int mca_bml_r2_finalize( void )
{
ompi_proc_t** procs;
size_t p, num_procs;
opal_list_item_t* w_item;
if (NULL != btl_names) {
free(btl_names);
btl_names = NULL;
}
/* Similar to mca_bml_r2_del_btl ... */
procs = ompi_proc_all(&num_procs);
if(NULL == procs)
goto CLEANUP;
for (w_item = opal_list_get_first(&mca_btl_base_modules_initialized);
w_item != opal_list_get_end(&mca_btl_base_modules_initialized);
w_item = opal_list_get_next(w_item)) {
mca_btl_base_selected_module_t *sm = (mca_btl_base_selected_module_t *) w_item;
mca_btl_base_module_t* btl = sm->btl_module;
/* unregister the BTL progress function if any */
bml_r2_remove_btl_progress(btl);
/* dont use this btl for any peers */
for( p = 0; p < num_procs; p++ ) {
ompi_proc_t* proc = procs[p];
mca_bml_r2_del_proc_btl(proc, sm->btl_module);
}
}
/* Release the procs as the ompi_proc_all increase their ref_count */
for( p = 0; p < num_procs; p++ ) {
OBJ_RELEASE(procs[p]);
}
free(procs);
CLEANUP:
mca_bml_r2.num_btl_modules = 0;
mca_bml_r2.num_btl_progress = 0;
if( NULL != mca_bml_r2.btl_modules) {
free( mca_bml_r2.btl_modules);
mca_bml_r2.btl_modules = NULL;
}
if( NULL != mca_bml_r2.btl_progress ) {
free( mca_bml_r2.btl_progress);
mca_bml_r2.btl_progress = NULL;
}
/* Do not close the BTL base here; the BML upper layer will take
care of that. */
return OMPI_SUCCESS;
}
/*
* Initial query function that is invoked during MPI_INIT, allowing
* this component to disqualify itself if it doesn't support the
* required level of thread support. This function is invoked exactly
* once.
*/
int mca_coll_sm_init_query(bool enable_progress_threads,
bool enable_mpi_threads)
{
ompi_proc_t *my_proc, **procs;
size_t i, size;
/* See if there are other procs in my job on this node. If not,
then don't bother going any further. */
if (NULL == (my_proc = ompi_proc_local()) ||
NULL == (procs = ompi_proc_all(&size))) {
opal_output_verbose(10, ompi_coll_base_framework.framework_output,
"coll:sm:init_query: weirdness on procs; disqualifying myself");
return OMPI_ERR_OUT_OF_RESOURCE;
}
/* if no session directory was created, then we cannot be used */
if (NULL == ompi_process_info.job_session_dir) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
if (size <= 1) {
opal_output_verbose(10, ompi_coll_base_framework.framework_output,
"coll:sm:init_query: comm size too small; disqualifying myself");
free(procs);
return OMPI_ERR_NOT_AVAILABLE;
}
for (i = 0; i < size; ++i) {
if (procs[i] != my_proc &&
procs[i]->proc_name.jobid == my_proc->proc_name.jobid &&
OPAL_PROC_ON_LOCAL_NODE(procs[i]->proc_flags)) {
break;
}
}
free(procs);
if (i >= size) {
opal_output_verbose(10, ompi_coll_base_framework.framework_output,
"coll:sm:init_query: no other local procs; disqualifying myself");
return OMPI_ERR_NOT_AVAILABLE;
}
/* Don't do much here because we don't really want to allocate any
shared memory until this component is selected to be used. */
opal_output_verbose(10, ompi_coll_base_framework.framework_output,
"coll:sm:init_query: pick me! pick me!");
return OMPI_SUCCESS;
}
int ompi_mpi_finalize(void)
{
int ret;
static int32_t finalize_has_already_started = 0;
opal_list_item_t *item;
struct timeval ompistart, ompistop;
ompi_rte_collective_t *coll;
ompi_proc_t** procs;
size_t nprocs;
/* Be a bit social if an erroneous program calls MPI_FINALIZE in
two different threads, otherwise we may deadlock in
ompi_comm_free() (or run into other nasty lions, tigers, or
bears) */
if (! opal_atomic_cmpset_32(&finalize_has_already_started, 0, 1)) {
/* Note that if we're already finalized, we cannot raise an
MPI exception. The best that we can do is write something
to stderr. */
char hostname[MAXHOSTNAMELEN];
pid_t pid = getpid();
gethostname(hostname, sizeof(hostname));
opal_show_help("help-mpi-runtime.txt",
"mpi_finalize:invoked_multiple_times",
true, hostname, pid);
return MPI_ERR_OTHER;
}
ompi_mpiext_fini();
/* Per MPI-2:4.8, we have to free MPI_COMM_SELF before doing
anything else in MPI_FINALIZE (to include setting up such that
MPI_FINALIZED will return true). */
if (NULL != ompi_mpi_comm_self.comm.c_keyhash) {
ompi_attr_delete_all(COMM_ATTR, &ompi_mpi_comm_self,
ompi_mpi_comm_self.comm.c_keyhash);
OBJ_RELEASE(ompi_mpi_comm_self.comm.c_keyhash);
ompi_mpi_comm_self.comm.c_keyhash = NULL;
}
/* Proceed with MPI_FINALIZE */
ompi_mpi_finalized = true;
/* As finalize is the last legal MPI call, we are allowed to force the release
* of the user buffer used for bsend, before going anywhere further.
