2 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the BSD-type
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
17 * Redistributions in binary form must reproduce the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer in the documentation and/or other materials provided
20 * with the distribution.
22 * Neither the name of the Network Appliance, Inc. nor the names of
23 * its contributors may be used to endorse or promote products
24 * derived from this software without specific prior written
27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
30 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
31 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
32 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
33 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
34 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
35 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
37 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
43 * Encapsulates the major functions managing:
50 #include <linux/interrupt.h>
51 #include <linux/slab.h>
52 #include <linux/prefetch.h>
53 #include <linux/sunrpc/addr.h>
54 #include <linux/sunrpc/svc_rdma.h>
55 #include <asm/bitops.h>
56 #include <linux/module.h> /* try_module_get()/module_put() */
57 #include <rdma/ib_cm.h>
59 #include "xprt_rdma.h"
65 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
66 # define RPCDBG_FACILITY RPCDBG_TRANS
73 static struct workqueue_struct *rpcrdma_receive_wq;
76 rpcrdma_alloc_wq(void)
78 struct workqueue_struct *recv_wq;
80 recv_wq = alloc_workqueue("xprtrdma_receive",
81 WQ_MEM_RECLAIM | WQ_UNBOUND | WQ_HIGHPRI,
86 rpcrdma_receive_wq = recv_wq;
91 rpcrdma_destroy_wq(void)
93 struct workqueue_struct *wq;
95 if (rpcrdma_receive_wq) {
96 wq = rpcrdma_receive_wq;
97 rpcrdma_receive_wq = NULL;
98 destroy_workqueue(wq);
103 rpcrdma_qp_async_error_upcall(struct ib_event *event, void *context)
105 struct rpcrdma_ep *ep = context;
107 pr_err("rpcrdma: %s on device %s ep %p\n",
108 ib_event_msg(event->event), event->device->name, context);
110 if (ep->rep_connected == 1) {
111 ep->rep_connected = -EIO;
112 rpcrdma_conn_func(ep);
113 wake_up_all(&ep->rep_connect_wait);
118 * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC
119 * @cq: completion queue (ignored)
124 rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
126 /* WARNING: Only wr_cqe and status are reliable at this point */
127 if (wc->status != IB_WC_SUCCESS && wc->status != IB_WC_WR_FLUSH_ERR)
128 pr_err("rpcrdma: Send: %s (%u/0x%x)\n",
129 ib_wc_status_msg(wc->status),
130 wc->status, wc->vendor_err);
133 /* Perform basic sanity checking to avoid using garbage
134 * to update the credit grant value.
137 rpcrdma_update_granted_credits(struct rpcrdma_rep *rep)
139 struct rpcrdma_msg *rmsgp = rdmab_to_msg(rep->rr_rdmabuf);
140 struct rpcrdma_buffer *buffer = &rep->rr_rxprt->rx_buf;
143 if (rep->rr_len < RPCRDMA_HDRLEN_ERR)
146 credits = be32_to_cpu(rmsgp->rm_credit);
148 credits = 1; /* don't deadlock */
149 else if (credits > buffer->rb_max_requests)
150 credits = buffer->rb_max_requests;
152 atomic_set(&buffer->rb_credits, credits);
156 * rpcrdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
157 * @cq: completion queue (ignored)
162 rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
164 struct ib_cqe *cqe = wc->wr_cqe;
165 struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep,
168 /* WARNING: Only wr_id and status are reliable at this point */
169 if (wc->status != IB_WC_SUCCESS)
172 /* status == SUCCESS means all fields in wc are trustworthy */
173 if (wc->opcode != IB_WC_RECV)
176 dprintk("RPC: %s: rep %p opcode 'recv', length %u: success\n",
177 __func__, rep, wc->byte_len);
179 rep->rr_len = wc->byte_len;
180 rep->rr_wc_flags = wc->wc_flags;
181 rep->rr_inv_rkey = wc->ex.invalidate_rkey;
183 ib_dma_sync_single_for_cpu(rep->rr_device,
184 rdmab_addr(rep->rr_rdmabuf),
185 rep->rr_len, DMA_FROM_DEVICE);
187 rpcrdma_update_granted_credits(rep);
190 queue_work(rpcrdma_receive_wq, &rep->rr_work);
194 if (wc->status != IB_WC_WR_FLUSH_ERR)
195 pr_err("rpcrdma: Recv: %s (%u/0x%x)\n",
196 ib_wc_status_msg(wc->status),
197 wc->status, wc->vendor_err);
198 rep->rr_len = RPCRDMA_BAD_LEN;
203 rpcrdma_update_connect_private(struct rpcrdma_xprt *r_xprt,
204 struct rdma_conn_param *param)
206 struct rpcrdma_create_data_internal *cdata = &r_xprt->rx_data;
207 const struct rpcrdma_connect_private *pmsg = param->private_data;
