[IPoIB] Abort broadcast MC group join if port rate is lower than

existing broadcast group and log an event to the system log.
Also expanded error logging to the system log.

git-svn-id: svn://openib.tc.cornell.edu/gen1@219 ad392aa1-c5ef-ae45-8dd8-e69d62a5ef86

diff --git a/trunk/ulp/ipoib/kernel/ipoib_log.mc b/trunk/ulp/ipoib/kernel/ipoib_log.mc
index 7f7c8a81eb4c581f166a8ed35409a4a6bc914e94..9676ff297d5d958591889308ee758ab2aeadea7f 100644 (file)
--- a/trunk/ulp/ipoib/kernel/ipoib_log.mc
+++ b/trunk/ulp/ipoib/kernel/ipoib_log.mc
@@ -128,3 +128,158 @@ Language=English
 %2: Driver Initialized succesfully.\r
 .\r
 \r
+MessageId=0x0041\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_OPEN_CA\r
+Language=English\r
+%2: Failed to open Channel Adapter.\r
+.\r
+\r
+MessageId=0x0042\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_ALLOC_PD\r
+Language=English\r
+%2: Failed to allocate Protection Domain.\r
+.\r
+\r
+MessageId=0x0043\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_CREATE_RECV_CQ\r
+Language=English\r
+%2: Failed to create receive Completion Queue.\r
+.\r
+\r
+MessageId=0x0044\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_CREATE_SEND_CQ\r
+Language=English\r
+%2: Failed to create send Completion Queue.\r
+.\r
+\r
+MessageId=0x0045\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_CREATE_QP\r
+Language=English\r
+%2: Failed to create Queue Pair.\r
+.\r
+\r
+MessageId=0x0046\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_QUERY_QP\r
+Language=English\r
+%2: Failed to get Queue Pair number.\r
+.\r
+\r
+MessageId=0x0047\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_REG_PHYS\r
+Language=English\r
+%2: Failed to create DMA Memory Region.\r
+.\r
+\r
+MessageId=0x0048\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_RECV_POOL\r
+Language=English\r
+%2: Failed to create receive descriptor pool.\r
+.\r
+\r
+MessageId=0x0049\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_RECV_PKT_POOL\r
+Language=English\r
+%2: Failed to create NDIS_PACKET pool for receive indications.\r
+.\r
+\r
+MessageId=0x004A\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_RECV_BUF_POOL\r
+Language=English\r
+%2: Failed to create NDIS_BUFFER pool for receive indications.\r
+.\r
+\r
+MessageId=0x004B\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_SEND_PKT_POOL\r
+Language=English\r
+%2: Failed to create NDIS_PACKET pool for send processing.\r
+.\r
+\r
+MessageId=0x004C\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_SEND_BUF_POOL\r
+Language=English\r
+%2: Failed to create NDIS_BUFFER pool for send processing.\r
+.\r
+\r
+MessageId=0x004D\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_RECV_PKT_ARRAY\r
+Language=English\r
+%2: Failed to allocate receive indication array.\r
+.\r
+\r
+MessageId=0x004E\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_PORT_INFO_TIMEOUT\r
+Language=English\r
+%2: Subnet Administrator query for port information timed out. \r
+Make sure the SA is functioning properly.  Increasing the number\r
+of retries and retry timeout adapter parameters may solve the\r
+issue.\r
+.\r
+\r
+MessageId=0x004F\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_PORT_INFO_REJECT\r
+Language=English\r
+%2: Subnet Administrator failed the query for port information.\r
+Make sure the SA is functioning properly and compatible.\r
+.\r
+\r
+MessageId=0x0050\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_QUERY_PORT_INFO\r
+Language=English\r
+%2: Subnet Administrator query for port information failed.\r
+.\r
+\r
+MessageId=0x0055\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_BCAST_GET\r
+Language=English\r
+%2: Subnet Administrator failed query for broadcast group information.\r
+.\r
+\r
+MessageId=0x0056\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_BCAST_JOIN\r
+Language=English\r
+%2: Subnet Administrator failed request to joing broadcast group.\r
+.\r
+\r
+MessageId=0x0057\r
+Facility=IPoIB\r
+Severity=Error\r
+SymbolicName=EVENT_IPOIB_BCAST_RATE\r
+Language=English\r
+%2: The local port rate is too slow for the existing broadcast MC group.\r
