static void
roq_eth_cleanup_ofa(struct roq_eth_priv *vdev)
{
int i;
if (vdev->send_cq)
ib_destroy_cq(vdev->send_cq);
if (vdev->recv_cq && vdev->recv_cq != vdev->send_cq)
ib_destroy_cq(vdev->recv_cq);
if (vdev->qps) {
for (i = 0; i < vdev->part_size; i++)
if (vdev->qps[i])
ib_destroy_qp(vdev->qps[i]);
kfree(vdev->qps);
}
if (vdev->qps_rem) {
for (i = 0; i < vdev->rem_part_size; i++)
if (vdev->qps_rem[i])
ib_destroy_qp(vdev->qps_rem[i]);
kfree(vdev->qps_rem);
}
if (vdev->kpd)
ib_dealloc_pd(vdev->kpd);
vdev->qps = vdev->qps_rem = NULL;
vdev->recv_cq = vdev->send_cq = NULL;
vdev->kpd = NULL;
return;
}
/*
* rpcrdma_ep_destroy
*
* Disconnect and destroy endpoint. After this, the only
* valid operations on the ep are to free it (if dynamically
* allocated) or re-create it.
*/
void
rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
{
int rc;
dprintk("RPC: %s: entering, connected is %d\n",
__func__, ep->rep_connected);
cancel_delayed_work_sync(&ep->rep_connect_worker);
if (ia->ri_id->qp) {
rpcrdma_ep_disconnect(ep, ia);
rdma_destroy_qp(ia->ri_id);
ia->ri_id->qp = NULL;
}
rpcrdma_free_regbuf(ia, ep->rep_padbuf);
rpcrdma_clean_cq(ep->rep_attr.recv_cq);
rc = ib_destroy_cq(ep->rep_attr.recv_cq);
if (rc)
dprintk("RPC: %s: ib_destroy_cq returned %i\n",
__func__, rc);
rpcrdma_clean_cq(ep->rep_attr.send_cq);
rc = ib_destroy_cq(ep->rep_attr.send_cq);
if (rc)
dprintk("RPC: %s: ib_destroy_cq returned %i\n",
__func__, rc);
}
void IBCompletionQueue::close()
{
if( _cqR )
{
ib_destroy_cq( _cqR, 0 );
_cqR = 0;
}
if( _cqW )
{
ib_destroy_cq( _cqW, 0 );
_cqW = 0;
}
if ( _cqWaitobjR )
{
CloseHandle( _cqWaitobjR );
_cqWaitobjR = 0;
}
if ( _cqWaitobjR )
{
CloseHandle( _cqWaitobjW );
_cqWaitobjW = 0;
}
}
/**
* iser_create_device_ib_res - creates Protection Domain (PD), Completion
* Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
* the adapator.
*
* returns 0 on success, -1 on failure
*/
static int iser_create_device_ib_res(struct iser_device *device)
{
device->pd = ib_alloc_pd(device->ib_device);
if (IS_ERR(device->pd))
goto pd_err;
device->rx_cq = ib_create_cq(device->ib_device,
iser_cq_callback,
iser_cq_event_callback,
(void *)device,
ISER_MAX_RX_CQ_LEN, 0);
if (IS_ERR(device->rx_cq))
goto rx_cq_err;
device->tx_cq = ib_create_cq(device->ib_device,
NULL, iser_cq_event_callback,
(void *)device,
ISER_MAX_TX_CQ_LEN, 0);
if (IS_ERR(device->tx_cq))
goto tx_cq_err;
if (ib_req_notify_cq(device->rx_cq, IB_CQ_NEXT_COMP))
goto cq_arm_err;
tasklet_init(&device->cq_tasklet,
iser_cq_tasklet_fn,
(unsigned long)device);
device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE |
IB_ACCESS_REMOTE_WRITE |
IB_ACCESS_REMOTE_READ);
if (IS_ERR(device->mr))
goto dma_mr_err;
INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device,
iser_event_handler);
if (ib_register_event_handler(&device->event_handler))
goto handler_err;
return 0;
handler_err:
ib_dereg_mr(device->mr);
dma_mr_err:
tasklet_kill(&device->cq_tasklet);
cq_arm_err:
ib_destroy_cq(device->tx_cq);
tx_cq_err:
ib_destroy_cq(device->rx_cq);
rx_cq_err:
ib_dealloc_pd(device->pd);
pd_err:
iser_err("failed to allocate an IB resource\n");
return -1;
}
void
sdp_tx_ring_destroy(struct sdp_sock *ssk)
{
sdp_dbg(ssk->socket, "tx ring destroy\n");
SDP_WLOCK(ssk);
callout_stop(&ssk->tx_ring.timer);
callout_stop(&ssk->nagle_timer);
SDP_WUNLOCK(ssk);
callout_drain(&ssk->tx_ring.timer);
callout_drain(&ssk->nagle_timer);
if (ssk->tx_ring.buffer) {
sdp_tx_ring_purge(ssk);
kfree(ssk->tx_ring.buffer);
ssk->tx_ring.buffer = NULL;
}
if (ssk->tx_ring.cq) {
if (ib_destroy_cq(ssk->tx_ring.cq)) {
sdp_warn(ssk->socket, "destroy cq(%p) failed\n",
ssk->tx_ring.cq);
} else {
ssk->tx_ring.cq = NULL;
}
}
WARN_ON(ring_head(ssk->tx_ring) != ring_tail(ssk->tx_ring));
}
void sdp_tx_ring_destroy(struct sdp_sock *ssk)
{
del_timer_sync(&ssk->tx_ring.timer);
if (ssk->nagle_timer.function)
del_timer_sync(&ssk->nagle_timer);
if (ssk->tx_ring.buffer) {
sdp_tx_ring_purge(ssk);
kfree(ssk->tx_ring.buffer);
ssk->tx_ring.buffer = NULL;
}
if (ssk->tx_ring.cq) {
if (ib_destroy_cq(ssk->tx_ring.cq)) {
sdp_warn(sk_ssk(ssk), "destroy cq(%p) failed\n",
ssk->tx_ring.cq);
} else {
ssk->tx_ring.cq = NULL;
}
}
tasklet_kill(&ssk->tx_ring.tasklet);
/* tx_cq is destroyed, so no more tx_irq, so no one will schedule this
* tasklet. */
SDP_WARN_ON(ring_head(ssk->tx_ring) != ring_tail(ssk->tx_ring));
}
/*
* rpcrdma_ep_destroy
*
* Disconnect and destroy endpoint. After this, the only
* valid operations on the ep are to free it (if dynamically
* allocated) or re-create it.
*
* The caller's error handling must be sure to not leak the endpoint
* if this function fails.
*/
int
rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
{
int rc;
dprintk("RPC: %s: entering, connected is %d\n",
__func__, ep->rep_connected);
if (ia->ri_id->qp) {
rc = rpcrdma_ep_disconnect(ep, ia);
if (rc)
dprintk("RPC: %s: rpcrdma_ep_disconnect"
" returned %i\n", __func__, rc);
rdma_destroy_qp(ia->ri_id);
ia->ri_id->qp = NULL;
}
/* padding - could be done in rpcrdma_buffer_destroy... */
if (ep->rep_pad_mr) {
rpcrdma_deregister_internal(ia, ep->rep_pad_mr, &ep->rep_pad);
ep->rep_pad_mr = NULL;
}
rpcrdma_clean_cq(ep->rep_cq);
rc = ib_destroy_cq(ep->rep_cq);
if (rc)
dprintk("RPC: %s: ib_destroy_cq returned %i\n",
__func__, rc);
return rc;
}
void
sdp_rx_ring_destroy(struct sdp_sock *ssk)
{
cancel_work_sync(&ssk->rx_comp_work);
rx_ring_destroy_lock(&ssk->rx_ring);
if (ssk->rx_ring.buffer) {
sdp_rx_ring_purge(ssk);
kfree(ssk->rx_ring.buffer);
ssk->rx_ring.buffer = NULL;
}
if (ssk->rx_ring.cq) {
if (ib_destroy_cq(ssk->rx_ring.cq)) {
sdp_warn(ssk->socket, "destroy cq(%p) failed\n",
ssk->rx_ring.cq);
} else {
ssk->rx_ring.cq = NULL;
}
}
WARN_ON(ring_head(ssk->rx_ring) != ring_tail(ssk->rx_ring));
}
/**
* iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
* CQ and PD created with the device associated with the adapator.