*/
(void)mca_pml_base_bsend_detach(NULL, NULL);
nprocs = 0;
procs = ompi_proc_all(&nprocs);
MCA_PML_CALL(del_procs(procs, nprocs));
free(procs);
#if OMPI_ENABLE_PROGRESS_THREADS == 0
opal_progress_set_event_flag(OPAL_EVLOOP_ONCE | OPAL_EVLOOP_NONBLOCK);
#endif
/* Redo ORTE calling opal_progress_event_users_increment() during
MPI lifetime, to get better latency when not using TCP */
opal_progress_event_users_increment();
/* check to see if we want timing information */
if (ompi_enable_timing != 0 && 0 == OMPI_PROC_MY_NAME->vpid) {
gettimeofday(&ompistart, NULL);
}
/* NOTE: MPI-2.1 requires that MPI_FINALIZE is "collective" across
*all* connected processes. This only means that all processes
have to call it. It does *not* mean that all connected
processes need to synchronize (either directly or indirectly).
For example, it is quite easy to construct complicated
scenarios where one job is "connected" to another job via
transitivity, but have no direct knowledge of each other.
Consider the following case: job A spawns job B, and job B
later spawns job C. A "connectedness" graph looks something
like this:
A <--> B <--> C
So what are we *supposed* to do in this case? If job A is
still connected to B when it calls FINALIZE, should it block
until jobs B and C also call FINALIZE?
After lengthy discussions many times over the course of this
project, the issue was finally decided at the Louisville Feb
2009 meeting: no.
Rationale:
- "Collective" does not mean synchronizing. It only means that
every process call it. Hence, in this scenario, every
process in A, B, and C must call FINALIZE.
- KEY POINT: if A calls FINALIZE, then it is erroneous for B or
C to try to communicate with A again.
- Hence, OMPI is *correct* to only effect a barrier across each
jobs' MPI_COMM_WORLD before exiting. Specifically, if A
calls FINALIZE long before B or C, it's *correct* if A exits
at any time (and doesn't notify B or C that it is exiting).
- Arguably, if B or C do try to communicate with the now-gone
A, OMPI should try to print a nice error ("you tried to
communicate with a job that is already gone...") instead of
segv or other Badness. However, that is an *extremely*
difficult problem -- sure, it's easy for A to tell B that it
is finalizing, but how can A tell C? A doesn't even know
about C. You'd need to construct a "connected" graph in a
distributed fashion, which is fraught with race conditions,
etc.
Hence, our conclusion is: OMPI is *correct* in its current
behavior (of only doing a barrier across its own COMM_WORLD)
before exiting. Any problems that occur are as a result of
erroneous MPI applications. We *could* tighten up the erroneous
cases and ensure that we print nice error messages / don't
crash, but that is such a difficult problem that we decided we
have many other, much higher priority issues to handle that deal
with non-erroneous cases. */
/* wait for everyone to reach this point
This is a grpcomm barrier instead of an MPI barrier because an
MPI barrier doesn't ensure that all messages have been transmitted
before exiting, so the possibility of a stranded message exists.
