Hi Nicolas,
On 17:26 Tue 14 Jul , Nicolas Morey-Chaisemartin wrote:
> Support for xmit_wait counters was missing in the perfmgr though it was read
> from the mad and event plugin interface already handles it.
> This patch adds support for it (tested and working with an event plugin)
>
> Tested-by: Jean-Vincent Ficet <[email protected]>
> Signed-off-by: Nicolas Morey-Chaisemartin <[email protected]>
> ---
> I think emails got mixed up the first time so here it is again.
>
> opensm/include/opensm/osm_perfmgr_db.h | 23 ++++++-
> opensm/opensm/osm_perfmgr.c | 30 +++++++-
> opensm/opensm/osm_perfmgr_db.c | 124 +++++++++++++++++++++++++++++--
> 3 files changed, 166 insertions(+), 11 deletions(-)
>
> diff --git a/opensm/include/opensm/osm_perfmgr_db.h
> b/opensm/include/opensm/osm_perfmgr_db.h
> index 42a47bd..35b5ac3 100644
> --- a/opensm/include/opensm/osm_perfmgr_db.h
> +++ b/opensm/include/opensm/osm_perfmgr_db.h
> @@ -109,6 +109,14 @@ typedef struct {
> } perfmgr_db_data_cnt_reading_t;
>
> /** =========================================================================
> + * Port select count reading
> + */
> +typedef struct {
> + uint64_t xmit_wait;
> + time_t time;
> +} perfmgr_db_sel_reading_t;
> +
Why do we need a separate structure for this counter?
Sasha
> +/** =========================================================================
> * Dump output options
> */
> typedef enum {
> @@ -125,6 +133,8 @@ typedef struct db_port {
> perfmgr_db_err_reading_t err_previous;
> perfmgr_db_data_cnt_reading_t dc_total;
> perfmgr_db_data_cnt_reading_t dc_previous;
> + perfmgr_db_sel_reading_t ps_total;
> + perfmgr_db_sel_reading_t ps_previous;
> time_t last_reset;
> } db_port_t;
>
> @@ -179,7 +189,16 @@ perfmgr_db_err_t perfmgr_db_get_prev_dc(perfmgr_db_t *
> db, uint64_t guid,
> reading);
> perfmgr_db_err_t perfmgr_db_clear_prev_dc(perfmgr_db_t * db, uint64_t guid,
> uint8_t port);
> -
> +perfmgr_db_err_t perfmgr_db_add_ps_reading(perfmgr_db_t * db, uint64_t guid,
> + uint8_t port,
> + perfmgr_db_sel_reading_t *
> + reading);
> +perfmgr_db_err_t perfmgr_db_get_prev_ps(perfmgr_db_t * db, uint64_t guid,
> + uint8_t port,
> + perfmgr_db_sel_reading_t *
> + reading);
> +perfmgr_db_err_t perfmgr_db_clear_prev_ps(perfmgr_db_t * db, uint64_t guid,
> + uint8_t port);
> void perfmgr_db_clear_counters(perfmgr_db_t * db);
> perfmgr_db_err_t perfmgr_db_dump(perfmgr_db_t * db, char *file,
> perfmgr_db_dump_t dump_type);
> @@ -196,6 +215,8 @@ 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_epc(ib_port_counters_ext_t * wire_read,
> perfmgr_db_data_cnt_reading_t * reading);
> +void perfmgr_db_fill_sel_read(ib_port_counters_t * wire_read,
> + perfmgr_db_sel_reading_t * reading);
>
> END_C_DECLS
>
> diff --git a/opensm/opensm/osm_perfmgr.c b/opensm/opensm/osm_perfmgr.c
> index ecfdbda..8a9eb12 100644
> --- a/opensm/opensm/osm_perfmgr.c
> +++ b/opensm/opensm/osm_perfmgr.c
