/*
* Where did OMPI bind this process? (layout string)
*/
static int get_layout_ompi_bound(char str[OMPI_AFFINITY_STRING_MAX])
{
int ret;
/* If OMPI did not bind, indicate that */
if (!ompi_rte_proc_is_bound) {
strncpy(str, ompi_nobind_str, OMPI_AFFINITY_STRING_MAX - 1);
return OMPI_SUCCESS;
}
/* Find out what OMPI bound us to and prettyprint it */
if (NULL == orte_proc_applied_binding) {
ret = OPAL_ERR_NOT_BOUND;
} else {
ret = opal_hwloc_base_cset2mapstr(str, OMPI_AFFINITY_STRING_MAX,
opal_hwloc_topology,
orte_proc_applied_binding);
}
if (OPAL_ERR_NOT_BOUND == ret) {
strncpy(str, not_bound_str, OMPI_AFFINITY_STRING_MAX - 1);
ret = OMPI_SUCCESS;
}
return ret;
}
/*
* Where is this process currently bound? (layout string)
*/
static int get_layout_current_binding(char str[OMPI_AFFINITY_STRING_MAX])
{
int ret;
hwloc_obj_t root;
hwloc_cpuset_t boundset, rootset;
bool bound = false;
/* get our root object */
root = hwloc_get_root_obj(opal_hwloc_topology);
rootset = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, root);
/* get our bindings */
boundset = hwloc_bitmap_alloc();
if (hwloc_get_cpubind(opal_hwloc_topology, boundset,
HWLOC_CPUBIND_PROCESS) < 0) {
/* we are NOT bound if get_cpubind fails, nor can we be bound
- the environment does not support it */
bound = false;
} else {
/* we are bound if the two cpusets are not equal, or if there
is only ONE PU available to us */
if (0 != hwloc_bitmap_compare(boundset, rootset) ||
opal_hwloc_base_single_cpu(rootset) ||
opal_hwloc_base_single_cpu(boundset)) {
bound = true;
}
}
/* If we are not bound, indicate that */
if (!bound) {
strncat(str, not_bound_str, OMPI_AFFINITY_STRING_MAX - 1);
ret = OMPI_SUCCESS;
}
/* If we are bound, print it out */
else {
ret = opal_hwloc_base_cset2mapstr(str, OMPI_AFFINITY_STRING_MAX,
opal_hwloc_topology,
boundset);
if (OPAL_ERR_NOT_BOUND == ret) {
strncpy(str, not_bound_str, OMPI_AFFINITY_STRING_MAX - 1);
ret = OMPI_SUCCESS;
}
}
hwloc_bitmap_free(boundset);
return ret;
}
static int bind_downwards(orte_job_t *jdata,
orte_node_t *node,
hwloc_obj_type_t target,
unsigned cache_level)
{
int j;
orte_job_map_t *map;
orte_proc_t *proc;
hwloc_obj_t trg_obj, nxt_obj;
hwloc_cpuset_t cpus;
unsigned int ncpus;
opal_hwloc_obj_data_t *data;
int total_cpus;
hwloc_cpuset_t totalcpuset;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: bind downward for job %s with bindings %s",
ORTE_JOBID_PRINT(jdata->jobid),
opal_hwloc_base_print_binding(jdata->map->binding));
/* initialize */
map = jdata->map;
totalcpuset = hwloc_bitmap_alloc();
/* cycle thru the procs */
for (j=0; j < node->procs->size; j++) {
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(node->procs, j))) {
continue;
}
/* ignore procs from other jobs */
if (proc->name.jobid != jdata->jobid) {
continue;
}
/* ignore procs that have already been bound - should
* never happen, but safer
*/
if (NULL != proc->cpu_bitmap) {
continue;
}
/* we don't know if the target is a direct child of this locale,
* or if it is some depth below it, so we have to conduct a bit
* of a search. Let hwloc find the min usage one for us.
*/
trg_obj = opal_hwloc_base_find_min_bound_target_under_obj(node->topology,
proc->locale,
target, cache_level);
if (NULL == trg_obj) {
/* there aren't any such targets under this object */
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
/* record the location */
proc->bind_location = trg_obj;
/* start with a clean slate */
hwloc_bitmap_zero(totalcpuset);
total_cpus = 0;
nxt_obj = trg_obj;
do {
if (NULL == nxt_obj) {
/* could not find enough cpus to meet request */
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
trg_obj = nxt_obj;
/* get the number of cpus under this location */
ncpus = opal_hwloc_base_get_npus(node->topology, trg_obj);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s GOT %d CPUS",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ncpus);
/* track the number bound */
if (NULL == (data = (opal_hwloc_obj_data_t*)trg_obj->userdata)) {
data = OBJ_NEW(opal_hwloc_obj_data_t);
trg_obj->userdata = data;
}
data->num_bound++;
/* error out if adding a proc would cause overload and that wasn't allowed,
* and it wasn't a default binding policy (i.e., the user requested it)
*/
if (ncpus < data->num_bound &&
!OPAL_BIND_OVERLOAD_ALLOWED(jdata->map->binding)) {
if (OPAL_BINDING_POLICY_IS_SET(jdata->map->binding)) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:binding-overload", true,
opal_hwloc_base_print_binding(map->binding), node->name,
data->num_bound, ncpus);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
} else {
/* if this is the default binding policy, then just don't
* bind this proc
*/
data->num_bound--; // maintain count
/* show the proc as not bound */
proc->bind_location = NULL;
hwloc_bitmap_zero(totalcpuset);
break;
}
}
/* bind the proc here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, trg_obj);
hwloc_bitmap_or(totalcpuset, totalcpuset, cpus);
/* track total #cpus */
total_cpus += ncpus;
/* move to the next location, in case we need it */
nxt_obj = trg_obj->next_cousin;
} while (total_cpus < orte_rmaps_base.cpus_per_rank);
hwloc_bitmap_list_asprintf(&proc->cpu_bitmap, totalcpuset);
if (4 < opal_output_get_verbosity(orte_rmaps_base_framework.framework_output)) {
char tmp1[1024], tmp2[1024];
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1),
node->topology, totalcpuset)) {
opal_output(orte_rmaps_base_framework.framework_output,
"%s PROC %s ON %s IS NOT BOUND",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), node->topology, totalcpuset);
opal_output(orte_rmaps_base_framework.framework_output,
"%s BOUND PROC %s[%s] TO %s: %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name,
tmp1, tmp2);
}
}
}
hwloc_bitmap_free(totalcpuset);
return ORTE_SUCCESS;
}
int orte_daemon(int argc, char *argv[])
{
int ret = 0;
opal_cmd_line_t *cmd_line = NULL;
char *rml_uri;
int i;
opal_buffer_t *buffer;
char hostname[100];
#if OPAL_ENABLE_FT_CR == 1
char *tmp_env_var = NULL;
#endif
/* initialize the globals */
memset(&orted_globals, 0, sizeof(orted_globals));
/* initialize the singleton died pipe to an illegal value so we can detect it was set */
orted_globals.singleton_died_pipe = -1;
/* init the failure orted vpid to an invalid value */
orted_globals.fail = ORTE_VPID_INVALID;
/* setup to check common command line options that just report and die */
cmd_line = OBJ_NEW(opal_cmd_line_t);
if (OPAL_SUCCESS != opal_cmd_line_create(cmd_line, orte_cmd_line_opts)) {
OBJ_RELEASE(cmd_line);
exit(1);
}
mca_base_cmd_line_setup(cmd_line);
if (ORTE_SUCCESS != (ret = opal_cmd_line_parse(cmd_line, false,
argc, argv))) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
fprintf(stderr, "Usage: %s [OPTION]...\n%s\n", argv[0], args);
free(args);
OBJ_RELEASE(cmd_line);
return ret;
}
/*
* Since this process can now handle MCA/GMCA parameters, make sure to
* process them.
*/
mca_base_cmd_line_process_args(cmd_line, &environ, &environ);
/* Ensure that enough of OPAL is setup for us to be able to run */
/*
* NOTE: (JJH)
* We need to allow 'mca_base_cmd_line_process_args()' to process command
* line arguments *before* calling opal_init_util() since the command
* line could contain MCA parameters that affect the way opal_init_util()
* functions. AMCA parameters are one such option normally received on the
* command line that affect the way opal_init_util() behaves.
* It is "safe" to call mca_base_cmd_line_process_args() before
* opal_init_util() since mca_base_cmd_line_process_args() does *not*
* depend upon opal_init_util() functionality.
*/
if (OPAL_SUCCESS != opal_init_util(&argc, &argv)) {
fprintf(stderr, "OPAL failed to initialize -- orted aborting\n");
exit(1);
}
/* save the environment for launch purposes. This MUST be
* done so that we can pass it to any local procs we
* spawn - otherwise, those local procs won't see any
* non-MCA envars that were set in the enviro when the
* orted was executed - e.g., by .csh
*/
orte_launch_environ = opal_argv_copy(environ);
/* purge any ess flag set in the environ when we were launched */
opal_unsetenv(OPAL_MCA_PREFIX"ess", &orte_launch_environ);
/* if orte_daemon_debug is set, let someone know we are alive right
* away just in case we have a problem along the way
*/
if (orted_globals.debug) {
gethostname(hostname, 100);
fprintf(stderr, "Daemon was launched on %s - beginning to initialize\n", hostname);
}
/* check for help request */
if (orted_globals.help) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
orte_show_help("help-orted.txt", "orted:usage", false,
argv[0], args);
free(args);
return 1;
}
#if defined(HAVE_SETSID)
/* see if we were directed to separate from current session */
if (orted_globals.set_sid) {
setsid();
}
#endif
/* see if they want us to spin until they can connect a debugger to us */
i=0;
while (orted_spin_flag) {
i++;
if (1000 < i) i=0;
}
#if OPAL_ENABLE_FT_CR == 1
/* Mark as a tool program */
(void) mca_base_var_env_name ("opal_cr_is_tool", &tmp_env_var);
opal_setenv(tmp_env_var,
"1",
true, &environ);
free(tmp_env_var);
#endif
/* if mapreduce set, flag it */
if (orted_globals.mapreduce) {
orte_map_reduce = true;
}
/* detach from controlling terminal
* otherwise, remain attached so output can get to us
*/
if(!orte_debug_flag &&
!orte_debug_daemons_flag &&
orted_globals.daemonize) {
opal_daemon_init(NULL);
}
/* Set the flag telling OpenRTE that I am NOT a
* singleton, but am "infrastructure" - prevents setting
* up incorrect infrastructure that only a singleton would
* require.
*/
if (orted_globals.hnp) {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_HNP))) {
ORTE_ERROR_LOG(ret);
return ret;
}
} else {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_DAEMON))) {
ORTE_ERROR_LOG(ret);
return ret;
}
}
/* finalize the OPAL utils. As they are opened again from orte_init->opal_init
* we continue to have a reference count on them. So we have to finalize them twice...