*/
coll = OBJ_NEW(ompi_rte_collective_t);
coll->id = ompi_process_info.peer_fini_barrier;
coll->active = true;
if (OMPI_SUCCESS != (ret = ompi_rte_barrier(coll))) {
OMPI_ERROR_LOG(ret);
return ret;
}
/* wait for barrier to complete */
OMPI_LAZY_WAIT_FOR_COMPLETION(coll->active);
OBJ_RELEASE(coll);
/* check for timing request - get stop time and report elapsed
time if so */
if (ompi_enable_timing && 0 == OMPI_PROC_MY_NAME->vpid) {
gettimeofday(&ompistop, NULL);
opal_output(0, "ompi_mpi_finalize[%ld]: time to execute barrier %ld usec",
(long)OMPI_PROC_MY_NAME->vpid,
(long int)((ompistop.tv_sec - ompistart.tv_sec)*1000000 +
(ompistop.tv_usec - ompistart.tv_usec)));
}
/*
* Shutdown the Checkpoint/Restart Mech.
*/
if (OMPI_SUCCESS != (ret = ompi_cr_finalize())) {
OMPI_ERROR_LOG(ret);
}
/* Shut down any bindings-specific issues: C++, F77, F90 */
/* Remove all memory associated by MPI_REGISTER_DATAREP (per
MPI-2:9.5.3, there is no way for an MPI application to
*un*register datareps, but we don't want the OMPI layer causing
memory leaks). */
while (NULL != (item = opal_list_remove_first(&ompi_registered_datareps))) {
OBJ_RELEASE(item);
}
OBJ_DESTRUCT(&ompi_registered_datareps);
/* Remove all F90 types from the hash tables. As the OBJ_DESTRUCT will
* call a special destructor able to release predefined types, we can
* simply call the OBJ_DESTRUCT on the hash table and all memory will
* be correctly released.
*/
OBJ_DESTRUCT( &ompi_mpi_f90_integer_hashtable );
OBJ_DESTRUCT( &ompi_mpi_f90_real_hashtable );
OBJ_DESTRUCT( &ompi_mpi_f90_complex_hashtable );
/* Free communication objects */
/* free file resources */
if (OMPI_SUCCESS != (ret = ompi_file_finalize())) {
return ret;
}
/* free window resources */
if (OMPI_SUCCESS != (ret = ompi_win_finalize())) {
return ret;
}
if (OMPI_SUCCESS != (ret = ompi_osc_base_finalize())) {
return ret;
}
/* free pml resource */
if(OMPI_SUCCESS != (ret = mca_pml_base_finalize())) {
return ret;
}
/* free communicator resources */
if (OMPI_SUCCESS != (ret = ompi_comm_finalize())) {
return ret;
}
/* free requests */
if (OMPI_SUCCESS != (ret = ompi_request_finalize())) {
return ret;
}
if (OMPI_SUCCESS != (ret = ompi_message_finalize())) {
return ret;
}
/* If requested, print out a list of memory allocated by ALLOC_MEM
but not freed by FREE_MEM */
if (0 != ompi_debug_show_mpi_alloc_mem_leaks) {
mca_mpool_base_tree_print();
}
/* Now that all MPI objects dealing with communications are gone,
shut down MCA types having to do with communications */
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_pml_base_framework) ) ) {
OMPI_ERROR_LOG(ret);
return ret;
}
/* shut down buffered send code */
mca_pml_base_bsend_fini();
#if OPAL_ENABLE_FT_CR == 1
/*
* Shutdown the CRCP Framework, must happen after PML shutdown
*/
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_crcp_base_framework) ) ) {
OMPI_ERROR_LOG(ret);
return ret;
}
#endif
/* Free secondary resources */
/* free attr resources */
if (OMPI_SUCCESS != (ret = ompi_attr_finalize())) {
return ret;
}
/* free group resources */
if (OMPI_SUCCESS != (ret = ompi_group_finalize())) {
return ret;
}
/* free proc resources */
if ( OMPI_SUCCESS != (ret = ompi_proc_finalize())) {
return ret;
}
/* finalize the pubsub functions */
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_pubsub_base_framework) ) ) {
return ret;
}
/* finalize the DPM framework */
if ( OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_dpm_base_framework))) {
return ret;
}
/* free internal error resources */
if (OMPI_SUCCESS != (ret = ompi_errcode_intern_finalize())) {
return ret;
}
/* free error code resources */
if (OMPI_SUCCESS != (ret = ompi_mpi_errcode_finalize())) {
return ret;
}