+.\r

diff --git a/trunk/ulp/ipoib/kernel/ipoib_port.c b/trunk/ulp/ipoib/kernel/ipoib_port.c
index 756cc9a19d6f9f87dc27f34b58d95d8752413b43..54f905c8059995907c185f3853ea32641d7121c0 100644 (file)
--- a/trunk/ulp/ipoib/kernel/ipoib_port.c
+++ b/trunk/ulp/ipoib/kernel/ipoib_port.c
@@ -48,7 +48,7 @@
  * request that specifies an MGID.  The ESM returns every McMemberRecord\r
  * instead.\r
  */\r
-#define PR_102801\r
+//#define PR_102801\r
 \r
 \r
 /* Amount of physical memory to register. */\r
@@ -132,6 +132,12 @@ static ib_api_status_t
 __ib_mgr_activate(\r
        IN                              ipoib_port_t* const                     p_port );\r
 \r
+static void\r
+__ib_mgr_set_rate(\r
+       IN                              ipoib_port_t* const                     p_port,\r
+       IN              const   uint8_t                                         link_width,\r
+       IN              const   uint8_t                                         link_speed );\r
+\r
 /******************************************************************************\r
 *\r
 * Buffer manager operations.\r
@@ -747,6 +753,8 @@ __ib_mgr_init(
                NULL, p_port, &p_port->ib_mgr.h_ca );\r
        if( status != IB_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_OPEN_CA, 1, status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("ib_open_ca returned %s\n", \r
                        p_port->p_adapter->p_ifc->get_err_str( status )) );\r
@@ -758,6 +766,8 @@ __ib_mgr_init(
                p_port->ib_mgr.h_ca, IB_PDT_UD, p_port, &p_port->ib_mgr.h_pd );\r
        if( status != IB_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_ALLOC_PD, 1, status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("ib_alloc_pd returned %s\n", \r
                        p_port->p_adapter->p_ifc->get_err_str( status )) );\r
@@ -774,6 +784,8 @@ __ib_mgr_init(
                __cq_event, &p_port->ib_mgr.h_recv_cq );\r
        if( status != IB_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_CREATE_RECV_CQ, 1, status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("ib_create_cq returned %s.\n", \r
                        p_port->p_adapter->p_ifc->get_err_str( status )) );\r
@@ -789,6 +801,8 @@ __ib_mgr_init(
                __cq_event, &p_port->ib_mgr.h_send_cq );\r
        if( status != IB_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_CREATE_SEND_CQ, 1, status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("ib_create_cq returned %s.\n", \r
                        p_port->p_adapter->p_ifc->get_err_str( status )) );\r
@@ -811,6 +825,8 @@ __ib_mgr_init(
                __qp_event, &p_port->ib_mgr.h_qp );\r
        if( status != IB_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_CREATE_QP, 1, status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("ib_create_qp returned %s\n", \r
                        p_port->p_adapter->p_ifc->get_err_str( status )) );\r
@@ -821,6 +837,8 @@ __ib_mgr_init(
                p_port->ib_mgr.h_qp, &qp_attr );\r
        if( status != IB_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_QUERY_QP, 1, status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("ib_query_qp returned %s\n", \r
                        p_port->p_adapter->p_ifc->get_err_str( status )) );\r
@@ -843,6 +861,8 @@ __ib_mgr_init(
                &p_port->ib_mgr.lkey, &rkey, &p_port->ib_mgr.h_mr );\r
        if( status != IB_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_REG_PHYS, 1, status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("ib_reg_phys returned %s\n", \r
                        p_port->p_adapter->p_ifc->get_err_str( status )) );\r
@@ -926,6 +946,8 @@ __buf_mgr_init(
 #endif /* IPOIB_INLINE_RECV */\r
        if( cl_status != CL_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_RECV_POOL, 1, cl_status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("cl_qpool_init for recvs returned %s\n",\r
                        cl_status_text[cl_status]) );\r
@@ -937,6 +959,8 @@ __buf_mgr_init(
                p_params->rq_depth, PROTOCOL_RESERVED_SIZE_IN_PACKET );\r
