int clear_bit(opal_bitmap_t *bm, int bit)
{
int err = opal_bitmap_clear_bit(bm, bit);
if ((err != 0) || opal_bitmap_is_set_bit(bm, bit)) {
fprintf(error_out, "ERROR: clear_bit for bit = %d \n\n", bit);
return ERR_CODE;
}
return 0;
}
int clear_bit(opal_bitmap_t *bm, int bit)
{
int err = opal_bitmap_clear_bit(bm, bit);
if ((err != 0)
|| (bm->bitmap[bit/SIZE_OF_CHAR] & (1 << bit % SIZE_OF_CHAR))) {
fprintf(error_out, "ERROR: clear_bit for bit = %d \n\n", bit);
return ERR_CODE;
}
return 0;
}
void mca_oob_tcp_component_lost_connection(int fd, short args, void *cbdata)
{
mca_oob_tcp_peer_op_t *pop = (mca_oob_tcp_peer_op_t*)cbdata;
uint64_t ui64;
orte_oob_base_peer_t *bpr;
int rc;
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s tcp:lost connection called for peer %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&pop->peer));
/* if we are terminating, or recovery isn't enabled, then don't attempt to reconnect */
if (!orte_enable_recovery || orte_orteds_term_ordered || orte_finalizing || orte_abnormal_term_ordered) {
goto cleanup;
}
/* Mark that we no longer support this peer */
memcpy(&ui64, (char*)&pop->peer, sizeof(uint64_t));
if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&orte_oob_base.peers,
ui64, (void**)&bpr) || NULL == bpr) {
bpr = OBJ_NEW(orte_oob_base_peer_t);
}
opal_bitmap_clear_bit(&bpr->addressable, mca_oob_tcp_component.super.idx);
if (OPAL_SUCCESS != (rc = opal_hash_table_set_value_uint64(&orte_oob_base.peers,
ui64, NULL))) {
ORTE_ERROR_LOG(rc);
}
cleanup:
/* activate the proc state */
if (ORTE_SUCCESS != orte_routed.route_lost(&pop->peer)) {
ORTE_ACTIVATE_PROC_STATE(&pop->peer, ORTE_PROC_STATE_LIFELINE_LOST);
} else {
ORTE_ACTIVATE_PROC_STATE(&pop->peer, ORTE_PROC_STATE_COMM_FAILED);
}
OBJ_RELEASE(pop);
}
void mca_oob_tcp_component_no_route(int fd, short args, void *cbdata)
{
mca_oob_tcp_msg_error_t *mop = (mca_oob_tcp_msg_error_t*)cbdata;
uint64_t ui64;
int rc;
orte_oob_base_peer_t *bpr;
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s tcp:no route called for peer %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&mop->hop));
/* mark that we cannot reach this hop */
memcpy(&ui64, (char*)&(mop->hop), sizeof(uint64_t));
if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&orte_oob_base.peers,
ui64, (void**)&bpr) || NULL == bpr) {
bpr = OBJ_NEW(orte_oob_base_peer_t);
}
opal_bitmap_clear_bit(&bpr->addressable, mca_oob_tcp_component.super.idx);
if (OPAL_SUCCESS != (rc = opal_hash_table_set_value_uint64(&orte_oob_base.peers,
ui64, NULL))) {
ORTE_ERROR_LOG(rc);
}
/* report the error back to the OOB and let it try other components
* or declare a problem
*/
if (!orte_finalizing && !orte_abnormal_term_ordered) {
/* if this was a lifeline, then alert */
if (ORTE_SUCCESS != orte_routed.route_lost(&mop->hop)) {
ORTE_ACTIVATE_PROC_STATE(&mop->hop, ORTE_PROC_STATE_LIFELINE_LOST);
} else {
ORTE_ACTIVATE_PROC_STATE(&mop->hop, ORTE_PROC_STATE_COMM_FAILED);
}
}
OBJ_RELEASE(mop);
}
void mca_oob_tcp_component_hop_unknown(int fd, short args, void *cbdata)
{
mca_oob_tcp_msg_error_t *mop = (mca_oob_tcp_msg_error_t*)cbdata;
uint64_t ui64;
orte_rml_send_t *snd;
orte_oob_base_peer_t *bpr;
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s tcp:unknown hop called for peer %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&mop->hop));
if (orte_finalizing || orte_abnormal_term_ordered) {
/* just ignore the problem */
OBJ_RELEASE(mop);
return;
}
/* mark that this component cannot reach this hop */
memcpy(&ui64, (char*)&(mop->hop), sizeof(uint64_t));
if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&orte_oob_base.peers,
ui64, (void**)&bpr) ||
NULL == bpr) {
/* the overall OOB has no knowledge of this hop. Only
* way this could happen is if the peer contacted us
* via this component, and it wasn't entered into the
* OOB framework hash table. We have no way of knowing
* what to do next, so just output an error message and
* abort */
opal_output(0, "%s ERROR: message to %s requires routing and the OOB has no knowledge of the reqd hop %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&mop->snd->hdr.dst),
ORTE_NAME_PRINT(&mop->hop));
ORTE_ACTIVATE_PROC_STATE(&mop->hop, ORTE_PROC_STATE_COMM_FAILED);
OBJ_RELEASE(mop);
return;
}
opal_bitmap_clear_bit(&bpr->addressable, mca_oob_tcp_component.super.idx);
/* mark that this component cannot reach this destination either */
memcpy(&ui64, (char*)&(mop->snd->hdr.dst), sizeof(uint64_t));
if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&orte_oob_base.peers,
ui64, (void**)&bpr) ||
NULL == bpr) {
opal_output(0, "%s ERROR: message to %s requires routing and the OOB has no knowledge of this process",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&mop->snd->hdr.dst));
ORTE_ACTIVATE_PROC_STATE(&mop->hop, ORTE_PROC_STATE_COMM_FAILED);
OBJ_RELEASE(mop);
return;
}
opal_bitmap_clear_bit(&bpr->addressable, mca_oob_tcp_component.super.idx);
/* post the message to the OOB so it can see
* if another component can transfer it
*/
MCA_OOB_TCP_HDR_NTOH(&mop->snd->hdr);
snd = OBJ_NEW(orte_rml_send_t);
snd->dst = mop->snd->hdr.dst;
snd->origin = mop->snd->hdr.origin;
snd->tag = mop->snd->hdr.tag;
snd->data = mop->snd->data;
snd->count = mop->snd->hdr.nbytes;
snd->cbfunc.iov = NULL;
snd->cbdata = NULL;
/* activate the OOB send state */
ORTE_OOB_SEND(snd);
/* protect the data */
mop->snd->data = NULL;
OBJ_RELEASE(mop);
}