*/
static void iser_free_device_ib_res(struct iser_device *device)
{
BUG_ON(device->mr == NULL);
tasklet_kill(&device->cq_tasklet);
(void)ib_unregister_event_handler(&device->event_handler);
(void)ib_dereg_mr(device->mr);
(void)ib_destroy_cq(device->tx_cq);
(void)ib_destroy_cq(device->rx_cq);
(void)ib_dealloc_pd(device->pd);
device->mr = NULL;
device->tx_cq = NULL;
device->rx_cq = NULL;
device->pd = NULL;
}
static void verbs_remove_device (struct ib_device *dev)
{
printk (KERN_INFO "IB remove device called. Name = %s\n", dev->name);
if (ah)
ib_destroy_ah (ah);
if (qp)
ib_destroy_qp (qp);
if (send_cq)
ib_destroy_cq (send_cq);
if (recv_cq)
ib_destroy_cq (recv_cq);
if (mr)
ib_dereg_mr (mr);
if (pd)
ib_dealloc_pd (pd);
}
static int get_port_caps(struct mlx5_ib_dev *dev)
{
struct ib_device_attr *dprops = NULL;
struct ib_port_attr *pprops = NULL;
int err = -ENOMEM;
int port;
struct ib_udata uhw = {.inlen = 0, .outlen = 0};
pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
if (!pprops)
goto out;
dprops = kmalloc(sizeof(*dprops), GFP_KERNEL);
if (!dprops)
goto out;
err = mlx5_ib_query_device(&dev->ib_dev, dprops, &uhw);
if (err) {
mlx5_ib_warn(dev, "query_device failed %d\n", err);
goto out;
}
for (port = 1; port <= MLX5_CAP_GEN(dev->mdev, num_ports); port++) {
err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
if (err) {
mlx5_ib_warn(dev, "query_port %d failed %d\n",
port, err);
break;
}
dev->mdev->port_caps[port - 1].pkey_table_len =
dprops->max_pkeys;
dev->mdev->port_caps[port - 1].gid_table_len =
pprops->gid_tbl_len;
mlx5_ib_dbg(dev, "pkey_table_len %d, gid_table_len %d\n",
dprops->max_pkeys, pprops->gid_tbl_len);
}
out:
kfree(pprops);
kfree(dprops);
return err;
}
static void destroy_umrc_res(struct mlx5_ib_dev *dev)
{
int err;
err = mlx5_mr_cache_cleanup(dev);
if (err)
mlx5_ib_warn(dev, "mr cache cleanup failed\n");
mlx5_ib_destroy_qp(dev->umrc.qp);
ib_destroy_cq(dev->umrc.cq);
ib_dealloc_pd(dev->umrc.pd);
}
ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
const char __user *buf, int in_len,
int out_len)
{
struct ib_uverbs_destroy_cq cmd;
struct ib_uverbs_destroy_cq_resp resp;
struct ib_cq *cq;
struct ib_ucq_object *uobj;
struct ib_uverbs_event_file *ev_file;
u64 user_handle;
int ret = -EINVAL;
if (copy_from_user(&cmd, buf, sizeof cmd))
return -EFAULT;
memset(&resp, 0, sizeof resp);
mutex_lock(&ib_uverbs_idr_mutex);
cq = idr_find(&ib_uverbs_cq_idr, cmd.cq_handle);
if (!cq || cq->uobject->context != file->ucontext)
goto out;
user_handle = cq->uobject->user_handle;
uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
ev_file = cq->cq_context;
ret = ib_destroy_cq(cq);
if (ret)
goto out;
idr_remove(&ib_uverbs_cq_idr, cmd.cq_handle);
mutex_lock(&file->mutex);
list_del(&uobj->uobject.list);
mutex_unlock(&file->mutex);
ib_uverbs_release_ucq(file, ev_file, uobj);
resp.comp_events_reported = uobj->comp_events_reported;
resp.async_events_reported = uobj->async_events_reported;
kfree(uobj);
if (copy_to_user((void __user *) (unsigned long) cmd.response,
&resp, sizeof resp))
ret = -EFAULT;
out:
mutex_unlock(&ib_uverbs_idr_mutex);
return ret ? ret : in_len;
}
/**
* Shut down CMRC connection gracefully
*
* @v cmrc Communication-Managed Reliable Connection
*
* The Infiniband data structures are not reference-counted or
* guarded. It is therefore unsafe to shut them down while we may be
* in the middle of a callback from the Infiniband stack (e.g. in a
* receive completion handler).
*
* This shutdown process will run some time after the call to
* ib_cmrc_close(), after control has returned out of the Infiniband
* core, and will shut down the Infiniband interfaces cleanly.
*
* The shutdown process holds an implicit reference on the CMRC
* connection, ensuring that the structure is not freed before the
* shutdown process has run.
*/
static void ib_cmrc_shutdown ( struct ib_cmrc_connection *cmrc ) {
DBGC ( cmrc, "CMRC %p shutting down\n", cmrc );
/* Shut down Infiniband interface */
ib_destroy_conn ( cmrc->ibdev, cmrc->qp, cmrc->conn );
ib_destroy_qp ( cmrc->ibdev, cmrc->qp );
ib_destroy_cq ( cmrc->ibdev, cmrc->cq );
ib_close ( cmrc->ibdev );
/* Drop the remaining reference */
ref_put ( &cmrc->refcnt );
}
static void isert_device_release(struct isert_device *isert_dev)
{
int err, i;
TRACE_ENTRY();
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_remove(isert_dev); /* remove from global list */
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 22)
/*
* cancel_work_sync() was introduced in 2.6.22. We can
* only wait until all scheduled work is done.
*/
flush_workqueue(cq_desc->cq_workqueue);
#else
cancel_work_sync(&cq_desc->cq_comp_work);
#endif
err = ib_destroy_cq(cq_desc->cq);
if (unlikely(err))
pr_err("Failed to destroy cq, err:%d\n", err);
destroy_workqueue(cq_desc->cq_workqueue);
}
err = ib_dereg_mr(isert_dev->mr);
if (unlikely(err))
pr_err("Failed to destroy mr, err:%d\n", err);
err = ib_dealloc_pd(isert_dev->pd);
if (unlikely(err))
pr_err("Failed to destroy pd, err:%d\n", err);
vfree(isert_dev->cq_desc);
isert_dev->cq_desc = NULL;
kfree(isert_dev->cq_qps);
isert_dev->cq_qps = NULL;
kfree(isert_dev);
TRACE_EXIT();
}
static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
{
if (!rdma)
return;
if (rdma->qp && !IS_ERR(rdma->qp))
ib_destroy_qp(rdma->qp);
if (rdma->pd && !IS_ERR(rdma->pd))
ib_dealloc_pd(rdma->pd);
if (rdma->cq && !IS_ERR(rdma->cq))
ib_destroy_cq(rdma->cq);
if (rdma->cm_id && !IS_ERR(rdma->cm_id))
rdma_destroy_id(rdma->cm_id);
kfree(rdma);
}
static int krping_setup_qp(struct krping_cb *cb, struct rdma_cm_id *cm_id)
{
int ret;
cb->pd = ib_alloc_pd(cm_id->device);
if (IS_ERR(cb->pd)) {
log(LOG_ERR, "ib_alloc_pd failed\n");
return PTR_ERR(cb->pd);
}
DEBUG_LOG(PFX "created pd %p\n", cb->pd);
cb->cq = ib_create_cq(cm_id->device, krping_cq_event_handler, NULL,
cb, cb->txdepth * 2, 0);
if (IS_ERR(cb->cq)) {
log(LOG_ERR, "ib_create_cq failed\n");
ret = PTR_ERR(cb->cq);
goto err1;
}
DEBUG_LOG(PFX "created cq %p\n", cb->cq);
if (!cb->wlat && !cb->rlat && !cb->bw) {
ret = ib_req_notify_cq(cb->cq, IB_CQ_NEXT_COMP);
if (ret) {
log(LOG_ERR, "ib_create_cq failed\n");
goto err2;
}
}
ret = krping_create_qp(cb);
if (ret) {
log(LOG_ERR, "krping_create_qp failed: %d\n", ret);
goto err2;
}
DEBUG_LOG(PFX "created qp %p\n", cb->qp);
return 0;
err2:
ib_destroy_cq(cb->cq);
err1:
ib_dealloc_pd(cb->pd);
return ret;
}
ib_cq_handle_t IBCompletionQueue::_createReadBack(
const IBAdapter* adapter,
const uint32_t size)
{
// build a completion queue for send part
ib_cq_create_t cqCreate;
cqCreate.h_wait_obj = 0;
cqCreate.pfn_comp_cb = pp_cq_comp_cb;
cqCreate.size = size;
ib_cq_handle_t _cq;
// Creates a completion queue
ib_api_status_t ibStatus = ib_create_cq( adapter->getHandle(),
&cqCreate, this, 0, &_cq);
if( ibStatus )
{
