/**
* Function for finding and opening either all MCA components, or the one
* that was specifically requested via a MCA parameter.
*/
int mca_btl_base_open(void)
{
int i;
if( ++mca_btl_base_already_opened > 1 ) return OMPI_SUCCESS;
/* Verbose output */
mca_base_param_reg_int_name("btl",
"base_verbose",
"Verbosity level of the BTL framework",
false, false,
0,
&mca_btl_base_verbose);
mca_btl_base_output = opal_output_open(NULL);
opal_output_set_verbosity(mca_btl_base_output, mca_btl_base_verbose);
/* Override the per-BTL "don't run if THREAD_MULTIPLE selected"
embargo? */
mca_base_param_reg_int_name("btl",
"base_thread_multiple_override",
"Enable BTLs that are not normally enabled when MPI_THREAD_MULTIPLE is enabled (THIS IS FOR DEVELOPERS ONLY! SHOULD NOT BE USED BY END USERS!)",
true, false,
0, &i);
mca_btl_base_thread_multiple_override = OPAL_INT_TO_BOOL(i);
/* Open up all available components */
if (OMPI_SUCCESS !=
mca_base_components_open("btl", mca_btl_base_output, mca_btl_base_static_components,
&mca_btl_base_components_opened, true)) {
return OMPI_ERROR;
}
/* Initialize the list so that in mca_btl_base_close(), we can
iterate over it (even if it's empty, as in the case of
ompi_info) */
OBJ_CONSTRUCT(&mca_btl_base_modules_initialized, opal_list_t);
/* register parameters */
mca_base_param_lookup_string(
mca_base_param_register_string("btl","base","include",NULL,NULL), &mca_btl_base_include);
mca_base_param_lookup_string(
mca_base_param_register_string("btl","base","exclude",NULL,NULL), &mca_btl_base_exclude);
mca_base_param_reg_int_name("btl", "base_warn_component_unused",
"This parameter is used to turn on warning messages when certain NICs are not used",
false, false, 1, &mca_btl_base_warn_component_unused);
/* All done */
return OMPI_SUCCESS;
}
/*
* Main MCA initialization.
*/
int mca_base_open(void)
{
int param_index;
char *value;
opal_output_stream_t lds;
char hostname[64];
if (!mca_base_opened) {
mca_base_opened = true;
} else {
return OPAL_SUCCESS;
}
/* Register some params */
#if OMPI_WANT_HOME_CONFIG_FILES
asprintf(&value, "%s%c%s"OPAL_PATH_SEP".openmpi"OPAL_PATH_SEP"components", opal_install_dirs.pkglibdir, OPAL_ENV_SEP, opal_home_directory() );
#else
# if defined(__WINDOWS__) && defined(_DEBUG)
asprintf(&value, "%s/debug", opal_install_dirs.pkglibdir);
# else
asprintf(&value, "%s", opal_install_dirs.pkglibdir);
# endif
#endif
mca_base_param_component_path =
mca_base_param_reg_string_name("mca", "component_path",
"Path where to look for Open MPI and ORTE components",
false, false, value, NULL);
free(value);
param_index = mca_base_param_reg_string_name("mca", "verbose",
"Top-level verbosity parameter",
false, false, NULL, NULL);
mca_base_param_reg_int_name("mca", "component_show_load_errors",
"Whether to show errors for components that failed to load or not",
false, false, 1, NULL);
mca_base_param_reg_int_name("mca", "component_disable_dlopen",
"Whether to attempt to disable opening dynamic components or not",
false, false, 0, NULL);
/* What verbosity level do we want? */
mca_base_param_lookup_string(param_index, &value);
memset(&lds, 0, sizeof(lds));
if (NULL != value) {
parse_verbose(value, &lds);
free(value);
} else {
set_defaults(&lds);
}
gethostname(hostname, 64);
asprintf(&lds.lds_prefix, "[%s:%05d] ", hostname, getpid());
opal_output_reopen(0, &lds);
opal_output_verbose(5, 0, "mca: base: opening components");
free(lds.lds_prefix);
/* Open up the component repository */
return mca_base_component_repository_init();
}
/**
* Removes the bproc directory
* @code /tmp/openmpi-bproc-<user>/ @endcode and all of its contents
* @retval ORTE_SUCCESS
* @retval error
*/
static int
odls_bproc_remove_dir()
{
char *frontend = NULL, *user = NULL, *filename = NULL;
int id;
/* get the username set by the bproc pls. We need to get it from here
* because on many bproc systems the method we use to get the username
* from the system on the backend fails and we only get the uid. */
id = mca_base_param_register_string("pls", "bproc", "username", NULL,
orte_system_info.user);
mca_base_param_lookup_string(id,&user);
asprintf(&filename, "openmpi-bproc-%s", user );
if( NULL == filename ) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERROR;
}
frontend = opal_os_path(false, "tmp", filename, NULL );
free(filename); /* Always free the filename */
if (NULL == frontend) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERROR;
}
/* we do our best to clean up the directory tree, but we ignore errors*/
odls_bproc_delete_dir_tree(frontend);
free(frontend);
return ORTE_SUCCESS;
}
/*
* utility routine for string parameter registration
*/
static int reg_string(const char* param_name,
const char* deprecated_param_name,
const char* param_desc,
const char* default_value, char **out_value,
int flags)
{
int index;
char *value;
index = mca_base_param_reg_string(&mca_btl_openib_component.super.btl_version,
param_name, param_desc, false, false,
default_value, &value);
if (NULL != deprecated_param_name) {
mca_base_param_reg_syn(index,
&mca_btl_openib_component.super.btl_version,
deprecated_param_name, true);
}
mca_base_param_lookup_string(index, &value);
if (0 != (flags & REGSTR_EMPTY_OK) && 0 == strlen(value)) {
opal_output(0, "Bad parameter value for parameter \"%s\"",
param_name);
return OMPI_ERR_BAD_PARAM;
}
*out_value = value;
return OMPI_SUCCESS;
}
static int slave_set_name(void)
{
char *jobid_str, *procid_str;
int id, rc;
orte_jobid_t jobid;
orte_vpid_t vpid;
id = mca_base_param_register_string("orte", "ess", "jobid", NULL, NULL);
mca_base_param_lookup_string(id, &jobid_str);
if (NULL == jobid_str) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (rc = orte_util_convert_string_to_jobid(&jobid, jobid_str))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
free(jobid_str);
id = mca_base_param_register_string("orte", "ess", "vpid", NULL, NULL);
mca_base_param_lookup_string(id, &procid_str);
if (NULL == procid_str) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (rc = orte_util_convert_string_to_vpid(&vpid, procid_str))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
free(procid_str);
ORTE_PROC_MY_NAME->jobid = jobid;
ORTE_PROC_MY_NAME->vpid = vpid;
ORTE_EPOCH_SET(ORTE_PROC_MY_NAME->epoch,orte_ess.proc_get_epoch(ORTE_PROC_MY_NAME));
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:slave set name to %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* get the non-name common environmental variables */
if (ORTE_SUCCESS != (rc = orte_ess_env_get())) {
ORTE_ERROR_LOG(rc);
return rc;
}
return ORTE_SUCCESS;
}
static inline char *mca_pml_v_param_register_string( const char* param_name,
char *default_value )
{
int id = mca_base_param_register_string("pml", "v", param_name, NULL, default_value);
char *param_value = default_value;
mca_base_param_lookup_string(id, ¶m_value);
return param_value;
}
static inline char* mca_btl_template_param_register_string(
const char* param_name,
const char* default_value)
{
char *param_value;
int id = mca_base_param_register_string("btl","template",param_name,NULL,default_value);
mca_base_param_lookup_string(id, ¶m_value);
return param_value;
}
/**
* Run a user-level debugger
*/
void orte_run_debugger(char *basename, opal_cmd_line_t *cmd_line,
int argc, char *argv[])
{
int i, id;
char **new_argv = NULL;
char *value, **lines;
/* Get the orte_base_debug MCA parameter and search for a debugger
that can run */
id = mca_base_param_find("orte", NULL, "base_user_debugger");
if (id < 0) {
opal_show_help("help-orterun.txt", "debugger-mca-param-not-found",
true);
exit(1);
}
value = NULL;
mca_base_param_lookup_string(id, &value);
if (NULL == value) {
opal_show_help("help-orterun.txt", "debugger-orte_base_user_debugger-empty",
true);
exit(1);
}
/* Look through all the values in the MCA param */
