1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
3 * Copyright (c) 2014-2017 Oracle. All rights reserved.
4 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the BSD-type
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
16 * Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
19 * Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials provided
22 * with the distribution.
24 * Neither the name of the Network Appliance, Inc. nor the names of
25 * its contributors may be used to endorse or promote products
26 * derived from this software without specific prior written
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45 * Encapsulates the major functions managing:
52 #include <linux/interrupt.h>
53 #include <linux/slab.h>
54 #include <linux/sunrpc/addr.h>
55 #include <linux/sunrpc/svc_rdma.h>
56 #include <linux/log2.h>
58 #include <asm-generic/barrier.h>
59 #include <asm/bitops.h>
61 #include <rdma/ib_cm.h>
63 #include "xprt_rdma.h"
64 #include <trace/events/rpcrdma.h>
70 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
71 # define RPCDBG_FACILITY RPCDBG_TRANS
77 static void rpcrdma_sendctx_put_locked(struct rpcrdma_sendctx *sc);
78 static void rpcrdma_reps_destroy(struct rpcrdma_buffer *buf);
79 static void rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt);
80 static void rpcrdma_mrs_destroy(struct rpcrdma_buffer *buf);
81 static void rpcrdma_mr_free(struct rpcrdma_mr *mr);
82 static struct rpcrdma_regbuf *
83 rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction,
85 static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb);
86 static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb);
87 static void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp);
89 /* Wait for outstanding transport work to finish. ib_drain_qp
90 * handles the drains in the wrong order for us, so open code
93 static void rpcrdma_xprt_drain(struct rpcrdma_xprt *r_xprt)
95 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
97 /* Flush Receives, then wait for deferred Reply work
100 ib_drain_rq(ia->ri_id->qp);
102 /* Deferred Reply processing might have scheduled
103 * local invalidations.
105 ib_drain_sq(ia->ri_id->qp);
109 * rpcrdma_qp_event_handler - Handle one QP event (error notification)
110 * @event: details of the event
111 * @context: ep that owns QP where event occurred
113 * Called from the RDMA provider (device driver) possibly in an interrupt
117 rpcrdma_qp_event_handler(struct ib_event *event, void *context)
119 struct rpcrdma_ep *ep = context;
120 struct rpcrdma_xprt *r_xprt = container_of(ep, struct rpcrdma_xprt,
123 trace_xprtrdma_qp_event(r_xprt, event);
127 * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC
128 * @cq: completion queue (ignored)
133 rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
135 struct ib_cqe *cqe = wc->wr_cqe;
136 struct rpcrdma_sendctx *sc =
137 container_of(cqe, struct rpcrdma_sendctx, sc_cqe);
139 /* WARNING: Only wr_cqe and status are reliable at this point */
140 trace_xprtrdma_wc_send(sc, wc);
141 rpcrdma_sendctx_put_locked(sc);
145 * rpcrdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
146 * @cq: completion queue (ignored)
151 rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
153 struct ib_cqe *cqe = wc->wr_cqe;
154 struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep,