EQERROR << "Can't create CQ" << std::endl;
ib_destroy_cq( _cq, 0 );
return 0;
}
return _cq;
}
/*
* Create a QP
*/
static int rds_iw_init_qp_attrs(struct ib_qp_init_attr *attr,
struct rds_iw_device *rds_iwdev,
struct rds_iw_work_ring *send_ring,
void (*send_cq_handler)(struct ib_cq *, void *),
struct rds_iw_work_ring *recv_ring,
void (*recv_cq_handler)(struct ib_cq *, void *),
void *context)
{
struct ib_device *dev = rds_iwdev->dev;
unsigned int send_size, recv_size;
int ret;
/* The offset of 1 is to accommodate the additional ACK WR. */
send_size = min_t(unsigned int, rds_iwdev->max_wrs, rds_iw_sysctl_max_send_wr + 1);
recv_size = min_t(unsigned int, rds_iwdev->max_wrs, rds_iw_sysctl_max_recv_wr + 1);
rds_iw_ring_resize(send_ring, send_size - 1);
rds_iw_ring_resize(recv_ring, recv_size - 1);
memset(attr, 0, sizeof(*attr));
attr->event_handler = rds_iw_qp_event_handler;
attr->qp_context = context;
attr->cap.max_send_wr = send_size;
attr->cap.max_recv_wr = recv_size;
attr->cap.max_send_sge = rds_iwdev->max_sge;
attr->cap.max_recv_sge = RDS_IW_RECV_SGE;
attr->sq_sig_type = IB_SIGNAL_REQ_WR;
attr->qp_type = IB_QPT_RC;
attr->send_cq = ib_create_cq(dev, send_cq_handler,
rds_iw_cq_event_handler,
context, send_size, 0);
if (IS_ERR(attr->send_cq)) {
ret = PTR_ERR(attr->send_cq);
attr->send_cq = NULL;
rdsdebug("ib_create_cq send failed: %d\n", ret);
goto out;
}
attr->recv_cq = ib_create_cq(dev, recv_cq_handler,
rds_iw_cq_event_handler,
context, recv_size, 0);
if (IS_ERR(attr->recv_cq)) {
ret = PTR_ERR(attr->recv_cq);
attr->recv_cq = NULL;
rdsdebug("ib_create_cq send failed: %d\n", ret);
goto out;
}
ret = ib_req_notify_cq(attr->send_cq, IB_CQ_NEXT_COMP);
if (ret) {
rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
goto out;
}
ret = ib_req_notify_cq(attr->recv_cq, IB_CQ_SOLICITED);
if (ret) {
rdsdebug("ib_req_notify_cq recv failed: %d\n", ret);
goto out;
}
out:
if (ret) {
if (attr->send_cq)
ib_destroy_cq(attr->send_cq);
if (attr->recv_cq)
ib_destroy_cq(attr->recv_cq);
}
return ret;
}
/* A vanilla 2.6.19 or older kernel without backported OFED kernel headers. */
static void isert_cq_comp_work_cb(void *ctx)
{
struct isert_cq *cq_desc = ctx;
#else
static void isert_cq_comp_work_cb(struct work_struct *work)
{
struct isert_cq *cq_desc =
container_of(work, struct isert_cq, cq_comp_work);
#endif
int ret;
TRACE_ENTRY();
ret = isert_poll_cq(cq_desc);
if (unlikely(ret < 0)) { /* poll error */
pr_err("ib_poll_cq failed\n");
goto out;
}
ib_req_notify_cq(cq_desc->cq,
IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS);
/*
* not all HCAs support IB_CQ_REPORT_MISSED_EVENTS,
* so we need to make sure we don't miss any events between
* last call to ib_poll_cq() and ib_req_notify_cq()
*/
isert_poll_cq(cq_desc);
out:
TRACE_EXIT();
return;
}
static void isert_cq_comp_handler(struct ib_cq *cq, void *context)
{
struct isert_cq *cq_desc = context;
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
queue_work(cq_desc->cq_workqueue, &cq_desc->cq_comp_work);
#else
queue_work_on(smp_processor_id(), cq_desc->cq_workqueue,
&cq_desc->cq_comp_work);
#endif
}
static const char *ib_event_type_str(enum ib_event_type ev_type)
{
switch (ev_type) {
case IB_EVENT_COMM_EST:
return "COMM_EST";
case IB_EVENT_QP_FATAL:
return "QP_FATAL";
case IB_EVENT_QP_REQ_ERR:
return "QP_REQ_ERR";
case IB_EVENT_QP_ACCESS_ERR:
return "QP_ACCESS_ERR";
case IB_EVENT_SQ_DRAINED:
return "SQ_DRAINED";
case IB_EVENT_PATH_MIG:
return "PATH_MIG";
case IB_EVENT_PATH_MIG_ERR:
return "PATH_MIG_ERR";
case IB_EVENT_QP_LAST_WQE_REACHED:
return "QP_LAST_WQE_REACHED";
case IB_EVENT_CQ_ERR:
return "CQ_ERR";
case IB_EVENT_SRQ_ERR:
return "SRQ_ERR";
case IB_EVENT_SRQ_LIMIT_REACHED:
return "SRQ_LIMIT_REACHED";
case IB_EVENT_PORT_ACTIVE:
return "PORT_ACTIVE";
case IB_EVENT_PORT_ERR:
return "PORT_ERR";
case IB_EVENT_LID_CHANGE:
return "LID_CHANGE";
case IB_EVENT_PKEY_CHANGE:
return "PKEY_CHANGE";
case IB_EVENT_SM_CHANGE:
return "SM_CHANGE";
case IB_EVENT_CLIENT_REREGISTER:
return "CLIENT_REREGISTER";
case IB_EVENT_DEVICE_FATAL:
return "DEVICE_FATAL";
default:
return "UNKNOWN";
}
}
static void isert_async_evt_handler(struct ib_event *async_ev, void *context)
{
struct isert_cq *cq = context;
struct isert_device *isert_dev = cq->dev;
struct ib_device *ib_dev = isert_dev->ib_dev;
char *dev_name = ib_dev->name;
enum ib_event_type ev_type = async_ev->event;
struct isert_connection *isert_conn;
TRACE_ENTRY();
switch (ev_type) {
case IB_EVENT_COMM_EST:
isert_conn = async_ev->element.qp->qp_context;
pr_info("conn:0x%p cm_id:0x%p dev:%s, QP evt: %s\n",
isert_conn, isert_conn->cm_id, dev_name,
ib_event_type_str(IB_EVENT_COMM_EST));
/* force "connection established" event */
rdma_notify(isert_conn->cm_id, IB_EVENT_COMM_EST);
break;
/* rest of QP-related events */
case IB_EVENT_QP_FATAL:
case IB_EVENT_QP_REQ_ERR:
case IB_EVENT_QP_ACCESS_ERR:
case IB_EVENT_SQ_DRAINED:
case IB_EVENT_PATH_MIG:
case IB_EVENT_PATH_MIG_ERR:
case IB_EVENT_QP_LAST_WQE_REACHED:
isert_conn = async_ev->element.qp->qp_context;
pr_err("conn:0x%p cm_id:0x%p dev:%s, QP evt: %s\n",
isert_conn, isert_conn->cm_id, dev_name,
ib_event_type_str(ev_type));
break;
/* CQ-related events */
case IB_EVENT_CQ_ERR:
pr_err("dev:%s CQ evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
/* SRQ events */
case IB_EVENT_SRQ_ERR:
case IB_EVENT_SRQ_LIMIT_REACHED:
pr_err("dev:%s SRQ evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
/* Port events */
case IB_EVENT_PORT_ACTIVE:
case IB_EVENT_PORT_ERR:
case IB_EVENT_LID_CHANGE:
case IB_EVENT_PKEY_CHANGE:
case IB_EVENT_SM_CHANGE:
case IB_EVENT_CLIENT_REREGISTER:
pr_err("dev:%s port:%d evt: %s\n",
dev_name, async_ev->element.port_num,
ib_event_type_str(ev_type));
break;
/* HCA events */
case IB_EVENT_DEVICE_FATAL:
pr_err("dev:%s HCA evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
default:
pr_err("dev:%s evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
}
TRACE_EXIT();
}
static struct isert_device *isert_device_create(struct ib_device *ib_dev)
{
struct isert_device *isert_dev;
struct ib_device_attr *dev_attr;
int cqe_num, err;
struct ib_pd *pd;
struct ib_mr *mr;
struct ib_cq *cq;
char wq_name[64];
int i, j;
TRACE_ENTRY();
isert_dev = kzalloc(sizeof(*isert_dev), GFP_KERNEL);
if (unlikely(isert_dev == NULL)) {
pr_err("Failed to allocate iser dev\n");
err = -ENOMEM;
goto out;
}
dev_attr = &isert_dev->device_attr;
err = ib_query_device(ib_dev, dev_attr);
if (unlikely(err)) {
pr_err("Failed to query device, err: %d\n", err);
goto fail_query;
}
isert_dev->num_cqs = min_t(int, num_online_cpus(),
ib_dev->num_comp_vectors);
isert_dev->cq_qps = kzalloc(sizeof(*isert_dev->cq_qps) * isert_dev->num_cqs,
GFP_KERNEL);
if (unlikely(isert_dev->cq_qps == NULL)) {
pr_err("Failed to allocate iser cq_qps\n");
err = -ENOMEM;