lines = opal_argv_split(value, ':');
free(value);
for (i = 0; NULL != lines[i]; ++i) {
if (ORTE_SUCCESS == process(lines[i], basename, cmd_line, argc, argv,
&new_argv)) {
break;
}
}
/* If we didn't find one, abort */
if (NULL == lines[i]) {
opal_show_help("help-orterun.txt", "debugger-not-found", true);
exit(1);
}
opal_argv_free(lines);
/* We found one */
execvp(new_argv[0], new_argv);
value = opal_argv_join(new_argv, ' ');
opal_show_help("help-orterun.txt", "debugger-exec-failed",
true, basename, value, new_argv[0]);
free(value);
opal_argv_free(new_argv);
exit(1);
}
static int parse_requested(int mca_param, bool *include_mode,
char ***requested_component_names)
{
int i;
char *requested, *requested_orig;
*requested_component_names = NULL;
*include_mode = true;
/* See if the user requested anything */
if (OPAL_ERROR == mca_base_param_lookup_string(mca_param, &requested)) {
return OPAL_ERROR;
}
if (NULL == requested || 0 == strlen(requested)) {
return OPAL_SUCCESS;
}
requested_orig = requested;
/* Are we including or excluding? We only allow the negate
character to be the *first* character of the value (but be nice
and allow any number of negate characters in the beginning). */
while (negate == requested[0] && '\0' != requested[0]) {
*include_mode = false;
++requested;
}
/* Double check to ensure that the user did not specify the negate
character anywhere else in the value. */
i = 0;
while ('\0' != requested[i]) {
if (negate == requested[i]) {
opal_show_help("help-mca-base.txt",
"framework-param:too-many-negates",
true, requested_orig);
free(requested_orig);
return OPAL_ERROR;
}
++i;
}
/* Split up the value into individual component names */
*requested_component_names = opal_argv_split(requested, ',');
/* All done */
free(requested_orig);
return OPAL_SUCCESS;
}
int orte_proc_info(void)
{
int id, tmp;
/* all other params are set elsewhere */
id = mca_base_param_register_int("seed", NULL, NULL, NULL, orte_process_info.seed);
mca_base_param_lookup_int(id, &tmp);
orte_process_info.seed = OPAL_INT_TO_BOOL(tmp);
/* if we are a seed, then make sure the daemon flag is NOT set so that
* framework components are properly selected
*/
if (orte_process_info.seed) {
orte_process_info.daemon = false;
}
id = mca_base_param_register_int("orte", "app", "num", NULL, -1);
mca_base_param_lookup_int(id, &tmp);
orte_process_info.app_num = tmp;
id = mca_base_param_register_string("gpr", "replica", "uri", NULL, orte_process_info.gpr_replica_uri);
mca_base_param_lookup_string(id, &(orte_process_info.gpr_replica_uri));
mca_base_param_set_internal(id, true);
id = mca_base_param_register_string("ns", "replica", "uri", NULL, orte_process_info.ns_replica_uri);
mca_base_param_lookup_string(id, &(orte_process_info.ns_replica_uri));
mca_base_param_set_internal(id, true);
id = mca_base_param_register_string("tmpdir", "base", NULL, NULL, orte_process_info.tmpdir_base);
mca_base_param_lookup_string(id, &(orte_process_info.tmpdir_base));
/* get the process id */
orte_process_info.pid = getpid();
return ORTE_SUCCESS;
}
/**
* Returns a path of the form:
* @code
* /tmp/openmpi-bproc-<user>/<universe>/<jobid>-<app_context>/<proc_rank>/
* @endcode
* which is used to put links to the pty/pipes in
* @param proc_rank the process's rank on the node
* @param jobid the jobid the proc belongs to
* @param app_context the application context number within the job
* @retval path
*/
static char *
odls_bproc_get_base_dir_name(int proc_rank, orte_jobid_t jobid,
orte_std_cntr_t app_context)
{
char *path = NULL, *user = NULL, *job = NULL;
int rc;
/* ensure that system info is set */
orte_sys_info();
if (NULL == orte_universe_info.name) { /* error condition */
ORTE_ERROR_LOG(ORTE_ERROR);
return NULL;
}
rc = orte_ns.convert_jobid_to_string(&job, jobid);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return NULL;
}
/* get the username set by the bproc pls. We need to get it from here
* because on many bproc systems the method we use to get the username
* from the system on the backend fails and we only get the uid. */
rc = mca_base_param_register_string("pls", "bproc", "username", NULL,
orte_system_info.user);
mca_base_param_lookup_string(rc,&user);
if (0 > asprintf(&path, OPAL_PATH_SEP"tmp"OPAL_PATH_SEP"openmpi-bproc-%s"OPAL_PATH_SEP"%s"OPAL_PATH_SEP"%s-%d"OPAL_PATH_SEP"%d",
user, orte_universe_info.name,
job, (int) app_context, proc_rank)) {
ORTE_ERROR_LOG(ORTE_ERROR);
path = NULL;
}
if(0 < mca_odls_bproc_component.debug) {
opal_output(0, "odls bproc io setup. Path: %s\n", path);
}
free(user);
free(job);
return path;
}
static int slurm_set_name(void)
{
int slurm_nodeid;
int rc;
int id;
orte_jobid_t jobid;
orte_vpid_t vpid;
char* jobid_string;
char* vpid_string;
char *nodeid;
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:slurm setting name"));
id = mca_base_param_register_string("orte", "ess", "jobid", NULL, NULL);
mca_base_param_lookup_string(id, &jobid_string);
if (NULL == jobid_string) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (rc = orte_util_convert_string_to_jobid(&jobid, jobid_string))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
id = mca_base_param_register_string("orte", "ess", "vpid", NULL, NULL);
mca_base_param_lookup_string(id, &vpid_string);
if (NULL == vpid_string) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (rc = orte_util_convert_string_to_vpid(&vpid, vpid_string))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
ORTE_PROC_MY_NAME->jobid = jobid;
/* fix up the vpid and make it the "real" vpid */
if (NULL == (nodeid = getenv("SLURM_NODEID"))) {
opal_output(0, "SLURM_NODEID not found - cannot define name");
return ORTE_ERR_NOT_FOUND;
}
slurm_nodeid = atoi(nodeid);
ORTE_PROC_MY_NAME->vpid = vpid + slurm_nodeid;
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:slurm set name to %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* fix up the system info nodename to match exactly what slurm returned */
if (NULL != orte_process_info.nodename) {
free(orte_process_info.nodename);
}
orte_process_info.nodename = get_slurm_nodename(slurm_nodeid);
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:slurm set nodename to %s",
orte_process_info.nodename));
/* get the non-name common environmental variables */
if (ORTE_SUCCESS != (rc = orte_ess_env_get())) {
ORTE_ERROR_LOG(rc);
return rc;
}
return ORTE_SUCCESS;
}
/**
* Function for finding and opening either all MCA components, or the one
* that was specifically requested via a MCA parameter.
*/
int orte_ras_base_open(void)
{
int value, rc, param;
orte_data_type_t tmp;
char *requested;
/* Debugging / verbose output */
orte_ras_base.ras_output = opal_output_open(NULL);
mca_base_param_reg_int_name("ras", "base_verbose",
"Enable debugging for the RAS framework (nonzero = enabled)",
false, false, 0, &value);
if (value != 0) {
orte_ras_base.ras_output = opal_output_open(NULL);
} else {
orte_ras_base.ras_output = -1;
}
/* Defaults */
orte_ras_base.ras_opened_valid = false;
orte_ras_base.ras_using_proxy = false;
orte_ras_base.ras_available_valid = false;
/** register the base system types with the DSS */
tmp = ORTE_RAS_NODE;
if (ORTE_SUCCESS != (rc = orte_dss.register_type(orte_ras_base_pack_node,
orte_ras_base_unpack_node,
(orte_dss_copy_fn_t)orte_ras_base_copy_node,
(orte_dss_compare_fn_t)orte_ras_base_compare_node,
(orte_dss_size_fn_t)orte_ras_base_size_node,
(orte_dss_print_fn_t)orte_ras_base_print_node,
(orte_dss_release_fn_t)orte_ras_base_std_obj_release,
ORTE_DSS_STRUCTURED,
"ORTE_RAS_NODE", &tmp))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* Some systems do not want any RAS support. In those cases,
* memory consumption is also an issue. For those systems, we
* avoid opening the RAS components by checking for a directive
* to use the "null" component.