> @@ -853,10 +853,12 @@ void osm_perfmgr_destroy(osm_perfmgr_t * pm)
> static void perfmgr_check_oob_clear(osm_perfmgr_t * pm,
> monitored_node_t * mon_node, uint8_t port,
> perfmgr_db_err_reading_t * cr,
> - perfmgr_db_data_cnt_reading_t * dc)
> + perfmgr_db_data_cnt_reading_t * dc,
> + perfmgr_db_sel_reading_t * ps)
> {
> perfmgr_db_err_reading_t prev_err;
> perfmgr_db_data_cnt_reading_t prev_dc;
> + perfmgr_db_sel_reading_t prev_ps;
>
> if (perfmgr_db_get_prev_err(pm->db, mon_node->guid, port, &prev_err)
> != PERFMGR_EVENT_DB_SUCCESS) {
> @@ -905,6 +907,23 @@ static void perfmgr_check_oob_clear(osm_perfmgr_t * pm,
> mon_node->name, mon_node->guid, port);
> perfmgr_db_clear_prev_dc(pm->db, mon_node->guid, port);
> }
> +
> + if (perfmgr_db_get_prev_ps(pm->db, mon_node->guid, port, &prev_ps)
> + != PERFMGR_EVENT_DB_SUCCESS) {
> + OSM_LOG(pm->log, OSM_LOG_VERBOSE,
> + "Failed to find previous select count "
> + "reading for %s (0x%" PRIx64 ") port %u\n",
> + mon_node->name, mon_node->guid, port);
> + return;
> + }
> +
> + if (ps->xmit_wait < prev_ps.xmit_wait) {
> + OSM_LOG(pm->log, OSM_LOG_ERROR,
> + "PerfMgr: ERR 4C17: Detected an out of band select
> counter "
> + "clear on node %s (0x%" PRIx64 ") port %u\n",
> + mon_node->name, mon_node->guid, port);
> + perfmgr_db_clear_prev_ps(pm->db, mon_node->guid, port);
> + }
> }
>
> /**********************************************************************
> @@ -1062,6 +1081,8 @@ static void pc_recv_process(void *context, void *data)
> uint8_t port = mad_context->perfmgr_context.port;
> perfmgr_db_err_reading_t err_reading;
> perfmgr_db_data_cnt_reading_t data_reading;
> + perfmgr_db_sel_reading_t select_reading;
> +
> cl_map_item_t *p_node;
> monitored_node_t *p_mon_node;
>
> @@ -1148,10 +1169,12 @@ static void pc_recv_process(void *context, void *data)
> */
> perfmgr_db_fill_data_cnt_read_pc(wire_read, &data_reading);
>
> + perfmgr_db_fill_sel_read(wire_read, &select_reading);
> +
> /* detect an out of band clear on the port */
> if (mad_context->perfmgr_context.mad_method != IB_MAD_METHOD_SET)
> perfmgr_check_oob_clear(pm, p_mon_node, port, &err_reading,
> - &data_reading);
> + &data_reading, &select_reading);
>
> /* log any critical events from this reading */
> perfmgr_log_events(pm, p_mon_node, port, &err_reading);
> @@ -1161,9 +1184,12 @@ static void pc_recv_process(void *context, void *data)
> &err_reading);
> perfmgr_db_add_dc_reading(pm->db, node_guid, port,
> &data_reading);
> + perfmgr_db_add_ps_reading(pm->db, node_guid, port,
> + &select_reading);
> } else {
> perfmgr_db_clear_prev_err(pm->db, node_guid, port);
> perfmgr_db_clear_prev_dc(pm->db, node_guid, port);
> + perfmgr_db_clear_prev_ps(pm->db, node_guid, port);
> }
>
> perfmgr_check_overflow(pm, p_mon_node, port, wire_read);
> diff --git a/opensm/opensm/osm_perfmgr_db.c b/opensm/opensm/osm_perfmgr_db.c
> index e5dfc19..132c2fb 100644
> --- a/opensm/opensm/osm_perfmgr_db.c