208 unsigned int rsize, wsize;
210 /* Default settings for RPC-over-RDMA Version One */
211 r_xprt->rx_ia.ri_reminv_expected = false;
212 r_xprt->rx_ia.ri_implicit_roundup = xprt_rdma_pad_optimize;
213 rsize = RPCRDMA_V1_DEF_INLINE_SIZE;
214 wsize = RPCRDMA_V1_DEF_INLINE_SIZE;
217 pmsg->cp_magic == rpcrdma_cmp_magic &&
218 pmsg->cp_version == RPCRDMA_CMP_VERSION) {
219 r_xprt->rx_ia.ri_reminv_expected = true;
220 r_xprt->rx_ia.ri_implicit_roundup = true;
221 rsize = rpcrdma_decode_buffer_size(pmsg->cp_send_size);
222 wsize = rpcrdma_decode_buffer_size(pmsg->cp_recv_size);
225 if (rsize < cdata->inline_rsize)
226 cdata->inline_rsize = rsize;
227 if (wsize < cdata->inline_wsize)
228 cdata->inline_wsize = wsize;
229 dprintk("RPC: %s: max send %u, max recv %u\n",
230 __func__, cdata->inline_wsize, cdata->inline_rsize);
231 rpcrdma_set_max_header_sizes(r_xprt);
235 rpcrdma_conn_upcall(struct rdma_cm_id *id, struct rdma_cm_event *event)
237 struct rpcrdma_xprt *xprt = id->context;
238 struct rpcrdma_ia *ia = &xprt->rx_ia;
239 struct rpcrdma_ep *ep = &xprt->rx_ep;
240 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
241 struct sockaddr *sap = (struct sockaddr *)&ep->rep_remote_addr;
243 struct ib_qp_attr *attr = &ia->ri_qp_attr;
244 struct ib_qp_init_attr *iattr = &ia->ri_qp_init_attr;
247 switch (event->event) {
248 case RDMA_CM_EVENT_ADDR_RESOLVED:
249 case RDMA_CM_EVENT_ROUTE_RESOLVED:
251 complete(&ia->ri_done);
253 case RDMA_CM_EVENT_ADDR_ERROR:
254 ia->ri_async_rc = -EHOSTUNREACH;
255 dprintk("RPC: %s: CM address resolution error, ep 0x%p\n",
257 complete(&ia->ri_done);
259 case RDMA_CM_EVENT_ROUTE_ERROR:
260 ia->ri_async_rc = -ENETUNREACH;
261 dprintk("RPC: %s: CM route resolution error, ep 0x%p\n",
263 complete(&ia->ri_done);
265 case RDMA_CM_EVENT_ESTABLISHED:
267 ib_query_qp(ia->ri_id->qp, attr,
268 IB_QP_MAX_QP_RD_ATOMIC | IB_QP_MAX_DEST_RD_ATOMIC,
270 dprintk("RPC: %s: %d responder resources"
272 __func__, attr->max_dest_rd_atomic,
273 attr->max_rd_atomic);
274 rpcrdma_update_connect_private(xprt, &event->param.conn);
276 case RDMA_CM_EVENT_CONNECT_ERROR:
277 connstate = -ENOTCONN;
279 case RDMA_CM_EVENT_UNREACHABLE:
280 connstate = -ENETDOWN;
282 case RDMA_CM_EVENT_REJECTED:
283 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
284 pr_info("rpcrdma: connection to %pIS:%u on %s rejected: %s\n",
285 sap, rpc_get_port(sap), ia->ri_device->name,
286 rdma_reject_msg(id, event->status));
288 connstate = -ECONNREFUSED;
289 if (event->status == IB_CM_REJ_STALE_CONN)
292 case RDMA_CM_EVENT_DISCONNECTED:
293 connstate = -ECONNABORTED;
295 case RDMA_CM_EVENT_DEVICE_REMOVAL:
298 dprintk("RPC: %s: %sconnected\n",
299 __func__, connstate > 0 ? "" : "dis");
300 atomic_set(&xprt->rx_buf.rb_credits, 1);
301 ep->rep_connected = connstate;
302 rpcrdma_conn_func(ep);
303 wake_up_all(&ep->rep_connect_wait);
306 dprintk("RPC: %s: %pIS:%u (ep 0x%p): %s\n",
307 __func__, sap, rpc_get_port(sap), ep,
308 rdma_event_msg(event->event));
312 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
313 if (connstate == 1) {
314 int ird = attr->max_dest_rd_atomic;
315 int tird = ep->rep_remote_cma.responder_resources;
317 pr_info("rpcrdma: connection to %pIS:%u on %s, memreg '%s', %d credits, %d responders%s\n",
318 sap, rpc_get_port(sap),
320 ia->ri_ops->ro_displayname,
321 xprt->rx_buf.rb_max_requests,
322 ird, ird < 4 && ird < tird / 2 ? " (low!)" : "");
323 } else if (connstate < 0) {
324 pr_info("rpcrdma: connection to %pIS:%u closed (%d)\n",
325 sap, rpc_get_port(sap), connstate);
332 static void rpcrdma_destroy_id(struct rdma_cm_id *id)
335 module_put(id->device->owner);
340 static struct rdma_cm_id *
341 rpcrdma_create_id(struct rpcrdma_xprt *xprt,
342 struct rpcrdma_ia *ia, struct sockaddr *addr)
344 unsigned long wtimeout = msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1;
345 struct rdma_cm_id *id;
348 init_completion(&ia->ri_done);
350 id = rdma_create_id(&init_net, rpcrdma_conn_upcall, xprt, RDMA_PS_TCP,
354 dprintk("RPC: %s: rdma_create_id() failed %i\n",
359 ia->ri_async_rc = -ETIMEDOUT;
360 rc = rdma_resolve_addr(id, NULL, addr, RDMA_RESOLVE_TIMEOUT);
362 dprintk("RPC: %s: rdma_resolve_addr() failed %i\n",
366 rc = wait_for_completion_interruptible_timeout(&ia->ri_done, wtimeout);
368 dprintk("RPC: %s: wait() exited: %i\n",
374 * Until xprtrdma supports DEVICE_REMOVAL, the provider must
375 * be pinned while there are active NFS/RDMA mounts to prevent
376 * hangs and crashes at umount time.