        if( ndis_status != NDIS_STATUS_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_RECV_PKT_POOL, 1, ndis_status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("NdisAllocatePacketPool returned %08X\n", ndis_status) );\r
                return IB_INSUFFICIENT_RESOURCES;\r
@@ -946,6 +970,8 @@ __buf_mgr_init(
                p_params->rq_depth );\r
        if( ndis_status != NDIS_STATUS_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_RECV_BUF_POOL, 1, ndis_status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("NdisAllocateBufferPool returned %08X\n", ndis_status) );\r
                return IB_INSUFFICIENT_RESOURCES;\r
@@ -956,6 +982,8 @@ __buf_mgr_init(
                1, PROTOCOL_RESERVED_SIZE_IN_PACKET );\r
        if( ndis_status != NDIS_STATUS_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_SEND_PKT_POOL, 1, ndis_status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("NdisAllocatePacketPool returned %08X\n", ndis_status) );\r
                return IB_INSUFFICIENT_RESOURCES;\r
@@ -965,6 +993,8 @@ __buf_mgr_init(
                &p_port->buf_mgr.h_send_buf_pool, 1 );\r
        if( ndis_status != NDIS_STATUS_SUCCESS )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_SEND_BUF_POOL, 1, ndis_status );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("NdisAllocateBufferPool returned %08X\n", ndis_status) );\r
                return IB_INSUFFICIENT_RESOURCES;\r
@@ -1249,6 +1279,8 @@ __recv_mgr_init(
                sizeof(NDIS_PACKET*) * p_port->p_adapter->params.rq_depth );\r
        if( !p_port->recv_mgr.recv_pkt_array )\r
        {\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_RECV_PKT_ARRAY, 0 );\r
                IPOIB_TRACE_EXIT( IPOIB_DBG_ERROR,\r
                        ("cl_malloc for PNDIS_PACKET array failed.\n") );\r
                return IB_INSUFFICIENT_MEMORY;\r
@@ -4436,7 +4468,7 @@ __port_info_cb(
                                ("Received port info: link width = %d.\n",\r
                                p_port_rec->port_info.link_width_active) );\r
 \r
-                       ipoib_set_rate( p_port->p_adapter,\r
+                       __ib_mgr_set_rate( p_port,\r
                                p_port_rec->port_info.link_width_active,\r
                                ib_port_info_get_link_speed_active( &p_port_rec->port_info ) );\r
 \r
@@ -4455,9 +4487,22 @@ __port_info_cb(
                        ("Instance destroying - Aborting.\n") );\r
                break;\r
 \r
+       case IB_TIMEOUT:\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_PORT_INFO_TIMEOUT, 0 );\r
+               IPOIB_TRACE( IPOIB_DBG_INFO, ("Port info query timed out.\n") );\r
+               break;\r
+\r
+       case IB_REMOTE_ERROR:\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_PORT_INFO_REJECT, 0 );\r
+               IPOIB_TRACE( IPOIB_DBG_INFO, ("Port info query rejected by SA.\n") );\r
+               break;\r
+\r
        default:\r
-               /* Flag the adapter as hung. */\r
-               p_port->p_adapter->hung = TRUE;\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_QUERY_PORT_INFO, 1, p_query_rec->status );\r
+               /* Hopefully we'll get an SM change event that will restart things. */\r
                IPOIB_TRACE( IPOIB_DBG_INFO, ("Port info query failed.\n") );\r
        }\r
 \r
@@ -4483,7 +4528,7 @@ __port_get_mcast(
 \r
        IPOIB_ENTER( IPOIB_DBG_MCAST );\r
 \r
-       info.method = IB_MAD_METHOD_GETTABLE;\r
+       info.method = IB_MAD_METHOD_GET;\r
        info.attr_id = IB_MAD_ATTR_MCMEMBER_RECORD;\r
        info.attr_size = sizeof(ib_member_rec_t);\r
        info.comp_mask = IB_MCR_COMPMASK_MGID;\r
@@ -4583,6 +4628,8 @@ __bcast_get_cb(
                IPOIB_TRACE( IPOIB_DBG_INIT, ("Instance destroying - Aborting.\n") );\r
        \r
        default:\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_BCAST_GET, 1, p_query_rec->status );\r
                /* Flag the adapter as hung. */\r
                p_port->p_adapter->hung = TRUE;\r
        }\r
@@ -4608,6 +4655,22 @@ __port_join_bcast(
 \r
