int mca_btl_ugni_ep_disconnect (mca_btl_base_endpoint_t *ep, bool send_disconnect)
{
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
mca_btl_ugni_device_t *device;
int rc;
if (MCA_BTL_UGNI_EP_STATE_INIT == ep->state) {
/* nothing to do */
return OPAL_SUCCESS;
}
device = ep->smsg_ep_handle.device;
while (device->dev_smsg_local_cq.active_operations) {
/* ensure all sends are complete before removing and procs */
rc = mca_btl_ugni_progress_local_smsg (ugni_module, device);
if (OPAL_SUCCESS != rc) {
break;
}
}
if (MCA_BTL_UGNI_EP_STATE_CONNECTED == ep->state && send_disconnect) {
rc = mca_btl_ugni_ep_send_disconnect (ep);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
BTL_VERBOSE(("could not send disconnect message to peer"));
}
/* wait for the disconnect messagse to go */
do {
/* ensure all sends are complete before removing and procs */
rc = mca_btl_ugni_progress_local_smsg (ugni_module, device);
if (OPAL_SUCCESS != rc) {
break;
}
} while (device->dev_smsg_local_cq.active_operations);
(void) opal_atomic_add_fetch_32 (&ep->smsg_ep_handle.device->smsg_connections, -1);
}
mca_btl_ugni_device_lock (device);
/* NTH: this call may not need the device lock. seems to work without it but
* the lock is here to be safe. */
(void) mca_btl_ugni_ep_handle_cleanup (&ep->smsg_ep_handle);
mca_btl_ugni_device_unlock (device);
if (ep->mailbox) {
opal_free_list_return (&ugni_module->smsg_mboxes, ((opal_free_list_item_t *) ep->mailbox));
ep->mailbox = NULL;
}
ep->state = MCA_BTL_UGNI_EP_STATE_INIT;
return OPAL_SUCCESS;
}
static int mca_btl_ugni_component_progress (void)
{
mca_btl_ugni_module_t *ugni_module;
unsigned int i;
int count = 0;
for (i = 0 ; i < mca_btl_ugni_component.ugni_num_btls ; ++i) {
ugni_module = mca_btl_ugni_component.modules + i;
mca_btl_ugni_progress_wait_list (ugni_module);
count += mca_btl_ugni_progress_datagram (ugni_module);
count += mca_btl_ugni_progress_local_smsg (ugni_module);
count += mca_btl_ugni_progress_remote_smsg (ugni_module);
count += mca_btl_ugni_progress_rdma (ugni_module, 0);
if (mca_btl_ugni_component.progress_thread_enabled) {
count += mca_btl_ugni_progress_rdma (ugni_module, 1);
}
/* post pending after progressing rdma */
mca_btl_ugni_post_pending (ugni_module);
}
return count;
}
static int mca_btl_ugni_component_progress (void)
{
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_component.modules;
int count = 0;
count += mca_btl_ugni_progress_remote_smsg (ugni_module);
if (ugni_module->active_datagrams) {
count += mca_btl_ugni_progress_datagram (ugni_module->devices);
}
for (int i = 0 ; i < mca_btl_ugni_component.virtual_device_count ; ++i) {
mca_btl_ugni_device_t *device = ugni_module->devices + i;
if (device->smsg_connections) {
count += mca_btl_ugni_progress_local_smsg (ugni_module, device);
mca_btl_ugni_progress_wait_list (ugni_module);
}
if (device->dev_rdma_local_cq.active_operations) {
count += mca_btl_ugni_progress_rdma (ugni_module, device, &device->dev_rdma_local_cq);
}
if (mca_btl_ugni_component.progress_thread_enabled && device->dev_rdma_local_irq_cq.active_operations) {
count += mca_btl_ugni_progress_rdma (ugni_module, device, &device->dev_rdma_local_irq_cq);
}
}
return count;
}
static int mca_btl_ugni_component_progress (void)
{
mca_btl_ugni_module_t *ugni_module;
static int64_t call_count = 0;
int64_t cur_call_count = OPAL_THREAD_ADD64(&call_count, 1);
unsigned int i;
int count = 0;
for (i = 0 ; i < mca_btl_ugni_component.ugni_num_btls ; ++i) {
ugni_module = mca_btl_ugni_component.modules + i;