*/
opal_finalize_util();
#if OPAL_HAVE_HWLOC
/* bind ourselves if so directed */
if (NULL != orte_daemon_cores) {
char **cores=NULL, tmp[128];
hwloc_obj_t pu;
hwloc_cpuset_t ours, pucpus, res;
int core;
/* could be a collection of comma-delimited ranges, so
* use our handy utility to parse it
*/
orte_util_parse_range_options(orte_daemon_cores, &cores);
if (NULL != cores) {
ours = hwloc_bitmap_alloc();
hwloc_bitmap_zero(ours);
pucpus = hwloc_bitmap_alloc();
res = hwloc_bitmap_alloc();
for (i=0; NULL != cores[i]; i++) {
core = strtoul(cores[i], NULL, 10);
if (NULL == (pu = opal_hwloc_base_get_pu(opal_hwloc_topology, core, OPAL_HWLOC_LOGICAL))) {
/* turn off the show help forwarding as we won't
* be able to cycle the event library to send
*/
orte_show_help_finalize();
/* the message will now come out locally */
orte_show_help("help-orted.txt", "orted:cannot-bind",
true, orte_process_info.nodename,
orte_daemon_cores);
ret = ORTE_ERR_NOT_SUPPORTED;
goto DONE;
}
hwloc_bitmap_and(pucpus, pu->online_cpuset, pu->allowed_cpuset);
hwloc_bitmap_or(res, ours, pucpus);
hwloc_bitmap_copy(ours, res);
}
/* if the result is all zeros, then don't bind */
if (!hwloc_bitmap_iszero(ours)) {
(void)hwloc_set_cpubind(opal_hwloc_topology, ours, 0);
if (opal_hwloc_report_bindings) {
opal_hwloc_base_cset2mapstr(tmp, sizeof(tmp), opal_hwloc_topology, ours);
opal_output(0, "Daemon %s is bound to cores %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
}
}
/* cleanup */
hwloc_bitmap_free(ours);
hwloc_bitmap_free(pucpus);
hwloc_bitmap_free(res);
opal_argv_free(cores);
}
}
#endif
if ((int)ORTE_VPID_INVALID != orted_globals.fail) {
orted_globals.abort=false;
/* some vpid was ordered to fail. The value can be positive
* or negative, depending upon the desired method for failure,
* so need to check both here
*/
if (0 > orted_globals.fail) {
orted_globals.fail = -1*orted_globals.fail;
orted_globals.abort = true;
}
/* are we the specified vpid? */
if ((int)ORTE_PROC_MY_NAME->vpid == orted_globals.fail) {
/* if the user specified we delay, then setup a timer
* and have it kill us
*/
if (0 < orted_globals.fail_delay) {
ORTE_TIMER_EVENT(orted_globals.fail_delay, 0, shutdown_callback, ORTE_SYS_PRI);
} else {
opal_output(0, "%s is executing clean %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orted_globals.abort ? "abort" : "abnormal termination");
/* do -not- call finalize as this will send a message to the HNP
* indicating clean termination! Instead, just forcibly cleanup
* the local session_dir tree and exit
*/
orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
/* if we were ordered to abort, do so */
if (orted_globals.abort) {
abort();
}
/* otherwise, return with non-zero status */
ret = ORTE_ERROR_DEFAULT_EXIT_CODE;
goto DONE;
}
}
}
/* insert our contact info into our process_info struct so we
* have it for later use and set the local daemon field to our name
*/
orte_process_info.my_daemon_uri = orte_rml.get_contact_info();
ORTE_PROC_MY_DAEMON->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_DAEMON->vpid = ORTE_PROC_MY_NAME->vpid;
/* if I am also the hnp, then update that contact info field too */
if (ORTE_PROC_IS_HNP) {
orte_process_info.my_hnp_uri = orte_rml.get_contact_info();
ORTE_PROC_MY_HNP->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_HNP->vpid = ORTE_PROC_MY_NAME->vpid;
}
/* setup the primary daemon command receive function */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON,
ORTE_RML_PERSISTENT, orte_daemon_recv, NULL);
/* output a message indicating we are alive, our name, and our pid
* for debugging purposes
*/
if (orte_debug_daemons_flag) {
fprintf(stderr, "Daemon %s checking in as pid %ld on host %s\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (long)orte_process_info.pid,
orte_process_info.nodename);
}
/* We actually do *not* want the orted to voluntarily yield() the
processor more than necessary. The orted already blocks when
it is doing nothing, so it doesn't use any more CPU cycles than
it should; but when it *is* doing something, we do not want it
to be unnecessarily delayed because it voluntarily yielded the
processor in the middle of its work.
For example: when a message arrives at the orted, we want the
OS to wake up the orted in a timely fashion (which most OS's
seem good about doing) and then we want the orted to process
the message as fast as possible. If the orted yields and lets
aggressive MPI applications get the processor back, it may be a
long time before the OS schedules the orted to run again
(particularly if there is no IO event to wake it up). Hence,
routed OOB messages (for example) may be significantly delayed
before being delivered to MPI processes, which can be
problematic in some scenarios (e.g., COMM_SPAWN, BTL's that
require OOB messages for wireup, etc.). */
opal_progress_set_yield_when_idle(false);
/* Change the default behavior of libevent such that we want to
continually block rather than blocking for the default timeout
and then looping around the progress engine again. There
should be nothing in the orted that cannot block in libevent
until "something" happens (i.e., there's no need to keep
cycling through progress because the only things that should
happen will happen in libevent). This is a minor optimization,
but what the heck... :-) */
opal_progress_set_event_flag(OPAL_EVLOOP_ONCE);
/* if requested, report my uri to the indicated pipe */
if (orted_globals.uri_pipe > 0) {
orte_job_t *jdata;
orte_proc_t *proc;
orte_node_t *node;
orte_app_context_t *app;
char *tmp, *nptr, *sysinfo;
int32_t ljob;
/* setup the singleton's job */
jdata = OBJ_NEW(orte_job_t);
orte_plm_base_create_jobid(jdata);
ljob = ORTE_LOCAL_JOBID(jdata->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, jdata);
/* must create a map for it (even though it has no
* info in it) so that the job info will be picked
* up in subsequent pidmaps or other daemons won't
* know how to route
*/
jdata->map = OBJ_NEW(orte_job_map_t);
/* setup an app_context for the singleton */
app = OBJ_NEW(orte_app_context_t);
app->app = strdup("singleton");
app->num_procs = 1;
opal_pointer_array_add(jdata->apps, app);
/* setup a proc object for the singleton - since we
* -must- be the HNP, and therefore we stored our
* node on the global node pool, and since the singleton
* -must- be on the same node as us, indicate that
*/
proc = OBJ_NEW(orte_proc_t);
proc->name.jobid = jdata->jobid;
proc->name.vpid = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_ALIVE);
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
/* obviously, it is on my node */
node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, 0);
proc->node = node;
OBJ_RETAIN(node); /* keep accounting straight */
opal_pointer_array_add(jdata->procs, proc);
jdata->num_procs = 1;
/* and it obviously is on the node */
OBJ_RETAIN(proc);
opal_pointer_array_add(node->procs, proc);
node->num_procs++;
/* and obviously it is one of my local procs */
OBJ_RETAIN(proc);
opal_pointer_array_add(orte_local_children, proc);
jdata->num_local_procs = 1;
/* set the trivial */
proc->local_rank = 0;
proc->node_rank = 0;
proc->app_rank = 0;
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_LOCAL);
/* create a string that contains our uri + sysinfo + PMIx server URI */
orte_util_convert_sysinfo_to_string(&sysinfo, orte_local_cpu_type, orte_local_cpu_model);
asprintf(&tmp, "%s[%s]%s", orte_process_info.my_daemon_uri, sysinfo, pmix_server_uri);
free(sysinfo);
/* pass that info to the singleton */
write(orted_globals.uri_pipe, tmp, strlen(tmp)+1); /* need to add 1 to get the NULL */
/* cleanup */
free(tmp);
/* since a singleton spawned us, we need to harvest
* any MCA params from the local environment so
* we can pass them along to any subsequent daemons
* we may start as the result of a comm_spawn
*/
for (i=0; NULL != environ[i]; i++) {
if (0 == strncmp(environ[i], OPAL_MCA_PREFIX, 9)) {
/* make a copy to manipulate */
tmp = strdup(environ[i]);
/* find the equal sign */
nptr = strchr(tmp, '=');
*nptr = '\0';
nptr++;
/* add the mca param to the orted cmd line */
opal_argv_append_nosize(&orted_cmd_line, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append_nosize(&orted_cmd_line, &tmp[9]);
opal_argv_append_nosize(&orted_cmd_line, nptr);
free(tmp);
}
}
}
/* if we were given a pipe to monitor for singleton termination, set that up */
if (orted_globals.singleton_died_pipe > 0) {
/* register shutdown handler */
pipe_handler = (opal_event_t*)malloc(sizeof(opal_event_t));
opal_event_set(orte_event_base, pipe_handler,
orted_globals.singleton_died_pipe,
OPAL_EV_READ,
pipe_closed,
pipe_handler);
opal_event_add(pipe_handler, NULL);
}
/* If I have a parent, then save his contact info so
* any messages we send can flow thru him.
*/
orte_parent_uri = NULL;
(void) mca_base_var_register ("orte", "orte", NULL, "parent_uri",
"URI for the parent if tree launch is enabled.",
MCA_BASE_VAR_TYPE_STRING, NULL, 0,
MCA_BASE_VAR_FLAG_INTERNAL,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_CONSTANT,
&orte_parent_uri);
if (NULL != orte_parent_uri) {
orte_process_name_t parent;
/* set the contact info into the hash table */
orte_rml.set_contact_info(orte_parent_uri);
ret = orte_rml_base_parse_uris(orte_parent_uri, &parent, NULL);
if (ORTE_SUCCESS != ret) {
ORTE_ERROR_LOG(ret);
free (orte_parent_uri);
orte_parent_uri = NULL;
goto DONE;
}
/* don't need this value anymore */
free(orte_parent_uri);
orte_parent_uri = NULL;
/* tell the routed module that we have a path
* back to the HNP
*/
if (ORTE_SUCCESS != (ret = orte_routed.update_route(ORTE_PROC_MY_HNP, &parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* set the lifeline to point to our parent so that we
* can handle the situation if that lifeline goes away
*/
if (ORTE_SUCCESS != (ret = orte_routed.set_lifeline(&parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
}
/* if we are not the HNP...the only time we will be an HNP
* is if we are launched by a singleton to provide support
* for it
*/
if (!ORTE_PROC_IS_HNP) {
/* send the information to the orted report-back point - this function
* will process the data, but also counts the number of
* orteds that reported back so the launch procedure can continue.