*/
param = mca_base_param_reg_string_name("ras", NULL, NULL,
false, false, NULL, NULL);
if (ORTE_ERROR == mca_base_param_lookup_string(param, &requested)) {
return ORTE_ERROR;
}
if (NULL != requested && 0 == strcmp(requested, "null")) {
/* the user has specifically requested that we use the "null"
* component. In this case, that means we do NOT open any
* components, and we simply use the default module we have
* already defined above
*/
orte_ras_base.ras_opened_valid = false;
orte_ras = orte_ras_no_op; /* use the no_op module */
return ORTE_SUCCESS;
}
/* check for timing tests */
param = mca_base_param_reg_int_name("orte", "timing",
"Request that critical timing loops be measured",
false, false, 0, &value);
if (value != 0) {
orte_ras_base.timing = true;
} else {
orte_ras_base.timing = false;
}
/* Open up all available components */
if (ORTE_SUCCESS !=
mca_base_components_open("ras", orte_ras_base.ras_output,
mca_ras_base_static_components,
&orte_ras_base.ras_opened, true)) {
return ORTE_ERROR;
}
/* if we are not on a HNP, select the proxy 'module' */
if (!orte_process_info.seed) {
orte_ras = orte_ras_base_proxy_module;
/* initialize the module */
orte_ras_base_proxy_init(&rc);
orte_ras_base.ras_using_proxy = true;
return ORTE_SUCCESS;
}
/* All done */
orte_ras_base.ras_opened_valid = true;
return ORTE_SUCCESS;
}
static int orte_rds_hostfile_query(orte_jobid_t job)
{
opal_list_t existing;
opal_list_t updates, rds_updates;
opal_list_item_t *item;
orte_rds_cell_desc_t *rds_item;
orte_rds_cell_attr_t *new_attr;
orte_ras_node_t *ras_item;
int rc;
if (orte_rds_hostfile_queried) {
/* if we have already been queried, then
* our info is on the registry, so just
* return. Note that this restriction
* may eventually be lifted - ideally,
* we might check to see if this is a
* new file name and go ahead with the
* query if so.
*/
return ORTE_SUCCESS;
}
orte_rds_hostfile_queried = true;
OBJ_CONSTRUCT(&existing, opal_list_t);
OBJ_CONSTRUCT(&updates, opal_list_t);
OBJ_CONSTRUCT(&rds_updates, opal_list_t);
rc = orte_ras_base_node_query(&existing);
if(ORTE_SUCCESS != rc) {
goto cleanup;
}
rc = mca_base_param_find("rds", "hostfile", "path");
mca_base_param_lookup_string(rc, &mca_rds_hostfile_component.path);
rc = orte_rds_hostfile_parse(mca_rds_hostfile_component.path, &existing, &updates);
if (ORTE_ERR_NOT_FOUND == rc) {
if(mca_rds_hostfile_component.default_hostfile) {
rc = ORTE_SUCCESS;
} else {
opal_show_help("help-rds-hostfile.txt", "rds:no-hostfile",
true,
mca_rds_hostfile_component.path);
}
goto cleanup;
} else if (ORTE_SUCCESS != rc) {
goto cleanup;
}
if ( !opal_list_is_empty(&updates) ) {
/* Convert RAS update list to RDS update list */
for ( ras_item = (orte_ras_node_t*)opal_list_get_first(&updates);
ras_item != (orte_ras_node_t*)opal_list_get_end(&updates);
ras_item = (orte_ras_node_t*)opal_list_get_next(ras_item)) {
rds_item = OBJ_NEW(orte_rds_cell_desc_t);
if (NULL == rds_item) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
rds_item->site = strdup("Hostfile");
rds_item->name = strdup(ras_item->node_name);
if (need_cellid) {
#if 0 /* JJH Repair when cellid's are fixed */
/* Create a new cellid for this hostfile */
rc = orte_ns.create_cellid(&local_cellid, rds_item->site, rds_item->name);
if (ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return rc;
}
#endif
local_cellid = 0;
need_cellid = false;
}
rds_item->cellid = local_cellid;
ras_item->node_cellid = local_cellid;
new_attr = OBJ_NEW(orte_rds_cell_attr_t);
if (NULL == new_attr) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
new_attr->keyval.key = strdup(ORTE_RDS_NAME);
new_attr->keyval.value = OBJ_NEW(orte_data_value_t);
if (NULL == new_attr->keyval.value) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
new_attr->keyval.value->type = ORTE_STRING;
new_attr->keyval.value->data = strdup(ras_item->node_name);
opal_list_append(&(rds_item->attributes), &new_attr->super);
new_attr = OBJ_NEW(orte_rds_cell_attr_t);
if (NULL == new_attr) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
new_attr->keyval.key = strdup(ORTE_CELLID_KEY);
new_attr->keyval.value = OBJ_NEW(orte_data_value_t);
if (NULL == new_attr->keyval.value) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
new_attr->keyval.value->type = ORTE_CELLID;
if (ORTE_SUCCESS != (rc = orte_dss.copy(&(new_attr->keyval.value->data), &(rds_item->cellid), ORTE_CELLID))) {
ORTE_ERROR_LOG(rc);
return rc;
}
opal_list_append(&(rds_item->attributes), &new_attr->super);
opal_list_append(&rds_updates, &rds_item->super);
}
/* Insert the new node into the RDS */
rc = orte_rds.store_resource(&rds_updates);
if (ORTE_SUCCESS != rc) {
goto cleanup;
}
/* Then the RAS, since we can assume that any
* resources listed in the hostfile have been
* already allocated for our use.
*/
rc = orte_ras_base_node_insert(&updates);
if (ORTE_SUCCESS != rc) {
goto cleanup;
}
/* and now, indicate that ORTE should override any oversubscribed conditions
* based on local hardware limits since the user (a) might not have
* provided us any info on the #slots for a node, and (b) the user
* might have been wrong! If we don't check the number of local physical
* processors, then we could be too aggressive on our sched_yield setting
* and cause performance problems.
