm_perfmgr.c

@@ -450,6 +450,7 @@ static ib_api_status_t perfmgr_send_pc_mad(osm_perfmgr_t * perfmgr,
 					   ib_net16_t dest_lid,
 					   ib_net32_t dest_qp, uint16_t pkey_ix,
 					   uint8_t port, uint8_t mad_method,
+					   uint16_t counter_select,
 					   osm_madw_context_t * p_context,
 					   uint8_t sl)
 {

@@ -469,7 +470,7 @@ static ib_api_status_t perfmgr_send_pc_mad(osm_perfmgr_t * perfmgr,
 	port_counter = (ib_port_counters_t *) & pm_mad->data;
 	memset(port_counter, 0, sizeof(*port_counter));
 	port_counter->port_select = port;
-	port_counter->counter_select = 0xFFFF;
+	port_counter->counter_select = cl_hton16(counter_select);
 
 	status = perfmgr_send_mad(perfmgr, p_madw);

@@ -613,7 +614,7 @@ static ib_api_status_t perfmgr_send_pce_mad(osm_perfmgr_t * perfmgr,
 	port_counter_ext = (ib_port_counters_ext_t *) & pm_mad->data;
 	memset(port_counter_ext, 0, sizeof(*port_counter_ext));
 	port_counter_ext->port_select = port;
-	port_counter_ext->counter_select = cl_hton16(0xFF);
+	port_counter_ext->counter_select = cl_hton16(0x00FF);
 
 	status = perfmgr_send_mad(perfmgr, p_madw);

@@ -715,6 +716,7 @@ static void perfmgr_query_counters(cl_map_item_t * p_map_item, void *context)
 			status = perfmgr_send_pc_mad(pm, lid, remote_qp,
 						     mon_node->port[port].pkey_ix,
 						     port, IB_MAD_METHOD_GET,
+						     0xffff,
 						     &mad_context,
 						     0); /* FIXME SL != 0 */
 			if (status != IB_SUCCESS)

@@ -1098,6 +1100,35 @@ static void perfmgr_check_oob_clear(osm_perfmgr_t * pm,
 		return;
 	}
 
+	OSM_LOG(pm->log, OSM_LOG_DEBUG,
+		"Errors vs previous node %s (0x%" PRIx64 ") port %u\n"
+		"SE:   %"PRIu64" ?< %"PRIu64"\n"
+		"LE:   %"PRIu64" ?< %"PRIu64"\n"
+		"LD:   %"PRIu64" ?< %"PRIu64"\n"
+		"RE:   %"PRIu64" ?< %"PRIu64"\n"
+		"RPE:  %"PRIu64" ?< %"PRIu64"\n"
+		"SRE:  %"PRIu64" ?< %"PRIu64"\n"
+		"XD:   %"PRIu64" ?< %"PRIu64"\n"
+		"XCE:  %"PRIu64" ?< %"PRIu64"\n"
+		"RCE:  %"PRIu64" ?< %"PRIu64"\n"
+		"LI:   %"PRIu64" ?< %"PRIu64"\n"
+		"BO:   %"PRIu64" ?< %"PRIu64"\n"
+		"VL15: %"PRIu64" ?< %"PRIu64"\n"
+		,
+		mon_node->name, mon_node->guid, port,
+		cr->symbol_err_cnt, prev_err.symbol_err_cnt,
+		cr->link_err_recover, prev_err.link_err_recover,
+		cr->link_downed, prev_err.link_downed,
+		cr->rcv_err, prev_err.rcv_err,
+		cr->rcv_rem_phys_err, prev_err.rcv_rem_phys_err,
+		cr->rcv_switch_relay_err, prev_err.rcv_switch_relay_err,
+		cr->xmit_discards, prev_err.xmit_discards,
+		cr->xmit_constraint_err, prev_err.xmit_constraint_err,
+		cr->rcv_constraint_err, prev_err.rcv_constraint_err,
+		cr->link_integrity, prev_err.link_integrity,
+		cr->buffer_overrun, prev_err.buffer_overrun,
+		cr->vl15_dropped, prev_err.vl15_dropped);
+
 	if (cr->symbol_err_cnt < prev_err.symbol_err_cnt ||
 	    cr->link_err_recover < prev_err.link_err_recover ||
 	    cr->link_downed < prev_err.link_downed ||

@@ -1158,6 +1189,7 @@ static void perfmgr_check_overflow(osm_perfmgr_t * pm,
 	osm_madw_context_t mad_context;
 	ib_api_status_t status;
 	ib_net32_t remote_qp;
+	uint16_t counter_select;
 
 	OSM_LOG_ENTER(pm->log);