* We need to do this at the last possible second as the HNP
* can turn right around and begin issuing orders to us
*/
buffer = OBJ_NEW(opal_buffer_t);
/* insert our name for rollup purposes */
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, ORTE_PROC_MY_NAME, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* for now, always include our contact info, even if we are using
* static ports. Eventually, this will be removed
*/
rml_uri = orte_rml.get_contact_info();
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &rml_uri, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* include our node name */
opal_dss.pack(buffer, &orte_process_info.nodename, 1, OPAL_STRING);
/* if requested, include any non-loopback aliases for this node */
if (orte_retain_aliases) {
char **aliases=NULL;
uint8_t naliases, ni;
char hostname[ORTE_MAX_HOSTNAME_SIZE];
/* if we stripped the prefix or removed the fqdn,
* include full hostname as an alias
*/
gethostname(hostname, ORTE_MAX_HOSTNAME_SIZE);
if (strlen(orte_process_info.nodename) < strlen(hostname)) {
opal_argv_append_nosize(&aliases, hostname);
}
opal_ifgetaliases(&aliases);
naliases = opal_argv_count(aliases);
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &naliases, 1, OPAL_UINT8))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
for (ni=0; ni < naliases; ni++) {
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &aliases[ni], 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
opal_argv_free(aliases);
}
#if OPAL_HAVE_HWLOC
{
char *coprocessors;
/* add the local topology */
if (NULL != opal_hwloc_topology &&
(1 == ORTE_PROC_MY_NAME->vpid || orte_hetero_nodes)) {
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &opal_hwloc_topology, 1, OPAL_HWLOC_TOPO))) {
ORTE_ERROR_LOG(ret);
}
}
/* detect and add any coprocessors */
coprocessors = opal_hwloc_base_find_coprocessors(opal_hwloc_topology);
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &coprocessors, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
/* see if I am on a coprocessor */
coprocessors = opal_hwloc_base_check_on_coprocessor();
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &coprocessors, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
}
#endif
/* send to the HNP's callback - will be routed if routes are available */
if (0 > (ret = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, buffer,
ORTE_RML_TAG_ORTED_CALLBACK,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
/* if we are tree-spawning, then we need to capture the MCA params
* from our cmd line so we can pass them along to the daemons we spawn -
* otherwise, only the first layer of daemons will ever see them
*/
if (orted_globals.tree_spawn) {
int j, k;
bool ignore;
char *no_keep[] = {
"orte_hnp_uri",
"orte_ess_jobid",
"orte_ess_vpid",
"orte_ess_num_procs",
"orte_parent_uri",
"mca_base_env_list",
NULL
};
for (i=0; i < argc; i++) {
if (0 == strcmp("-"OPAL_MCA_CMD_LINE_ID, argv[i]) ||
0 == strcmp("--"OPAL_MCA_CMD_LINE_ID, argv[i]) ) {
ignore = false;
/* see if this is something we cannot pass along */
for (k=0; NULL != no_keep[k]; k++) {
if (0 == strcmp(no_keep[k], argv[i+1])) {
ignore = true;
break;
}
}
if (!ignore) {
/* see if this is already present so we at least can
* avoid growing the cmd line with duplicates
*/
if (NULL != orted_cmd_line) {
for (j=0; NULL != orted_cmd_line[j]; j++) {
if (0 == strcmp(argv[i+1], orted_cmd_line[j])) {
/* already here - ignore it */
ignore = true;
break;
}
}
}
if (!ignore) {
opal_argv_append_nosize(&orted_cmd_line, argv[i]);
opal_argv_append_nosize(&orted_cmd_line, argv[i+1]);
opal_argv_append_nosize(&orted_cmd_line, argv[i+2]);
}
}
i += 2;
}
}
}
if (orte_debug_daemons_flag) {
opal_output(0, "%s orted: up and running - waiting for commands!", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
}
ret = ORTE_SUCCESS;
/* loop the event lib until an exit event is detected */
while (orte_event_base_active) {
opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
}
/* ensure all local procs are dead */
orte_odls.kill_local_procs(NULL);
DONE:
/* update the exit status, in case it wasn't done */
ORTE_UPDATE_EXIT_STATUS(ret);
/* cleanup and leave */
orte_finalize();
if (orte_debug_flag) {
fprintf(stderr, "exiting with status %d\n", orte_exit_status);
}
exit(orte_exit_status);
}
static void set(orte_job_t *jobdat,
orte_proc_t *child,
char ***environ_copy,
int write_fd)
{
hwloc_cpuset_t cpuset;
hwloc_obj_t root;
opal_hwloc_topo_data_t *sum;
orte_app_context_t *context;
int rc=ORTE_ERROR;
char *msg, *param;
char *cpu_bitmap;
opal_output_verbose(2, orte_rtc_base_framework.framework_output,
"%s hwloc:set on child %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == child) ? "NULL" : ORTE_NAME_PRINT(&child->name));
if (NULL == jobdat || NULL == child) {
/* nothing for us to do */
opal_output_verbose(2, orte_rtc_base_framework.framework_output,
"%s hwloc:set jobdat %s child %s - nothing to do",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == jobdat) ? "NULL" : ORTE_JOBID_PRINT(jobdat->jobid),
(NULL == child) ? "NULL" : ORTE_NAME_PRINT(&child->name));
return;
}
context = (orte_app_context_t*)opal_pointer_array_get_item(jobdat->apps, child->app_idx);
/* Set process affinity, if given */
cpu_bitmap = NULL;
if (!orte_get_attribute(&child->attributes, ORTE_PROC_CPU_BITMAP, (void**)&cpu_bitmap, OPAL_STRING) ||
NULL == cpu_bitmap || 0 == strlen(cpu_bitmap)) {
/* if the daemon is bound, then we need to "free" this proc */
if (NULL != orte_daemon_cores) {
root = hwloc_get_root_obj(opal_hwloc_topology);
if (NULL == root->userdata) {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "incorrectly bound",
orte_process_info.nodename, context->app,
__FILE__, __LINE__);
}
sum = (opal_hwloc_topo_data_t*)root->userdata;
/* bind this proc to all available processors */
rc = hwloc_set_cpubind(opal_hwloc_topology, sum->available, 0);
/* if we got an error and this wasn't a default binding policy, then report it */
if (rc < 0 && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
if (errno == ENOSYS) {
msg = "hwloc indicates cpu binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates cpu binding cannot be enforced";
} else {
char *tmp;
(void)hwloc_bitmap_list_asprintf(&tmp, sum->available);
asprintf(&msg, "hwloc_set_cpubind returned \"%s\" for bitmap \"%s\"",
opal_strerror(rc), tmp);
free(tmp);
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding)) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
return;
}
}
}
if (0 == rc && opal_hwloc_report_bindings) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, environ_copy);
free(param);
}
} else {
/* convert the list to a cpuset */
cpuset = hwloc_bitmap_alloc();
if (0 != (rc = hwloc_bitmap_list_sscanf(cpuset, cpu_bitmap))) {
/* See comment above about "This may be a small memory leak" */
asprintf(&msg, "hwloc_bitmap_sscanf returned \"%s\" for the string \"%s\"",
opal_strerror(rc), cpu_bitmap);
if (NULL == msg) {
msg = "failed to convert bitmap list to hwloc bitmap";
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding) &&
OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
/* If binding is required and a binding directive was explicitly
* given (i.e., we are not binding due to a default policy),
* send an error up the pipe (which exits -- it doesn't return).
*/
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename,
context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
free(cpu_bitmap);
return;
}
}
/* bind as specified */
rc = hwloc_set_cpubind(opal_hwloc_topology, cpuset, 0);
/* if we got an error and this wasn't a default binding policy, then report it */
if (rc < 0 && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
char *tmp = NULL;
if (errno == ENOSYS) {
msg = "hwloc indicates cpu binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates cpu binding cannot be enforced";
} else {
asprintf(&msg, "hwloc_set_cpubind returned \"%s\" for bitmap \"%s\"",
opal_strerror(rc), cpu_bitmap);
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding)) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
if (NULL != tmp) {
free(tmp);
free(msg);
}
return;
}
if (NULL != tmp) {
free(tmp);
free(msg);
}
}
if (0 == rc && opal_hwloc_report_bindings) {
char tmp1[1024], tmp2[1024];
hwloc_cpuset_t mycpus;
/* get the cpus we are bound to */
mycpus = hwloc_bitmap_alloc();
if (hwloc_get_cpubind(opal_hwloc_topology,
mycpus,
HWLOC_CPUBIND_PROCESS) < 0) {
opal_output(0, "MCW rank %d is not bound",
child->name.vpid);
} else {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), opal_hwloc_topology, mycpus)) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), opal_hwloc_topology, mycpus);
opal_output(0, "MCW rank %d bound to %s: %s",
child->name.vpid, tmp1, tmp2);
}
}
hwloc_bitmap_free(mycpus);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, environ_copy);
free(param);
}
/* set memory affinity policy - if we get an error, don't report
* anything unless the user actually specified the binding policy
*/
rc = opal_hwloc_base_set_process_membind_policy();
if (ORTE_SUCCESS != rc && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
if (errno == ENOSYS) {
msg = "hwloc indicates memory binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates memory binding cannot be enforced";
} else {
msg = "failed to bind memory";
}
if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"memory binding error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "memory not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
free(cpu_bitmap);
return;
}
}
}
if (NULL != cpu_bitmap) {
free(cpu_bitmap);
}
}
int orte_daemon(int argc, char *argv[])
{
int ret = 0;
opal_cmd_line_t *cmd_line = NULL;
int i;
opal_buffer_t *buffer;
char hostname[OPAL_MAXHOSTNAMELEN];
#if OPAL_ENABLE_FT_CR == 1
char *tmp_env_var = NULL;
#endif
/* initialize the globals */
memset(&orted_globals, 0, sizeof(orted_globals));
/* initialize the singleton died pipe to an illegal value so we can detect it was set */
orted_globals.singleton_died_pipe = -1;
bucket = OBJ_NEW(opal_buffer_t);
/* setup to check common command line options that just report and die */
cmd_line = OBJ_NEW(opal_cmd_line_t);
if (OPAL_SUCCESS != opal_cmd_line_create(cmd_line, orte_cmd_line_opts)) {
OBJ_RELEASE(cmd_line);
exit(1);
}
mca_base_cmd_line_setup(cmd_line);
if (ORTE_SUCCESS != (ret = opal_cmd_line_parse(cmd_line, false, false,
argc, argv))) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
fprintf(stderr, "Usage: %s [OPTION]...\n%s\n", argv[0], args);
free(args);
OBJ_RELEASE(cmd_line);
return ret;
}
/*
* Since this process can now handle MCA/GMCA parameters, make sure to
* process them.