        IPOIB_ENTER( IPOIB_DBG_MCAST );\r
 \r
+       /* Check that the rate is realizable for our port. */\r
+       if( p_port->ib_mgr.rate < (p_member_rec->rate & 0x3F) )\r
+       {\r
+               /*\r
+                * The MC group rate is higher than our port's rate.  Log an error\r
+                * and stop.  A port transition will drive the retry.\r
+                */\r
+               IPOIB_TRACE( IPOIB_DBG_WARN,\r
+                       ("Unrealizable join due to rate mismatch.\n") );\r
+               NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                       EVENT_IPOIB_BCAST_RATE, 2,\r
+                       (uint32_t)(p_member_rec->rate & 0x3F),\r
+                       (uint32_t)p_port->ib_mgr.rate );\r
+               return;\r
+       }\r
+\r
        /* Join the broadcast group. */\r
        cl_memclr( &mcast_req, sizeof(mcast_req) );\r
        /* Copy the results of the Get to use as parameters. */\r
@@ -4729,7 +4792,7 @@ ipoib_port_down(
        p_port->state = IB_QPS_ERROR;\r
 \r
        NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
-               EVENT_IPOIB_PORT_DOWN, 0);\r
+               EVENT_IPOIB_PORT_DOWN, 0 );\r
 \r
        if( p_port->ib_mgr.h_query )\r
        {\r
@@ -4810,6 +4873,8 @@ __bcast_cb(
                }\r
                else\r
                {\r
+                       NdisWriteErrorLogEntry( p_port->p_adapter->h_adapter,\r
+                               EVENT_IPOIB_BCAST_JOIN, 1, p_mcast_rec->status );\r
                        /* Flag the adapter as hung. */\r
                        p_port->p_adapter->hung = TRUE;\r
                }\r
@@ -4943,6 +5008,60 @@ __ib_mgr_activate(
        return IB_SUCCESS;\r
 }\r
 \r
+static void\r
+__ib_mgr_set_rate(\r
+       IN                              ipoib_port_t* const                     p_port,\r
+       IN              const   uint8_t                                         link_width,\r
+       IN              const   uint8_t                                         link_speed )\r
+{\r
+       IPOIB_ENTER( IPOIB_DBG_INIT );\r
+\r
+       /* Set the link speed based on the IB link speed (1x vs 4x, etc). */\r
+       switch( link_width * link_width * link_speed )\r
+       {\r
+       case 1:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_2_5_GBS;\r
+               break;\r
+\r
+       case 2:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_5_GBS;\r
+               break;\r
+\r
+       case 4:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_10_GBS;\r
+               break;\r
+\r
+       case 8:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_20_GBS;\r
+               break;\r
+\r
+       case 16:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_40_GBS;\r
+               break;\r
+\r
+       case 64:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_30_GBS;\r
+               break;\r
+\r
+       case 128:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_60_GBS;\r
+               break;\r
+\r
+       case 256:\r
+               p_port->ib_mgr.rate = IB_PATH_RECORD_RATE_120_GBS;\r
+               break;\r
+\r
+       default:\r
+               IPOIB_TRACE( IPOIB_DBG_ERROR,\r
+                       ("Invalid link rate (%d).\n", link_width) );\r
+               p_port->ib_mgr.rate = 0;\r
+       }\r
+\r
+       ipoib_set_rate( p_port->p_adapter, link_width, link_speed );\r
+\r
+       IPOIB_EXIT( IPOIB_DBG_INIT );\r
+}\r
+\r
 \r
 /* Transition to a passive level thread. */\r
 ib_api_status_t\r

diff --git a/trunk/ulp/ipoib/kernel/ipoib_port.h b/trunk/ulp/ipoib/kernel/ipoib_port.h
index 8d78ce69647f3fc13a4d5381ec063c8c9030261a..c9c8cc4e521645f68f47f3827cbb588bace8b1b9 100644 (file)
--- a/trunk/ulp/ipoib/kernel/ipoib_port.h
+++ b/trunk/ulp/ipoib/kernel/ipoib_port.h
@@ -90,6 +90,7 @@ typedef struct _ipoib_ib_mgr
        ib_mr_handle_t                  h_mr;\r
        net32_t                                 lkey;\r
 \r
+       uint8_t                                 rate;\r
        ib_member_rec_t                 bcast_rec;\r
 \r
 }      ipoib_ib_mgr_t;\r