if ((cur_call_count & 0x7) == 0) {
count += mca_btl_ugni_progress_datagram (ugni_module);
}
if (ugni_module->connected_peer_count) {
mca_btl_ugni_progress_wait_list (ugni_module);
count += mca_btl_ugni_progress_local_smsg (ugni_module);
count += mca_btl_ugni_progress_remote_smsg (ugni_module);
}
if (ugni_module->active_rdma_count) {
count += mca_btl_ugni_progress_rdma (ugni_module, 0);
}
if (mca_btl_ugni_component.progress_thread_enabled) {
count += mca_btl_ugni_progress_rdma (ugni_module, 1);
}
/* post pending after progressing rdma */
mca_btl_ugni_post_pending (ugni_module);
}
return count;
}
static int mca_btl_ugni_component_progress (void)
{
mca_btl_ugni_module_t *ugni_module;
unsigned int i;
int count = 0;
for (i = 0 ; i < mca_btl_ugni_component.ugni_num_btls ; ++i) {
ugni_module = mca_btl_ugni_component.modules + i;
mca_btl_ugni_retry_failed (ugni_module);
mca_btl_ugni_progress_wait_list (ugni_module);
count += mca_btl_ugni_progress_datagram (ugni_module);
count += mca_btl_ugni_progress_local_smsg (ugni_module);
count += mca_btl_ugni_progress_remote_smsg (ugni_module);
count += mca_btl_ugni_progress_rdma (ugni_module);
}
return count;
}
static int
mca_btl_ugni_module_finalize (struct mca_btl_base_module_t *btl)
{
mca_btl_ugni_module_t *ugni_module = (mca_btl_ugni_module_t *)btl;
mca_btl_base_endpoint_t *ep;
uint64_t key;
void *node;
int rc;
while (ugni_module->active_send_count) {
/* ensure all sends are complete before closing the module */
rc = mca_btl_ugni_progress_local_smsg (ugni_module);
if (OMPI_SUCCESS != rc) {
break;
}
}
OBJ_DESTRUCT(&ugni_module->eager_frags_send);
OBJ_DESTRUCT(&ugni_module->eager_frags_recv);
OBJ_DESTRUCT(&ugni_module->smsg_frags);
OBJ_DESTRUCT(&ugni_module->rdma_frags);
OBJ_DESTRUCT(&ugni_module->rdma_int_frags);
OBJ_DESTRUCT(&ugni_module->ep_wait_list);
/* close all open connections and release endpoints */
if (ugni_module->initialized) {
rc = opal_hash_table_get_first_key_uint64 (&ugni_module->id_to_endpoint, &key, (void **) &ep, &node);
while (OPAL_SUCCESS == rc) {
if (NULL != ep) {
mca_btl_ugni_release_ep (ep);
}
rc = opal_hash_table_get_next_key_uint64 (&ugni_module->id_to_endpoint, &key, (void **) &ep, node, &node);
}
/* destroy all cqs */
rc = GNI_CqDestroy (ugni_module->rdma_local_cq);
if (GNI_RC_SUCCESS != rc) {
BTL_ERROR(("error tearing down local BTE/FMA CQ"));
}
rc = GNI_CqDestroy (ugni_module->smsg_local_cq);
if (GNI_RC_SUCCESS != rc) {
BTL_ERROR(("error tearing down local SMSG CQ"));
}
rc = GNI_CqDestroy (ugni_module->smsg_remote_cq);
if (GNI_RC_SUCCESS != rc) {
BTL_ERROR(("error tearing down remote SMSG CQ"));
}
/* cancel wildcard post */
rc = GNI_EpPostDataCancelById (ugni_module->wildcard_ep,
MCA_BTL_UGNI_CONNECT_WILDCARD_ID |
OMPI_PROC_MY_NAME->vpid);
if (GNI_RC_SUCCESS != rc) {
BTL_VERBOSE(("btl/ugni error cancelling wildcard post"));
}
/* tear down wildcard endpoint */
rc = GNI_EpDestroy (ugni_module->wildcard_ep);
if (GNI_RC_SUCCESS != rc) {
BTL_VERBOSE(("btl/ugni error destroying endpoint"));
}
if (NULL != ugni_module->smsg_mpool) {
(void) mca_mpool_base_module_destroy (ugni_module->smsg_mpool);
ugni_module->smsg_mpool = NULL;
}
if (NULL != ugni_module->super.btl_mpool) {
(void) mca_mpool_base_module_destroy (ugni_module->super.btl_mpool);
ugni_module->super.btl_mpool = NULL;
}
}
OBJ_DESTRUCT(&ugni_module->pending_smsg_frags_bb);
OBJ_DESTRUCT(&ugni_module->id_to_endpoint);
OBJ_DESTRUCT(&ugni_module->endpoints);
OBJ_DESTRUCT(&ugni_module->failed_frags);
ugni_module->initialized = false;
return OMPI_SUCCESS;
}