*/
mca_base_cmd_line_process_args(cmd_line, &environ, &environ);
/* Ensure that enough of OPAL is setup for us to be able to run */
/*
* NOTE: (JJH)
* We need to allow 'mca_base_cmd_line_process_args()' to process command
* line arguments *before* calling opal_init_util() since the command
* line could contain MCA parameters that affect the way opal_init_util()
* functions. AMCA parameters are one such option normally received on the
* command line that affect the way opal_init_util() behaves.
* It is "safe" to call mca_base_cmd_line_process_args() before
* opal_init_util() since mca_base_cmd_line_process_args() does *not*
* depend upon opal_init_util() functionality.
*/
if (OPAL_SUCCESS != opal_init_util(&argc, &argv)) {
fprintf(stderr, "OPAL failed to initialize -- orted aborting\n");
exit(1);
}
/* save the environment for launch purposes. This MUST be
* done so that we can pass it to any local procs we
* spawn - otherwise, those local procs won't see any
* non-MCA envars that were set in the enviro when the
* orted was executed - e.g., by .csh
*/
orte_launch_environ = opal_argv_copy(environ);
/* purge any ess/pmix flags set in the environ when we were launched */
opal_unsetenv(OPAL_MCA_PREFIX"ess", &orte_launch_environ);
opal_unsetenv(OPAL_MCA_PREFIX"pmix", &orte_launch_environ);
/* if orte_daemon_debug is set, let someone know we are alive right
* away just in case we have a problem along the way
*/
if (orted_globals.debug) {
gethostname(hostname, sizeof(hostname));
fprintf(stderr, "Daemon was launched on %s - beginning to initialize\n", hostname);
}
/* check for help request */
if (orted_globals.help) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
orte_show_help("help-orted.txt", "orted:usage", false,
argv[0], args);
free(args);
return 1;
}
#if defined(HAVE_SETSID)
/* see if we were directed to separate from current session */
if (orted_globals.set_sid) {
setsid();
}
#endif
/* see if they want us to spin until they can connect a debugger to us */
i=0;
while (orted_spin_flag) {
i++;
if (1000 < i) i=0;
}
#if OPAL_ENABLE_FT_CR == 1
/* Mark as a tool program */
(void) mca_base_var_env_name ("opal_cr_is_tool", &tmp_env_var);
opal_setenv(tmp_env_var,
"1",
true, &environ);
free(tmp_env_var);
#endif
/* detach from controlling terminal
* otherwise, remain attached so output can get to us
*/
if(!orte_debug_flag &&
!orte_debug_daemons_flag &&
orted_globals.daemonize) {
opal_daemon_init(NULL);
}
/* Set the flag telling OpenRTE that I am NOT a
* singleton, but am "infrastructure" - prevents setting
* up incorrect infrastructure that only a singleton would
* require.
*/
if (orted_globals.hnp) {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_HNP))) {
ORTE_ERROR_LOG(ret);
return ret;
}
} else {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_DAEMON))) {
ORTE_ERROR_LOG(ret);
return ret;
}
}
/* finalize the OPAL utils. As they are opened again from orte_init->opal_init
* we continue to have a reference count on them. So we have to finalize them twice...
*/
opal_finalize_util();
/* bind ourselves if so directed */
if (NULL != orte_daemon_cores) {
char **cores=NULL, tmp[128];
hwloc_obj_t pu;
hwloc_cpuset_t ours, res;
int core;
/* could be a collection of comma-delimited ranges, so
* use our handy utility to parse it
*/
orte_util_parse_range_options(orte_daemon_cores, &cores);
if (NULL != cores) {
ours = hwloc_bitmap_alloc();
hwloc_bitmap_zero(ours);
res = hwloc_bitmap_alloc();
for (i=0; NULL != cores[i]; i++) {
core = strtoul(cores[i], NULL, 10);
if (NULL == (pu = opal_hwloc_base_get_pu(opal_hwloc_topology, core, OPAL_HWLOC_LOGICAL))) {
/* turn off the show help forwarding as we won't
* be able to cycle the event library to send
*/
orte_show_help_finalize();
/* the message will now come out locally */
orte_show_help("help-orted.txt", "orted:cannot-bind",
true, orte_process_info.nodename,
orte_daemon_cores);
ret = ORTE_ERR_NOT_SUPPORTED;
hwloc_bitmap_free(ours);
hwloc_bitmap_free(res);
goto DONE;
}
hwloc_bitmap_or(res, ours, pu->cpuset);
hwloc_bitmap_copy(ours, res);
}
/* if the result is all zeros, then don't bind */
if (!hwloc_bitmap_iszero(ours)) {
(void)hwloc_set_cpubind(opal_hwloc_topology, ours, 0);
if (opal_hwloc_report_bindings) {
opal_hwloc_base_cset2mapstr(tmp, sizeof(tmp), opal_hwloc_topology, ours);
opal_output(0, "Daemon %s is bound to cores %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
}
}
/* cleanup */
hwloc_bitmap_free(ours);
hwloc_bitmap_free(res);
opal_argv_free(cores);
}
}
if ((int)ORTE_VPID_INVALID != orted_debug_failure) {
orted_globals.abort=false;
/* some vpid was ordered to fail. The value can be positive
* or negative, depending upon the desired method for failure,
* so need to check both here
*/
if (0 > orted_debug_failure) {
orted_debug_failure = -1*orted_debug_failure;
orted_globals.abort = true;
}
/* are we the specified vpid? */
if ((int)ORTE_PROC_MY_NAME->vpid == orted_debug_failure) {
/* if the user specified we delay, then setup a timer
* and have it kill us
*/
if (0 < orted_debug_failure_delay) {
ORTE_TIMER_EVENT(orted_debug_failure_delay, 0, shutdown_callback, ORTE_SYS_PRI);
} else {
opal_output(0, "%s is executing clean %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orted_globals.abort ? "abort" : "abnormal termination");
/* do -not- call finalize as this will send a message to the HNP
* indicating clean termination! Instead, just forcibly cleanup
* the local session_dir tree and exit
*/
orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
/* if we were ordered to abort, do so */
if (orted_globals.abort) {
abort();
}
/* otherwise, return with non-zero status */
ret = ORTE_ERROR_DEFAULT_EXIT_CODE;
goto DONE;
}
}
}
/* insert our contact info into our process_info struct so we
* have it for later use and set the local daemon field to our name
*/
orte_oob_base_get_addr(&orte_process_info.my_daemon_uri);
if (NULL == orte_process_info.my_daemon_uri) {
/* no way to communicate */
ret = ORTE_ERROR;
goto DONE;
}
ORTE_PROC_MY_DAEMON->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_DAEMON->vpid = ORTE_PROC_MY_NAME->vpid;
/* if I am also the hnp, then update that contact info field too */
if (ORTE_PROC_IS_HNP) {
orte_process_info.my_hnp_uri = strdup(orte_process_info.my_daemon_uri);
ORTE_PROC_MY_HNP->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_HNP->vpid = ORTE_PROC_MY_NAME->vpid;
}
/* setup the primary daemon command receive function */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON,
ORTE_RML_PERSISTENT, orte_daemon_recv, NULL);
/* output a message indicating we are alive, our name, and our pid
* for debugging purposes
*/
if (orte_debug_daemons_flag) {
fprintf(stderr, "Daemon %s checking in as pid %ld on host %s\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (long)orte_process_info.pid,
orte_process_info.nodename);
}
/* We actually do *not* want the orted to voluntarily yield() the
processor more than necessary. The orted already blocks when
it is doing nothing, so it doesn't use any more CPU cycles than
it should; but when it *is* doing something, we do not want it
to be unnecessarily delayed because it voluntarily yielded the
processor in the middle of its work.