*/
rc = orte_ras_base_set_oversubscribe_override(job);
if (ORTE_SUCCESS != rc) {
goto cleanup;
}
}
cleanup:
if (NULL != mca_rds_hostfile_component.path) {
free(mca_rds_hostfile_component.path);
mca_rds_hostfile_component.path = NULL;
}
while(NULL != (item = opal_list_remove_first(&existing))) {
OBJ_RELEASE(item);
}
while(NULL != (item = opal_list_remove_first(&updates))) {
OBJ_RELEASE(item);
}
while (NULL != (rds_item = (orte_rds_cell_desc_t*)opal_list_remove_first(&rds_updates))) {
while (NULL != (new_attr = (orte_rds_cell_attr_t*)opal_list_remove_first(&(rds_item->attributes)))) {
OBJ_RELEASE(new_attr);
}
OBJ_RELEASE(rds_item);
}
OBJ_DESTRUCT(&existing);
OBJ_DESTRUCT(&updates);
OBJ_DESTRUCT(&rds_updates);
return rc;
}
/**
* Run a user-level debugger
*/
void orte_run_debugger(char *basename, opal_cmd_line_t *cmd_line,
int argc, char *argv[], int num_procs)
{
int i, id;
char **new_argv = NULL;
char *value, **lines, *env_name;
/* Get the orte_base_debug MCA parameter and search for a debugger
that can run */
id = mca_base_param_find("orte", NULL, "base_user_debugger");
if (id < 0) {
orte_show_help("help-orterun.txt", "debugger-mca-param-not-found",
true);
exit(1);
}
value = NULL;
mca_base_param_lookup_string(id, &value);
if (NULL == value) {
orte_show_help("help-orterun.txt", "debugger-orte_base_user_debugger-empty",
true);
exit(1);
}
/* Look through all the values in the MCA param */
lines = opal_argv_split(value, ':');
free(value);
for (i = 0; NULL != lines[i]; ++i) {
if (ORTE_SUCCESS == process(lines[i], basename, cmd_line, argc, argv,
&new_argv, num_procs)) {
break;
}
}
/* If we didn't find one, abort */
if (NULL == lines[i]) {
orte_show_help("help-orterun.txt", "debugger-not-found", true);
exit(1);
}
opal_argv_free(lines);
/* We found one */
/* cleanup the MPIR arrays in case the debugger doesn't set them */
memset((char*)MPIR_executable_path, 0, MPIR_MAX_PATH_LENGTH);
memset((char*)MPIR_server_arguments, 0, MPIR_MAX_ARG_LENGTH);
/* Set an MCA param so that everyone knows that they are being
launched under a debugger; not all debuggers are consistent
about setting MPIR_being_debugged in both the launcher and the
MPI processes */
env_name = mca_base_param_environ_variable("orte",
"in_parallel_debugger", NULL);
if (NULL != env_name) {
opal_setenv(env_name, "1", true, &environ);
free(env_name);
}
/* Launch the debugger */
execvp(new_argv[0], new_argv);
value = opal_argv_join(new_argv, ' ');
orte_show_help("help-orterun.txt", "debugger-exec-failed",
true, basename, value, new_argv[0]);
free(value);
opal_argv_free(new_argv);
exit(1);
}
int orcm_init_util(void)
{
int ret, i;
char *error;
char *destdir, *tmp, *mcp, *new_mcp;
/* Setup OPAL */
if( ORTE_SUCCESS != (ret = opal_init(NULL, NULL)) ) {
error = "opal_init_util";
goto error;
}
/* register handler for errnum -> string conversion */
opal_error_register("OPENRCM", ORCM_ERR_BASE, ORCM_ERR_MAX, orcm_err2str);
/* register where the OPENRCM show_help files are located */
if (NULL != (destdir = getenv("ORCM_DESTDIR"))) {
asprintf(&tmp, "%s%s", destdir, ORCM_PKGHELPDIR);
} else {
tmp = strdup(ORCM_PKGHELPDIR);
}
if (ORTE_SUCCESS != (ret = opal_show_help_add_dir(tmp))) {
error = "register show_help_dir";
goto error;
}
free(tmp);
/* Add ORCM's component directory into the
mca_base_param_component_path */
i = mca_base_param_find("mca", NULL, "component_path");
if (i < 0) {
ret = ORCM_ERR_NOT_FOUND;
error = "Could not find mca_component_path";
goto error;
}
mca_base_param_lookup_string(i, &mcp);
if (NULL == mcp) {
ret = ORCM_ERR_NOT_FOUND;
error = "Could not find mca_component_path";
goto error;
}
if (NULL != destdir) {
asprintf(&new_mcp, "%s%s:%s", destdir, ORCM_PKGLIBDIR, mcp);
} else {
asprintf(&new_mcp, "%s:%s", ORCM_PKGLIBDIR, mcp);
}
mca_base_param_set_string(i, new_mcp);
free(new_mcp);
free(mcp);
orcm_util_initialized = true;
return ORCM_SUCCESS;
error:
if (ORCM_ERR_SILENT != ret) {
orte_show_help("help-openrcm-runtime.txt",
"orcm_init:startup:internal-failure",
true, error, ORTE_ERROR_NAME(ret), ret);
}
return ret;
}
/*
* Open up all directories in a given path and search for components of
* the specified type (and possibly of a given name).
*
* Note that we use our own path iteration functionality (vs. ltdl's
* lt_dladdsearchdir() functionality) because we need to look at
* companion .ompi_info files in the same directory as the library to
* generate dependencies, etc. If we use the plain lt_dlopen()
* functionality, we would not get the directory name of the file
* finally opened in recursive dependency traversals.
*/
static void find_dyn_components(const char *path, const char *type_name,
const char **name, bool include_mode,
opal_list_t *found_components)
{
ltfn_data_holder_t params;
char *path_to_use, *dir, *end;
component_file_item_t *file;
opal_list_item_t *cur;
strncpy(params.type, type_name, MCA_BASE_MAX_TYPE_NAME_LEN);
params.type[MCA_BASE_MAX_TYPE_NAME_LEN] = '\0';
params.name[0] = '\0';
/* If path is NULL, iterate over the set of directories specified by
the MCA param mca_base_component_path. If path is not NULL, then
use that as the path. */
if (NULL == path) {
mca_base_param_lookup_string(mca_base_param_component_path, &path_to_use);
if (NULL == path_to_use) {
/* If there's no path, then there's nothing to search -- we're
done */
return;
}
} else {
path_to_use = strdup(path);
}
/* Iterate over all the files in the directories in the path and
make a master array of all the matching filenames that we
find. */
OBJ_CONSTRUCT(&found_files, opal_list_t);
dir = path_to_use;
if (NULL != dir) {
do {
end = strchr(dir, OPAL_ENV_SEP);
if (NULL != end) {
*end = '\0';
}
if (0 != lt_dlforeachfile(dir, save_filename, ¶ms)) {
break;
}
dir = end + 1;
} while (NULL != end);
}
/* Iterate through all the filenames that we found. Since one
component may [try to] call another to be loaded, only try to load
the UNVISITED files. Also, ignore the return code -- basically,
give every file one chance to try to load. If they load, great.
If not, great. */
for (cur = opal_list_get_first(&found_files);
opal_list_get_end(&found_files) != cur;
cur = opal_list_get_next(cur)) {
file = (component_file_item_t *) cur;
if( UNVISITED == file->status ) {
bool op = true;
file->status = CHECKING_CYCLE;
op = use_component(include_mode, name, file->name);
if( true == op ) {
open_component(file, found_components);
}
}
}
/* So now we have a final list of loaded components. We can free all
the file information. */
for (cur = opal_list_remove_first(&found_files);
NULL != cur;
cur = opal_list_remove_first(&found_files)) {
OBJ_RELEASE(cur);
}
/* All done, now let's cleanup */
free(path_to_use);
OBJ_DESTRUCT(&found_files);
}
/*
* Open up all directories in a given path and search for components of
* the specified type (and possibly of a given name).
*
* Note that we use our own path iteration functionality (vs. ltdl's
* lt_dladdsearchdir() functionality) because we need to look at
* companion .ompi_info files in the same directory as the library to
* generate dependencies, etc. If we use the plain lt_dlopen()
* functionality, we would not get the directory name of the file
* finally opened in recursive dependency traversals.
*/
static void find_dyn_components(const char *path, const char *type_name,
const char *name,
opal_list_t *found_components)
{
ltfn_data_holder_t params;
char *path_to_use, *dir, *end, *param;
component_file_item_t *file;
opal_list_item_t *cur;
strcpy(params.type, type_name);
if (NULL == name) {
params.name[0] = '\0';
opal_output_verbose(40, 0, "mca: base: component_find: looking for all dynamic %s MCA components",
type_name, NULL);
} else {
strcpy(params.name, name);
opal_output_verbose(40, 0,
"mca: base: component_find: looking for dynamic %s MCA component named \"%s\"",
type_name, name, NULL);
}
/* If path is NULL, iterate over the set of directories specified by
the MCA param mca_base_component_path. If path is not NULL, then
use that as the path. */
param = NULL;
if (NULL == path) {
mca_base_param_lookup_string(mca_base_param_component_path, ¶m);
if (NULL == param) {
/* If there's no path, then there's nothing to search -- we're
done */
return;
} else {
path_to_use = strdup(param);
}
} else {
path_to_use = strdup(path);
}
/* Iterate over all the files in the directories in the path and
make a master array of all the matching filenames that we
find. */
OBJ_CONSTRUCT(&found_files, opal_list_t);
dir = path_to_use;
if (NULL != dir) {
do {
end = strchr(dir, OPAL_ENV_SEP);
if (NULL != end) {
*end = '\0';
}
if (0 != lt_dlforeachfile(dir, save_filename, ¶ms)) {
break;
}
dir = end + 1;
} while (NULL != end);
}
/* Iterate through all the filenames that we found. Since one
component may [try to] call another to be loaded, only try to load
the UNVISITED files. Also, ignore the return code -- basically,
give every file one chance to try to load. If they load, great.