/* free errhandler resources */
if (OMPI_SUCCESS != (ret = ompi_errhandler_finalize())) {
return ret;
}
/* Free all other resources */
/* free op resources */
if (OMPI_SUCCESS != (ret = ompi_op_finalize())) {
return ret;
}
/* free ddt resources */
if (OMPI_SUCCESS != (ret = ompi_datatype_finalize())) {
return ret;
}
/* free info resources */
if (OMPI_SUCCESS != (ret = ompi_info_finalize())) {
return ret;
}
/* Close down MCA modules */
/* io is opened lazily, so it's only necessary to close it if it
was actually opened */
if (0 < ompi_io_base_framework.framework_refcnt) {
/* May have been "opened" multiple times. We want it closed now */
ompi_io_base_framework.framework_refcnt = 1;
if (OMPI_SUCCESS != mca_base_framework_close(&ompi_io_base_framework)) {
return ret;
}
}
(void) mca_base_framework_close(&ompi_topo_base_framework);
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_osc_base_framework))) {
return ret;
}
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_coll_base_framework))) {
return ret;
}
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_bml_base_framework))) {
return ret;
}
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_mpool_base_framework))) {
return ret;
}
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_rcache_base_framework))) {
return ret;
}
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_allocator_base_framework))) {
return ret;
}
if (NULL != ompi_mpi_main_thread) {
OBJ_RELEASE(ompi_mpi_main_thread);
ompi_mpi_main_thread = NULL;
}
/* Leave the RTE */
if (OMPI_SUCCESS != (ret = ompi_rte_finalize())) {
return ret;
}
/* now close the rte framework */
if (OMPI_SUCCESS != (ret = mca_base_framework_close(&ompi_rte_base_framework) ) ) {
OMPI_ERROR_LOG(ret);
return ret;
}
if (OPAL_SUCCESS != (ret = opal_finalize_util())) {
return ret;
}
/* All done */
return MPI_SUCCESS;
}
int mca_pml_ob1_ft_event( int state )
{
static bool first_continue_pass = false;
ompi_proc_t** procs = NULL;
size_t num_procs;
int ret, p;
ompi_rte_collective_t *coll, *modex;
coll = OBJ_NEW(ompi_rte_collective_t);
coll->id = ompi_process_info.peer_init_barrier;
if(OPAL_CRS_CHECKPOINT == state) {
if( opal_cr_timing_barrier_enabled ) {
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_CRCPBR1);
ompi_rte_barrier(coll);
OMPI_WAIT_FOR_COMPLETION(coll->active);
}
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_P2P0);
}
else if(OPAL_CRS_CONTINUE == state) {
first_continue_pass = !first_continue_pass;
if( !first_continue_pass ) {
if( opal_cr_timing_barrier_enabled ) {
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_COREBR0);
ompi_rte_barrier(coll);
OMPI_WAIT_FOR_COMPLETION(coll->active);
}
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_P2P2);
}
if( orte_cr_continue_like_restart && !first_continue_pass ) {
/*
* Get a list of processes
*/
procs = ompi_proc_all(&num_procs);
if(NULL == procs) {
ret = OMPI_ERR_OUT_OF_RESOURCE;
goto clean;
}
/*
* Refresh the proc structure, and publish our proc info in the modex.
* NOTE: Do *not* call ompi_proc_finalize as there are many places in
* the code that point to indv. procs in this strucutre. For our
* needs here we only need to fix up the modex, bml and pml
* references.
*/
if (OMPI_SUCCESS != (ret = ompi_proc_refresh())) {
opal_output(0,
"pml:ob1: ft_event(Restart): proc_refresh Failed %d",
ret);
for(p = 0; p < (int)num_procs; ++p) {
OBJ_RELEASE(procs[p]);
}
free (procs);
goto clean;
}
}
}
else if(OPAL_CRS_RESTART_PRE == state ) {
/* Nothing here */
}
else if(OPAL_CRS_RESTART == state ) {
/*
* Get a list of processes
*/
procs = ompi_proc_all(&num_procs);
if(NULL == procs) {
ret = OMPI_ERR_OUT_OF_RESOURCE;
goto clean;
}
/*
* Clean out the modex information since it is invalid now.