For example: when a message arrives at the orted, we want the
OS to wake up the orted in a timely fashion (which most OS's
seem good about doing) and then we want the orted to process
the message as fast as possible. If the orted yields and lets
aggressive MPI applications get the processor back, it may be a
long time before the OS schedules the orted to run again
(particularly if there is no IO event to wake it up). Hence,
routed OOB messages (for example) may be significantly delayed
before being delivered to MPI processes, which can be
problematic in some scenarios (e.g., COMM_SPAWN, BTL's that
require OOB messages for wireup, etc.). */
opal_progress_set_yield_when_idle(false);
/* Change the default behavior of libevent such that we want to
continually block rather than blocking for the default timeout
and then looping around the progress engine again. There
should be nothing in the orted that cannot block in libevent
until "something" happens (i.e., there's no need to keep
cycling through progress because the only things that should
happen will happen in libevent). This is a minor optimization,
but what the heck... :-) */
opal_progress_set_event_flag(OPAL_EVLOOP_ONCE);
/* if requested, report my uri to the indicated pipe */
if (orted_globals.uri_pipe > 0) {
orte_job_t *jdata;
orte_proc_t *proc;
orte_node_t *node;
orte_app_context_t *app;
char *tmp, *nptr, *sysinfo;
char **singenv=NULL, *string_key, *env_str;
/* setup the singleton's job */
jdata = OBJ_NEW(orte_job_t);
/* default to ompi for now */
opal_argv_append_nosize(&jdata->personality, "ompi");
orte_plm_base_create_jobid(jdata);
opal_hash_table_set_value_uint32(orte_job_data, jdata->jobid, jdata);
/* must create a map for it (even though it has no
* info in it) so that the job info will be picked
* up in subsequent pidmaps or other daemons won't
* know how to route
*/
jdata->map = OBJ_NEW(orte_job_map_t);
/* setup an app_context for the singleton */
app = OBJ_NEW(orte_app_context_t);
app->app = strdup("singleton");
app->num_procs = 1;
opal_pointer_array_add(jdata->apps, app);
jdata->num_apps = 1;
/* setup a proc object for the singleton - since we
* -must- be the HNP, and therefore we stored our
* node on the global node pool, and since the singleton
* -must- be on the same node as us, indicate that
*/
proc = OBJ_NEW(orte_proc_t);
proc->name.jobid = jdata->jobid;
proc->name.vpid = 0;
proc->parent = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_ALIVE);
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
/* obviously, it is on my node */
node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, 0);
proc->node = node;
OBJ_RETAIN(node); /* keep accounting straight */
opal_pointer_array_add(jdata->procs, proc);
jdata->num_procs = 1;
/* add the node to the job map */
OBJ_RETAIN(node);
opal_pointer_array_add(jdata->map->nodes, node);
jdata->map->num_nodes++;
/* and it obviously is on the node */
OBJ_RETAIN(proc);
opal_pointer_array_add(node->procs, proc);
node->num_procs++;
/* and obviously it is one of my local procs */
OBJ_RETAIN(proc);
opal_pointer_array_add(orte_local_children, proc);
jdata->num_local_procs = 1;
/* set the trivial */
proc->local_rank = 0;
proc->node_rank = 0;
proc->app_rank = 0;
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_LOCAL);
/* set the ORTE_JOB_TRANSPORT_KEY from the environment */
orte_pre_condition_transports(jdata, NULL);
/* register the singleton's nspace with our PMIx server */
if (ORTE_SUCCESS != (ret = orte_pmix_server_register_nspace(jdata, false))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* use setup fork to create the envars needed by the singleton */
if (OPAL_SUCCESS != (ret = opal_pmix.server_setup_fork(&proc->name, &singenv))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* append the transport key to the envars needed by the singleton */
if (!orte_get_attribute(&jdata->attributes, ORTE_JOB_TRANSPORT_KEY, (void**)&string_key, OPAL_STRING) || NULL == string_key) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto DONE;
}
asprintf(&env_str, OPAL_MCA_PREFIX"orte_precondition_transports=%s", string_key);
opal_argv_append_nosize(&singenv, env_str);
free(env_str);
nptr = opal_argv_join(singenv, '*');
opal_argv_free(singenv);
/* create a string that contains our uri + sysinfo + PMIx server URI envars */
orte_util_convert_sysinfo_to_string(&sysinfo, orte_local_cpu_type, orte_local_cpu_model);
asprintf(&tmp, "%s[%s]%s", orte_process_info.my_daemon_uri, sysinfo, nptr);
free(sysinfo);
free(nptr);
/* pass that info to the singleton */
if (OPAL_SUCCESS != (ret = opal_fd_write(orted_globals.uri_pipe, strlen(tmp)+1, tmp))) { ; /* need to add 1 to get the NULL */
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* cleanup */
free(tmp);
close(orted_globals.uri_pipe);
/* since a singleton spawned us, we need to harvest
* any MCA params from the local environment so
* we can pass them along to any subsequent daemons
* we may start as the result of a comm_spawn
*/
for (i=0; NULL != environ[i]; i++) {
if (0 == strncmp(environ[i], OPAL_MCA_PREFIX, 9)) {
/* make a copy to manipulate */
tmp = strdup(environ[i]);
/* find the equal sign */
nptr = strchr(tmp, '=');
*nptr = '\0';
nptr++;
/* add the mca param to the orted cmd line */
opal_argv_append_nosize(&orted_cmd_line, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append_nosize(&orted_cmd_line, &tmp[9]);
opal_argv_append_nosize(&orted_cmd_line, nptr);
free(tmp);
}
}
}
/* if we were given a pipe to monitor for singleton termination, set that up */
if (orted_globals.singleton_died_pipe > 0) {
/* register shutdown handler */
pipe_handler = (opal_event_t*)malloc(sizeof(opal_event_t));
opal_event_set(orte_event_base, pipe_handler,
orted_globals.singleton_died_pipe,
OPAL_EV_READ,
pipe_closed,
pipe_handler);
opal_event_add(pipe_handler, NULL);
}
/* If I have a parent, then save his contact info so
* any messages we send can flow thru him.
*/
orte_parent_uri = NULL;
(void) mca_base_var_register ("orte", "orte", NULL, "parent_uri",
"URI for the parent if tree launch is enabled.",
MCA_BASE_VAR_TYPE_STRING, NULL, 0,
MCA_BASE_VAR_FLAG_INTERNAL,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_CONSTANT,
&orte_parent_uri);
if (NULL != orte_parent_uri) {
orte_process_name_t parent;
opal_value_t val;
/* set the contact info into our local database */
ret = orte_rml_base_parse_uris(orte_parent_uri, &parent, NULL);
if (ORTE_SUCCESS != ret) {
ORTE_ERROR_LOG(ret);
free (orte_parent_uri);
orte_parent_uri = NULL;
goto DONE;
}
OBJ_CONSTRUCT(&val, opal_value_t);
val.key = OPAL_PMIX_PROC_URI;
val.type = OPAL_STRING;
val.data.string = orte_parent_uri;
if (OPAL_SUCCESS != (ret = opal_pmix.store_local(&parent, &val))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&val);
goto DONE;
}
val.key = NULL;
val.data.string = NULL;
OBJ_DESTRUCT(&val);
/* don't need this value anymore */
free(orte_parent_uri);
orte_parent_uri = NULL;
/* tell the routed module that we have a path
* back to the HNP
*/
if (ORTE_SUCCESS != (ret = orte_routed.update_route(NULL, ORTE_PROC_MY_HNP, &parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* set the lifeline to point to our parent so that we
* can handle the situation if that lifeline goes away
*/
if (ORTE_SUCCESS != (ret = orte_routed.set_lifeline(NULL, &parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
}
/* if we are not the HNP...the only time we will be an HNP
* is if we are launched by a singleton to provide support
* for it
*/
if (!ORTE_PROC_IS_HNP) {
orte_process_name_t target;
target.jobid = ORTE_PROC_MY_NAME->jobid;
if (orte_fwd_mpirun_port || orte_static_ports) {
/* setup the rollup callback */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_ORTED_CALLBACK,
ORTE_RML_PERSISTENT, rollup, NULL);
target.vpid = ORTE_PROC_MY_NAME->vpid;
/* since we will be waiting for any children to send us
* their rollup info before sending to our parent, save
* a little time in the launch phase by "warming up" the
* connection to our parent while we wait for our children */
buffer = OBJ_NEW(opal_buffer_t); // zero-byte message
if (0 > (ret = orte_rml.send_buffer_nb(orte_mgmt_conduit,
ORTE_PROC_MY_PARENT, buffer,
ORTE_RML_TAG_WARMUP_CONNECTION,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
} else {
target.vpid = 0;
}
/* send the information to the orted report-back point - this function
* will process the data, but also counts the number of
* orteds that reported back so the launch procedure can continue.
* We need to do this at the last possible second as the HNP
* can turn right around and begin issuing orders to us
*/
buffer = OBJ_NEW(opal_buffer_t);
/* insert our name for rollup purposes */
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, ORTE_PROC_MY_NAME, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* get any connection info we may have pushed */
{
opal_value_t *val = NULL, *kv;
opal_list_t *modex;
int32_t flag;
if (OPAL_SUCCESS != (ret = opal_pmix.get(ORTE_PROC_MY_NAME, NULL, NULL, &val)) || NULL == val) {
/* just pack a marker indicating we don't have any to share */
flag = 0;
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &flag, 1, OPAL_INT32))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
} else {
/* the data is returned as a list of key-value pairs in the opal_value_t */
if (OPAL_PTR == val->type) {
modex = (opal_list_t*)val->data.ptr;
flag = (int32_t)opal_list_get_size(modex);
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &flag, 1, OPAL_INT32))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
OPAL_LIST_FOREACH(kv, modex, opal_value_t) {
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &kv, 1, OPAL_VALUE))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
OPAL_LIST_RELEASE(modex);
} else {
opal_output(0, "VAL KEY: %s", (NULL == val->key) ? "NULL" : val->key);
/* single value */
flag = 1;
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &flag, 1, OPAL_INT32))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &val, 1, OPAL_VALUE))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
OBJ_RELEASE(val);
}
static int bind_downwards(orte_job_t *jdata,
hwloc_obj_type_t target,
unsigned cache_level)
{
int i, j;
orte_job_map_t *map;
orte_node_t *node;
orte_proc_t *proc;
hwloc_obj_t trg_obj, nxt_obj;
hwloc_cpuset_t cpus;
unsigned int ncpus;
struct hwloc_topology_support *support;
opal_hwloc_obj_data_t *data;
int total_cpus;
hwloc_cpuset_t totalcpuset;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: bind downward for job %s with bindings %s",
ORTE_JOBID_PRINT(jdata->jobid),
opal_hwloc_base_print_binding(jdata->map->binding));
/* initialize */
map = jdata->map;
totalcpuset = hwloc_bitmap_alloc();
for (i=0; i < map->nodes->size; i++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(map->nodes, i))) {
continue;
}
if (!orte_do_not_launch) {
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
* interprets the current behavior. So check both flags to be sure.
*/
if (!support->cpubind->set_thisproc_cpubind &&
!support->cpubind->set_thisthread_cpubind) {
if (!OPAL_BINDING_REQUIRED(opal_hwloc_binding_policy)) {
/* we are not required to bind, so ignore this */
continue;
}
orte_show_help("help-orte-rmaps-base.txt", "rmaps:cpubind-not-supported", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
/* check if topology supports membind - have to be careful here
* as hwloc treats this differently than I (at least) would have
* expected. Per hwloc, Linux memory binding is at the thread,
* and not process, level. Thus, hwloc sets the "thisproc" flag
* to "false" on all Linux systems, and uses the "thisthread" flag
* to indicate binding capability
*/
if (!support->membind->set_thisproc_membind &&
!support->membind->set_thisthread_membind) {
if (OPAL_HWLOC_BASE_MBFA_WARN == opal_hwloc_base_mbfa && !membind_warned) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported", true, node->name);
membind_warned = true;
} else if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported-fatal", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
}
}
/* clear the topology of any prior usage numbers */
opal_hwloc_base_clear_usage(node->topology);
/* cycle thru the procs */
for (j=0; j < node->procs->size; j++) {
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(node->procs, j))) {
continue;
}
/* ignore procs from other jobs */
if (proc->name.jobid != jdata->jobid) {
continue;
}
/* ignore procs that have already been bound - should
* never happen, but safer
*/
if (NULL != proc->cpu_bitmap) {
continue;
}
/* we don't know if the target is a direct child of this locale,
* or if it is some depth below it, so we have to conduct a bit
* of a search. Let hwloc find the min usage one for us.