> +++ b/opensm/opensm/osm_perfmgr_db.c
> @@ -486,6 +486,102 @@ Exit:
> return (rc);
> }
>
> +static inline void
> +debug_dump_ps_reading(perfmgr_db_t * db, uint64_t guid, uint8_t port_num,
> + db_port_t * port, perfmgr_db_sel_reading_t * cur)
> +{
> + osm_log_t *log = db->perfmgr->log;
> + if (!osm_log_is_active(log, OSM_LOG_DEBUG))
> + return;
> +
> + osm_log(log, OSM_LOG_DEBUG,
> + "xd %" PRIu64 " <-- %" PRIu64 " (%" PRIu64 ")\n",
> + cur->xmit_wait, port->ps_previous.xmit_wait,
> + port->ps_total.xmit_wait);
> +}
> +
> +/**********************************************************************
> + * perfmgr_db_sel_reading_t functions
> + **********************************************************************/
> +perfmgr_db_err_t
> +perfmgr_db_add_ps_reading(perfmgr_db_t * db, uint64_t guid, uint8_t port,
> + perfmgr_db_sel_reading_t * reading)
> +{
> + db_port_t *p_port = NULL;
> + db_node_t *node = NULL;
> + perfmgr_db_sel_reading_t *previous = NULL;
> + perfmgr_db_err_t rc = PERFMGR_EVENT_DB_SUCCESS;
> + osm_epi_ps_event_t epi_ps_data;
> +
> + cl_plock_excl_acquire(&db->lock);
> + node = get(db, guid);
> + if ((rc = bad_node_port(node, port)) != PERFMGR_EVENT_DB_SUCCESS)
> + goto Exit;
> +
> + p_port = &node->ports[port];
> + previous = &node->ports[port].ps_previous;
> +
> + debug_dump_ps_reading(db, guid, port, p_port, reading);
> +
> + epi_ps_data.time_diff_s = reading->time - previous->time;
> + osm_epi_create_port_id(&epi_ps_data.port_id, guid, port,
> + node->node_name);
> +
> + /* calculate changes from previous reading */
> + epi_ps_data.xmit_wait = reading->xmit_wait - previous->xmit_wait;
> + p_port->ps_total.xmit_wait += epi_ps_data.xmit_wait;
> +
> + p_port->ps_previous = *reading;
> + osm_opensm_report_event(db->perfmgr->osm,
> + OSM_EVENT_ID_PORT_SELECT, &epi_ps_data);
> +
> +Exit:
> + cl_plock_release(&db->lock);
> + return (rc);
> +}
> +
> +perfmgr_db_err_t perfmgr_db_get_prev_ps(perfmgr_db_t * db, uint64_t guid,
> + uint8_t port,
> + perfmgr_db_sel_reading_t * reading)
> +{
> + db_node_t *node = NULL;
> + perfmgr_db_err_t rc = PERFMGR_EVENT_DB_SUCCESS;
> +
> + cl_plock_acquire(&db->lock);
> +
> + node = get(db, guid);
> + if ((rc = bad_node_port(node, port)) != PERFMGR_EVENT_DB_SUCCESS)
> + goto Exit;
> +
> + *reading = node->ports[port].ps_previous;
> +
> +Exit:
> + cl_plock_release(&db->lock);
> + return (rc);
> +}
> +
> +perfmgr_db_err_t
> +perfmgr_db_clear_prev_ps(perfmgr_db_t * db, uint64_t guid, uint8_t port)
> +{
> + db_node_t *node = NULL;
> + perfmgr_db_sel_reading_t *previous = NULL;
> + perfmgr_db_err_t rc = PERFMGR_EVENT_DB_SUCCESS;
> +
> + cl_plock_excl_acquire(&db->lock);
> + node = get(db, guid);
> + if ((rc = bad_node_port(node, port)) != PERFMGR_EVENT_DB_SUCCESS)
> + goto Exit;
> +
> + previous = &node->ports[port].ps_previous;
> +
> + memset(previous, 0, sizeof(*previous));
> + node->ports[port].ps_previous.time = time(NULL);