If not, great. */
for (cur = opal_list_get_first(&found_files);
opal_list_get_end(&found_files) != cur;
cur = opal_list_get_next(cur)) {
file = (component_file_item_t *) cur;
if (UNVISITED == file->status) {
open_component(file, found_components);
}
}
/* So now we have a final list of loaded components. We can free all
the file information. */
for (cur = opal_list_remove_first(&found_files);
NULL != cur;
cur = opal_list_remove_first(&found_files)) {
OBJ_RELEASE(cur);
}
/* All done */
if (NULL != param) {
free(param);
}
if (NULL != path_to_use) {
free(path_to_use);
}
OBJ_DESTRUCT(&found_files);
}
int
orte_sds_slurm_set_name(void)
{
int rc;
int id;
int vpid_start;
int num_procs;
char* name_string = NULL;
int slurm_nodeid;
/* start by getting our cellid, jobid, and vpid (which is the
starting vpid for the list of daemons) */
id = mca_base_param_register_string("ns", "nds", "name", NULL, NULL);
mca_base_param_lookup_string(id, &name_string);
if(name_string != NULL) {
if (ORTE_SUCCESS !=
(rc = orte_ns.convert_string_to_process_name(&(orte_process_info.my_name),
name_string))) {
ORTE_ERROR_LOG(rc);
free(name_string);
return rc;
}
free(name_string);
} else {
orte_cellid_t cellid;
orte_jobid_t jobid;
orte_vpid_t vpid;
char* cellid_string;
char* jobid_string;
char* vpid_string;
id = mca_base_param_register_string("ns", "nds", "cellid", NULL, NULL);
mca_base_param_lookup_string(id, &cellid_string);
if (NULL == cellid_string) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (rc = orte_ns.convert_string_to_cellid(&cellid, cellid_string))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
id = mca_base_param_register_string("ns", "nds", "jobid", NULL, NULL);
mca_base_param_lookup_string(id, &jobid_string);
if (NULL == jobid_string) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (rc = orte_ns.convert_string_to_jobid(&jobid, jobid_string))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
id = mca_base_param_register_string("ns", "nds", "vpid", NULL, NULL);
mca_base_param_lookup_string(id, &vpid_string);
if (NULL == vpid_string) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (rc = orte_ns.convert_string_to_vpid(&vpid, vpid_string))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
if (ORTE_SUCCESS != (rc = orte_ns.create_process_name(&(orte_process_info.my_name),
cellid,
jobid,
vpid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
/* fix up the base name and make it the "real" name */
slurm_nodeid = atoi(getenv("SLURM_NODEID"));
orte_process_info.my_name->vpid += slurm_nodeid;
/* fix up the system info nodename to match exactly what slurm returned */
if (NULL != orte_system_info.nodename) {
free(orte_system_info.nodename);
}
orte_system_info.nodename = get_slurm_nodename(slurm_nodeid);
id = mca_base_param_register_int("ns", "nds", "vpid_start", NULL, -1);
mca_base_param_lookup_int(id, &vpid_start);
if (vpid_start < 0) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
id = mca_base_param_register_int("ns", "nds", "num_procs", NULL, -1);
mca_base_param_lookup_int(id, &num_procs);
if (num_procs < 0) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
orte_process_info.vpid_start = (orte_vpid_t)vpid_start;
orte_process_info.num_procs = (size_t)num_procs;
return ORTE_SUCCESS;
}
int ompi_show_all_mca_params(int32_t rank, int requested, char *nodename) {
opal_list_t *info;
opal_list_item_t *i;
mca_base_param_info_t *item;
char *value_string;
int value_int;
FILE *fp = NULL;
time_t timestamp;
mca_base_param_source_t source;
char *src_file;
char *src_string;
if (rank != 0) {
return OMPI_SUCCESS;
}
timestamp = time(NULL);
/* Open the file if one is specified */
if (0 != strlen(ompi_mpi_show_mca_params_file)) {
if ( NULL == (fp = fopen(ompi_mpi_show_mca_params_file, "w")) ) {
opal_output(0, "Unable to open file <%s> to write MCA parameters", ompi_mpi_show_mca_params_file);
return OMPI_ERR_FILE_OPEN_FAILURE;
}
fprintf(fp, "#\n");
fprintf(fp, "# This file was automatically generated on %s", ctime(×tamp));
fprintf(fp, "# by MPI_COMM_WORLD rank %d (out of a total of %d) on %s\n", rank, requested, nodename );
fprintf(fp, "#\n");
}
mca_base_param_dump(&info, false);
for (i = opal_list_get_first(info);
i != opal_list_get_last(info);
i = opal_list_get_next(i)) {
item = (mca_base_param_info_t*) i;
/* If this is an internal param, don't print it */
if (item->mbpp_internal) {
continue;
}
/* get the source - where the param was last set */
if (OPAL_SUCCESS !=
mca_base_param_lookup_source(item->mbpp_index, &source, &src_file)) {
continue;
}
/* is this a default value and we are not displaying
* defaults, ignore this one
*/
if (MCA_BASE_PARAM_SOURCE_DEFAULT == source && !show_default_mca_params) {
continue;
}
/* is this a file value and we are not displaying files,
* ignore it
*/
if (MCA_BASE_PARAM_SOURCE_FILE == source && !show_file_mca_params) {
continue;
}
/* is this an enviro value and we are not displaying enviros,
* ignore it
*/
if (MCA_BASE_PARAM_SOURCE_ENV == source && !show_enviro_mca_params) {
continue;
}
/* is this an API value and we are not displaying APIs,
* ignore it
*/
if (MCA_BASE_PARAM_SOURCE_OVERRIDE == source && !show_override_mca_params) {
continue;
}
/* Get the parameter name, and convert it to a printable string */
if (MCA_BASE_PARAM_TYPE_STRING == item->mbpp_type) {
mca_base_param_lookup_string(item->mbpp_index, &value_string);
if (NULL == value_string) {
value_string = strdup("");
}
} else {
mca_base_param_lookup_int(item->mbpp_index, &value_int);
asprintf(&value_string, "%d", value_int);
}
switch(source) {
case MCA_BASE_PARAM_SOURCE_DEFAULT:
src_string = "default value";
break;
case MCA_BASE_PARAM_SOURCE_ENV:
src_string = "environment";
break;
case MCA_BASE_PARAM_SOURCE_FILE:
src_string = "file";
break;
case MCA_BASE_PARAM_SOURCE_OVERRIDE:
src_string = "API override";
break;
default:
src_string = NULL;
break;
}
/* Print the parameter */
if (0 != strlen(ompi_mpi_show_mca_params_file)) {
if (NULL == src_file) {
fprintf(fp, "%s=%s (%s)\n", item->mbpp_full_name, value_string,
(NULL != src_string ? src_string : "unknown"));
} else {
fprintf(fp, "%s=%s (%s:%s)\n", item->mbpp_full_name, value_string,
(NULL != src_string ? src_string : "unknown"), src_file);
}
} else {
if (NULL == src_file) {
opal_output(0, "%s=%s (%s)\n", item->mbpp_full_name, value_string,
(NULL != src_string ? src_string : "unknown"));
} else {
opal_output(0, "%s=%s (%s:%s)\n", item->mbpp_full_name, value_string,
(NULL != src_string ? src_string : "unknown"), src_file);
}
}
free(value_string);
}
/* Close file, cleanup allocated memory*/
if (0 != strlen(ompi_mpi_show_mca_params_file)) {
fclose(fp);
}
mca_base_param_dump_release(info);
return OMPI_SUCCESS;
}
int mca_fcoll_base_file_select (struct mca_io_ompio_file_t *file,
mca_base_component_t *preferred)
{
int priority;
int best_priority;
opal_list_item_t *item;
opal_list_item_t *next_item;
char *names, **name_array;
int num_names;
mca_base_component_priority_list_item_t *cpli;
mca_fcoll_base_component_t *component;
mca_fcoll_base_component_t *best_component;
mca_fcoll_base_module_t *module;
opal_list_t queried;
queried_module_t *om;
opal_list_t *selectable;
char *str;
int err = MPI_SUCCESS;
int i;
bool was_selectable_constructed = false;
/* Check and see if a preferred component was provided. If it was
provided then it should be used (if possible) */
if (NULL != preferred) {
/* We have a preferred component. Check if it is available
and if so, whether it wants to run */
str = &(preferred->mca_component_name[0]);
opal_output_verbose(10, mca_fcoll_base_output,
"fcoll:base:file_select: Checking preferred component: %s",
str);
/* query the component for its priority and get its module
structure. This is necessary to proceed */
component = (mca_fcoll_base_component_t *)preferred;
module = component->fcollm_file_query (file, &priority);
if (NULL != module &&
NULL != module->fcoll_module_init) {
/* this query seems to have returned something legitimate
* and we can now go ahead and initialize the
* file with it * but first, the functions which
* are null need to be filled in */
/*fill_null_pointers (module);*/
file->f_fcoll = module;
file->f_fcoll_component = preferred;
return module->fcoll_module_init(file);
}