*/
trg_obj = opal_hwloc_base_find_min_bound_target_under_obj(node->topology,
proc->locale,
target, cache_level);
if (NULL == trg_obj) {
/* there aren't any such targets under this object */
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
/* start with a clean slate */
hwloc_bitmap_zero(totalcpuset);
total_cpus = 0;
nxt_obj = trg_obj;
do {
if (NULL == nxt_obj) {
/* could not find enough cpus to meet request */
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
trg_obj = nxt_obj;
/* get the number of cpus under this location */
ncpus = opal_hwloc_base_get_npus(node->topology, trg_obj);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s GOT %d CPUS",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ncpus);
/* track the number bound */
if (NULL == (data = (opal_hwloc_obj_data_t*)trg_obj->userdata)) {
data = OBJ_NEW(opal_hwloc_obj_data_t);
trg_obj->userdata = data;
}
data->num_bound++;
/* error out if adding a proc would cause overload and that wasn't allowed */
if (ncpus < data->num_bound &&
!OPAL_BIND_OVERLOAD_ALLOWED(jdata->map->binding)) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:binding-overload", true,
opal_hwloc_base_print_binding(map->binding), node->name,
data->num_bound, ncpus);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
/* bind the proc here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, trg_obj);
hwloc_bitmap_or(totalcpuset, totalcpuset, cpus);
total_cpus += ncpus;
/* move to the next location, in case we need it */
nxt_obj = trg_obj->next_cousin;
} while (total_cpus < orte_rmaps_base.cpus_per_rank);
hwloc_bitmap_list_asprintf(&proc->cpu_bitmap, totalcpuset);
if (4 < opal_output_get_verbosity(orte_rmaps_base_framework.framework_output)) {
char tmp1[1024], tmp2[1024];
opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), totalcpuset);
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), totalcpuset);
opal_output(orte_rmaps_base_framework.framework_output,
"%s BOUND PROC %s[%s] TO %s: %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name,
tmp1, tmp2);
}
}
}
hwloc_bitmap_free(totalcpuset);
return ORTE_SUCCESS;
}
static int orcmd_init(void)
{
int ret = ORTE_ERROR;
char *error = NULL;
opal_buffer_t buf, *clusterbuf, *uribuf;
orte_job_t *jdata;
orte_node_t *node;
orte_proc_t *proc;
opal_list_t config;
orcm_scheduler_t *scheduler;
orcm_node_t *mynode=NULL;
int32_t n;
if (initialized) {
return ORCM_SUCCESS;
}
initialized = true;
/* Initialize the ORTE data type support */
if (ORTE_SUCCESS != (ret = orte_ess_base_std_prolog())) {
error = "orte_std_prolog";
goto error;
}
/* setup the global job and node arrays */
orte_job_data = OBJ_NEW(opal_pointer_array_t);
if (ORTE_SUCCESS != (ret = opal_pointer_array_init(orte_job_data,
1,
ORTE_GLOBAL_ARRAY_MAX_SIZE,
1))) {
ORTE_ERROR_LOG(ret);
error = "setup job array";
goto error;
}
orte_node_pool = OBJ_NEW(opal_pointer_array_t);
if (ORTE_SUCCESS != (ret = opal_pointer_array_init(orte_node_pool,
ORTE_GLOBAL_ARRAY_BLOCK_SIZE,
ORTE_GLOBAL_ARRAY_MAX_SIZE,
ORTE_GLOBAL_ARRAY_BLOCK_SIZE))) {
ORTE_ERROR_LOG(ret);
error = "setup node array";
goto error;
}
orte_node_topologies = OBJ_NEW(opal_pointer_array_t);
if (ORTE_SUCCESS != (ret = opal_pointer_array_init(orte_node_topologies,
ORTE_GLOBAL_ARRAY_BLOCK_SIZE,
ORTE_GLOBAL_ARRAY_MAX_SIZE,
ORTE_GLOBAL_ARRAY_BLOCK_SIZE))) {
ORTE_ERROR_LOG(ret);
error = "setup node topologies array";
goto error;
}
/* create a job tracker for the daemons */
jdata = OBJ_NEW(orte_job_t);
jdata->jobid = 0;
ORTE_PROC_MY_NAME->jobid = 0;
opal_pointer_array_set_item(orte_job_data, 0, jdata);
/* read the site configuration */
OBJ_CONSTRUCT(&config, opal_list_t);
if (ORCM_SUCCESS != (ret = orcm_cfgi.read_config(&config))) {
error = "getting config";
goto error;
}
/* define the cluster and collect contact info for all
* aggregators - we'll need to know how to talk to any
* of them in case of failures
*/
OBJ_CONSTRUCT(&buf, opal_buffer_t);
if (ORCM_SUCCESS != (ret = orcm_cfgi.define_system(&config,
&mynode,
&orte_process_info.num_procs,
&buf))) {
OBJ_DESTRUCT(&buf);
error = "define system";
goto error;
}
/* if my name didn't get set, then we didn't find our node
* in the config - report it and die
*/
if (NULL == mynode) {
orte_show_help("help-ess-orcm.txt", "node-not-found", true,
orcm_cfgi_base.config_file,
orte_process_info.nodename);
OBJ_DESTRUCT(&buf);
return ORTE_ERR_SILENT;
}
/* define a node and proc object for ourselves as some parts
* of ORTE and ORCM require it */
if (NULL == (node = OBJ_NEW(orte_node_t))) {
ret = ORTE_ERR_OUT_OF_RESOURCE;
error = "out of memory";
goto error;
}
node->name = strdup(orte_process_info.nodename);
opal_pointer_array_set_item(orte_node_pool, ORTE_PROC_MY_NAME->vpid, node);
if (NULL == (proc = OBJ_NEW(orte_proc_t))) {
ret = ORTE_ERR_OUT_OF_RESOURCE;
error = "out of memory";
goto error;
}
proc->name.jobid = ORTE_PROC_MY_NAME->jobid;
proc->name.vpid = ORTE_PROC_MY_NAME->vpid;
OBJ_RETAIN(proc);
node->daemon = proc;
OBJ_RETAIN(node);
proc->node = node;
opal_pointer_array_set_item(jdata->procs, ORTE_PROC_MY_NAME->vpid, proc);
/* For now, we only support a single scheduler daemon in the system.
* This *may* change someday in the future */
scheduler = (orcm_scheduler_t*)opal_list_get_first(orcm_schedulers);
/* If we are in test mode, then we don't *require* that a scheduler
* be defined in the system - otherwise, we do */
if (NULL == scheduler) {
if (mca_sst_orcmd_component.scheduler_reqd) {
error = "no scheduler found";
ret = ORTE_ERR_NOT_FOUND;
goto error;
}
} else {
ORTE_PROC_MY_SCHEDULER->jobid = scheduler->controller.daemon.jobid;
ORTE_PROC_MY_SCHEDULER->vpid = scheduler->controller.daemon.vpid;
}
/* register the ORTE-level params at this time now that the
* config has had a chance to push things into the environ
*/
if (ORTE_SUCCESS != (ret = orte_register_params())) {
OBJ_DESTRUCT(&buf);
error = "orte_register_params";
goto error;
}
/* setup callback for SIGPIPE */
setup_sighandler(SIGPIPE, &epipe_handler, epipe_signal_callback);
/* Set signal handlers to catch kill signals so we can properly clean up
* after ourselves.
*/
setup_sighandler(SIGTERM, &term_handler, shutdown_signal);
setup_sighandler(SIGINT, &int_handler, shutdown_signal);
/** setup callbacks for signals we should ignore */
setup_sighandler(SIGUSR1, &sigusr1_handler, signal_callback);
setup_sighandler(SIGUSR2, &sigusr2_handler, signal_callback);
signals_set = true;
#if OPAL_HAVE_HWLOC
{
hwloc_obj_t obj;
unsigned i, j;
/* get the local topology */
if (NULL == opal_hwloc_topology) {
if (OPAL_SUCCESS != opal_hwloc_base_get_topology()) {
OBJ_DESTRUCT(&buf);
error = "topology discovery";
goto error;
}
}
/* remove the hostname from the topology. Unfortunately, hwloc
* decided to add the source hostname to the "topology", thus
* rendering it unusable as a pure topological description. So
* we remove that information here.
*/
obj = hwloc_get_root_obj(opal_hwloc_topology);
for (i=0; i < obj->infos_count; i++) {
if (NULL == obj->infos[i].name ||
NULL == obj->infos[i].value) {
continue;
}
if (0 == strncmp(obj->infos[i].name, "HostName", strlen("HostName"))) {
free(obj->infos[i].name);
free(obj->infos[i].value);
/* left justify the array */
for (j=i; j < obj->infos_count-1; j++) {
obj->infos[j] = obj->infos[j+1];
}
obj->infos[obj->infos_count-1].name = NULL;
obj->infos[obj->infos_count-1].value = NULL;
obj->infos_count--;
break;
}
}
if (15 < opal_output_get_verbosity(orcm_sst_base_framework.framework_output)) {
opal_output(0, "%s Topology Info:", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
opal_dss.dump(0, opal_hwloc_topology, OPAL_HWLOC_TOPO);
}
/* if we were asked to bind to specific core(s), do so now */
if (NULL != orte_daemon_cores) {
char **cores=NULL, tmp[128];
hwloc_obj_t pu;
hwloc_cpuset_t ours, pucpus, res;
int core;
/* could be a collection of comma-delimited ranges, so
* use our handy utility to parse it
*/
orte_util_parse_range_options(orte_daemon_cores, &cores);
if (NULL != cores) {
ours = hwloc_bitmap_alloc();
hwloc_bitmap_zero(ours);
pucpus = hwloc_bitmap_alloc();
res = hwloc_bitmap_alloc();
for (i=0; NULL != cores[i]; i++) {
core = strtoul(cores[i], NULL, 10);
if (NULL == (pu = opal_hwloc_base_get_pu(opal_hwloc_topology, core, OPAL_HWLOC_LOGICAL))) {
orte_show_help("help-orted.txt", "orted:cannot-bind",
true, orte_process_info.nodename,
orte_daemon_cores);
ret = ORTE_ERR_NOT_SUPPORTED;
OBJ_DESTRUCT(&buf);
error = "cannot bind";
goto error;
}
hwloc_bitmap_and(pucpus, pu->online_cpuset, pu->allowed_cpuset);
hwloc_bitmap_or(res, ours, pucpus);
hwloc_bitmap_copy(ours, res);
}
/* if the result is all zeros, then don't bind */
if (!hwloc_bitmap_iszero(ours)) {
(void)hwloc_set_cpubind(opal_hwloc_topology, ours, 0);
if (opal_hwloc_report_bindings) {
opal_hwloc_base_cset2mapstr(tmp, sizeof(tmp), opal_hwloc_topology, ours);
opal_output(0, "Daemon %s is bound to cores %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
}
}
/* cleanup */
hwloc_bitmap_free(ours);
hwloc_bitmap_free(pucpus);
hwloc_bitmap_free(res);
opal_argv_free(cores);
}
}
}
#endif
/* open and select the pstat framework */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&opal_pstat_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "opal_pstat_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = opal_pstat_base_select())) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "opal_pstat_base_select";
goto error;
}
/* open and setup the state machine */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_state_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_state_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_state_base_select())) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_state_base_select";
goto error;
}
/* open the notifier */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_notifier_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_notifier_base_open";
goto error;
}
/* open the errmgr */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_errmgr_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_errmgr_base_open";
goto error;
}
/* Setup the communication infrastructure */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_oob_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_oob_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_oob_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_oob_base_select";
goto error;
}
if (!opal_list_get_size(&orte_oob_base.actives)) {
ret = ORTE_ERROR;
error = "orte_oob: Found 0 active transports";
goto error;
}
/* Runtime Messaging Layer */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_rml_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_rml_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_rml_base_select())) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_rml_base_select";
goto error;
}
/* select the notifier*/
if (ORTE_SUCCESS != (ret = orte_notifier_base_select())) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_notifier_base_select";
goto error;
}
/* select the errmgr */
if (ORTE_SUCCESS != (ret = orte_errmgr_base_select())) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_errmgr_base_select";
goto error;
}
/* Routed system */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_routed_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_rml_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_routed_base_select())) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orte_routed_base_select";
goto error;
}
/* database */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orcm_db_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orcm_db_base_open";
goto error;
}
/* always restrict daemons to local database components */
if (ORTE_SUCCESS != (ret = orcm_db_base_select())) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "orcm_db_base_select";
goto error;
}
/* datastore - ensure we don't pickup the pmi component, but
* don't override anything set by user
*/
if (NULL == getenv(OPAL_MCA_PREFIX"dstore")) {
putenv(OPAL_MCA_PREFIX"dstore=^pmi");
}
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&opal_dstore_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "opal_dstore_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = opal_dstore_base_select())) {
ORTE_ERROR_LOG(ret);
error = "opal_dstore_base_select";
goto error;
}
/* create the handle */
if (0 > (opal_dstore_internal = opal_dstore.open("INTERNAL", NULL, NULL))) {
error = "opal dstore internal";
ret = ORTE_ERR_FATAL;
goto error;
}
/* extract the cluster description and setup the routed info - the orcm routed component
* will know what to do. */
n = 1;
if (OPAL_SUCCESS != (ret = opal_dss.unpack(&buf, &clusterbuf, &n, OPAL_BUFFER))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "extract cluster buf";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_routed.init_routes(ORTE_PROC_MY_NAME->jobid, clusterbuf))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
OBJ_RELEASE(clusterbuf);
error = "orte_routed.init_routes";
goto error;
}
OBJ_RELEASE(clusterbuf);
/* extract the uri buffer and load the hash tables */
n = 1;
if (OPAL_SUCCESS != (ret = opal_dss.unpack(&buf, &uribuf, &n, OPAL_BUFFER))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
error = "extract uri buffer";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_rml_base_update_contact_info(uribuf))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&buf);
OBJ_RELEASE(uribuf);
error = "load hash tables";
goto error;
}
OBJ_DESTRUCT(&buf);
OBJ_RELEASE(uribuf);
/*
* Group communications
*/
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_grpcomm_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orte_grpcomm_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_grpcomm_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_grpcomm_base_select";
goto error;
}
/* Open/select the odls */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_odls_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orte_odls_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_odls_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_odls_base_select";
goto error;
}
/* enable communication with the rml */
if (ORTE_SUCCESS != (ret = orte_rml.enable_comm())) {
ORTE_ERROR_LOG(ret);
error = "orte_rml.enable_comm";
goto error;
}
/* setup the FileM */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_filem_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orte_filem_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_filem_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_filem_base_select";
goto error;
}
/*
* Initalize the CR setup
* Note: Always do this, even in non-FT builds.
