1
1
openmpi/ompi/mca/btl/mx/btl_mx_proc.c
Ralph Castain 9613b3176c Effectively revert the orte_output system and return to direct use of opal_output at all levels. Retain the orte_show_help subsystem to allow aggregation of show_help messages at the HNP.
After much work by Jeff and myself, and quite a lot of discussion, it has become clear that we simply cannot resolve the infinite loops caused by RML-involved subsystems calling orte_output. The original rationale for the change to orte_output has also been reduced by shifting the output of XML-formatted vs human readable messages to an alternative approach.

I have globally replaced the orte_output/ORTE_OUTPUT calls in the code base, as well as the corresponding .h file name. I have test compiled and run this on the various environments within my reach, so hopefully this will prove minimally disruptive.

This commit was SVN r18619.
2008-06-09 14:53:58 +00:00

245 строки
8.1 KiB
C

/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "opal/class/opal_hash_table.h"
#include "orte/util/name_fns.h"
#include "ompi/runtime/ompi_module_exchange.h"
#include "btl_mx.h"
#include "btl_mx_proc.h"
static void mca_btl_mx_proc_construct(mca_btl_mx_proc_t* proc);
static void mca_btl_mx_proc_destruct(mca_btl_mx_proc_t* proc);
OBJ_CLASS_INSTANCE(mca_btl_mx_proc_t,
opal_list_item_t, mca_btl_mx_proc_construct,
mca_btl_mx_proc_destruct);
void mca_btl_mx_proc_construct(mca_btl_mx_proc_t* proc)
{
proc->proc_ompi = 0;
proc->proc_endpoints = NULL;
proc->proc_endpoint_count = 0;
proc->mx_peers_count = 0;
proc->mx_peers = NULL;
OBJ_CONSTRUCT(&proc->proc_lock, opal_mutex_t);
/* add to list of all proc instance */
OPAL_THREAD_LOCK(&mca_btl_mx_component.mx_lock);
opal_list_append(&mca_btl_mx_component.mx_procs, &proc->super);
OPAL_THREAD_UNLOCK(&mca_btl_mx_component.mx_lock);
}
/*
* Cleanup MX proc instance
*/
void mca_btl_mx_proc_destruct(mca_btl_mx_proc_t* proc)
{
/* remove from list of all proc instances */
OPAL_THREAD_LOCK(&mca_btl_mx_component.mx_lock);
opal_list_remove_item(&mca_btl_mx_component.mx_procs, &proc->super);
OPAL_THREAD_UNLOCK(&mca_btl_mx_component.mx_lock);
/* release resources */
if( NULL != proc->proc_endpoints ) {
free(proc->proc_endpoints);
proc->proc_endpoints = NULL;
}
if( NULL != proc->mx_peers ) {
free(proc->mx_peers);
proc->mx_peers = NULL;
}
}
/*
* Look for an existing MX process instances based on the associated
* ompi_proc_t instance.
*/
static mca_btl_mx_proc_t* mca_btl_mx_proc_lookup_ompi(ompi_proc_t* ompi_proc)
{
mca_btl_mx_proc_t* mx_proc;
OPAL_THREAD_LOCK(&mca_btl_mx_component.mx_lock);
for( mx_proc = (mca_btl_mx_proc_t*)opal_list_get_first(&mca_btl_mx_component.mx_procs);
mx_proc != (mca_btl_mx_proc_t*)opal_list_get_end(&mca_btl_mx_component.mx_procs);
mx_proc = (mca_btl_mx_proc_t*)opal_list_get_next(mx_proc) ) {
if(mx_proc->proc_ompi == ompi_proc) {
OPAL_THREAD_UNLOCK(&mca_btl_mx_component.mx_lock);
return mx_proc;
}
}
OPAL_THREAD_UNLOCK(&mca_btl_mx_component.mx_lock);
return NULL;
}
/*
* Create a MX process structure. There is a one-to-one correspondence
* between a ompi_proc_t and a mca_btl_mx_proc_t instance. We cache
* additional data (specifically the list of mca_btl_mx_endpoint_t instances,
* and published addresses) associated w/ a given destination on this
* datastructure.
*/
mca_btl_mx_proc_t* mca_btl_mx_proc_create(ompi_proc_t* ompi_proc)
{
mca_btl_mx_proc_t* module_proc = NULL;
mca_btl_mx_addr_t *mx_peers;
int rc;
size_t size;
/* Check if we have already created a MX proc
* structure for this ompi process */
module_proc = mca_btl_mx_proc_lookup_ompi(ompi_proc);
if( module_proc != NULL ) {
/* Gotcha! */
return module_proc;
}
/* query for the peer address info */
rc = ompi_modex_recv( &mca_btl_mx_component.super.btl_version,