156 struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
158 /* WARNING: Only wr_cqe and status are reliable at this point */
159 trace_xprtrdma_wc_receive(wc);
160 --r_xprt->rx_ep.rep_receive_count;
161 if (wc->status != IB_WC_SUCCESS)
164 /* status == SUCCESS means all fields in wc are trustworthy */
165 rpcrdma_set_xdrlen(&rep->rr_hdrbuf, wc->byte_len);
166 rep->rr_wc_flags = wc->wc_flags;
167 rep->rr_inv_rkey = wc->ex.invalidate_rkey;
169 ib_dma_sync_single_for_cpu(rdmab_device(rep->rr_rdmabuf),
170 rdmab_addr(rep->rr_rdmabuf),
171 wc->byte_len, DMA_FROM_DEVICE);
173 rpcrdma_post_recvs(r_xprt, false);
174 rpcrdma_reply_handler(rep);
178 rpcrdma_recv_buffer_put(rep);
182 rpcrdma_update_connect_private(struct rpcrdma_xprt *r_xprt,
183 struct rdma_conn_param *param)
185 const struct rpcrdma_connect_private *pmsg = param->private_data;
186 unsigned int rsize, wsize;
188 /* Default settings for RPC-over-RDMA Version One */
189 r_xprt->rx_ia.ri_implicit_roundup = xprt_rdma_pad_optimize;
190 rsize = RPCRDMA_V1_DEF_INLINE_SIZE;
191 wsize = RPCRDMA_V1_DEF_INLINE_SIZE;
194 pmsg->cp_magic == rpcrdma_cmp_magic &&
195 pmsg->cp_version == RPCRDMA_CMP_VERSION) {
196 r_xprt->rx_ia.ri_implicit_roundup = true;
197 rsize = rpcrdma_decode_buffer_size(pmsg->cp_send_size);
198 wsize = rpcrdma_decode_buffer_size(pmsg->cp_recv_size);
201 if (rsize < r_xprt->rx_ep.rep_inline_recv)
202 r_xprt->rx_ep.rep_inline_recv = rsize;
203 if (wsize < r_xprt->rx_ep.rep_inline_send)
204 r_xprt->rx_ep.rep_inline_send = wsize;
205 dprintk("RPC: %s: max send %u, max recv %u\n", __func__,
206 r_xprt->rx_ep.rep_inline_send,
207 r_xprt->rx_ep.rep_inline_recv);
208 rpcrdma_set_max_header_sizes(r_xprt);
212 * rpcrdma_cm_event_handler - Handle RDMA CM events
213 * @id: rdma_cm_id on which an event has occurred
214 * @event: details of the event
216 * Called with @id's mutex held. Returns 1 if caller should
217 * destroy @id, otherwise 0.
220 rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
222 struct rpcrdma_xprt *r_xprt = id->context;
223 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
224 struct rpcrdma_ep *ep = &r_xprt->rx_ep;
225 struct rpc_xprt *xprt = &r_xprt->rx_xprt;
229 trace_xprtrdma_cm_event(r_xprt, event);
230 switch (event->event) {
231 case RDMA_CM_EVENT_ADDR_RESOLVED:
232 case RDMA_CM_EVENT_ROUTE_RESOLVED:
234 complete(&ia->ri_done);
236 case RDMA_CM_EVENT_ADDR_ERROR:
237 ia->ri_async_rc = -EPROTO;
238 complete(&ia->ri_done);
240 case RDMA_CM_EVENT_ROUTE_ERROR:
241 ia->ri_async_rc = -ENETUNREACH;
242 complete(&ia->ri_done);
244 case RDMA_CM_EVENT_DEVICE_REMOVAL:
245 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
246 pr_info("rpcrdma: removing device %s for %s:%s\n",
247 ia->ri_id->device->name,
248 rpcrdma_addrstr(r_xprt), rpcrdma_portstr(r_xprt));
250 set_bit(RPCRDMA_IAF_REMOVING, &ia->ri_flags);
251 ep->rep_connected = -ENODEV;
252 xprt_force_disconnect(xprt);
253 wait_for_completion(&ia->ri_remove_done);
256 /* Return 1 to ensure the core destroys the id. */
258 case RDMA_CM_EVENT_ESTABLISHED:
259 ++xprt->connect_cookie;
260 ep->rep_connected = 1;
261 rpcrdma_update_connect_private(r_xprt, &event->param.conn);
262 wake_up_all(&ep->rep_connect_wait);
264 case RDMA_CM_EVENT_CONNECT_ERROR:
265 ep->rep_connected = -ENOTCONN;
267 case RDMA_CM_EVENT_UNREACHABLE:
268 ep->rep_connected = -ENETUNREACH;
