From 50646b07ce7e616957362f5eb471304c0b7a2fec Mon Sep 17 00:00:00 2001 From: Ralph Castain Date: Wed, 10 May 2017 09:17:06 -0700 Subject: [PATCH] Update the RML OFI by copying the updated files from @anandhis branch Signed-off-by: Ralph Castain --- orte/mca/rml/ofi/rml_ofi_component.c | 20 +++++++++++++------- orte/mca/rml/ofi/rml_ofi_send.c | 10 +++++----- 2 files changed, 18 insertions(+), 12 deletions(-) diff --git a/orte/mca/rml/ofi/rml_ofi_component.c b/orte/mca/rml/ofi/rml_ofi_component.c index ab2dc1c425..2e0213e495 100644 --- a/orte/mca/rml/ofi/rml_ofi_component.c +++ b/orte/mca/rml/ofi/rml_ofi_component.c @@ -970,7 +970,6 @@ static orte_rml_base_module_t* open_conduit(opal_list_t *attributes) return NULL; } - /* someone may require this specific component, so look for "ofi" */ if (orte_get_attribute(attributes, ORTE_RML_INCLUDE_COMP_ATTRIB, (void**)&comp_attrib, OPAL_STRING) && NULL != comp_attrib) { @@ -998,17 +997,23 @@ static orte_rml_base_module_t* open_conduit(opal_list_t *attributes) } } } - /*[Debug] to check for daemon commn over ofi-ethernet, enable the default conduit ORTE_MGMT_CONDUIT over ofi */ if (orte_get_attribute(attributes, ORTE_RML_TRANSPORT_TYPE, (void**)&comp_attrib, OPAL_STRING) && NULL != comp_attrib) { - opal_output_verbose(20,orte_rml_base_framework.framework_output, - "%s - Forcibly returning ofi socket provider for ethernet transport request", - ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)); + opal_output_verbose(20,orte_rml_base_framework.framework_output, + "%s - ORTE_RML_TRANSPORT_TYPE = %s ", + ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), comp_attrib); comps = opal_argv_split(comp_attrib, ','); for (i=0; NULL != comps[i]; i++) { - if (0 == strcmp(comps[i], "ethernet")) { + /* changing below to check for oob, as trying to use ofi for only mgmt conduit */ + if (0 == strcmp(comps[i], "oob")) { + /* changing below to check for fabric, as trying to use ofi for only coll conduit + if (0 == strcmp(comps[i], "fabric")) { */ + /*if (0 == strcmp(comps[i], "ethernet")) { */ /* we are a candidate, */ + opal_output_verbose(20,orte_rml_base_framework.framework_output, + "%s - Forcibly returning ofi socket provider for ethernet transport request", + ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)); opal_argv_free(comps); OBJ_CONSTRUCT(&provider, opal_list_t); orte_set_attribute(&provider, ORTE_RML_PROVIDER_ATTRIB, @@ -1018,7 +1023,7 @@ static orte_rml_base_module_t* open_conduit(opal_list_t *attributes) } opal_argv_free(comps); } - /*[Debug] */ + /* end [Debug] */ /* Alternatively, check the attributes to see if we qualify - we only handle * "pt2pt" */ @@ -1241,6 +1246,7 @@ void convert_to_sockaddr( char *ofiuri, struct sockaddr_in* ep_sockaddr) ep_sockaddr->sin_family = atoi( sin_fly ); port = atoi( sin_port); ep_sockaddr->sin_port = htons(port); + ep_sockaddr->sin_addr.s_addr = inet_addr(sin_addr); opal_output_verbose(1,orte_rml_base_framework.framework_output, "%s OFI convert_to_sockaddr() port = 0x%x decimal-%d, InternetAddr = %s ", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),ntohs(ep_sockaddr->sin_port),ntohs(ep_sockaddr->sin_port), diff --git a/orte/mca/rml/ofi/rml_ofi_send.c b/orte/mca/rml/ofi/rml_ofi_send.c index 718c13a017..d0115664be 100644 --- a/orte/mca/rml/ofi/rml_ofi_send.c +++ b/orte/mca/rml/ofi/rml_ofi_send.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2015-2016 Intel, Inc. All rights reserved. + * Copyright (c) 2015-2017 Intel, Inc. All rights reserved. * $COPYRIGHT$ * * Additional copyrights may follow @@ -72,7 +72,6 @@ OBJ_CLASS_INSTANCE(ofi_recv_msg_queue_t, opal_list_item_t, ofi_recv_msg_queue_cons, ofi_recv_msg_queue_des); - static void send_self_exe(int fd, short args, void* data) { orte_self_send_xfer_t *xfer = (orte_self_send_xfer_t*)data; @@ -523,7 +522,7 @@ static void send_msg(int fd, short args, void *cbdata) } } if ( OPAL_SUCCESS == ret) { - //Anandhi added for debug purpose + //[Debug] printing additional info of IP switch ( orte_rml_ofi.ofi_prov[ofi_prov_id].fabric_info->addr_format) { case FI_SOCKADDR_IN : @@ -531,14 +530,14 @@ static void send_msg(int fd, short args, void *cbdata) /*[debug] - print the sockaddr - port and s_addr */ ep_sockaddr = (struct sockaddr_in*)dest_ep_name; opal_output_verbose(1,orte_rml_base_framework.framework_output, - "%s peer %s epnamelen is %lu, port = %d (or) 0x%x, InternetAddr = 0x%s ", + "%s peer %s epnamelen is %d, port = %d (or) 0x%x, InternetAddr = 0x%s ", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),ORTE_NAME_PRINT(peer), orte_rml_ofi.ofi_prov[ofi_prov_id].epnamelen,ntohs(ep_sockaddr->sin_port), ntohs(ep_sockaddr->sin_port),inet_ntoa(ep_sockaddr->sin_addr)); /*[end debug]*/ break; } - //Anandhi end debug + //[Debug] end debug opal_output_verbose(10, orte_rml_base_framework.framework_output, "%s OPAL_MODEX_RECV succeded, %s peer ep name obtained. length=%lu", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), @@ -552,6 +551,7 @@ static void send_msg(int fd, short args, void *cbdata) snd->status = ORTE_ERR_ADDRESSEE_UNKNOWN; ORTE_RML_SEND_COMPLETE(snd); + return; } } else {