* ompi_rte_purge_proc_attrs();
* This happens at the ORTE level, so doing it again here will cause
* some issues with socket caching.
*/
/*
* Refresh the proc structure, and publish our proc info in the modex.
* NOTE: Do *not* call ompi_proc_finalize as there are many places in
* the code that point to indv. procs in this strucutre. For our
* needs here we only need to fix up the modex, bml and pml
* references.
*/
if (OMPI_SUCCESS != (ret = ompi_proc_refresh())) {
opal_output(0,
"pml:ob1: ft_event(Restart): proc_refresh Failed %d",
ret);
for(p = 0; p < (int)num_procs; ++p) {
OBJ_RELEASE(procs[p]);
}
free (procs);
goto clean;
}
}
else if(OPAL_CRS_TERM == state ) {
;
}
else {
;
}
/* Call the BML
* BML is expected to call ft_event in
* - BTL(s)
* - MPool(s)
*/
if( OMPI_SUCCESS != (ret = mca_bml.bml_ft_event(state))) {
opal_output(0, "pml:base: ft_event: BML ft_event function failed: %d\n",
ret);
}
if(OPAL_CRS_CHECKPOINT == state) {
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_P2P1);
if( opal_cr_timing_barrier_enabled ) {
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_P2PBR0);
/* JJH Cannot barrier here due to progress engine -- ompi_rte_barrier();*/
}
}
else if(OPAL_CRS_CONTINUE == state) {
if( !first_continue_pass ) {
if( opal_cr_timing_barrier_enabled ) {
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_P2PBR1);
ompi_rte_barrier(coll);
OMPI_WAIT_FOR_COMPLETION(coll->active);
}
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_P2P3);
}
if( orte_cr_continue_like_restart && !first_continue_pass ) {
/*
* Exchange the modex information once again.
* BTLs will have republished their modex information.
*/
modex = OBJ_NEW(ompi_rte_collective_t);
modex->id = ompi_process_info.peer_modex;
if (OMPI_SUCCESS != (ret = orte_grpcomm.modex(modex))) {
opal_output(0,
"pml:ob1: ft_event(Restart): Failed orte_grpcomm.modex() = %d",
ret);
OBJ_RELEASE(modex);
goto clean;
}
OMPI_WAIT_FOR_COMPLETION(modex->active);
OBJ_RELEASE(modex);
/*
* Startup the PML stack now that the modex is running again
* Add the new procs (BTLs redo modex recv's)
*/
if( OMPI_SUCCESS != (ret = mca_pml_ob1_add_procs(procs, num_procs) ) ) {
opal_output(0, "pml:ob1: ft_event(Restart): Failed in add_procs (%d)", ret);
goto clean;
}
/* Is this barrier necessary ? JJH */
if (OMPI_SUCCESS != (ret = ompi_rte_barrier(coll))) {
opal_output(0, "pml:ob1: ft_event(Restart): Failed in ompi_rte_barrier (%d)", ret);
goto clean;
}
OMPI_WAIT_FOR_COMPLETION(coll->active);
if( NULL != procs ) {
for(p = 0; p < (int)num_procs; ++p) {
OBJ_RELEASE(procs[p]);
}
free(procs);
procs = NULL;
}
}
if( !first_continue_pass ) {
if( opal_cr_timing_barrier_enabled ) {
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_P2PBR2);
ompi_rte_barrier(coll);
OMPI_WAIT_FOR_COMPLETION(coll->active);
}
OPAL_CR_SET_TIMER(OPAL_CR_TIMER_CRCP1);
}
}
else if(OPAL_CRS_RESTART_PRE == state ) {
/* Nothing here */
}
else if(OPAL_CRS_RESTART == state ) {
/*
* Exchange the modex information once again.
* BTLs will have republished their modex information.