goto fail_cq_qps;
}
isert_dev->cq_desc = vmalloc(sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
if (unlikely(isert_dev->cq_desc == NULL)) {
pr_err("Failed to allocate %ld bytes for iser cq_desc\n",
sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
err = -ENOMEM;
goto fail_alloc_cq_desc;
}
pd = ib_alloc_pd(ib_dev);
if (unlikely(IS_ERR(pd))) {
err = PTR_ERR(pd);
pr_err("Failed to alloc iser dev pd, err:%d\n", err);
goto fail_pd;
}
mr = ib_get_dma_mr(pd, IB_ACCESS_LOCAL_WRITE);
if (unlikely(IS_ERR(mr))) {
err = PTR_ERR(mr);
pr_err("Failed to get dma mr, err: %d\n", err);
goto fail_mr;
}
cqe_num = min(isert_dev->device_attr.max_cqe, ISER_CQ_ENTRIES);
cqe_num = cqe_num / isert_dev->num_cqs;
#ifdef CONFIG_SCST_EXTRACHECKS
if (isert_dev->device_attr.max_cqe == 0)
pr_err("Zero max_cqe encountered: you may have a compilation problem\n");
#endif
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
cq_desc->dev = isert_dev;
cq_desc->idx = i;
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb, NULL);
#else
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb);
#endif
snprintf(wq_name, sizeof(wq_name), "isert_cq_%p", cq_desc);
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = create_singlethread_workqueue(wq_name);
#else
#if LINUX_VERSION_CODE == KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_RESCUER, 1);
#else
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_MEM_RECLAIM, 1);
#endif
#endif
if (unlikely(!cq_desc->cq_workqueue)) {
pr_err("Failed to alloc iser cq work queue for dev:%s\n",
ib_dev->name);
err = -ENOMEM;
goto fail_cq;
}
#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 2, 0)
cq = ib_create_cq(ib_dev,
isert_cq_comp_handler,
isert_async_evt_handler,
cq_desc, /* context */
cqe_num,
i); /* completion vector */
#else
{
struct ib_cq_init_attr ia = {
.cqe = cqe_num,
.comp_vector = i,
};
cq = ib_create_cq(ib_dev,
isert_cq_comp_handler,
isert_async_evt_handler,
cq_desc, /* context */
&ia);
}
#endif
if (unlikely(IS_ERR(cq))) {
cq_desc->cq = NULL;
err = PTR_ERR(cq);
pr_err("Failed to create iser dev cq, err:%d\n", err);
goto fail_cq;
}
cq_desc->cq = cq;
err = ib_req_notify_cq(cq, IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS);
if (unlikely(err)) {
pr_err("Failed to request notify cq, err: %d\n", err);
goto fail_cq;
}
}
isert_dev->ib_dev = ib_dev;
isert_dev->pd = pd;
isert_dev->mr = mr;
INIT_LIST_HEAD(&isert_dev->conn_list);
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_add(isert_dev);
pr_info("iser created device:%p\n", isert_dev);
return isert_dev;
fail_cq:
for (j = 0; j <= i; ++j) {
if (isert_dev->cq_desc[j].cq)
ib_destroy_cq(isert_dev->cq_desc[j].cq);
if (isert_dev->cq_desc[j].cq_workqueue)
destroy_workqueue(isert_dev->cq_desc[j].cq_workqueue);
}
ib_dereg_mr(mr);
fail_mr:
ib_dealloc_pd(pd);
fail_pd:
vfree(isert_dev->cq_desc);
fail_alloc_cq_desc:
kfree(isert_dev->cq_qps);
fail_cq_qps:
fail_query:
kfree(isert_dev);
out:
TRACE_EXIT_RES(err);
return ERR_PTR(err);
}
static void isert_device_release(struct isert_device *isert_dev)
{
int err, i;
TRACE_ENTRY();
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_remove(isert_dev); /* remove from global list */
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 22)
/*
* cancel_work_sync() was introduced in 2.6.22. We can
* only wait until all scheduled work is done.
*/
flush_workqueue(cq_desc->cq_workqueue);
#else
cancel_work_sync(&cq_desc->cq_comp_work);
#endif
err = ib_destroy_cq(cq_desc->cq);
if (unlikely(err))
pr_err("Failed to destroy cq, err:%d\n", err);
destroy_workqueue(cq_desc->cq_workqueue);
}
err = ib_dereg_mr(isert_dev->mr);
if (unlikely(err))
pr_err("Failed to destroy mr, err:%d\n", err);
err = ib_dealloc_pd(isert_dev->pd);
if (unlikely(err))
pr_err("Failed to destroy pd, err:%d\n", err);
vfree(isert_dev->cq_desc);
isert_dev->cq_desc = NULL;
kfree(isert_dev->cq_qps);
isert_dev->cq_qps = NULL;
kfree(isert_dev);
TRACE_EXIT();
}
/*
* This needs to be very careful to not leave IS_ERR pointers around for
* cleanup to trip over.
*/
static int
rdsv3_ib_setup_qp(struct rdsv3_connection *conn)
{
struct rdsv3_ib_connection *ic = conn->c_transport_data;
struct ib_device *dev = ic->i_cm_id->device;
struct ib_qp_init_attr attr;
struct rdsv3_ib_device *rds_ibdev;
ibt_send_wr_t *wrp;
ibt_wr_ds_t *sgl;
int ret, i;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp", "Enter conn: %p", conn);
/*
* rdsv3_ib_add_one creates a rdsv3_ib_device object per IB device,
* and allocates a protection domain, memory range and FMR pool
* for each. If that fails for any reason, it will not register
* the rds_ibdev at all.
*/
rds_ibdev = ib_get_client_data(dev, &rdsv3_ib_client);
if (!rds_ibdev) {
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"RDS/IB: No client_data for device %s", dev->name);
return (-EOPNOTSUPP);
}
ic->rds_ibdev = rds_ibdev;
if (rds_ibdev->max_wrs < ic->i_send_ring.w_nr + 1)
rdsv3_ib_ring_resize(&ic->i_send_ring, rds_ibdev->max_wrs - 1);
if (rds_ibdev->max_wrs < ic->i_recv_ring.w_nr + 1)
rdsv3_ib_ring_resize(&ic->i_recv_ring, rds_ibdev->max_wrs - 1);
/* Protection domain and memory range */
ic->i_pd = rds_ibdev->pd;
/*
* IB_CQ_VECTOR_LEAST_ATTACHED and/or the corresponding feature is
* not implmeneted in Hermon yet, but we can pass it to ib_create_cq()
* anyway.
*/
ic->i_cq = ib_create_cq(dev, rdsv3_ib_cq_comp_handler,
rdsv3_ib_cq_event_handler, conn,
ic->i_recv_ring.w_nr + ic->i_send_ring.w_nr + 1,
rdsv3_af_grp_get_sched(ic->rds_ibdev->aft_hcagp));
if (IS_ERR(ic->i_cq)) {
ret = PTR_ERR(ic->i_cq);
ic->i_cq = NULL;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"ib_create_cq failed: %d", ret);
goto out;
}
if (rdsv3_enable_snd_cq) {
ic->i_snd_cq = ib_create_cq(dev, rdsv3_ib_snd_cq_comp_handler,
rdsv3_ib_cq_event_handler, conn, ic->i_send_ring.w_nr + 1,
rdsv3_af_grp_get_sched(ic->rds_ibdev->aft_hcagp));
if (IS_ERR(ic->i_snd_cq)) {
ret = PTR_ERR(ic->i_snd_cq);
(void) ib_destroy_cq(ic->i_cq);
ic->i_cq = NULL;
ic->i_snd_cq = NULL;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"ib_create_cq send cq failed: %d", ret);
goto out;
}
}
/* XXX negotiate max send/recv with remote? */
(void) memset(&attr, 0, sizeof (attr));
attr.event_handler = rdsv3_ib_qp_event_handler;
attr.qp_context = conn;
/* + 1 to allow for the single ack message */
attr.cap.max_send_wr = ic->i_send_ring.w_nr + 1;
attr.cap.max_recv_wr = ic->i_recv_ring.w_nr + 1;
attr.cap.max_send_sge = rds_ibdev->max_sge;
attr.cap.max_recv_sge = RDSV3_IB_RECV_SGE;
attr.sq_sig_type = IB_SIGNAL_REQ_WR;
attr.qp_type = IB_QPT_RC;
if (rdsv3_enable_snd_cq) {
attr.send_cq = ic->i_snd_cq;
} else {
attr.send_cq = ic->i_cq;
}
attr.recv_cq = ic->i_cq;
/*
* XXX this can fail if max_*_wr is too large? Are we supposed
* to back off until we get a value that the hardware can support?
