static UCS_CLASS_CLEANUP_FUNC(uct_cm_iface_t)
{
ucs_trace_func("");
ucs_async_unset_event_handler(self->cmdev->fd);
uct_cm_enter(self);
while (self->num_outstanding > 0) {
ib_cm_destroy_id(uct_cm_iface_outstanding_pop(self));
}
ib_cm_destroy_id(self->listen_id);
ib_cm_close_device(self->cmdev);
uct_cm_leave(self);
ucs_free(self->outstanding);
}
static UCS_CLASS_CLEANUP_FUNC(uct_cm_iface_t)
{
ucs_trace_func("");
ucs_async_unset_event_handler(self->cmdev->fd);
uct_cm_enter(self);
while (self->num_outstanding > 0) {
ib_cm_destroy_id(uct_cm_iface_outstanding_pop(self));
}
ib_cm_destroy_id(self->listen_id);
ib_cm_close_device(self->cmdev);
uct_cm_leave(self);
/* At this point all outstanding have been removed, and no further events
* can be added.
*/
ucs_callbackq_remove_all(&uct_cm_iface_worker(self)->progress_q,
uct_cm_iface_progress, self);
ucs_free(self->outstanding);
}
ssize_t uct_cm_ep_am_bcopy(uct_ep_h tl_ep, uint8_t am_id,
uct_pack_callback_t pack_cb, void *arg)
{
uct_cm_iface_t *iface = ucs_derived_of(tl_ep->iface, uct_cm_iface_t);
uct_cm_ep_t *ep = ucs_derived_of(tl_ep, uct_cm_ep_t);
struct ib_cm_sidr_req_param req;
struct ibv_sa_path_rec path;
struct ib_cm_id *id;
ucs_status_t status;
uct_cm_hdr_t *hdr;
size_t payload_len;
size_t total_len;
int ret;
UCT_CHECK_AM_ID(am_id);
uct_cm_enter(iface);
if (iface->num_outstanding >= iface->config.max_outstanding) {
status = UCS_ERR_NO_RESOURCE;
goto err;
}
/* Allocate temporary contiguous buffer */
hdr = ucs_malloc(IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE, "cm_send_buf");
if (hdr == NULL) {
status = UCS_ERR_NO_MEMORY;
goto err;
}
payload_len = pack_cb(hdr + 1, arg);
hdr->am_id = am_id;
hdr->length = payload_len;
total_len = sizeof(*hdr) + payload_len;
status = uct_cm_ep_fill_path_rec(ep, &path);
if (status != UCS_OK) {
goto err_free;
}
/* Fill SIDR request */
memset(&req, 0, sizeof req);
req.path = &path;
req.service_id = ep->dest_addr.id;
req.timeout_ms = iface->config.timeout_ms;
req.private_data = hdr;
req.private_data_len = total_len;
req.max_cm_retries = iface->config.retry_count;
/* Create temporary ID for this message. Will be released when getting REP. */
ret = ib_cm_create_id(iface->cmdev, &id, NULL);
if (ret) {
ucs_error("ib_cm_create_id() failed: %m");
status = UCS_ERR_IO_ERROR;
goto err_free;
}
uct_cm_dump_path(&path);
ret = ib_cm_send_sidr_req(id, &req);
if (ret) {
ucs_error("ib_cm_send_sidr_req() failed: %m");
status = UCS_ERR_IO_ERROR;
goto err_destroy_id;
}
iface->outstanding[iface->num_outstanding++] = id;
UCT_TL_EP_STAT_OP(&ep->super, AM, BCOPY, payload_len);
uct_cm_leave(iface);
uct_cm_iface_trace_data(iface, UCT_AM_TRACE_TYPE_SEND, hdr,
"TX: SIDR_REQ [dlid %d svc 0x%"PRIx64"]",
ntohs(path.dlid), req.service_id);
ucs_free(hdr);
return payload_len;
err_destroy_id:
ib_cm_destroy_id(id);
err_free:
ucs_free(hdr);
err:
uct_cm_leave(iface);
return status;
}
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 void uct_cm_iface_event_handler(void *arg)
{
uct_cm_iface_t *iface = arg;
struct ib_cm_event *event;
struct ib_cm_id *id;
int destroy_id;
int ret;
ucs_trace_func("");
for (;;) {
/* Fetch all events */
ret = ib_cm_get_event(iface->cmdev, &event);
if (ret) {
if (errno != EAGAIN) {
ucs_warn("ib_cm_get_event() failed: %m");
}
return;
}
id = event->cm_id;
/* Handle the event */
switch (event->event) {
case IB_CM_SIDR_REQ_ERROR:
ucs_error("SIDR request error, status: %s",
ibv_wc_status_str(event->param.send_status));
destroy_id = 1;
break;
case IB_CM_SIDR_REQ_RECEIVED:
uct_cm_iface_handle_sidr_req(iface, event);
destroy_id = 1; /* Destroy the ID created by the driver */
break;
case IB_CM_SIDR_REP_RECEIVED:
ucs_trace_data("RX: SIDR_REP [id %p{%u}]", id, id->handle);
uct_cm_iface_outstanding_remove(iface, id);
destroy_id = 1; /* Destroy the ID which was used for sending */
break;
default:
ucs_warn("Unexpected CM event: %d", event->event);
destroy_id = 0;
break;
}
/* Acknowledge CM event, remember the id, in case we would destroy it */
ret = ib_cm_ack_event(event);
if (ret) {
ucs_warn("ib_cm_ack_event() failed: %m");
}
/* If there is an id which should be destroyed, do it now, after
* acknowledging all events.
*/
if (destroy_id) {
ret = ib_cm_destroy_id(id);
if (ret) {
ucs_error("ib_cm_destroy_id() failed: %m");
}
}
uct_cm_iface_notify(iface);
}
}