*/
modex = OBJ_NEW(ompi_rte_collective_t);
modex->id = ompi_process_info.peer_modex;
if (OMPI_SUCCESS != (ret = orte_grpcomm.modex(modex))) {
opal_output(0,
"pml:ob1: ft_event(Restart): Failed orte_grpcomm.modex() = %d",
ret);
OBJ_RELEASE(modex);
goto clean;
}
OMPI_WAIT_FOR_COMPLETION(modex->active);
OBJ_RELEASE(modex);
/*
* Startup the PML stack now that the modex is running again
* Add the new procs (BTLs redo modex recv's)
*/
if( OMPI_SUCCESS != (ret = mca_pml_ob1_add_procs(procs, num_procs) ) ) {
opal_output(0, "pml:ob1: ft_event(Restart): Failed in add_procs (%d)", ret);
goto clean;
}
/* Is this barrier necessary ? JJH */
if (OMPI_SUCCESS != (ret = ompi_rte_barrier(coll))) {
opal_output(0, "pml:ob1: ft_event(Restart): Failed in ompi_rte_barrier (%d)", ret);
goto clean;
}
OMPI_WAIT_FOR_COMPLETION(coll->active);
if( NULL != procs ) {
for(p = 0; p < (int)num_procs; ++p) {
OBJ_RELEASE(procs[p]);
}
free(procs);
procs = NULL;
}
}
else if(OPAL_CRS_TERM == state ) {
;
}
else {
;
}
ret = OMPI_SUCCESS;
clean:
OBJ_RELEASE(coll);
return ret;
}
static int mca_bml_r2_del_btl(mca_btl_base_module_t* btl)
{
ompi_proc_t** procs;
size_t i, m, p, num_procs;
opal_list_item_t* item;
mca_btl_base_module_t** modules;
bool found = false;
if(opal_list_get_size(&mca_btl_base_modules_initialized) == 2) {
opal_output(0, "only one BTL left, can't failover");
return OMPI_SUCCESS;
}
procs = ompi_proc_all(&num_procs);
if(NULL == procs)
return OMPI_SUCCESS;
/* Get rid of the associated progress function */
bml_r2_remove_btl_progress(btl);
/* dont use this btl for any peers */
for( p = 0; p < num_procs; p++ ) {
ompi_proc_t* proc = procs[p];
mca_bml_r2_del_proc_btl(proc, btl);
}
/* remove from the btl list */
for (item = opal_list_get_first(&mca_btl_base_modules_initialized);
item != opal_list_get_end(&mca_btl_base_modules_initialized);
item = opal_list_get_next(item)) {
mca_btl_base_selected_module_t *sm = (mca_btl_base_selected_module_t *) item;
if(sm->btl_module == btl) {
opal_list_remove_item(&mca_btl_base_modules_initialized, item);
free(sm);
found = true;
break;
}
}
if(!found) {
/* doesn't even exist */
goto CLEANUP;
}
/* remove from bml list */
modules = (mca_btl_base_module_t**)malloc(sizeof(mca_btl_base_module_t*) * mca_bml_r2.num_btl_modules-1);
for(i=0,m=0; i<mca_bml_r2.num_btl_modules; i++) {
if(mca_bml_r2.btl_modules[i] != btl) {
modules[m++] = mca_bml_r2.btl_modules[i];
}
}
free(mca_bml_r2.btl_modules);
mca_bml_r2.btl_modules = modules;
mca_bml_r2.num_btl_modules = m;
/* cleanup */
btl->btl_finalize(btl);
CLEANUP:
/* Decrease the ref_count increased by the call to ompi_proc_all */
for( p = 0; p < num_procs; p++ ) {
OBJ_RELEASE(procs[p]);
}
free(procs);
return OMPI_SUCCESS;
}
int mca_bml_r2_ft_event(int state)
{
#if OPAL_ENABLE_FT_CR == 1
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;
const char **btl_list;
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 (opal_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 (opal_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;
}
if( OMPI_SUCCESS != (ret = mca_base_framework_close(&opal_btl_base_framework)) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to close BTL framework\n");
return ret;
}
}
/* Matches OPAL_CRS_RESTART */
else if (opal_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.
*/
opal_pmix.fence(NULL, 0);
/*
* Re-open the BTL framework to get the full list of components.
*/
if( OMPI_SUCCESS != (ret = mca_base_framework_open(&opal_btl_base_framework, 0)) ) {
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(OPAL_ENABLE_PROGRESS_THREADS,
OMPI_ENABLE_THREAD_MULTIPLE) ) ) {
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_endpoints[OMPI_PROC_ENDPOINT_TAG_BML]) {
OBJ_RELEASE(procs[p]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML]);
procs[p]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_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;
}
if( OMPI_SUCCESS != (ret = mca_base_framework_close(&opal_btl_base_framework)) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to close BTL 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.