*/
ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
if (ret) {
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"rdma_create_qp failed: %d", ret);
goto out;
}
ret = rdsv3_ib_alloc_hdrs(dev, ic);
if (ret != 0) {
ret = -ENOMEM;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"rdsv3_ib_alloc_hdrs failed: %d", ret);
goto out;
}
ic->i_sends = kmem_alloc(ic->i_send_ring.w_nr *
sizeof (struct rdsv3_ib_send_work), KM_NOSLEEP);
if (ic->i_sends == NULL) {
ret = -ENOMEM;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"send allocation failed: %d", ret);
goto out;
}
(void) memset(ic->i_sends, 0, ic->i_send_ring.w_nr *
sizeof (struct rdsv3_ib_send_work));
ic->i_send_wrs =
kmem_alloc(ic->i_send_ring.w_nr * (sizeof (ibt_send_wr_t) +
RDSV3_IB_MAX_SGE * sizeof (ibt_wr_ds_t)), KM_NOSLEEP);
if (ic->i_send_wrs == NULL) {
ret = -ENOMEM;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"Send WR allocation failed: %d", ret);
goto out;
}
sgl = (ibt_wr_ds_t *)((uint8_t *)ic->i_send_wrs +
(ic->i_send_ring.w_nr * sizeof (ibt_send_wr_t)));
for (i = 0; i < ic->i_send_ring.w_nr; i++) {
wrp = &ic->i_send_wrs[i];
wrp->wr_sgl = &sgl[i * RDSV3_IB_MAX_SGE];
}
ic->i_recvs = kmem_alloc(ic->i_recv_ring.w_nr *
sizeof (struct rdsv3_ib_recv_work), KM_NOSLEEP);
if (ic->i_recvs == NULL) {
ret = -ENOMEM;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"recv allocation failed: %d", ret);
goto out;
}
(void) memset(ic->i_recvs, 0, ic->i_recv_ring.w_nr *
sizeof (struct rdsv3_ib_recv_work));
ic->i_recv_wrs =
kmem_alloc(ic->i_recv_ring.w_nr * sizeof (ibt_recv_wr_t),
KM_NOSLEEP);
if (ic->i_recv_wrs == NULL) {
ret = -ENOMEM;
RDSV3_DPRINTF2("rdsv3_ib_setup_qp",
"Recv WR allocation failed: %d", ret);
goto out;
}
rdsv3_ib_recv_init_ack(ic);
RDSV3_DPRINTF2("rdsv3_ib_setup_qp", "conn %p pd %p mr %p cq %p",
conn, ic->i_pd, ic->i_mr, ic->i_cq);
out:
return (ret);
}
static struct isert_device *isert_device_create(struct ib_device *ib_dev)
{
struct isert_device *isert_dev;
struct ib_device_attr *dev_attr;
int cqe_num, err;
struct ib_pd *pd;
struct ib_mr *mr;
struct ib_cq *cq;
char wq_name[64];
int i, j;
TRACE_ENTRY();
isert_dev = kzalloc(sizeof(*isert_dev), GFP_KERNEL);
if (unlikely(isert_dev == NULL)) {
pr_err("Failed to allocate iser dev\n");
err = -ENOMEM;
goto out;
}
dev_attr = &isert_dev->device_attr;
err = ib_query_device(ib_dev, dev_attr);
if (unlikely(err)) {
pr_err("Failed to query device, err: %d\n", err);
goto fail_query;
}
isert_dev->num_cqs = min_t(int, num_online_cpus(),
ib_dev->num_comp_vectors);
isert_dev->cq_qps = kzalloc(sizeof(*isert_dev->cq_qps) * isert_dev->num_cqs,
GFP_KERNEL);
if (unlikely(isert_dev->cq_qps == NULL)) {
pr_err("Failed to allocate iser cq_qps\n");
err = -ENOMEM;
goto fail_cq_qps;
}
isert_dev->cq_desc = vmalloc(sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
if (unlikely(isert_dev->cq_desc == NULL)) {
pr_err("Failed to allocate %ld bytes for iser cq_desc\n",
sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
err = -ENOMEM;
goto fail_alloc_cq_desc;
}
pd = ib_alloc_pd(ib_dev);
if (unlikely(IS_ERR(pd))) {
err = PTR_ERR(pd);
pr_err("Failed to alloc iser dev pd, err:%d\n", err);
goto fail_pd;
}
mr = ib_get_dma_mr(pd, IB_ACCESS_LOCAL_WRITE);
if (unlikely(IS_ERR(mr))) {
err = PTR_ERR(mr);
pr_err("Failed to get dma mr, err: %d\n", err);
goto fail_mr;
}
cqe_num = min(isert_dev->device_attr.max_cqe, ISER_CQ_ENTRIES);
cqe_num = cqe_num / isert_dev->num_cqs;
#ifdef CONFIG_SCST_EXTRACHECKS
if (isert_dev->device_attr.max_cqe == 0)
pr_err("Zero max_cqe encountered: you may have a compilation problem\n");
#endif
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
cq_desc->dev = isert_dev;
cq_desc->idx = i;
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb, NULL);
#else
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb);
#endif
snprintf(wq_name, sizeof(wq_name), "isert_cq_%p", cq_desc);
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = create_singlethread_workqueue(wq_name);
#else
#if LINUX_VERSION_CODE == KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_RESCUER, 1);
#else
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_MEM_RECLAIM, 1);
#endif
#endif
if (unlikely(!cq_desc->cq_workqueue)) {
pr_err("Failed to alloc iser cq work queue for dev:%s\n",
ib_dev->name);
err = -ENOMEM;
goto fail_cq;
}
cq = ib_create_cq(ib_dev,
isert_cq_comp_handler,
isert_async_evt_handler,
cq_desc, /* context */
cqe_num,
i); /* completion vector */
if (unlikely(IS_ERR(cq))) {
cq_desc->cq = NULL;
err = PTR_ERR(cq);
pr_err("Failed to create iser dev cq, err:%d\n", err);
goto fail_cq;
}
cq_desc->cq = cq;
err = ib_req_notify_cq(cq, IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS);
if (unlikely(err)) {
pr_err("Failed to request notify cq, err: %d\n", err);
goto fail_cq;
}
}
isert_dev->ib_dev = ib_dev;
isert_dev->pd = pd;
isert_dev->mr = mr;
INIT_LIST_HEAD(&isert_dev->conn_list);
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_add(isert_dev);
pr_info("iser created device:%p\n", isert_dev);
return isert_dev;
fail_cq:
for (j = 0; j <= i; ++j) {
if (isert_dev->cq_desc[j].cq)
ib_destroy_cq(isert_dev->cq_desc[j].cq);
if (isert_dev->cq_desc[j].cq_workqueue)
destroy_workqueue(isert_dev->cq_desc[j].cq_workqueue);
}
ib_dereg_mr(mr);
fail_mr:
ib_dealloc_pd(pd);
fail_pd:
vfree(isert_dev->cq_desc);
fail_alloc_cq_desc:
kfree(isert_dev->cq_qps);
fail_cq_qps:
fail_query:
kfree(isert_dev);
out:
TRACE_EXIT_RES(err);
return ERR_PTR(err);
}
/*
* Create unconnected endpoint.
*/
int
rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
struct rpcrdma_create_data_internal *cdata)
{
struct ib_device_attr devattr;
int rc, err;
rc = ib_query_device(ia->ri_id->device, &devattr);
if (rc) {
dprintk("RPC: %s: ib_query_device failed %d\n",
__func__, rc);
return rc;
}
/* check provider's send/recv wr limits */
if (cdata->max_requests > devattr.max_qp_wr)
cdata->max_requests = devattr.max_qp_wr;
ep->rep_attr.event_handler = rpcrdma_qp_async_error_upcall;
ep->rep_attr.qp_context = ep;
/* send_cq and recv_cq initialized below */
ep->rep_attr.srq = NULL;
ep->rep_attr.cap.max_send_wr = cdata->max_requests;
switch (ia->ri_memreg_strategy) {
case RPCRDMA_FRMR:
/* Add room for frmr register and invalidate WRs */
ep->rep_attr.cap.max_send_wr *= 3;
if (ep->rep_attr.cap.max_send_wr > devattr.max_qp_wr)
return -EINVAL;
break;
case RPCRDMA_MEMWINDOWS_ASYNC:
case RPCRDMA_MEMWINDOWS:
/* Add room for mw_binds+unbinds - overkill! */
ep->rep_attr.cap.max_send_wr++;
ep->rep_attr.cap.max_send_wr *= (2 * RPCRDMA_MAX_SEGS);
if (ep->rep_attr.cap.max_send_wr > devattr.max_qp_wr)
return -EINVAL;
break;
default:
break;
}
ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
ep->rep_attr.cap.max_send_sge = (cdata->padding ? 4 : 2);
ep->rep_attr.cap.max_recv_sge = 1;
ep->rep_attr.cap.max_inline_data = 0;
ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
ep->rep_attr.qp_type = IB_QPT_RC;
ep->rep_attr.port_num = ~0;
dprintk("RPC: %s: requested max: dtos: send %d recv %d; "
"iovs: send %d recv %d\n",
__func__,
ep->rep_attr.cap.max_send_wr,
ep->rep_attr.cap.max_recv_wr,
ep->rep_attr.cap.max_send_sge,
ep->rep_attr.cap.max_recv_sge);
/* set trigger for requesting send completion */
ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 /* - 1*/;
switch (ia->ri_memreg_strategy) {
case RPCRDMA_MEMWINDOWS_ASYNC:
case RPCRDMA_MEMWINDOWS:
ep->rep_cqinit -= RPCRDMA_MAX_SEGS;
break;
default:
break;
}
if (ep->rep_cqinit <= 2)
ep->rep_cqinit = 0;
INIT_CQCOUNT(ep);
ep->rep_ia = ia;
init_waitqueue_head(&ep->rep_connect_wait);
/*
* Create a single cq for receive dto and mw_bind (only ever
* care about unbind, really). Send completions are suppressed.