> +
> +Exit:
> + cl_plock_release(&db->lock);
> + return (rc);
> +}
> +
> static void clear_counters(cl_map_item_t * const p_map_item, void *context)
> {
> db_node_t *node = (db_node_t *) p_map_item;
> @@ -517,6 +613,8 @@ static void clear_counters(cl_map_item_t * const
> p_map_item, void *context)
> node->ports[i].dc_total.multicast_rcv_pkts = 0;
> node->ports[i].dc_total.time = ts;
>
> + node->ports[i].ps_total.xmit_wait = 0;
> +
> node->ports[i].last_reset = ts;
> }
> }
> @@ -546,7 +644,7 @@ 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",
> @@ -565,8 +663,7 @@ static void dump_node_mr(db_node_t * node, FILE * fp)
> "rcv_pkts",
> "unicast_xmit_pkts",
> "unicast_rcv_pkts",
> - "multicast_xmit_pkts",
> - "multicast_rcv_pkts");
> + "multicast_xmit_pkts", "multicast_rcv_pkts", "xmit_wait");
> for (i = (node->esp0) ? 0 : 1; i < node->num_ports; i++) {
> char *since = ctime(&node->ports[i].last_reset);
> since[strlen(since) - 1] = '\0'; /* remove \n */
> @@ -577,8 +674,8 @@ 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,
> - node->node_guid, i, since,
> + "\t%" PRIu64 "\t%" PRIu64 "\t%" PRIu64 "\n",
> + node->node_name, node->node_guid, i, since,
> node->ports[i].err_total.symbol_err_cnt,
> node->ports[i].err_total.link_err_recover,
> node->ports[i].err_total.link_downed,
> @@ -598,7 +695,8 @@ static void dump_node_mr(db_node_t * node, FILE * fp)
> node->ports[i].dc_total.unicast_xmit_pkts,
> node->ports[i].dc_total.unicast_rcv_pkts,
> node->ports[i].dc_total.multicast_xmit_pkts,
> - node->ports[i].dc_total.multicast_rcv_pkts);
> + node->ports[i].dc_total.multicast_rcv_pkts,
> + node->ports[i].ps_total.xmit_wait);
> }
> }
>
> @@ -634,7 +732,8 @@ static void dump_node_hr(db_node_t * node, FILE * fp)
> " unicast_xmit_pkts : %" PRIu64 "\n"
> " unicast_rcv_pkts : %" PRIu64 "\n"
> " multicast_xmit_pkts : %" PRIu64 "\n"
> - " multicast_rcv_pkts : %" PRIu64 "\n",
> + " multicast_rcv_pkts : %" PRIu64 "\n"
> + " xmit_wait : %" PRIu64 "\n",
> node->node_name,
> node->node_guid,
> i,
> @@ -658,7 +757,8 @@ static void dump_node_hr(db_node_t * node, FILE * fp)
> node->ports[i].dc_total.unicast_xmit_pkts,
> node->ports[i].dc_total.unicast_rcv_pkts,
> node->ports[i].dc_total.multicast_xmit_pkts,
> - node->ports[i].dc_total.multicast_rcv_pkts);
> + node->ports[i].dc_total.multicast_rcv_pkts,
> + node->ports[i].ps_total.xmit_wait);
> }
> }
>
> @@ -809,4 +909,12 @@ perfmgr_db_fill_data_cnt_read_epc(ib_port_counters_ext_t
> * wire_read,
> reading->multicast_rcv_pkts = cl_ntoh64(wire_read->multicast_rcv_pkts);
> reading->time = time(NULL);
> }
> +
> +void
> +perfmgr_db_fill_sel_read(ib_port_counters_t * wire_read,
> + perfmgr_db_sel_reading_t * reading)
> +{
> + reading->xmit_wait = cl_ntoh32(wire_read->xmit_wait);
> + reading->time = time(NULL);
> +}
> #endif /* ENABLE_OSM_PERF_MGR */
>
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