378 if (!ia->ri_async_rc && !try_module_get(id->device->owner)) {
379 dprintk("RPC: %s: Failed to get device module\n",
381 ia->ri_async_rc = -ENODEV;
383 rc = ia->ri_async_rc;
387 ia->ri_async_rc = -ETIMEDOUT;
388 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
390 dprintk("RPC: %s: rdma_resolve_route() failed %i\n",
394 rc = wait_for_completion_interruptible_timeout(&ia->ri_done, wtimeout);
396 dprintk("RPC: %s: wait() exited: %i\n",
400 rc = ia->ri_async_rc;
406 module_put(id->device->owner);
413 * Exported functions.
417 * Open and initialize an Interface Adapter.
418 * o initializes fields of struct rpcrdma_ia, including
419 * interface and provider attributes and protection zone.
422 rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
424 struct rpcrdma_ia *ia = &xprt->rx_ia;
427 ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
428 if (IS_ERR(ia->ri_id)) {
429 rc = PTR_ERR(ia->ri_id);
432 ia->ri_device = ia->ri_id->device;
434 ia->ri_pd = ib_alloc_pd(ia->ri_device, 0);
435 if (IS_ERR(ia->ri_pd)) {
436 rc = PTR_ERR(ia->ri_pd);
437 pr_err("rpcrdma: ib_alloc_pd() returned %d\n", rc);
443 if (frwr_is_supported(ia)) {
444 ia->ri_ops = &rpcrdma_frwr_memreg_ops;
448 case RPCRDMA_MTHCAFMR:
449 if (fmr_is_supported(ia)) {
450 ia->ri_ops = &rpcrdma_fmr_memreg_ops;
455 pr_err("rpcrdma: Unsupported memory registration mode: %d\n",
464 ib_dealloc_pd(ia->ri_pd);
467 rpcrdma_destroy_id(ia->ri_id);
474 * Clean up/close an IA.
475 * o if event handles and PD have been initialized, free them.
479 rpcrdma_ia_close(struct rpcrdma_ia *ia)
481 dprintk("RPC: %s: entering\n", __func__);
482 if (ia->ri_id != NULL && !IS_ERR(ia->ri_id)) {
484 rdma_destroy_qp(ia->ri_id);
485 rpcrdma_destroy_id(ia->ri_id);
489 /* If the pd is still busy, xprtrdma missed freeing a resource */
490 if (ia->ri_pd && !IS_ERR(ia->ri_pd))
491 ib_dealloc_pd(ia->ri_pd);
495 * Create unconnected endpoint.
498 rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
499 struct rpcrdma_create_data_internal *cdata)
501 struct rpcrdma_connect_private *pmsg = &ep->rep_cm_private;
502 unsigned int max_qp_wr, max_sge;
503 struct ib_cq *sendcq, *recvcq;
506 max_sge = min_t(unsigned int, ia->ri_device->attrs.max_sge,
507 RPCRDMA_MAX_SEND_SGES);
508 if (max_sge < RPCRDMA_MIN_SEND_SGES) {
509 pr_warn("rpcrdma: HCA provides only %d send SGEs\n", max_sge);
512 ia->ri_max_send_sges = max_sge - RPCRDMA_MIN_SEND_SGES;
514 if (ia->ri_device->attrs.max_qp_wr <= RPCRDMA_BACKWARD_WRS) {
515 dprintk("RPC: %s: insufficient wqe's available\n",
519 max_qp_wr = ia->ri_device->attrs.max_qp_wr - RPCRDMA_BACKWARD_WRS - 1;
521 /* check provider's send/recv wr limits */
522 if (cdata->max_requests > max_qp_wr)
523 cdata->max_requests = max_qp_wr;
525 ep->rep_attr.event_handler = rpcrdma_qp_async_error_upcall;
526 ep->rep_attr.qp_context = ep;
527 ep->rep_attr.srq = NULL;
528 ep->rep_attr.cap.max_send_wr = cdata->max_requests;
529 ep->rep_attr.cap.max_send_wr += RPCRDMA_BACKWARD_WRS;
530 ep->rep_attr.cap.max_send_wr += 1; /* drain cqe */
531 rc = ia->ri_ops->ro_open(ia, ep, cdata);
534 ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
535 ep->rep_attr.cap.max_recv_wr += RPCRDMA_BACKWARD_WRS;
536 ep->rep_attr.cap.max_recv_wr += 1; /* drain cqe */
537 ep->rep_attr.cap.max_send_sge = max_sge;
538 ep->rep_attr.cap.max_recv_sge = 1;
539 ep->rep_attr.cap.max_inline_data = 0;
540 ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
541 ep->rep_attr.qp_type = IB_QPT_RC;
542 ep->rep_attr.port_num = ~0;