* Use single threaded tasklet upcalls to maintain ordering.
*/
ep->rep_cq = ib_create_cq(ia->ri_id->device, rpcrdma_cq_event_upcall,
rpcrdma_cq_async_error_upcall, NULL,
ep->rep_attr.cap.max_recv_wr +
ep->rep_attr.cap.max_send_wr + 1, 0);
if (IS_ERR(ep->rep_cq)) {
rc = PTR_ERR(ep->rep_cq);
dprintk("RPC: %s: ib_create_cq failed: %i\n",
__func__, rc);
goto out1;
}
rc = ib_req_notify_cq(ep->rep_cq, IB_CQ_NEXT_COMP);
if (rc) {
dprintk("RPC: %s: ib_req_notify_cq failed: %i\n",
__func__, rc);
goto out2;
}
ep->rep_attr.send_cq = ep->rep_cq;
ep->rep_attr.recv_cq = ep->rep_cq;
/* Initialize cma parameters */
/* RPC/RDMA does not use private data */
ep->rep_remote_cma.private_data = NULL;
ep->rep_remote_cma.private_data_len = 0;
/* Client offers RDMA Read but does not initiate */
ep->rep_remote_cma.initiator_depth = 0;
if (ia->ri_memreg_strategy == RPCRDMA_BOUNCEBUFFERS)
ep->rep_remote_cma.responder_resources = 0;
else if (devattr.max_qp_rd_atom > 32) /* arbitrary but <= 255 */
ep->rep_remote_cma.responder_resources = 32;
else
ep->rep_remote_cma.responder_resources = devattr.max_qp_rd_atom;
ep->rep_remote_cma.retry_count = 7;
ep->rep_remote_cma.flow_control = 0;
ep->rep_remote_cma.rnr_retry_count = 0;
return 0;
out2:
err = ib_destroy_cq(ep->rep_cq);
if (err)
dprintk("RPC: %s: ib_destroy_cq returned %i\n",
__func__, err);
out1:
return rc;
}
static int create_umr_res(struct mlx5_ib_dev *dev)
{
struct ib_qp_init_attr *init_attr = NULL;
struct ib_qp_attr *attr = NULL;
struct ib_pd *pd;
struct ib_cq *cq;
struct ib_qp *qp;
struct ib_cq_init_attr cq_attr = {};
int ret;
attr = kzalloc(sizeof(*attr), GFP_KERNEL);
init_attr = kzalloc(sizeof(*init_attr), GFP_KERNEL);
if (!attr || !init_attr) {
ret = -ENOMEM;
goto error_0;
}
pd = ib_alloc_pd(&dev->ib_dev);
if (IS_ERR(pd)) {
mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n");
ret = PTR_ERR(pd);
goto error_0;
}
cq_attr.cqe = 128;
cq = ib_create_cq(&dev->ib_dev, mlx5_umr_cq_handler, NULL, NULL,
&cq_attr);
if (IS_ERR(cq)) {
mlx5_ib_dbg(dev, "Couldn't create CQ for sync UMR QP\n");
ret = PTR_ERR(cq);
goto error_2;
}
ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
init_attr->send_cq = cq;
init_attr->recv_cq = cq;
init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
init_attr->cap.max_send_wr = MAX_UMR_WR;
init_attr->cap.max_send_sge = 1;
init_attr->qp_type = MLX5_IB_QPT_REG_UMR;
init_attr->port_num = 1;
qp = mlx5_ib_create_qp(pd, init_attr, NULL);
if (IS_ERR(qp)) {
mlx5_ib_dbg(dev, "Couldn't create sync UMR QP\n");
ret = PTR_ERR(qp);
goto error_3;
}
qp->device = &dev->ib_dev;
qp->real_qp = qp;
qp->uobject = NULL;
qp->qp_type = MLX5_IB_QPT_REG_UMR;
attr->qp_state = IB_QPS_INIT;
attr->port_num = 1;
ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_PKEY_INDEX |
IB_QP_PORT, NULL);
if (ret) {
mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n");
goto error_4;
}
memset(attr, 0, sizeof(*attr));
attr->qp_state = IB_QPS_RTR;
attr->path_mtu = IB_MTU_256;
ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
if (ret) {
mlx5_ib_dbg(dev, "Couldn't modify umr QP to rtr\n");
goto error_4;
}
memset(attr, 0, sizeof(*attr));
attr->qp_state = IB_QPS_RTS;
ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
if (ret) {
mlx5_ib_dbg(dev, "Couldn't modify umr QP to rts\n");
goto error_4;
}
dev->umrc.qp = qp;
dev->umrc.cq = cq;
dev->umrc.pd = pd;
sema_init(&dev->umrc.sem, MAX_UMR_WR);
ret = mlx5_mr_cache_init(dev);
if (ret) {
mlx5_ib_warn(dev, "mr cache init failed %d\n", ret);
goto error_4;
}
kfree(attr);
kfree(init_attr);
return 0;
error_4:
mlx5_ib_destroy_qp(qp);
error_3:
ib_destroy_cq(cq);
error_2:
ib_dealloc_pd(pd);
error_0:
kfree(attr);
kfree(init_attr);
return ret;
}
static void krping_free_qp(struct krping_cb *cb)
{
ib_destroy_qp(cb->qp);
ib_destroy_cq(cb->cq);
ib_dealloc_pd(cb->pd);
}
/*
* Create unconnected endpoint.
*/
int
rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
struct rpcrdma_create_data_internal *cdata)
{
struct ib_device_attr *devattr = &ia->ri_devattr;
struct ib_cq *sendcq, *recvcq;
struct ib_cq_init_attr cq_attr = {};
unsigned int max_qp_wr;
int rc, err;
if (devattr->max_sge < RPCRDMA_MAX_IOVS) {
dprintk("RPC: %s: insufficient sge's available\n",
__func__);
return -ENOMEM;
}
if (devattr->max_qp_wr <= RPCRDMA_BACKWARD_WRS) {
dprintk("RPC: %s: insufficient wqe's available\n",
__func__);
return -ENOMEM;
}
max_qp_wr = devattr->max_qp_wr - RPCRDMA_BACKWARD_WRS;
/* check provider's send/recv wr limits */
if (cdata->max_requests > max_qp_wr)
cdata->max_requests = max_qp_wr;
ep->rep_attr.event_handler = rpcrdma_qp_async_error_upcall;
ep->rep_attr.qp_context = ep;
ep->rep_attr.srq = NULL;
ep->rep_attr.cap.max_send_wr = cdata->max_requests;
ep->rep_attr.cap.max_send_wr += RPCRDMA_BACKWARD_WRS;
rc = ia->ri_ops->ro_open(ia, ep, cdata);
if (rc)
return rc;
ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
ep->rep_attr.cap.max_recv_wr += RPCRDMA_BACKWARD_WRS;
ep->rep_attr.cap.max_send_sge = RPCRDMA_MAX_IOVS;
ep->rep_attr.cap.max_recv_sge = 1;
ep->rep_attr.cap.max_inline_data = 0;
ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
ep->rep_attr.qp_type = IB_QPT_RC;
ep->rep_attr.port_num = ~0;
dprintk("RPC: %s: requested max: dtos: send %d recv %d; "
"iovs: send %d recv %d\n",
__func__,
ep->rep_attr.cap.max_send_wr,
ep->rep_attr.cap.max_recv_wr,
ep->rep_attr.cap.max_send_sge,
ep->rep_attr.cap.max_recv_sge);
/* set trigger for requesting send completion */
ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 - 1;
if (ep->rep_cqinit <= 2)
ep->rep_cqinit = 0; /* always signal? */
INIT_CQCOUNT(ep);
init_waitqueue_head(&ep->rep_connect_wait);
INIT_DELAYED_WORK(&ep->rep_connect_worker, rpcrdma_connect_worker);
cq_attr.cqe = ep->rep_attr.cap.max_send_wr + 1;
sendcq = ib_create_cq(ia->ri_device, rpcrdma_sendcq_upcall,
rpcrdma_cq_async_error_upcall, NULL, &cq_attr);
if (IS_ERR(sendcq)) {
rc = PTR_ERR(sendcq);
dprintk("RPC: %s: failed to create send CQ: %i\n",
__func__, rc);
goto out1;
}
rc = ib_req_notify_cq(sendcq, IB_CQ_NEXT_COMP);
if (rc) {
dprintk("RPC: %s: ib_req_notify_cq failed: %i\n",
__func__, rc);
goto out2;
}
cq_attr.cqe = ep->rep_attr.cap.max_recv_wr + 1;
recvcq = ib_create_cq(ia->ri_device, rpcrdma_recvcq_upcall,
rpcrdma_cq_async_error_upcall, NULL, &cq_attr);
if (IS_ERR(recvcq)) {
rc = PTR_ERR(recvcq);
dprintk("RPC: %s: failed to create recv CQ: %i\n",
__func__, rc);
goto out2;
}
rc = ib_req_notify_cq(recvcq, IB_CQ_NEXT_COMP);
if (rc) {
dprintk("RPC: %s: ib_req_notify_cq failed: %i\n",
__func__, rc);
ib_destroy_cq(recvcq);
goto out2;
}
ep->rep_attr.send_cq = sendcq;
ep->rep_attr.recv_cq = recvcq;
/* Initialize cma parameters */
/* RPC/RDMA does not use private data */
ep->rep_remote_cma.private_data = NULL;
ep->rep_remote_cma.private_data_len = 0;
/* Client offers RDMA Read but does not initiate */