/* His preferred component is present, but is unable to
* run. This is not a good sign. We should try selecting
* some other component We let it fall through and select
* from the list of available components
*/
} /*end of selection for preferred component */
/*
* We fall till here if one of the two things happened:
* 1. The preferred component was provided but for some reason was
* not able to be selected
* 2. No preferred component was provided
*
* All we need to do is to go through the list of available
* components and find the one which has the highest priority and
* use that for this file
*/
/* Check if anything was requested by means on the name parameters */
names = NULL;
mca_base_param_lookup_string (mca_fcoll_base_param, &names);
if (NULL != names && 0 < strlen(names)) {
name_array = opal_argv_split (names, ',');
num_names = opal_argv_count (name_array);
opal_output_verbose(10, mca_fcoll_base_output,
"fcoll:base:file_Select: Checking all available module");
/* since there are somethings which the mca requested through the
if the intersection is NULL, then we barf saying that the requested
modules are not being available */
selectable = OBJ_NEW(opal_list_t);
was_selectable_constructed = true;
/* go through the compoents_available list and check against the names
* to see whether this can be added or not */
for (item = opal_list_get_first(&mca_fcoll_base_components_available);
item != opal_list_get_end(&mca_fcoll_base_components_available);
item = opal_list_get_next(item)) {
/* convert the opal_list_item_t returned into the proper type */
cpli = (mca_base_component_priority_list_item_t *) item;
component = (mca_fcoll_base_component_t *) cpli->super.cli_component;
opal_output_verbose(10, mca_fcoll_base_output,
"select: initialising %s component %s",
component->fcollm_version.mca_type_name,
component->fcollm_version.mca_component_name);
/* check if this name is present in the mca_base_params */
for (i=0; i < num_names; i++) {
if (0 == strcmp(name_array[i], component->fcollm_version.mca_component_name)) {
/* this is present, and should be added o the selectable list */
/* We need to create a seperate object to initialise this list with
* since we cannot have the same item in 2 lists */
module = component->fcollm_file_query (file, &priority);
if (NULL != module &&
NULL != module->fcoll_module_init) {
file->f_fcoll = module;
file->f_fcoll_component = (mca_base_component_t *)component;
return module->fcoll_module_init(file);
}
/*
selectable_item = OBJ_NEW (mca_base_component_priority_list_item_t);
*selectable_item = *cpli;
opal_list_append (selectable, (opal_list_item_t *)selectable_item);
break;*/
}
}
}
/* check for a NULL intersection between the available list and the
* list which was asked for */
if (0 == opal_list_get_size(selectable)) {
was_selectable_constructed = true;
OBJ_RELEASE (selectable);
opal_output_verbose (10, mca_fcoll_base_output,
"fcoll:base:file_select: preferred modules were not available");
return OMPI_ERROR;
}
} else { /* if there was no name_array, then we need to simply initialize
selectable to mca_fcoll_base_components_available */
selectable = &mca_fcoll_base_components_available;
}
best_component = NULL;
best_priority = -1;
OBJ_CONSTRUCT(&queried, opal_list_t);
for (item = opal_list_get_first(selectable);
item != opal_list_get_end(selectable);
item = opal_list_get_next(item)) {
/*
* convert the opal_list_item_t returned into the proper type
*/
cpli = (mca_base_component_priority_list_item_t *) item;
component = (mca_fcoll_base_component_t *) cpli->super.cli_component;
opal_output_verbose(10, mca_fcoll_base_output,
"select: initialising %s component %s",
component->fcollm_version.mca_type_name,
component->fcollm_version.mca_component_name);
/*
* we can call the query function only if there is a function :-)
*/
if (NULL == component->fcollm_file_query) {
opal_output_verbose(10, mca_fcoll_base_output,
"select: no query, ignoring the component");
} else {
/*
* call the query function and see what it returns
*/
module = component->fcollm_file_query (file, &priority);
if (NULL == module ||
NULL == module->fcoll_module_init) {
/*
* query did not return any action which can be used
*/
opal_output_verbose(10, mca_fcoll_base_output,
"select: query returned failure");
} else {
opal_output_verbose(10, mca_fcoll_base_output,
"select: query returned priority %d",
priority);
/*
* is this the best component we have found till now?
*/
if (priority > best_priority) {
best_priority = priority;
best_component = component;
}
om = OBJ_NEW(queried_module_t);
/*
* check if we have run out of space
*/
if (NULL == om) {
OBJ_DESTRUCT(&queried);
return OMPI_ERR_OUT_OF_RESOURCE;
}
om->om_component = component;
om->om_module = module;
opal_list_append(&queried, (opal_list_item_t *)om);
} /* end else of if (NULL == module) */
} /* end else of if (NULL == component->fcollm_init) */
} /* end for ... end of traversal */
/* We have to remove empty out the selectable list if the selectable
* list was constructed as a duplicate and not as a pointer to the
* mca_base_components_available list. So, check and destroy */
if (was_selectable_constructed) {
/* remove all the items first */
for (item = opal_list_get_first(&mca_fcoll_base_components_available);
item != opal_list_get_end(&mca_fcoll_base_components_available);
item = next_item) {
next_item = opal_list_get_next(item);
OBJ_RELEASE (item);
}
/* release the list itself */
OBJ_RELEASE (selectable);
was_selectable_constructed = false;
}
/*
* Now we have alist of components which successfully returned
* their module struct. One of these components has the best
* priority. The rest have to be comm_unqueried to counter the
* effects of file_query'ing them. Finalize happens only on
* components which should are initialized.
*/
if (NULL == best_component) {
/*
* This typically means that there was no component which was
* able to run properly this time. So, we need to abort
*/
OBJ_DESTRUCT(&queried);
return OMPI_ERROR;
}
/*
* We now have a list of components which have successfully
* returned their priorities from the query. We now have to
* unquery() those components which have not been selected and
* init() the component which was selected
*/
for (item = opal_list_remove_first(&queried);
NULL != item;
item = opal_list_remove_first(&queried)) {
om = (queried_module_t *) item;
if (om->om_component == best_component) {
/*
* this is the chosen component, we have to initialise the
* module of this component.
*
* ANJU: a component might not have all the functions
* defined. Whereever a function pointer is null in the
* module structure we need to fill it in with the base
* structure function pointers. This is yet to be done
*/
/*
* We don return here coz we still need to go through and
* elease the other objects
*/
/*fill_null_pointers (om->om_module);*/
file->f_fcoll = om->om_module;
err = om->om_module->fcoll_module_init(file);
file->f_fcoll_component = (mca_base_component_t *)best_component;
/*
printf ("SELECTED: %s\n", best_component->fcollm_version.mca_component_name);
*/
} else {
/*
* this is not the "choosen one", finalize
*/
if (NULL != om->om_component->fcollm_file_unquery) {
/* unquery the component only if they have some clean
* up job to do. Components which are queried but do
* not actually do anything typically do not have a
* unquery. Hence this check is necessary
*/
(void) om->om_component->fcollm_file_unquery(file);
opal_output_verbose(10, mca_fcoll_base_output,
"select: component %s is not selected",
om->om_component->fcollm_version.mca_component_name);
} /* end if */
} /* if not best component */
OBJ_RELEASE(om);
} /* traversing through the entire list */
opal_output_verbose(10, mca_fcoll_base_output,
"select: component %s selected",
best_component->fcollm_version.mca_component_name);
OBJ_DESTRUCT(&queried);
return err;
}
int mca_io_base_delete(char *filename, struct ompi_info_t *info)
{
int err, num_names;
char *names, **name_array;
opal_list_t *selectable;
opal_list_item_t *item;
avail_io_t *avail, selected;
/* Announce */
opal_output_verbose(10, mca_io_base_output,
"io:base:delete: deleting file: %s",
filename);
/* See if a set of component was requested by the MCA parameter.