544 dprintk("RPC: %s: requested max: dtos: send %d recv %d; "
545 "iovs: send %d recv %d\n",
547 ep->rep_attr.cap.max_send_wr,
548 ep->rep_attr.cap.max_recv_wr,
549 ep->rep_attr.cap.max_send_sge,
550 ep->rep_attr.cap.max_recv_sge);
552 /* set trigger for requesting send completion */
553 ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 - 1;
554 if (ep->rep_cqinit <= 2)
555 ep->rep_cqinit = 0; /* always signal? */
556 rpcrdma_init_cqcount(ep, 0);
557 init_waitqueue_head(&ep->rep_connect_wait);
558 INIT_DELAYED_WORK(&ep->rep_connect_worker, rpcrdma_connect_worker);
560 sendcq = ib_alloc_cq(ia->ri_device, NULL,
561 ep->rep_attr.cap.max_send_wr + 1,
563 if (IS_ERR(sendcq)) {
564 rc = PTR_ERR(sendcq);
565 dprintk("RPC: %s: failed to create send CQ: %i\n",
570 recvcq = ib_alloc_cq(ia->ri_device, NULL,
571 ep->rep_attr.cap.max_recv_wr + 1,
573 if (IS_ERR(recvcq)) {
574 rc = PTR_ERR(recvcq);
575 dprintk("RPC: %s: failed to create recv CQ: %i\n",
580 ep->rep_attr.send_cq = sendcq;
581 ep->rep_attr.recv_cq = recvcq;
583 /* Initialize cma parameters */
584 memset(&ep->rep_remote_cma, 0, sizeof(ep->rep_remote_cma));
586 /* Prepare RDMA-CM private message */
587 pmsg->cp_magic = rpcrdma_cmp_magic;
588 pmsg->cp_version = RPCRDMA_CMP_VERSION;
589 pmsg->cp_flags |= ia->ri_ops->ro_send_w_inv_ok;
590 pmsg->cp_send_size = rpcrdma_encode_buffer_size(cdata->inline_wsize);
591 pmsg->cp_recv_size = rpcrdma_encode_buffer_size(cdata->inline_rsize);
592 ep->rep_remote_cma.private_data = pmsg;
593 ep->rep_remote_cma.private_data_len = sizeof(*pmsg);
595 /* Client offers RDMA Read but does not initiate */
596 ep->rep_remote_cma.initiator_depth = 0;
597 if (ia->ri_device->attrs.max_qp_rd_atom > 32) /* arbitrary but <= 255 */
598 ep->rep_remote_cma.responder_resources = 32;
600 ep->rep_remote_cma.responder_resources =
601 ia->ri_device->attrs.max_qp_rd_atom;
603 /* Limit transport retries so client can detect server
604 * GID changes quickly. RPC layer handles re-establishing
605 * transport connection and retransmission.
607 ep->rep_remote_cma.retry_count = 6;
609 /* RPC-over-RDMA handles its own flow control. In addition,
610 * make all RNR NAKs visible so we know that RPC-over-RDMA
611 * flow control is working correctly (no NAKs should be seen).
613 ep->rep_remote_cma.flow_control = 0;
614 ep->rep_remote_cma.rnr_retry_count = 0;
627 * Disconnect and destroy endpoint. After this, the only
628 * valid operations on the ep are to free it (if dynamically
629 * allocated) or re-create it.
632 rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
634 dprintk("RPC: %s: entering, connected is %d\n",
635 __func__, ep->rep_connected);
637 cancel_delayed_work_sync(&ep->rep_connect_worker);
640 rpcrdma_ep_disconnect(ep, ia);
641 rdma_destroy_qp(ia->ri_id);
642 ia->ri_id->qp = NULL;
645 ib_free_cq(ep->rep_attr.recv_cq);
646 ib_free_cq(ep->rep_attr.send_cq);
650 * Connect unconnected endpoint.
653 rpcrdma_ep_connect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
655 struct rpcrdma_xprt *r_xprt = container_of(ia, struct rpcrdma_xprt,
657 struct rdma_cm_id *id, *old;
658 struct sockaddr *sap;
662 if (ep->rep_connected != 0) {
664 dprintk("RPC: %s: reconnecting...\n", __func__);
666 rpcrdma_ep_disconnect(ep, ia);
668 sap = (struct sockaddr *)&r_xprt->rx_data.addr;
669 id = rpcrdma_create_id(r_xprt, ia, sap);
674 /* TEMP TEMP TEMP - fail if new device:
675 * Deregister/remarshal *all* requests!
676 * Close and recreate adapter, pd, etc!
677 * Re-determine all attributes still sane!
678 * More stuff I haven't thought of!