@@ -1207,9 +1239,20 @@ static void perfmgr_check_overflow(osm_perfmgr_t * pm,
 		mad_context.perfmgr_context.node_guid = mon_node->guid;
 		mad_context.perfmgr_context.port = port;
 		mad_context.perfmgr_context.mad_method = IB_MAD_METHOD_SET;
-		/* clear port counters */
+
+		/* apparently some HW uses the same counters for the 32 and 64
+		 * bit versions and a clear of them in the PortCounters
+		 * attribute also clears the ExtendedPortCounters equivalant
+		 * counters
+		 */
+		if (pce_supported(mon_node, port))
+			counter_select = 0x0fff;
+		else
+			counter_select = 0xffff;
+
 		status = perfmgr_send_pc_mad(pm, lid, remote_qp, pkey_ix,
 					     port, IB_MAD_METHOD_SET,
+					     counter_select,
 					     &mad_context,
 					     0); /* FIXME SL != 0 */
 		if (status != IB_SUCCESS)

@@ -1513,6 +1556,27 @@ static void perfmgr_check_data_cnt_oob_clear(osm_perfmgr_t * pm,
 		return;
 	}
 
+	OSM_LOG(pm->log, OSM_LOG_DEBUG,
+		"Data vs previous node %s (0x%" PRIx64 ") port %u\n"
+		"TX:    %"PRIu64" ?< %"PRIu64"\n"
+		"RX:    %"PRIu64" ?< %"PRIu64"\n"
+		"TXP:   %"PRIu64" ?< %"PRIu64"\n"
+		"RXP:   %"PRIu64" ?< %"PRIu64"\n"
+		"UTXP:  %"PRIu64" ?< %"PRIu64"\n"
+		"URXP:  %"PRIu64" ?< %"PRIu64"\n"
+		"MTXP:  %"PRIu64" ?< %"PRIu64"\n"
+		"MRXP:  %"PRIu64" ?< %"PRIu64"\n"
+		,
+		mon_node->name, mon_node->guid, port,
+		dc->xmit_data, prev_dc.xmit_data,
+		dc->rcv_data, prev_dc.rcv_data,
+		dc->xmit_pkts, prev_dc.xmit_pkts,
+		dc->rcv_pkts, prev_dc.rcv_pkts,
+		dc->unicast_xmit_pkts, prev_dc.unicast_xmit_pkts,
+		dc->unicast_rcv_pkts, prev_dc.unicast_rcv_pkts,
+		dc->multicast_xmit_pkts, prev_dc.multicast_xmit_pkts,
+		dc->multicast_rcv_pkts, prev_dc.multicast_rcv_pkts);
+
 	if (dc->xmit_data < prev_dc.xmit_data ||
 	    dc->rcv_data < prev_dc.rcv_data ||
 	    dc->xmit_pkts < prev_dc.xmit_pkts ||

@@ -1526,6 +1590,7 @@ static void perfmgr_check_data_cnt_oob_clear(osm_perfmgr_t * pm,
 			"PerfMgr: ERR 540B: Detected an out of band data counter "
 			"clear on node %s (0x%" PRIx64 ") port %u\n",
 			mon_node->name, mon_node->guid, port);
+
 		perfmgr_db_clear_prev_dc(pm->db, mon_node->guid, port);
 	}
 }

@@ -1617,15 +1682,13 @@ static void pc_recv_process(void *context, void *data)
 						  ietf_supported(p_mon_node,
 								 port));
 
-		/* detect an out of band clear on the port */
-		if (mad_context->perfmgr_context.mad_method !=
-		    IB_MAD_METHOD_SET)
-			perfmgr_check_data_cnt_oob_clear(pm, p_mon_node, port,
-						    &data_reading);
-
 		/* add counter */
 		if (mad_context->perfmgr_context.mad_method
 		    == IB_MAD_METHOD_GET) {
+			/* detect an out of band clear on the port */
+			perfmgr_check_data_cnt_oob_clear(pm, p_mon_node, port,
+						    &data_reading);
+
 			perfmgr_db_add_dc_reading(pm->db, node_guid, port,
 						  &data_reading,
 						  ietf_supported(p_mon_node,

@@ -1634,7 +1697,6 @@ static void pc_recv_process(void *context, void *data)
 			perfmgr_db_clear_prev_dc(pm->db, node_guid, port);
 		}
 
-		/* check overflow */
 		perfmgr_check_pce_overflow(pm, p_mon_node,
 					   p_mon_node->port[port].pkey_ix,
 					   port, ext_wire_read);

@@ -1648,15 +1710,13 @@ static void pc_recv_process(void *context, void *data)
 		if (!pce_sup)
 			perfmgr_db_fill_data_cnt_read_pc(wire_read, &data_reading);
 
-		/* detect an out of band clear on the port */
-		if (mad_context->perfmgr_context.mad_method != IB_MAD_METHOD_SET) {
+		if (mad_context->perfmgr_context.mad_method == IB_MAD_METHOD_GET) {
+			/* detect an out of band clear on the port */
 			perfmgr_check_oob_clear(pm, p_mon_node, port, &err_reading);
 			if (!pce_sup)
 				perfmgr_check_data_cnt_oob_clear(pm, p_mon_node, port,
 							    &data_reading);
-		}
 
-		if (mad_context->perfmgr_context.mad_method == IB_MAD_METHOD_GET) {
 			/* log errors from this reading */
 			if (pm->subn->opt.perfmgr_log_errors)
 				perfmgr_log_errors(pm, p_mon_node, port, &err_reading);

-- 
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help