ep->rep_remote_cma.initiator_depth = 0;
if (devattr->max_qp_rd_atom > 32) /* arbitrary but <= 255 */
ep->rep_remote_cma.responder_resources = 32;
else
ep->rep_remote_cma.responder_resources =
devattr->max_qp_rd_atom;
ep->rep_remote_cma.retry_count = 7;
ep->rep_remote_cma.flow_control = 0;
ep->rep_remote_cma.rnr_retry_count = 0;
return 0;
out2:
err = ib_destroy_cq(sendcq);
if (err)
dprintk("RPC: %s: ib_destroy_cq returned %i\n",
__func__, err);
out1:
if (ia->ri_dma_mr)
ib_dereg_mr(ia->ri_dma_mr);
return rc;
}
ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
const char __user *buf, int in_len,
int out_len)
{
struct ib_uverbs_create_cq cmd;
struct ib_uverbs_create_cq_resp resp;
struct ib_udata udata;
struct ib_ucq_object *uobj;
struct ib_uverbs_event_file *ev_file = NULL;
struct ib_cq *cq;
int ret;
if (out_len < sizeof resp)
return -ENOSPC;
if (copy_from_user(&cmd, buf, sizeof cmd))
return -EFAULT;
INIT_UDATA(&udata, buf + sizeof cmd,
(unsigned long) cmd.response + sizeof resp,
in_len - sizeof cmd, out_len - sizeof resp);
if (cmd.comp_vector >= file->device->num_comp_vectors)
return -EINVAL;
uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
if (!uobj)
return -ENOMEM;
if (cmd.comp_channel >= 0) {
ev_file = ib_uverbs_lookup_comp_file(cmd.comp_channel);
if (!ev_file) {
ret = -EINVAL;
goto err;
}
}
uobj->uobject.user_handle = cmd.user_handle;
uobj->uobject.context = file->ucontext;
uobj->uverbs_file = file;
uobj->comp_events_reported = 0;
uobj->async_events_reported = 0;
INIT_LIST_HEAD(&uobj->comp_list);
INIT_LIST_HEAD(&uobj->async_list);
cq = file->device->ib_dev->create_cq(file->device->ib_dev, cmd.cqe,
file->ucontext, &udata);
if (IS_ERR(cq)) {
ret = PTR_ERR(cq);
goto err;
}
cq->device = file->device->ib_dev;
cq->uobject = &uobj->uobject;
cq->comp_handler = ib_uverbs_comp_handler;
cq->event_handler = ib_uverbs_cq_event_handler;
cq->cq_context = ev_file;
atomic_set(&cq->usecnt, 0);
mutex_lock(&ib_uverbs_idr_mutex);
retry:
if (!idr_pre_get(&ib_uverbs_cq_idr, GFP_KERNEL)) {
ret = -ENOMEM;
goto err_up;
}
ret = idr_get_new(&ib_uverbs_cq_idr, cq, &uobj->uobject.id);
if (ret == -EAGAIN)
goto retry;
if (ret)
goto err_up;
memset(&resp, 0, sizeof resp);
resp.cq_handle = uobj->uobject.id;
resp.cqe = cq->cqe;
if (copy_to_user((void __user *) (unsigned long) cmd.response,
&resp, sizeof resp)) {
ret = -EFAULT;
goto err_idr;
}
mutex_lock(&file->mutex);
list_add_tail(&uobj->uobject.list, &file->ucontext->cq_list);
mutex_unlock(&file->mutex);
mutex_unlock(&ib_uverbs_idr_mutex);
return in_len;
err_idr:
idr_remove(&ib_uverbs_cq_idr, uobj->uobject.id);
err_up:
mutex_unlock(&ib_uverbs_idr_mutex);
ib_destroy_cq(cq);
err:
if (ev_file)
ib_uverbs_release_ucq(file, ev_file, uobj);
kfree(uobj);
return ret;
}
/*
* Create unconnected endpoint.
*/
int
rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
struct rpcrdma_create_data_internal *cdata)
{
struct ib_device_attr devattr;
struct ib_cq *sendcq, *recvcq;
int rc, err;
rc = ib_query_device(ia->ri_id->device, &devattr);
if (rc) {
dprintk("RPC: %s: ib_query_device failed %d\n",
__func__, rc);
return rc;
}
/* check provider's send/recv wr limits */
if (cdata->max_requests > devattr.max_qp_wr)
cdata->max_requests = devattr.max_qp_wr;
ep->rep_attr.event_handler = rpcrdma_qp_async_error_upcall;
ep->rep_attr.qp_context = ep;
/* send_cq and recv_cq initialized below */
ep->rep_attr.srq = NULL;
ep->rep_attr.cap.max_send_wr = cdata->max_requests;
switch (ia->ri_memreg_strategy) {
case RPCRDMA_FRMR: {
int depth = 7;
/* Add room for frmr register and invalidate WRs.
* 1. FRMR reg WR for head
* 2. FRMR invalidate WR for head
* 3. N FRMR reg WRs for pagelist
* 4. N FRMR invalidate WRs for pagelist
* 5. FRMR reg WR for tail
* 6. FRMR invalidate WR for tail
* 7. The RDMA_SEND WR
*/
/* Calculate N if the device max FRMR depth is smaller than
* RPCRDMA_MAX_DATA_SEGS.
*/
if (ia->ri_max_frmr_depth < RPCRDMA_MAX_DATA_SEGS) {
int delta = RPCRDMA_MAX_DATA_SEGS -
ia->ri_max_frmr_depth;
do {
depth += 2; /* FRMR reg + invalidate */
delta -= ia->ri_max_frmr_depth;
} while (delta > 0);
}
ep->rep_attr.cap.max_send_wr *= depth;
if (ep->rep_attr.cap.max_send_wr > devattr.max_qp_wr) {
cdata->max_requests = devattr.max_qp_wr / depth;
if (!cdata->max_requests)
return -EINVAL;
ep->rep_attr.cap.max_send_wr = cdata->max_requests *
depth;
}
break;
}
default:
break;
}
ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
ep->rep_attr.cap.max_send_sge = (cdata->padding ? 4 : 2);
ep->rep_attr.cap.max_recv_sge = 1;
ep->rep_attr.cap.max_inline_data = 0;
ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
ep->rep_attr.qp_type = IB_QPT_RC;
ep->rep_attr.port_num = ~0;
dprintk("RPC: %s: requested max: dtos: send %d recv %d; "
"iovs: send %d recv %d\n",
__func__,
ep->rep_attr.cap.max_send_wr,
ep->rep_attr.cap.max_recv_wr,
ep->rep_attr.cap.max_send_sge,
ep->rep_attr.cap.max_recv_sge);
/* set trigger for requesting send completion */
ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 - 1;
if (ep->rep_cqinit <= 2)
ep->rep_cqinit = 0;
INIT_CQCOUNT(ep);
ep->rep_ia = ia;
init_waitqueue_head(&ep->rep_connect_wait);
INIT_DELAYED_WORK(&ep->rep_connect_worker, rpcrdma_connect_worker);
sendcq = ib_create_cq(ia->ri_id->device, rpcrdma_sendcq_upcall,
rpcrdma_cq_async_error_upcall, ep,
ep->rep_attr.cap.max_send_wr + 1, 0);
if (IS_ERR(sendcq)) {
rc = PTR_ERR(sendcq);
dprintk("RPC: %s: failed to create send CQ: %i\n",
__func__, rc);
goto out1;
}
rc = ib_req_notify_cq(sendcq, IB_CQ_NEXT_COMP);
if (rc) {
dprintk("RPC: %s: ib_req_notify_cq failed: %i\n",
__func__, rc);
goto out2;
}
recvcq = ib_create_cq(ia->ri_id->device, rpcrdma_recvcq_upcall,
rpcrdma_cq_async_error_upcall, ep,
ep->rep_attr.cap.max_recv_wr + 1, 0);
if (IS_ERR(recvcq)) {
rc = PTR_ERR(recvcq);
dprintk("RPC: %s: failed to create recv CQ: %i\n",
__func__, rc);
goto out2;
}
rc = ib_req_notify_cq(recvcq, IB_CQ_NEXT_COMP);
if (rc) {
dprintk("RPC: %s: ib_req_notify_cq failed: %i\n",
__func__, rc);
ib_destroy_cq(recvcq);
goto out2;
}
ep->rep_attr.send_cq = sendcq;
ep->rep_attr.recv_cq = recvcq;
/* Initialize cma parameters */
/* RPC/RDMA does not use private data */
ep->rep_remote_cma.private_data = NULL;
ep->rep_remote_cma.private_data_len = 0;
/* Client offers RDMA Read but does not initiate */
ep->rep_remote_cma.initiator_depth = 0;
if (devattr.max_qp_rd_atom > 32) /* arbitrary but <= 255 */
ep->rep_remote_cma.responder_resources = 32;
else
ep->rep_remote_cma.responder_resources = devattr.max_qp_rd_atom;
ep->rep_remote_cma.retry_count = 7;
ep->rep_remote_cma.flow_control = 0;
ep->rep_remote_cma.rnr_retry_count = 0;
return 0;
out2:
err = ib_destroy_cq(sendcq);
if (err)
dprintk("RPC: %s: ib_destroy_cq returned %i\n",
__func__, err);
out1:
return rc;
}
/*
* This needs to be very careful to not leave IS_ERR pointers around for
* cleanup to trip over.