* If we don't some user level tools may hang.
*/
opal_cr_set_enabled(false);
if (ORTE_SUCCESS != (ret = orte_cr_init())) {
ORTE_ERROR_LOG(ret);
error = "orte_cr_init";
goto error;
}
/* setup the ANALYTICS framework */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orcm_analytics_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orcm_analytics_base_open";
goto error;
}
/* setup the EVGEN framework */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orcm_evgen_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orcm_evgen_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orcm_evgen_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orcm_evgen_select";
goto error;
}
/* setup the SENSOR framework */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orcm_sensor_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orcm_sensor_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orcm_sensor_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orcm_sensor_select";
goto error;
}
/* start the local sensors */
orcm_sensor.start(ORTE_PROC_MY_NAME->jobid);
/* setup the PWRMGMT framework */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orcm_pwrmgmt_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orcm_pwrmgmt_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orcm_pwrmgmt_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orcm_pwrmgmt_select";
goto error;
}
/* setup the DFS framework */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_dfs_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orte_dfs_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_dfs_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_dfs_select";
goto error;
}
/* open and setup the DIAG framework */
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orcm_diag_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orcm_diag_base_open";
goto error;
}
if (ORCM_SUCCESS != (ret = orcm_diag_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orcm_diag_select";
goto error;
}
return ORTE_SUCCESS;
error:
orte_show_help("help-orcm-runtime.txt",
"orcm_init:startup:internal-failure",
true, error, ORTE_ERROR_NAME(ret), ret);
return ORTE_ERR_SILENT;
}
static int do_child(orte_app_context_t* context,
orte_proc_t *child,
char **environ_copy,
orte_job_t *jobdat, int write_fd,
orte_iof_base_io_conf_t opts)
{
int i, rc;
sigset_t sigs;
long fd, fdmax = sysconf(_SC_OPEN_MAX);
char *param, *msg;
if (orte_forward_job_control) {
/* Set a new process group for this child, so that a
SIGSTOP can be sent to it without being sent to the
orted. */
setpgid(0, 0);
}
/* Setup the pipe to be close-on-exec */
fcntl(write_fd, F_SETFD, FD_CLOEXEC);
if (NULL != child) {
/* setup stdout/stderr so that any error messages that we
may print out will get displayed back at orterun.
NOTE: Definitely do this AFTER we check contexts so
that any error message from those two functions doesn't
come out to the user. IF we didn't do it in this order,
THEN a user who gives us a bad executable name or
working directory would get N error messages, where
N=num_procs. This would be very annoying for large
jobs, so instead we set things up so that orterun
always outputs a nice, single message indicating what
happened
*/
if (ORTE_SUCCESS != (i = orte_iof_base_setup_child(&opts,
&environ_copy))) {
ORTE_ERROR_LOG(i);
send_error_show_help(write_fd, 1,
"help-orte-odls-default.txt",
"iof setup failed",
orte_process_info.nodename, context->app);
/* Does not return */
}
#if OPAL_HAVE_HWLOC
{
hwloc_cpuset_t cpuset;
hwloc_obj_t root;
opal_hwloc_topo_data_t *sum;
/* Set process affinity, if given */
if (NULL == child->cpu_bitmap) {
/* if the daemon is bound, then we need to "free" this proc */
if (NULL != orte_daemon_cores) {
root = hwloc_get_root_obj(opal_hwloc_topology);
if (NULL == root->userdata) {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "incorrectly bound",
orte_process_info.nodename, context->app,
__FILE__, __LINE__);
}
sum = (opal_hwloc_topo_data_t*)root->userdata;
/* bind this proc to all available processors */
hwloc_set_cpubind(opal_hwloc_topology, sum->available, 0);
}
if (opal_hwloc_report_bindings) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
}
} else {
if (0 == strlen(child->cpu_bitmap)) {
/* this proc is not bound */
if (opal_hwloc_report_bindings) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
}
/* if the daemon is bound, then we need to "free" this proc */
if (NULL != orte_daemon_cores) {
root = hwloc_get_root_obj(opal_hwloc_topology);
if (NULL == root->userdata) {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "incorrectly bound",
orte_process_info.nodename, context->app,
__FILE__, __LINE__);
}
sum = (opal_hwloc_topo_data_t*)root->userdata;
/* bind this proc to all available processors */
hwloc_set_cpubind(opal_hwloc_topology, sum->available, 0);
}
/* provide a nice string representation of what we bound to */
(void) mca_base_var_env_name ("orte_base_applied_binding", ¶m);
opal_setenv(param, child->cpu_bitmap, true, &environ_copy);
free(param);
goto PROCEED;
}
/* convert the list to a cpu bitmap */
cpuset = hwloc_bitmap_alloc();
if (0 != (rc = hwloc_bitmap_list_sscanf(cpuset, child->cpu_bitmap))) {
/* See comment above about "This may be a small memory leak" */
asprintf(&msg, "hwloc_bitmap_sscanf returned \"%s\" for the string \"%s\"",
opal_strerror(rc), child->cpu_bitmap);
if (NULL == msg) {
msg = "failed to convert bitmap list to hwloc bitmap";
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding) &&
OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
/* If binding is required and a binding directive was explicitly
* given (i.e., we are not binding due to a default policy),
* send an error up the pipe (which exits -- it doesn't return).