681 if (ia->ri_device != id->device) {
682 printk("RPC: %s: can't reconnect on "
683 "different device!\n", __func__);
684 rpcrdma_destroy_id(id);
689 rc = rdma_create_qp(id, ia->ri_pd, &ep->rep_attr);
691 dprintk("RPC: %s: rdma_create_qp failed %i\n",
693 rpcrdma_destroy_id(id);
701 rdma_destroy_qp(old);
702 rpcrdma_destroy_id(old);
704 dprintk("RPC: %s: connecting...\n", __func__);
705 rc = rdma_create_qp(ia->ri_id, ia->ri_pd, &ep->rep_attr);
707 dprintk("RPC: %s: rdma_create_qp failed %i\n",
709 /* do not update ep->rep_connected */
714 ep->rep_connected = 0;
716 rc = rdma_connect(ia->ri_id, &ep->rep_remote_cma);
718 dprintk("RPC: %s: rdma_connect() failed with %i\n",
723 wait_event_interruptible(ep->rep_connect_wait, ep->rep_connected != 0);
724 if (ep->rep_connected <= 0) {
725 if (ep->rep_connected == -EAGAIN)
727 rc = ep->rep_connected;
731 dprintk("RPC: %s: connected\n", __func__);
732 extras = r_xprt->rx_buf.rb_bc_srv_max_requests;
734 rpcrdma_ep_post_extra_recv(r_xprt, extras);
738 ep->rep_connected = rc;
743 * rpcrdma_ep_disconnect
745 * This is separate from destroy to facilitate the ability
746 * to reconnect without recreating the endpoint.
748 * This call is not reentrant, and must not be made in parallel
749 * on the same endpoint.
752 rpcrdma_ep_disconnect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
756 rc = rdma_disconnect(ia->ri_id);
758 /* returns without wait if not connected */
759 wait_event_interruptible(ep->rep_connect_wait,
760 ep->rep_connected != 1);
761 dprintk("RPC: %s: after wait, %sconnected\n", __func__,
762 (ep->rep_connected == 1) ? "still " : "dis");
764 dprintk("RPC: %s: rdma_disconnect %i\n", __func__, rc);
765 ep->rep_connected = rc;
768 ib_drain_qp(ia->ri_id->qp);
772 rpcrdma_mr_recovery_worker(struct work_struct *work)
774 struct rpcrdma_buffer *buf = container_of(work, struct rpcrdma_buffer,
775 rb_recovery_worker.work);
776 struct rpcrdma_mw *mw;
778 spin_lock(&buf->rb_recovery_lock);
779 while (!list_empty(&buf->rb_stale_mrs)) {
780 mw = rpcrdma_pop_mw(&buf->rb_stale_mrs);
781 spin_unlock(&buf->rb_recovery_lock);
783 dprintk("RPC: %s: recovering MR %p\n", __func__, mw);
784 mw->mw_xprt->rx_ia.ri_ops->ro_recover_mr(mw);
786 spin_lock(&buf->rb_recovery_lock);
788 spin_unlock(&buf->rb_recovery_lock);
792 rpcrdma_defer_mr_recovery(struct rpcrdma_mw *mw)
794 struct rpcrdma_xprt *r_xprt = mw->mw_xprt;
795 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
797 spin_lock(&buf->rb_recovery_lock);
798 rpcrdma_push_mw(mw, &buf->rb_stale_mrs);
799 spin_unlock(&buf->rb_recovery_lock);
801 schedule_delayed_work(&buf->rb_recovery_worker, 0);
805 rpcrdma_create_mrs(struct rpcrdma_xprt *r_xprt)
807 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
808 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
813 for (count = 0; count < 32; count++) {
814 struct rpcrdma_mw *mw;
817 mw = kzalloc(sizeof(*mw), GFP_KERNEL);
821 rc = ia->ri_ops->ro_init_mr(ia, mw);
827 mw->mw_xprt = r_xprt;
829 list_add(&mw->mw_list, &free);
830 list_add(&mw->mw_all, &all);
833 spin_lock(&buf->rb_mwlock);
834 list_splice(&free, &buf->rb_mws);
835 list_splice(&all, &buf->rb_all);
836 r_xprt->rx_stats.mrs_allocated += count;
837 spin_unlock(&buf->rb_mwlock);
839 dprintk("RPC: %s: created %u MRs\n", __func__, count);
843 rpcrdma_mr_refresh_worker(struct work_struct *work)
845 struct rpcrdma_buffer *buf = container_of(work, struct rpcrdma_buffer,
846 rb_refresh_worker.work);
847 struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt,
850 rpcrdma_create_mrs(r_xprt);
854 rpcrdma_create_req(struct rpcrdma_xprt *r_xprt)
856 struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
857 struct rpcrdma_req *req;
859 req = kzalloc(sizeof(*req), GFP_KERNEL);