*/
static int rds_ib_setup_qp(struct rds_connection *conn)
{
struct rds_ib_connection *ic = conn->c_transport_data;
struct ib_device *dev = ic->i_cm_id->device;
struct ib_qp_init_attr attr;
struct ib_cq_init_attr cq_attr = {};
struct rds_ib_device *rds_ibdev;
int ret, fr_queue_space;
/*
* It's normal to see a null device if an incoming connection races
* with device removal, so we don't print a warning.
*/
rds_ibdev = rds_ib_get_client_data(dev);
if (!rds_ibdev)
return -EOPNOTSUPP;
/* The fr_queue_space is currently set to 512, to add extra space on
* completion queue and send queue. This extra space is used for FRMR
* registration and invalidation work requests
*/
fr_queue_space = rds_ibdev->use_fastreg ?
(RDS_IB_DEFAULT_FR_WR + 1) +
(RDS_IB_DEFAULT_FR_INV_WR + 1)
: 0;
/* add the conn now so that connection establishment has the dev */
rds_ib_add_conn(rds_ibdev, conn);
if (rds_ibdev->max_wrs < ic->i_send_ring.w_nr + 1)
rds_ib_ring_resize(&ic->i_send_ring, rds_ibdev->max_wrs - 1);
if (rds_ibdev->max_wrs < ic->i_recv_ring.w_nr + 1)
rds_ib_ring_resize(&ic->i_recv_ring, rds_ibdev->max_wrs - 1);
/* Protection domain and memory range */
ic->i_pd = rds_ibdev->pd;
ic->i_scq_vector = ibdev_get_unused_vector(rds_ibdev);
cq_attr.cqe = ic->i_send_ring.w_nr + fr_queue_space + 1;
cq_attr.comp_vector = ic->i_scq_vector;
ic->i_send_cq = ib_create_cq(dev, rds_ib_cq_comp_handler_send,
rds_ib_cq_event_handler, conn,
&cq_attr);
if (IS_ERR(ic->i_send_cq)) {
ret = PTR_ERR(ic->i_send_cq);
ic->i_send_cq = NULL;
ibdev_put_vector(rds_ibdev, ic->i_scq_vector);
rdsdebug("ib_create_cq send failed: %d\n", ret);
goto rds_ibdev_out;
}
ic->i_rcq_vector = ibdev_get_unused_vector(rds_ibdev);
cq_attr.cqe = ic->i_recv_ring.w_nr;
cq_attr.comp_vector = ic->i_rcq_vector;
ic->i_recv_cq = ib_create_cq(dev, rds_ib_cq_comp_handler_recv,
rds_ib_cq_event_handler, conn,
&cq_attr);
if (IS_ERR(ic->i_recv_cq)) {
ret = PTR_ERR(ic->i_recv_cq);
ic->i_recv_cq = NULL;
ibdev_put_vector(rds_ibdev, ic->i_rcq_vector);
rdsdebug("ib_create_cq recv failed: %d\n", ret);
goto send_cq_out;
}
ret = ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP);
if (ret) {
rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
goto recv_cq_out;
}
ret = ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED);
if (ret) {
rdsdebug("ib_req_notify_cq recv failed: %d\n", ret);
goto recv_cq_out;
}
/* XXX negotiate max send/recv with remote? */
memset(&attr, 0, sizeof(attr));
attr.event_handler = rds_ib_qp_event_handler;
attr.qp_context = conn;
/* + 1 to allow for the single ack message */
attr.cap.max_send_wr = ic->i_send_ring.w_nr + fr_queue_space + 1;
attr.cap.max_recv_wr = ic->i_recv_ring.w_nr + 1;
attr.cap.max_send_sge = rds_ibdev->max_sge;
attr.cap.max_recv_sge = RDS_IB_RECV_SGE;
attr.sq_sig_type = IB_SIGNAL_REQ_WR;
attr.qp_type = IB_QPT_RC;
attr.send_cq = ic->i_send_cq;
attr.recv_cq = ic->i_recv_cq;
atomic_set(&ic->i_fastreg_wrs, RDS_IB_DEFAULT_FR_WR);
atomic_set(&ic->i_fastunreg_wrs, RDS_IB_DEFAULT_FR_INV_WR);
/*
* XXX this can fail if max_*_wr is too large? Are we supposed
* to back off until we get a value that the hardware can support?
*/
ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
if (ret) {
rdsdebug("rdma_create_qp failed: %d\n", ret);
goto recv_cq_out;
}
ic->i_send_hdrs = ib_dma_alloc_coherent(dev,
ic->i_send_ring.w_nr *
sizeof(struct rds_header),
&ic->i_send_hdrs_dma, GFP_KERNEL);
if (!ic->i_send_hdrs) {
ret = -ENOMEM;
rdsdebug("ib_dma_alloc_coherent send failed\n");
goto qp_out;
}
ic->i_recv_hdrs = ib_dma_alloc_coherent(dev,
ic->i_recv_ring.w_nr *
sizeof(struct rds_header),
&ic->i_recv_hdrs_dma, GFP_KERNEL);
if (!ic->i_recv_hdrs) {
ret = -ENOMEM;
rdsdebug("ib_dma_alloc_coherent recv failed\n");
goto send_hdrs_dma_out;
}
ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header),
&ic->i_ack_dma, GFP_KERNEL);
if (!ic->i_ack) {
ret = -ENOMEM;
rdsdebug("ib_dma_alloc_coherent ack failed\n");
goto recv_hdrs_dma_out;
}
ic->i_sends = vzalloc_node(ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work),
ibdev_to_node(dev));
if (!ic->i_sends) {
ret = -ENOMEM;
rdsdebug("send allocation failed\n");
goto ack_dma_out;
}
ic->i_recvs = vzalloc_node(ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work),
ibdev_to_node(dev));
if (!ic->i_recvs) {
ret = -ENOMEM;
rdsdebug("recv allocation failed\n");
goto sends_out;
}
rds_ib_recv_init_ack(ic);
rdsdebug("conn %p pd %p cq %p %p\n", conn, ic->i_pd,
ic->i_send_cq, ic->i_recv_cq);
return ret;
sends_out:
vfree(ic->i_sends);
ack_dma_out:
ib_dma_free_coherent(dev, sizeof(struct rds_header),
ic->i_ack, ic->i_ack_dma);
recv_hdrs_dma_out:
ib_dma_free_coherent(dev, ic->i_recv_ring.w_nr *
sizeof(struct rds_header),
ic->i_recv_hdrs, ic->i_recv_hdrs_dma);
send_hdrs_dma_out:
ib_dma_free_coherent(dev, ic->i_send_ring.w_nr *
sizeof(struct rds_header),
ic->i_send_hdrs, ic->i_send_hdrs_dma);
qp_out:
rdma_destroy_qp(ic->i_cm_id);
recv_cq_out:
if (!ib_destroy_cq(ic->i_recv_cq))
ic->i_recv_cq = NULL;
send_cq_out:
if (!ib_destroy_cq(ic->i_send_cq))
ic->i_send_cq = NULL;
rds_ibdev_out:
rds_ib_remove_conn(rds_ibdev, conn);
rds_ib_dev_put(rds_ibdev);
return ret;
}