Don't check for error. */
names = NULL;
mca_base_param_lookup_string(mca_io_base_param, &names);
/* Compute the intersection of all of my available components with
the components from all the other processes in this file */
/* JMS CONTINUE HERE */
/* See if there were any listed in the MCA parameter; parse them
and check them all */
err = OMPI_ERROR;
if (NULL != names && 0 < strlen(names)) {
name_array = opal_argv_split(names, ',');
num_names = opal_argv_count(name_array);
opal_output_verbose(10, mca_io_base_output,
"io:base:delete: Checking specific modules: %s",
names);
selectable = check_components(&mca_io_base_components_available,
filename, info, name_array, num_names);
opal_argv_free(name_array);
}
/* Nope -- a specific [set of] component[s] was not requested. Go
check them all. */
else {
opal_output_verbose(10, mca_io_base_output,
"io:base:delete: Checking all available modules");
selectable = check_components(&mca_io_base_components_available,
filename, info, NULL, 0);
}
/* Upon return from the above, the modules list will contain the
list of modules that returned (priority >= 0). If we have no
io modules available, it's an error */
if (NULL == selectable) {
/* There's no modules available. Doh! */
/* show_help */
return OMPI_ERROR;
}
/* Do some kind of collective operation to find a module that
everyone has available */
#if 1
/* For the moment, just take the top module off the list */
/* MSC actually take the buttom */
item = opal_list_remove_last(selectable);
avail = (avail_io_t *) item;
selected = *avail;
OBJ_RELEASE(avail);
#else
/* JMS CONTINUE HERE */
#endif
/* Everything left in the selectable list is therefore unwanted,
and we call their unquery() method (because they all had
query() invoked, but will never have init() invoked in this
scope). */
for (item = opal_list_remove_first(selectable); item != NULL;
item = opal_list_remove_first(selectable)) {
avail = (avail_io_t *) item;
unquery(avail, filename, info);
OBJ_RELEASE(item);
}
OBJ_RELEASE(selectable);
/* Finally -- delete the file with the selected component */
if (OMPI_SUCCESS != (err = delete_file(&selected, filename, info))) {
return err;
}
/* Announce the winner */
opal_output_verbose(10, mca_io_base_output,
"io:base:delete: Selected io component %s",
selected.ai_component.v2_0_0.io_version.mca_component_name);
return OMPI_SUCCESS;
}
void ompi_info_show_mca_params(opal_list_t *info,
const char *type, const char *component,
bool want_internal)
{
opal_list_item_t *i;
mca_base_param_info_t *p;
char *value_string, *empty = "";
char *message, *content, *tmp;
int value_int, j;
mca_base_param_source_t source;
char *src_file;
for (i = opal_list_get_first(info); i != opal_list_get_last(info);
i = opal_list_get_next(i)) {
p = (mca_base_param_info_t*) i;
if (NULL != p->mbpp_type_name && 0 == strcmp(type, p->mbpp_type_name)) {
if (0 == strcmp(component, ompi_info_component_all) ||
NULL == p->mbpp_component_name ||
(NULL != p->mbpp_component_name &&
0 == strcmp(component, p->mbpp_component_name))) {
/* Find the source of the value */
if (OPAL_SUCCESS !=
mca_base_param_lookup_source(p->mbpp_index, &source, &src_file)) {
continue;
}
/* Make a char *for the default value. Invoke a
* lookup because it may transform the char *("~/" ->
* "<home dir>/") or get the value from the
* environment, a file, etc.
*/
if (MCA_BASE_PARAM_TYPE_STRING == p->mbpp_type) {
mca_base_param_lookup_string(p->mbpp_index,
&value_string);
/* Can't let the char *be NULL because we
* assign it to a std::string, below
*/
if (NULL == value_string) {
value_string = strdup(empty);
}
} else {
mca_base_param_lookup_int(p->mbpp_index, &value_int);
asprintf(&value_string, "%d", value_int);
}
/* Build up the strings to ompi_info_output. */
if (ompi_info_pretty) {
asprintf(&message, "MCA %s", p->mbpp_type_name);
/* Put in the real, full name (which may be
* different than the categorization).
*/
asprintf(&content, "%s \"%s\" (%s: <%s>, data source: ",
p->mbpp_read_only ? "information" : "parameter",
p->mbpp_full_name,
p->mbpp_read_only ? "value" : "current value",
(0 == strlen(value_string)) ? "none" : value_string);
/* Indicate where the param was set from */
switch(source) {
case MCA_BASE_PARAM_SOURCE_DEFAULT:
asprintf(&tmp, "%sdefault value", content);
free(content);
content = tmp;
break;
case MCA_BASE_PARAM_SOURCE_ENV:
asprintf(&tmp, "%senvironment or cmdline", content);
free(content);
content = tmp;
break;
case MCA_BASE_PARAM_SOURCE_FILE:
asprintf(&tmp, "%sfile [%s]", content, src_file);
free(content);
content = tmp;
break;
case MCA_BASE_PARAM_SOURCE_OVERRIDE:
asprintf(&tmp, "%sAPI override", content);
free(content);
content = tmp;
break;
default:
break;
}
/* Is this parameter deprecated? */
if (p->mbpp_deprecated) {
asprintf(&tmp, "%s, deprecated", content);
free(content);
content = tmp;
}
/* Does this parameter have any synonyms? */
if (p->mbpp_synonyms_len > 0) {
asprintf(&tmp, "%s, synonyms: ", content);
free(content);
content = tmp;
for (j = 0; j < p->mbpp_synonyms_len; ++j) {
if (j > 0) {
asprintf(&tmp, "%s, %s", content, p->mbpp_synonyms[j]->mbpp_full_name);
free(content);
content = tmp;
} else {
asprintf(&tmp, "%s%s", content, p->mbpp_synonyms[j]->mbpp_full_name);
free(content);
content = tmp;
}
}
}
/* Is this parameter a synonym of something else? */
else if (NULL != p->mbpp_synonym_parent) {
asprintf(&tmp, "%s, synonym of: %s", content, p->mbpp_synonym_parent->mbpp_full_name);
free(content);
content = tmp;
}
asprintf(&tmp, "%s)", content);
free(content);
content = tmp;
ompi_info_out(message, message, content);
free(message);
free(content);
/* If we have a help message, ompi_info_output it */
if (NULL != p->mbpp_help_msg) {
ompi_info_out("", "", p->mbpp_help_msg);
}
} else {
/* build the message*/
asprintf(&tmp, "mca:%s:%s:param:%s:", p->mbpp_type_name,
(NULL == p->mbpp_component_name) ? "base" : p->mbpp_component_name,
p->mbpp_full_name);
/* Output the value */
asprintf(&message, "%svalue", tmp);
ompi_info_out(message, message, value_string);
free(message);
/* Indicate where the param was set from */
asprintf(&message, "%sdata_source", tmp);
switch(source) {
case MCA_BASE_PARAM_SOURCE_DEFAULT:
content = strdup("default value");
break;
case MCA_BASE_PARAM_SOURCE_ENV:
content = strdup("environment-cmdline");
break;
case MCA_BASE_PARAM_SOURCE_FILE:
asprintf(&content, "file: %s", src_file);
break;
case MCA_BASE_PARAM_SOURCE_OVERRIDE:
content = strdup("API override");
break;
default:
break;
}
ompi_info_out(message, message, content);
free(message);
free(content);
/* Output whether it's read only or writable */
asprintf(&message, "%sstatus", tmp);
content = p->mbpp_read_only ? "read-only" : "writable";
ompi_info_out(message, message, content);
free(message);
/* If it has a help message, ompi_info_output that */
if (NULL != p->mbpp_help_msg) {
asprintf(&message, "%shelp", tmp);
content = p->mbpp_help_msg;
ompi_info_out(message, message, content);
free(message);
}
/* Is this parameter deprecated? */
asprintf(&message, "%sdeprecated", tmp);
content = p->mbpp_deprecated ? "yes" : "no";
ompi_info_out(message, message, content);
free(message);
/* Does this parameter have any synonyms? */
if (p->mbpp_synonyms_len > 0) {
for (j = 0; j < p->mbpp_synonyms_len; ++j) {
asprintf(&message, "%ssynonym:name", tmp);
content = p->mbpp_synonyms[j]->mbpp_full_name;
ompi_info_out(message, message, content);
free(message);
}
}
/* Is this parameter a synonym of something else? */
else if (NULL != p->mbpp_synonym_parent) {
asprintf(&message, "%ssynonym_of:name", tmp);
content = p->mbpp_synonym_parent->mbpp_full_name;
ompi_info_out(message, message, content);
free(message);
}
}
/* If we allocated the string, then free it */
if (NULL != value_string) {
free(value_string);
}
}
}
}
}
int mca_bml_r2_ft_event(int state)
{
static bool first_continue_pass = false;
ompi_proc_t** procs = NULL;
size_t num_procs;
size_t btl_idx;
int ret, p;
int loc_state;
int param_type = -1;
char *param_list = NULL;
if(OPAL_CRS_CHECKPOINT == state) {
/* Do nothing for now */
}
else if(OPAL_CRS_CONTINUE == state) {
first_continue_pass = !first_continue_pass;
/* Since nothing in Checkpoint, we are fine here (unless required by BTL) */
if( ompi_cr_continue_like_restart && !first_continue_pass) {
procs = ompi_proc_all(&num_procs);
if(NULL == procs) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
}
else if(OPAL_CRS_RESTART_PRE == state ) {
/* Nothing here */
}
else if(OPAL_CRS_RESTART == state ) {
procs = ompi_proc_all(&num_procs);
if(NULL == procs) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
else if(OPAL_CRS_TERM == state ) {
;
}
else {
;
}
/* Never call the ft_event functions attached to the BTLs on the second
* pass of RESTART since on the first pass they were unloaded and therefore
* no longer exist.
