static UCS_CLASS_INIT_FUNC(uct_cm_iface_t, uct_pd_h pd, uct_worker_h worker,
const char *dev_name, size_t rx_headroom,
const uct_iface_config_t *tl_config)
{
uct_cm_iface_config_t *config = ucs_derived_of(tl_config, uct_cm_iface_config_t);
ucs_status_t status;
int ret;
ucs_trace_func("");
UCS_CLASS_CALL_SUPER_INIT(uct_ib_iface_t, &uct_cm_iface_ops, pd, worker,
dev_name, rx_headroom, 0 /* rx_priv_len */,
0 /* rx_hdr_len */, 1 /* tx_cq_len */,
IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE, /* mss */
&config->super);
if (worker->async == NULL) {
ucs_error("cm must have async!=NULL");
return UCS_ERR_INVALID_PARAM;
}
self->service_id = (uint32_t)(ucs_generate_uuid((uintptr_t)self) &
(~IB_CM_ASSIGN_SERVICE_ID_MASK));
self->num_outstanding = 0;
self->config.timeout_ms = (int)(config->timeout * 1e3 + 0.5);
self->config.max_outstanding = config->max_outstanding;
self->config.retry_count = ucs_min(config->retry_count, UINT8_MAX);
self->notify_q.head = NULL;
ucs_queue_head_init(&self->notify_q);
self->outstanding = ucs_calloc(self->config.max_outstanding,
sizeof(*self->outstanding),
"cm_outstanding");
if (self->outstanding == NULL) {
status = UCS_ERR_NO_MEMORY;
goto err;
}
self->cmdev = ib_cm_open_device(uct_ib_iface_device(&self->super)->ibv_context);
if (self->cmdev == NULL) {
ucs_error("ib_cm_open_device() failed: %m. Check if ib_ucm.ko module is loaded.");
status = UCS_ERR_NO_DEVICE;
goto err_free_outstanding;
}
status = ucs_sys_fcntl_modfl(self->cmdev->fd, O_NONBLOCK, 0);
if (status != UCS_OK) {
goto err_close_device;
}
ret = ib_cm_create_id(self->cmdev, &self->listen_id, self);
if (ret) {
ucs_error("ib_cm_create_id() failed: %m");
status = UCS_ERR_NO_DEVICE;
goto err_close_device;
}
ret = ib_cm_listen(self->listen_id, self->service_id, 0);
if (ret) {
ucs_error("ib_cm_listen() failed: %m");
status = UCS_ERR_INVALID_ADDR;
goto err_destroy_id;
}
if (config->async_mode == UCS_ASYNC_MODE_SIGNAL) {
ucs_warn("ib_cm fd does not support SIGIO");
}
status = ucs_async_set_event_handler(config->async_mode, self->cmdev->fd,
POLLIN, uct_cm_iface_event_handler, self,
worker->async);
if (status != UCS_OK) {
ucs_error("failed to set event handler");
goto err_destroy_id;
}
ucs_debug("listening for SIDR service_id 0x%x on fd %d", self->service_id,
self->cmdev->fd);
return UCS_OK;
err_destroy_id:
ib_cm_destroy_id(self->listen_id);
err_close_device:
ib_cm_close_device(self->cmdev);
err_free_outstanding:
ucs_free(self->outstanding);
err:
return status;
}
static UCS_CLASS_INIT_FUNC(uct_ugni_udt_iface_t, uct_md_h md, uct_worker_h worker,
const uct_iface_params_t *params,
const uct_iface_config_t *tl_config)
{
uct_ugni_iface_config_t *config = ucs_derived_of(tl_config, uct_ugni_iface_config_t);
ucs_status_t status;
uct_ugni_udt_desc_t *desc;
gni_return_t ugni_rc;
int rc;
UCS_CLASS_CALL_SUPER_INIT(uct_ugni_iface_t, md, worker, params,
&uct_ugni_udt_iface_ops,
&config->super UCS_STATS_ARG(NULL));
/* Setting initial configuration */
self->config.udt_seg_size = GNI_DATAGRAM_MAXSIZE;
self->config.rx_headroom = params->rx_headroom;
self->release_desc.cb = uct_ugni_udt_iface_release_desc;
status = ucs_async_pipe_create(&self->event_pipe);
if (UCS_OK != status) {
ucs_error("Pipe creation failed");
goto exit;
}
status = ucs_mpool_init(&self->free_desc,
0,
uct_ugni_udt_get_diff(self) + self->config.udt_seg_size * 2,
uct_ugni_udt_get_diff(self),
UCS_SYS_CACHE_LINE_SIZE, /* alignment */
128, /* grow */
config->mpool.max_bufs, /* max buffers */
&uct_ugni_udt_desc_mpool_ops,
"UGNI-UDT-DESC");
if (UCS_OK != status) {
ucs_error("Mpool creation failed");
goto clean_pipe;
}
ugni_rc = GNI_EpCreate(uct_ugni_udt_iface_nic_handle(self), NULL, &self->ep_any);
if (GNI_RC_SUCCESS != ugni_rc) {
ucs_error("GNI_EpCreate failed, Error status: %s %d",
gni_err_str[ugni_rc], ugni_rc);
status = UCS_ERR_NO_DEVICE;
goto clean_free_desc;
}
UCT_TL_IFACE_GET_TX_DESC(&self->super.super, &self->free_desc,
desc, goto clean_ep);
ucs_debug("First wildcard desc is %p", desc);
/* Init any desc */
self->desc_any = desc;
status = uct_ugni_udt_ep_any_post(self);
if (UCS_OK != status) {
/* We can't continue if we can't post the first receive */
ucs_error("Failed to post wildcard request");
goto clean_any_desc;
}
status = ucs_async_set_event_handler(self->super.super.worker->async->mode,
ucs_async_pipe_rfd(&self->event_pipe),
POLLIN,
uct_ugni_proccess_datagram_pipe,
self, self->super.super.worker->async);
if (UCS_OK != status) {
goto clean_cancel_desc;
}
pthread_mutex_init(&self->device_lock, NULL);
pthread_cond_init(&self->device_condition, NULL);
self->events_ready = 0;
rc = pthread_create(&self->event_thread, NULL, uct_ugni_udt_device_thread, self);
if(0 != rc) {
goto clean_remove_event;
}
return UCS_OK;
clean_remove_event:
ucs_async_pipe_destroy(&self->event_pipe);
clean_cancel_desc:
uct_ugni_udt_clean_wildcard(self);
clean_any_desc:
ucs_mpool_put(self->desc_any);
clean_ep:
ugni_rc = GNI_EpDestroy(self->ep_any);
if (GNI_RC_SUCCESS != ugni_rc) {
ucs_warn("GNI_EpDestroy failed, Error status: %s %d",
gni_err_str[ugni_rc], ugni_rc);
}
clean_free_desc:
ucs_mpool_cleanup(&self->free_desc, 1);
clean_pipe:
ucs_async_pipe_destroy(&self->event_pipe);
exit:
uct_ugni_cleanup_base_iface(&self->super);
ucs_error("Failed to activate interface");
return status;
}