*/
opal_pmix.fence(NULL, 0);
/*
* 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_var_find("ompi", "btl", NULL, NULL);
btl_list = NULL;
mca_base_var_get_value(param_type, &btl_list, NULL, NULL);
opal_output_verbose(11, ompi_cr_output,
"Restart (Previous BTL MCA): <%s>\n", btl_list ? btl_list[0] : "");
if( OMPI_SUCCESS != (ret = mca_base_framework_open(&opal_btl_base_framework, 0)) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to open BTL framework\n");
return ret;
}
/* The reregistered paramter is guaranteed to have the same index */
btl_list = NULL;
mca_base_var_get_value(param_type, &btl_list, NULL, NULL);
opal_output_verbose(11, ompi_cr_output,
"Restart (New BTL MCA): <%s>\n", btl_list ? btl_list[0] : "");
if( NULL != btl_list ) {
free(btl_list);
btl_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(OPAL_ENABLE_PROGRESS_THREADS,
OMPI_ENABLE_THREAD_MULTIPLE) ) ) {
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_endpoints[OMPI_PROC_ENDPOINT_TAG_BML]) {
OBJ_RELEASE(procs[p]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML]);
procs[p]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML] = NULL;
}
OBJ_RELEASE(procs[p]);
}
if( NULL != procs ) {
free(procs);
procs = NULL;
}
}
else if(OPAL_CRS_TERM == state ) {
;
}
else {
;
}
#endif
return OMPI_SUCCESS;
}
int mca_bml_r2_del_btl(mca_btl_base_module_t* btl)
{
ompi_proc_t** procs;
size_t i, j, m, p, num_procs;
opal_list_item_t* item;
mca_btl_base_module_t** modules;
mca_btl_base_component_progress_fn_t * btl_progress_new;
procs = ompi_proc_all(&num_procs);
if(NULL == procs)
return OMPI_SUCCESS;
/* dont use this btl for any peers */
for(p=0; p<num_procs; p++) {
ompi_proc_t* proc = procs[p];
mca_bml_r2_del_proc_btl(proc, btl);
}
/* remove from the btl list */
for (item = opal_list_get_first(&mca_btl_base_modules_initialized);
item != opal_list_get_end(&mca_btl_base_modules_initialized);
item = opal_list_get_next(item)) {
mca_btl_base_selected_module_t *sm = (mca_btl_base_selected_module_t *) item;
if(sm->btl_module == btl) {
opal_list_remove_item(&mca_btl_base_modules_initialized, item);
free(sm);
break;
}
}
/* remove from bml list */
modules = (mca_btl_base_module_t**)malloc(sizeof(mca_btl_base_module_t*) * mca_bml_r2.num_btl_modules-1);
for(i=0,m=0; i<mca_bml_r2.num_btl_modules; i++) {
if(mca_bml_r2.btl_modules[i] != btl) {
modules[m++] = mca_bml_r2.btl_modules[i];
}
}
free(mca_bml_r2.btl_modules);
mca_bml_r2.btl_modules = modules;
mca_bml_r2.num_btl_modules = m;
/* remove progress function so btl_progress isn't
called on the failed BTL */
if(mca_bml_r2.num_btl_progress <= 1) {
/* nothing left to send on! */
opal_output(0, "%s:%d:%s: only one BTL, can't fail-over!",
__FILE__, __LINE__, __func__);
return OMPI_ERROR;
}
/* figure out which progress functions to keep */
btl_progress_new = (mca_btl_base_component_progress_fn_t*)
malloc(sizeof(mca_btl_base_component_progress_fn_t) *
(mca_bml_r2.num_btl_progress - 1));
j = 0;
for(i = 0; i < mca_bml_r2.num_btl_progress; i++) {
if(btl->btl_component->btl_progress != mca_bml_r2.btl_progress[i]) {
btl_progress_new[j] = mca_bml_r2.btl_progress[i];
j++;
}
}
free(mca_bml_r2.btl_progress);
mca_bml_r2.btl_progress = btl_progress_new;
mca_bml_r2.num_btl_progress--;
/* cleanup */
btl->btl_finalize(btl);
free(procs);
return OMPI_SUCCESS;
}