*/
if( OPAL_CRS_RESTART != state ) {
if( OPAL_CRS_CONTINUE == state && !first_continue_pass ) {
;
} else {
/* Since we only ever call into the BTLs once during the first restart
* pass, just lie to them on this pass for a bit of local clarity.
*/
if( OPAL_CRS_RESTART_PRE == state ) {
loc_state = OPAL_CRS_RESTART;
} else {
loc_state = state;
}
/*
* Call ft_event in:
* - BTL modules
* - MPool modules
*
* These should be cleaning out stale state, and memory references in
* preparation for being shut down.
*/
for(btl_idx = 0; btl_idx < mca_bml_r2.num_btl_modules; btl_idx++) {
/*
* Notify Mpool
*/
if( NULL != (mca_bml_r2.btl_modules[btl_idx])->btl_mpool &&
NULL != (mca_bml_r2.btl_modules[btl_idx])->btl_mpool->mpool_ft_event ) {
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event: Notify the %s MPool.\n",
(mca_bml_r2.btl_modules[btl_idx])->btl_mpool->mpool_component->mpool_version.mca_component_name);
if(OMPI_SUCCESS != (ret = (mca_bml_r2.btl_modules[btl_idx])->btl_mpool->mpool_ft_event(loc_state) ) ) {
continue;
}
}
/*
* Notify BTL
*/
if( NULL != (mca_bml_r2.btl_modules[btl_idx])->btl_ft_event) {
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event: Notify the %s BTL.\n",
(mca_bml_r2.btl_modules[btl_idx])->btl_component->btl_version.mca_component_name);
if(OMPI_SUCCESS != (ret = (mca_bml_r2.btl_modules[btl_idx])->btl_ft_event(loc_state) ) ) {
continue;
}
}
}
} /* OPAL_CRS_CONTINUE == state && !first_continue_pass */
}
if(OPAL_CRS_CHECKPOINT == state) {
;
}
else if(OPAL_CRS_CONTINUE == state) {
/* Matches OPAL_CRS_RESTART_PRE */
if( ompi_cr_continue_like_restart && first_continue_pass) {
if( OMPI_SUCCESS != (ret = mca_bml_r2_finalize()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to finalize BML framework\n");
return ret;
}
}
/* Matches OPAL_CRS_RESTART */
else if( ompi_cr_continue_like_restart && !first_continue_pass ) {
/*
* Barrier to make all processes have been successfully restarted before
* we try to remove some restart only files.
*/
if (OMPI_SUCCESS != (ret = orte_grpcomm.barrier())) {
opal_output(0, "bml:r2: ft_event(Restart): Failed in orte_grpcomm.barrier (%d)", ret);
return ret;
}
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event(Restart): Cleanup restart files\n");
opal_crs_base_cleanup_flush();
/*
* Re-open the BTL framework to get the full list of components.
*/
if( OMPI_SUCCESS != (ret = mca_btl_base_open()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to open BTL framework\n");
return ret;
}
/*
* Re-select the BTL components/modules
* This will cause the BTL components to discover the available
* network options on this machine, and post proper modex informaiton.
*/
if( OMPI_SUCCESS != (ret = mca_btl_base_select(OMPI_ENABLE_PROGRESS_THREADS,
OMPI_ENABLE_MPI_THREADS) ) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to select in BTL framework\n");
return ret;
}
/*
* Clear some structures so we can properly repopulate them
*/
mca_bml_r2.btls_added = false;
for(p = 0; p < (int)num_procs; ++p) {
if( NULL != procs[p]->proc_bml) {
OBJ_RELEASE(procs[p]->proc_bml);
procs[p]->proc_bml = NULL;
}
OBJ_RELEASE(procs[p]);
}
if( NULL != procs ) {
free(procs);
procs = NULL;
}
}
}
else if(OPAL_CRS_RESTART_PRE == state ) {
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event(Restart): Finalize BML\n");
/*
* Finalize the BML
* - Flush progress functions
* - Flush module references
* - mca_btl_base_close()
* Need to do this because we may have BTL components that were
* unloaded in the first selection that may be available now.
* Conversely we may have BTL components loaded now that
* are not available now.
*/
if( OMPI_SUCCESS != (ret = mca_bml_r2_finalize()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to finalize BML framework\n");
return ret;
}
}
else if(OPAL_CRS_RESTART == state ) {
/*
* Barrier to make all processes have been successfully restarted before
* we try to remove some restart only files.
*/
if (OMPI_SUCCESS != (ret = orte_grpcomm.barrier())) {
opal_output(0, "bml:r2: ft_event(Restart): Failed in orte_grpcomm.barrier (%d)", ret);
return ret;
}
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event(Restart): Cleanup restart files\n");
opal_crs_base_cleanup_flush();
/*
* Re-open the BTL framework to get the full list of components.
* - but first clear the MCA value that was there
*/
param_type = mca_base_param_find("btl", NULL, NULL);
mca_base_param_lookup_string(param_type, ¶m_list);
opal_output_verbose(11, ompi_cr_output,
"Restart (Previous BTL MCA): <%s>\n", param_list);
if( NULL != param_list ) {
free(param_list);
param_list = NULL;
}
/* Deregister the old value, and refresh the file cache to grab any updates */
mca_base_param_deregister(param_type);
mca_base_param_recache_files(false);
if( OMPI_SUCCESS != (ret = mca_btl_base_open()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to open BTL framework\n");
return ret;
}
param_type = mca_base_param_find("btl", NULL, NULL);
mca_base_param_lookup_string(param_type, ¶m_list);
opal_output_verbose(11, ompi_cr_output,
"Restart (New BTL MCA): <%s>\n", param_list);
if( NULL != param_list ) {
free(param_list);
param_list = NULL;
}
/*
* Re-select the BTL components/modules
* This will cause the BTL components to discover the available
* network options on this machine, and post proper modex informaiton.
*/
if( OMPI_SUCCESS != (ret = mca_btl_base_select(OMPI_ENABLE_PROGRESS_THREADS,
OMPI_ENABLE_MPI_THREADS) ) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to select in BTL framework\n");
return ret;
}
/*
* Clear some structures so we can properly repopulate them
*/
mca_bml_r2.btls_added = false;
for(p = 0; p < (int)num_procs; ++p) {
if( NULL != procs[p]->proc_bml) {
OBJ_RELEASE(procs[p]->proc_bml);
procs[p]->proc_bml = NULL;
}
OBJ_RELEASE(procs[p]);
}
if( NULL != procs ) {
free(procs);
procs = NULL;
}
}
else if(OPAL_CRS_TERM == state ) {
;
}
else {
;
}
return OMPI_SUCCESS;
}