ompi_proc, (void*)&mx_peers, &size );
if( OMPI_SUCCESS != rc ) {
opal_output( 0, "mca_pml_base_modex_recv failed for peer %s",
ORTE_NAME_PRINT(&ompi_proc->proc_name) );
return NULL;
}
if( size < sizeof(mca_btl_mx_addr_t) ) { /* no available connection */
return NULL;
}
if( (size % sizeof(mca_btl_mx_addr_t)) != 0 ) {
opal_output( 0, "invalid mx address for peer %s",
ORTE_NAME_PRINT(&ompi_proc->proc_name) );
return NULL;
}
module_proc = OBJ_NEW(mca_btl_mx_proc_t);
module_proc->proc_ompi = ompi_proc;
module_proc->mx_peers_count = size / sizeof(mca_btl_mx_addr_t);
#if OMPI_ENABLE_HETEROGENEOUS_SUPPORT
for (rc = 0 ; rc < module_proc->mx_peers_count ; ++rc) {
BTL_MX_ADDR_NTOH(mx_peers[rc]);
}
#endif
module_proc->mx_peers = mx_peers;
return module_proc;
}
/*
* Note that this routine must be called with the lock on the process
* already held. Insert a btl instance into the proc array and assign
* it an address.
*/
int mca_btl_mx_proc_insert( mca_btl_mx_proc_t* module_proc,
mca_btl_mx_endpoint_t* module_endpoint )
{
mca_btl_mx_module_t* mx_btl;
int i, j;
/**
* Check if there is any Myrinet network between myself and the peer
*/
for( i = 0; i < mca_btl_mx_component.mx_num_btls; i++ ) {
mx_btl = mca_btl_mx_component.mx_btls[i];
for( j = 0; j < module_proc->mx_peers_count; j++ ) {
if( mx_btl->mx_unique_network_id == module_proc->mx_peers[j].unique_network_id ) {
/* There is at least one connection between these two nodes */
goto create_peer_endpoint;
}
}
}
module_proc->mx_peers_count = 0;
/**
* No Myrinet connectivity. Let the PML layer figure out another
* way to communicate with the peer.
*/
return OMPI_ERROR;
create_peer_endpoint:
mx_btl = module_endpoint->endpoint_btl;
for( j = 0; j < module_proc->mx_peers_count; j++ ) {
if( mx_btl->mx_unique_network_id == module_proc->mx_peers[j].unique_network_id ) {
module_endpoint->mx_peer.nic_id = module_proc->mx_peers[j].nic_id;
module_endpoint->mx_peer.endpoint_id = module_proc->mx_peers[j].endpoint_id;
break;
}
}
if( NULL == module_proc->proc_endpoints ) {
module_proc->proc_endpoints = (mca_btl_base_endpoint_t**)
malloc(module_proc->mx_peers_count * sizeof(mca_btl_base_endpoint_t*));
if( NULL == module_proc->proc_endpoints ) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
module_endpoint->endpoint_proc = module_proc;
return OMPI_SUCCESS;
}
int mca_btl_mx_proc_connect( mca_btl_mx_endpoint_t* module_endpoint )
{
int num_retry = 0;
mx_return_t mx_status;
mx_endpoint_addr_t mx_remote_addr;
mca_btl_mx_proc_t* module_proc = module_endpoint->endpoint_proc;
module_endpoint->status = MCA_BTL_MX_CONNECTION_PENDING;
retry_connect:
mx_status = mx_connect( module_endpoint->endpoint_btl->mx_endpoint,
module_endpoint->mx_peer.nic_id, module_endpoint->mx_peer.endpoint_id,
mca_btl_mx_component.mx_filter, mca_btl_mx_component.mx_timeout, &mx_remote_addr );
if( MX_SUCCESS != mx_status ) {
if( MX_TIMEOUT == mx_status )
if( num_retry++ < mca_btl_mx_component.mx_connection_retries )
goto retry_connect;
{
char peer_name[MX_MAX_HOSTNAME_LEN];
if( MX_SUCCESS != mx_nic_id_to_hostname( module_endpoint->mx_peer.nic_id, peer_name ) )
sprintf( peer_name, "unknown %lx nic_id", (long)module_endpoint->mx_peer.nic_id );
opal_output( 0, "mx_connect fail for %s with key %x (error %s)\n\tUnique ID (local %x remote %x)\n",
peer_name, mca_btl_mx_component.mx_filter, mx_strerror(mx_status),
module_endpoint->endpoint_btl->mx_unique_network_id,
module_endpoint->mx_peer.unique_network_id );
}
module_endpoint->status = MCA_BTL_MX_NOT_REACHEABLE;
return OMPI_ERROR;
}
module_endpoint->mx_peer_addr = mx_remote_addr;
module_endpoint->status = MCA_BTL_MX_CONNECTED;
module_proc->proc_endpoints[module_proc->proc_endpoint_count++] = module_endpoint;
return OMPI_SUCCESS;
}