861 return ERR_PTR(-ENOMEM);
863 INIT_LIST_HEAD(&req->rl_free);
864 spin_lock(&buffer->rb_reqslock);
865 list_add(&req->rl_all, &buffer->rb_allreqs);
866 spin_unlock(&buffer->rb_reqslock);
867 req->rl_cqe.done = rpcrdma_wc_send;
868 req->rl_buffer = &r_xprt->rx_buf;
869 INIT_LIST_HEAD(&req->rl_registered);
870 req->rl_send_wr.next = NULL;
871 req->rl_send_wr.wr_cqe = &req->rl_cqe;
872 req->rl_send_wr.sg_list = req->rl_send_sge;
873 req->rl_send_wr.opcode = IB_WR_SEND;
878 rpcrdma_create_rep(struct rpcrdma_xprt *r_xprt)
880 struct rpcrdma_create_data_internal *cdata = &r_xprt->rx_data;
881 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
882 struct rpcrdma_rep *rep;
886 rep = kzalloc(sizeof(*rep), GFP_KERNEL);
890 rep->rr_rdmabuf = rpcrdma_alloc_regbuf(cdata->inline_rsize,
891 DMA_FROM_DEVICE, GFP_KERNEL);
892 if (IS_ERR(rep->rr_rdmabuf)) {
893 rc = PTR_ERR(rep->rr_rdmabuf);
897 rep->rr_device = ia->ri_device;
898 rep->rr_cqe.done = rpcrdma_wc_receive;
899 rep->rr_rxprt = r_xprt;
900 INIT_WORK(&rep->rr_work, rpcrdma_reply_handler);
901 rep->rr_recv_wr.next = NULL;
902 rep->rr_recv_wr.wr_cqe = &rep->rr_cqe;
903 rep->rr_recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov;
904 rep->rr_recv_wr.num_sge = 1;
914 rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt)
916 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
919 buf->rb_max_requests = r_xprt->rx_data.max_requests;
920 buf->rb_bc_srv_max_requests = 0;
921 atomic_set(&buf->rb_credits, 1);
922 spin_lock_init(&buf->rb_mwlock);
923 spin_lock_init(&buf->rb_lock);
924 spin_lock_init(&buf->rb_recovery_lock);
925 INIT_LIST_HEAD(&buf->rb_mws);
926 INIT_LIST_HEAD(&buf->rb_all);
927 INIT_LIST_HEAD(&buf->rb_stale_mrs);
928 INIT_DELAYED_WORK(&buf->rb_refresh_worker,
929 rpcrdma_mr_refresh_worker);
930 INIT_DELAYED_WORK(&buf->rb_recovery_worker,
931 rpcrdma_mr_recovery_worker);
933 rpcrdma_create_mrs(r_xprt);
935 INIT_LIST_HEAD(&buf->rb_send_bufs);
936 INIT_LIST_HEAD(&buf->rb_allreqs);
937 spin_lock_init(&buf->rb_reqslock);
938 for (i = 0; i < buf->rb_max_requests; i++) {
939 struct rpcrdma_req *req;
941 req = rpcrdma_create_req(r_xprt);
943 dprintk("RPC: %s: request buffer %d alloc"
944 " failed\n", __func__, i);
948 req->rl_backchannel = false;
949 list_add(&req->rl_free, &buf->rb_send_bufs);
952 INIT_LIST_HEAD(&buf->rb_recv_bufs);
953 for (i = 0; i < buf->rb_max_requests + RPCRDMA_MAX_BC_REQUESTS; i++) {
954 struct rpcrdma_rep *rep;
956 rep = rpcrdma_create_rep(r_xprt);
958 dprintk("RPC: %s: reply buffer %d alloc failed\n",
963 list_add(&rep->rr_list, &buf->rb_recv_bufs);
968 rpcrdma_buffer_destroy(buf);
972 static struct rpcrdma_req *
973 rpcrdma_buffer_get_req_locked(struct rpcrdma_buffer *buf)
975 struct rpcrdma_req *req;
977 req = list_first_entry(&buf->rb_send_bufs,
978 struct rpcrdma_req, rl_free);
979 list_del(&req->rl_free);
983 static struct rpcrdma_rep *
984 rpcrdma_buffer_get_rep_locked(struct rpcrdma_buffer *buf)
986 struct rpcrdma_rep *rep;
988 rep = list_first_entry(&buf->rb_recv_bufs,
989 struct rpcrdma_rep, rr_list);
990 list_del(&rep->rr_list);
995 rpcrdma_destroy_rep(struct rpcrdma_rep *rep)
997 rpcrdma_free_regbuf(rep->rr_rdmabuf);
1002 rpcrdma_destroy_req(struct rpcrdma_req *req)
1004 rpcrdma_free_regbuf(req->rl_recvbuf);
1005 rpcrdma_free_regbuf(req->rl_sendbuf);
1006 rpcrdma_free_regbuf(req->rl_rdmabuf);
1011 rpcrdma_destroy_mrs(struct rpcrdma_buffer *buf)
1013 struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt,
1015 struct rpcrdma_ia *ia = rdmab_to_ia(buf);
1016 struct rpcrdma_mw *mw;
1020 spin_lock(&buf->rb_mwlock);
1021 while (!list_empty(&buf->rb_all)) {
1022 mw = list_entry(buf->rb_all.next, struct rpcrdma_mw, mw_all);
1023 list_del(&mw->mw_all);
1025 spin_unlock(&buf->rb_mwlock);