*/
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename,
context->app, msg,
__FILE__, __LINE__);
} else {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
goto PROCEED;
}
}
/* bind as specified */
rc = hwloc_set_cpubind(opal_hwloc_topology, cpuset, 0);
/* if we got an error and this wasn't a default binding policy, then report it */
if (rc < 0 && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
char *tmp = NULL;
if (errno == ENOSYS) {
msg = "hwloc indicates cpu binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates cpu binding cannot be enforced";
} else {
asprintf(&msg, "hwloc_set_cpubind returned \"%s\" for bitmap \"%s\"",
opal_strerror(rc), child->cpu_bitmap);
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding)) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
if (NULL != tmp) {
free(tmp);
free(msg);
}
goto PROCEED;
}
if (NULL != tmp) {
free(tmp);
free(msg);
}
}
if (0 == rc && opal_hwloc_report_bindings) {
char tmp1[1024], tmp2[1024];
hwloc_cpuset_t mycpus;
/* get the cpus we are bound to */
mycpus = hwloc_bitmap_alloc();
if (hwloc_get_cpubind(opal_hwloc_topology,
mycpus,
HWLOC_CPUBIND_PROCESS) < 0) {
opal_output(0, "MCW rank %d is not bound",
child->name.vpid);
} else {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), opal_hwloc_topology, mycpus)) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), opal_hwloc_topology, mycpus);
opal_output(0, "MCW rank %d bound to %s: %s",
child->name.vpid, tmp1, tmp2);
}
}
hwloc_bitmap_free(mycpus);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
}
/* set memory affinity policy - if we get an error, don't report
* anything unless the user actually specified the binding policy
*/
rc = opal_hwloc_base_set_process_membind_policy();
if (ORTE_SUCCESS != rc && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
if (errno == ENOSYS) {
msg = "hwloc indicates memory binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates memory binding cannot be enforced";
} else {
msg = "failed to bind memory";
}
if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"memory binding error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "memory not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
goto PROCEED;
}
}
}
}
#endif
} else if (!(ORTE_JOB_CONTROL_FORWARD_OUTPUT & jobdat->controls)) {
/* tie stdin/out/err/internal to /dev/null */
int fdnull;
for (i=0; i < 3; i++) {
fdnull = open("/dev/null", O_RDONLY, 0);
if (fdnull > i && i != write_fd) {
dup2(fdnull, i);
}
close(fdnull);
}
fdnull = open("/dev/null", O_RDONLY, 0);
if (fdnull > opts.p_internal[1]) {
dup2(fdnull, opts.p_internal[1]);
}
close(fdnull);
}
#if OPAL_HAVE_HWLOC
PROCEED:
#endif
/* if the user requested it, set the system resource limits */
if (OPAL_SUCCESS != (rc = opal_util_init_sys_limits(&msg))) {
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"set limit",
orte_process_info.nodename, context->app,
__FILE__, __LINE__, msg);
}
/* ensure we only do this once */
(void) mca_base_var_env_name("opal_set_max_sys_limits", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
/* close all file descriptors w/ exception of stdin/stdout/stderr,
the pipe used for the IOF INTERNAL messages, and the pipe up to
the parent. */
for(fd=3; fd<fdmax; fd++) {
if (fd != opts.p_internal[1] && fd != write_fd) {
close(fd);
}
}
if (context->argv == NULL) {
context->argv = malloc(sizeof(char*)*2);
context->argv[0] = strdup(context->app);
context->argv[1] = NULL;
}
/* Set signal handlers back to the default. Do this close to
the exev() because the event library may (and likely will)
reset them. If we don't do this, the event library may
have left some set that, at least on some OS's, don't get
reset via fork() or exec(). Hence, the launched process
could be unkillable (for example). */
set_handler_default(SIGTERM);
set_handler_default(SIGINT);
set_handler_default(SIGHUP);
set_handler_default(SIGPIPE);
set_handler_default(SIGCHLD);
/* Unblock all signals, for many of the same reasons that we
set the default handlers, above. This is noticable on
Linux where the event library blocks SIGTERM, but we don't
want that blocked by the launched process. */
sigprocmask(0, 0, &sigs);
sigprocmask(SIG_UNBLOCK, &sigs, 0);
/* Exec the new executable */
if (10 < opal_output_get_verbosity(orte_odls_base_framework.framework_output)) {
int jout;
opal_output(0, "%s STARTING %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), context->app);
for (jout=0; NULL != context->argv[jout]; jout++) {
opal_output(0, "%s\tARGV[%d]: %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), jout, context->argv[jout]);
}
for (jout=0; NULL != environ_copy[jout]; jout++) {
opal_output(0, "%s\tENVIRON[%d]: %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), jout, environ_copy[jout]);
}
}
execve(context->app, context->argv, environ_copy);
send_error_show_help(write_fd, 1,
"help-orte-odls-default.txt", "execve error",
orte_process_info.nodename, context->app, strerror(errno));
/* Does not return */
}
int orte_ess_base_proc_binding(void)
{
hwloc_obj_t node, obj;
hwloc_cpuset_t cpus, nodeset;
hwloc_obj_type_t target;
unsigned int cache_level = 0;
struct hwloc_topology_support *support;
char *map;
int ret;
char *error=NULL;
hwloc_cpuset_t mycpus;
/* Determine if we were pre-bound or not */
if (NULL != getenv(OPAL_MCA_PREFIX"orte_bound_at_launch")) {
orte_proc_is_bound = true;
if (NULL != (map = getenv(OPAL_MCA_PREFIX"orte_base_applied_binding"))) {
orte_proc_applied_binding = hwloc_bitmap_alloc();
if (0 != (ret = hwloc_bitmap_list_sscanf(orte_proc_applied_binding, map))) {
error = "applied_binding parse";
goto error;
}
}
}
/* see if we were bound when launched */
if (!orte_proc_is_bound) {
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Not bound at launch",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* we were not bound at launch */
if (NULL == opal_hwloc_topology) {
/* there is nothing we can do, so just return */
return ORTE_SUCCESS;
}
support = (struct hwloc_topology_support*)hwloc_topology_get_support(opal_hwloc_topology);
/* get our node object */
node = hwloc_get_root_obj(opal_hwloc_topology);
nodeset = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, node);
/* get our bindings */
cpus = hwloc_bitmap_alloc();
if (hwloc_get_cpubind(opal_hwloc_topology, cpus, HWLOC_CPUBIND_PROCESS) < 0) {
/* we are NOT bound if get_cpubind fails, nor can we be bound - the
* environment does not support it
*/
hwloc_bitmap_free(cpus);
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Binding not supported",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto MOVEON;
}
/* we are bound if the two cpusets are not equal,
* or if there is only ONE cpu available to us
*/
if (0 != hwloc_bitmap_compare(cpus, nodeset) ||
opal_hwloc_base_single_cpu(nodeset) ||
opal_hwloc_base_single_cpu(cpus)) {
/* someone external set it - indicate it is set
* so that we know
*/
orte_proc_is_bound = true;
hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus);
hwloc_bitmap_free(cpus);
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Process was externally bound",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
} else if (support->cpubind->set_thisproc_cpubind &&
OPAL_BINDING_POLICY_IS_SET(opal_hwloc_binding_policy) &&
OPAL_BIND_TO_NONE != OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
/* the system is capable of doing processor affinity, but it
* has not yet been set - see if a slot_list was given
*/
hwloc_bitmap_zero(cpus);
if (OPAL_BIND_TO_CPUSET == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
if (OPAL_SUCCESS != (ret = opal_hwloc_base_slot_list_parse(opal_hwloc_base_slot_list,
opal_hwloc_topology,
OPAL_HWLOC_LOGICAL, cpus))) {
error = "Setting processor affinity failed";
hwloc_bitmap_free(cpus);
goto error;
}
if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) {
error = "Setting processor affinity failed";
hwloc_bitmap_free(cpus);
goto error;
}
hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus);
hwloc_bitmap_free(cpus);
orte_proc_is_bound = true;
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Process bound according to slot_list",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
} else {
/* cleanup */
hwloc_bitmap_free(cpus);
/* get the node rank */
if (ORTE_NODE_RANK_INVALID == orte_process_info.my_node_rank) {
/* this is not an error - could be due to being
* direct launched - so just ignore and leave
* us unbound
*/
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Process not bound - no node rank available",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto MOVEON;
}
/* if the binding policy is hwthread, then we bind to the nrank-th
* hwthread on this node
*/
if (OPAL_BIND_TO_HWTHREAD == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(opal_hwloc_topology, HWLOC_OBJ_PU,
0, orte_process_info.my_node_rank, OPAL_HWLOC_LOGICAL))) {
ret = ORTE_ERR_NOT_FOUND;
error = "Getting hwthread object";
goto error;
}
cpus = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, obj);
if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) {
ret = ORTE_ERROR;
error = "Setting processor affinity failed";
goto error;
}
hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus);
hwloc_bitmap_free(cpus);
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Process bound to hwthread",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
} else if (OPAL_BIND_TO_CORE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
/* if the binding policy is core, then we bind to the nrank-th
* core on this node
*/
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(opal_hwloc_topology, HWLOC_OBJ_CORE,
0, orte_process_info.my_node_rank, OPAL_HWLOC_LOGICAL))) {
ret = ORTE_ERR_NOT_FOUND;
error = "Getting core object";
goto error;
}
cpus = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, obj);
if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) {
error = "Setting processor affinity failed";
ret = ORTE_ERROR;
goto error;
}
hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus);
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Process bound to core",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
} else {
/* for all higher binding policies, we bind to the specified
* object that the nrank-th core belongs to
*/
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(opal_hwloc_topology, HWLOC_OBJ_CORE,
0, orte_process_info.my_node_rank, OPAL_HWLOC_LOGICAL))) {
ret = ORTE_ERR_NOT_FOUND;
error = "Getting core object";
goto error;
}
if (OPAL_BIND_TO_L1CACHE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
target = HWLOC_OBJ_CACHE;
cache_level = 1;
} else if (OPAL_BIND_TO_L2CACHE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
target = HWLOC_OBJ_CACHE;
cache_level = 2;
} else if (OPAL_BIND_TO_L3CACHE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
target = HWLOC_OBJ_CACHE;
cache_level = 3;
} else if (OPAL_BIND_TO_SOCKET == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
target = HWLOC_OBJ_SOCKET;
} else if (OPAL_BIND_TO_NUMA == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) {
target = HWLOC_OBJ_NODE;
} else {
ret = ORTE_ERR_NOT_FOUND;
error = "Binding policy not known";
goto error;
}
for (obj = obj->parent; NULL != obj; obj = obj->parent) {
if (target == obj->type) {
if (HWLOC_OBJ_CACHE == target && cache_level != obj->attr->cache.depth) {
continue;
}
/* this is the place! */
cpus = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, obj);
if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) {
ret = ORTE_ERROR;
error = "Setting processor affinity failed";
goto error;
}
hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus);
orte_proc_is_bound = true;
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Process bound to %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
hwloc_obj_type_string(target)));
break;
}
}
if (!orte_proc_is_bound) {
ret = ORTE_ERROR;
error = "Setting processor affinity failed";
goto error;
}
}
}
}
} else {
OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,
"%s Process bound at launch",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
}
MOVEON:
/* get or update our local cpuset - it will get used multiple
* times, so it's more efficient to keep a global copy
*/
opal_hwloc_base_get_local_cpuset();
/* get the cpus we are bound to */
mycpus = hwloc_bitmap_alloc();
if (hwloc_get_cpubind(opal_hwloc_topology,
mycpus,
HWLOC_CPUBIND_PROCESS) < 0) {
if (NULL != orte_process_info.cpuset) {
free(orte_process_info.cpuset);
orte_process_info.cpuset = NULL;
}
if (opal_hwloc_report_bindings || 4 < opal_output_get_verbosity(orte_ess_base_framework.framework_output)) {
opal_output(0, "MCW rank %d is not bound",
ORTE_PROC_MY_NAME->vpid);
}
} else {
/* store/update the string representation of our local binding */
if (NULL != orte_process_info.cpuset) {
free(orte_process_info.cpuset);
orte_process_info.cpuset = NULL;
}
hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, mycpus);
/* report the binding, if requested */
if (opal_hwloc_report_bindings || 4 < opal_output_get_verbosity(orte_ess_base_framework.framework_output)) {
char tmp1[1024], tmp2[1024];
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), opal_hwloc_topology, mycpus)) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", ORTE_PROC_MY_NAME->vpid);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), opal_hwloc_topology, mycpus);
opal_output(0, "MCW rank %d bound to %s: %s",
ORTE_PROC_MY_NAME->vpid, tmp1, tmp2);
}
}
}
hwloc_bitmap_free(mycpus);
/* push our cpuset so others can calculate our locality */
if (NULL != orte_process_info.cpuset) {
OPAL_MODEX_SEND_VALUE(ret, OPAL_PMIX_GLOBAL, OPAL_PMIX_CPUSET,
orte_process_info.cpuset, OPAL_STRING);
}
return ORTE_SUCCESS;
error:
if (ORTE_ERR_SILENT != ret) {
orte_show_help("help-orte-runtime",
"orte_init:startup:internal-failure",
true, error, ORTE_ERROR_NAME(ret), ret);
}
return ORTE_ERR_SILENT;
}