1026 ia->ri_ops->ro_release_mr(mw);
1028 spin_lock(&buf->rb_mwlock);
1030 spin_unlock(&buf->rb_mwlock);
1031 r_xprt->rx_stats.mrs_allocated = 0;
1033 dprintk("RPC: %s: released %u MRs\n", __func__, count);
1037 rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf)
1039 cancel_delayed_work_sync(&buf->rb_recovery_worker);
1041 while (!list_empty(&buf->rb_recv_bufs)) {
1042 struct rpcrdma_rep *rep;
1044 rep = rpcrdma_buffer_get_rep_locked(buf);
1045 rpcrdma_destroy_rep(rep);
1047 buf->rb_send_count = 0;
1049 spin_lock(&buf->rb_reqslock);
1050 while (!list_empty(&buf->rb_allreqs)) {
1051 struct rpcrdma_req *req;
1053 req = list_first_entry(&buf->rb_allreqs,
1054 struct rpcrdma_req, rl_all);
1055 list_del(&req->rl_all);
1057 spin_unlock(&buf->rb_reqslock);
1058 rpcrdma_destroy_req(req);
1059 spin_lock(&buf->rb_reqslock);
1061 spin_unlock(&buf->rb_reqslock);
1062 buf->rb_recv_count = 0;
1064 rpcrdma_destroy_mrs(buf);
1068 rpcrdma_get_mw(struct rpcrdma_xprt *r_xprt)
1070 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1071 struct rpcrdma_mw *mw = NULL;
1073 spin_lock(&buf->rb_mwlock);
1074 if (!list_empty(&buf->rb_mws))
1075 mw = rpcrdma_pop_mw(&buf->rb_mws);
1076 spin_unlock(&buf->rb_mwlock);
1083 dprintk("RPC: %s: no MWs available\n", __func__);
1084 schedule_delayed_work(&buf->rb_refresh_worker, 0);
1086 /* Allow the reply handler and refresh worker to run */
1093 rpcrdma_put_mw(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mw *mw)
1095 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1097 spin_lock(&buf->rb_mwlock);
1098 rpcrdma_push_mw(mw, &buf->rb_mws);
1099 spin_unlock(&buf->rb_mwlock);
1102 static struct rpcrdma_rep *
1103 rpcrdma_buffer_get_rep(struct rpcrdma_buffer *buffers)
1105 /* If an RPC previously completed without a reply (say, a
1106 * credential problem or a soft timeout occurs) then hold off
1107 * on supplying more Receive buffers until the number of new
1108 * pending RPCs catches up to the number of posted Receives.
1110 if (unlikely(buffers->rb_send_count < buffers->rb_recv_count))
1113 if (unlikely(list_empty(&buffers->rb_recv_bufs)))
1115 buffers->rb_recv_count++;
1116 return rpcrdma_buffer_get_rep_locked(buffers);
1120 * Get a set of request/reply buffers.
1122 * Reply buffer (if available) is attached to send buffer upon return.
1124 struct rpcrdma_req *
1125 rpcrdma_buffer_get(struct rpcrdma_buffer *buffers)
1127 struct rpcrdma_req *req;
1129 spin_lock(&buffers->rb_lock);
1130 if (list_empty(&buffers->rb_send_bufs))
1132 buffers->rb_send_count++;
1133 req = rpcrdma_buffer_get_req_locked(buffers);
1134 req->rl_reply = rpcrdma_buffer_get_rep(buffers);
1135 spin_unlock(&buffers->rb_lock);
1139 spin_unlock(&buffers->rb_lock);
1140 pr_warn("RPC: %s: out of request buffers\n", __func__);
1145 * Put request/reply buffers back into pool.
1146 * Pre-decrement counter/array index.
1149 rpcrdma_buffer_put(struct rpcrdma_req *req)
1151 struct rpcrdma_buffer *buffers = req->rl_buffer;
1152 struct rpcrdma_rep *rep = req->rl_reply;
1154 req->rl_send_wr.num_sge = 0;
1155 req->rl_reply = NULL;
1157 spin_lock(&buffers->rb_lock);
1158 buffers->rb_send_count--;
1159 list_add_tail(&req->rl_free, &buffers->rb_send_bufs);
1161 buffers->rb_recv_count--;
1162 list_add_tail(&rep->rr_list, &buffers->rb_recv_bufs);
1164 spin_unlock(&buffers->rb_lock);
1168 * Recover reply buffers from pool.
1169 * This happens when recovering from disconnect.
1172 rpcrdma_recv_buffer_get(struct rpcrdma_req *req)
1174 struct rpcrdma_buffer *buffers = req->rl_buffer;
1176 spin_lock(&buffers->rb_lock);
1177 req->rl_reply = rpcrdma_buffer_get_rep(buffers);
1178 spin_unlock(&buffers->rb_lock);
1182 * Put reply buffers back into pool when not attached to
1183 * request. This happens in error conditions.
1186 rpcrdma_recv_buffer_put(struct rpcrdma_rep *rep)
1188 struct rpcrdma_buffer *buffers = &rep->rr_rxprt->rx_buf;
1190 spin_lock(&buffers->rb_lock);
1191 buffers->rb_recv_count--;
1192 list_add_tail(&rep->rr_list, &buffers->rb_recv_bufs);
1193 spin_unlock(&buffers->rb_lock);
1197 * rpcrdma_alloc_regbuf - allocate and DMA-map memory for SEND/RECV buffers
1198 * @size: size of buffer to be allocated, in bytes
1199 * @direction: direction of data movement
1202 * Returns an ERR_PTR, or a pointer to a regbuf, a buffer that
1203 * can be persistently DMA-mapped for I/O.
1205 * xprtrdma uses a regbuf for posting an outgoing RDMA SEND, or for
1206 * receiving the payload of RDMA RECV operations. During Long Calls
1207 * or Replies they may be registered externally via ro_map.
1209 struct rpcrdma_regbuf *
1210 rpcrdma_alloc_regbuf(size_t size, enum dma_data_direction direction,
1213 struct rpcrdma_regbuf *rb;
1215 rb = kmalloc(sizeof(*rb) + size, flags);
1217 return ERR_PTR(-ENOMEM);
1219 rb->rg_device = NULL;
1220 rb->rg_direction = direction;
1221 rb->rg_iov.length = size;
1227 * __rpcrdma_map_regbuf - DMA-map a regbuf
1228 * @ia: controlling rpcrdma_ia
1229 * @rb: regbuf to be mapped
1232 __rpcrdma_dma_map_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
1234 if (rb->rg_direction == DMA_NONE)
1237 rb->rg_iov.addr = ib_dma_map_single(ia->ri_device,
1238 (void *)rb->rg_base,
1241 if (ib_dma_mapping_error(ia->ri_device, rdmab_addr(rb)))
1244 rb->rg_device = ia->ri_device;
1245 rb->rg_iov.lkey = ia->ri_pd->local_dma_lkey;
1250 rpcrdma_dma_unmap_regbuf(struct rpcrdma_regbuf *rb)
1252 if (!rpcrdma_regbuf_is_mapped(rb))
1255 ib_dma_unmap_single(rb->rg_device, rdmab_addr(rb),
1256 rdmab_length(rb), rb->rg_direction);
1257 rb->rg_device = NULL;
1261 * rpcrdma_free_regbuf - deregister and free registered buffer
1262 * @rb: regbuf to be deregistered and freed
1265 rpcrdma_free_regbuf(struct rpcrdma_regbuf *rb)
1270 rpcrdma_dma_unmap_regbuf(rb);
1275 * Prepost any receive buffer, then post send.
1277 * Receive buffer is donated to hardware, reclaimed upon recv completion.
1280 rpcrdma_ep_post(struct rpcrdma_ia *ia,
1281 struct rpcrdma_ep *ep,
1282 struct rpcrdma_req *req)
1284 struct ib_send_wr *send_wr = &req->rl_send_wr;
1285 struct ib_send_wr *send_wr_fail;
1288 if (req->rl_reply) {
1289 rc = rpcrdma_ep_post_recv(ia, req->rl_reply);
1292 req->rl_reply = NULL;
1295 dprintk("RPC: %s: posting %d s/g entries\n",
1296 __func__, send_wr->num_sge);
1298 rpcrdma_set_signaled(ep, send_wr);
1299 rc = ib_post_send(ia->ri_id->qp, send_wr, &send_wr_fail);
1301 goto out_postsend_err;
1305 pr_err("rpcrdma: RDMA Send ib_post_send returned %i\n", rc);
1310 rpcrdma_ep_post_recv(struct rpcrdma_ia *ia,
1311 struct rpcrdma_rep *rep)
1313 struct ib_recv_wr *recv_wr_fail;
1316 if (!rpcrdma_dma_map_regbuf(ia, rep->rr_rdmabuf))
1318 rc = ib_post_recv(ia->ri_id->qp, &rep->rr_recv_wr, &recv_wr_fail);
1324 pr_err("rpcrdma: failed to DMA map the Receive buffer\n");
1328 pr_err("rpcrdma: ib_post_recv returned %i\n", rc);
1333 * rpcrdma_ep_post_extra_recv - Post buffers for incoming backchannel requests
1334 * @r_xprt: transport associated with these backchannel resources
1335 * @min_reqs: minimum number of incoming requests expected
1337 * Returns zero if all requested buffers were posted, or a negative errno.
1340 rpcrdma_ep_post_extra_recv(struct rpcrdma_xprt *r_xprt, unsigned int count)
1342 struct rpcrdma_buffer *buffers = &r_xprt->rx_buf;
1343 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
1344 struct rpcrdma_rep *rep;
1348 spin_lock(&buffers->rb_lock);
1349 if (list_empty(&buffers->rb_recv_bufs))
1351 rep = rpcrdma_buffer_get_rep_locked(buffers);
1352 spin_unlock(&buffers->rb_lock);
1354 rc = rpcrdma_ep_post_recv(ia, rep);
1362 spin_unlock(&buffers->rb_lock);
1363 pr_warn("%s: no extra receive buffers\n", __func__);
1367 rpcrdma_recv_buffer_put(rep);