270 case RDMA_CM_EVENT_REJECTED:
271 dprintk("rpcrdma: connection to %s:%s rejected: %s\n",
272 rpcrdma_addrstr(r_xprt), rpcrdma_portstr(r_xprt),
273 rdma_reject_msg(id, event->status));
274 ep->rep_connected = -ECONNREFUSED;
275 if (event->status == IB_CM_REJ_STALE_CONN)
276 ep->rep_connected = -EAGAIN;
278 case RDMA_CM_EVENT_DISCONNECTED:
279 ep->rep_connected = -ECONNABORTED;
281 xprt_force_disconnect(xprt);
282 wake_up_all(&ep->rep_connect_wait);
288 dprintk("RPC: %s: %s:%s on %s/frwr: %s\n", __func__,
289 rpcrdma_addrstr(r_xprt), rpcrdma_portstr(r_xprt),
290 ia->ri_id->device->name, rdma_event_msg(event->event));
294 static struct rdma_cm_id *
295 rpcrdma_create_id(struct rpcrdma_xprt *xprt, struct rpcrdma_ia *ia)
297 unsigned long wtimeout = msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1;
298 struct rdma_cm_id *id;
301 trace_xprtrdma_conn_start(xprt);
303 init_completion(&ia->ri_done);
304 init_completion(&ia->ri_remove_done);
306 id = rdma_create_id(xprt->rx_xprt.xprt_net, rpcrdma_cm_event_handler,
307 xprt, RDMA_PS_TCP, IB_QPT_RC);
311 ia->ri_async_rc = -ETIMEDOUT;
312 rc = rdma_resolve_addr(id, NULL,
313 (struct sockaddr *)&xprt->rx_xprt.addr,
314 RDMA_RESOLVE_TIMEOUT);
317 rc = wait_for_completion_interruptible_timeout(&ia->ri_done, wtimeout);
319 trace_xprtrdma_conn_tout(xprt);
323 rc = ia->ri_async_rc;
327 ia->ri_async_rc = -ETIMEDOUT;
328 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
331 rc = wait_for_completion_interruptible_timeout(&ia->ri_done, wtimeout);
333 trace_xprtrdma_conn_tout(xprt);
336 rc = ia->ri_async_rc;
348 * Exported functions.
352 * rpcrdma_ia_open - Open and initialize an Interface Adapter.
353 * @xprt: transport with IA to (re)initialize
355 * Returns 0 on success, negative errno if an appropriate
356 * Interface Adapter could not be found and opened.
359 rpcrdma_ia_open(struct rpcrdma_xprt *xprt)
361 struct rpcrdma_ia *ia = &xprt->rx_ia;
364 ia->ri_id = rpcrdma_create_id(xprt, ia);
365 if (IS_ERR(ia->ri_id)) {
366 rc = PTR_ERR(ia->ri_id);
370 ia->ri_pd = ib_alloc_pd(ia->ri_id->device, 0);
371 if (IS_ERR(ia->ri_pd)) {
372 rc = PTR_ERR(ia->ri_pd);
373 pr_err("rpcrdma: ib_alloc_pd() returned %d\n", rc);
377 switch (xprt_rdma_memreg_strategy) {
379 if (frwr_is_supported(ia->ri_id->device))
383 pr_err("rpcrdma: Device %s does not support memreg mode %d\n",
384 ia->ri_id->device->name, xprt_rdma_memreg_strategy);
392 rpcrdma_ia_close(ia);
397 * rpcrdma_ia_remove - Handle device driver unload
398 * @ia: interface adapter being removed
400 * Divest transport H/W resources associated with this adapter,
401 * but allow it to be restored later.
404 rpcrdma_ia_remove(struct rpcrdma_ia *ia)
406 struct rpcrdma_xprt *r_xprt = container_of(ia, struct rpcrdma_xprt,
408 struct rpcrdma_ep *ep = &r_xprt->rx_ep;
409 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
410 struct rpcrdma_req *req;
412 cancel_work_sync(&buf->rb_refresh_worker);
414 /* This is similar to rpcrdma_ep_destroy, but:
415 * - Don't cancel the connect worker.
416 * - Don't call rpcrdma_ep_disconnect, which waits
417 * for another conn upcall, which will deadlock.
418 * - rdma_disconnect is unneeded, the underlying
419 * connection is already gone.
422 rpcrdma_xprt_drain(r_xprt);
423 rdma_destroy_qp(ia->ri_id);
424 ia->ri_id->qp = NULL;
426 ib_free_cq(ep->rep_attr.recv_cq);
427 ep->rep_attr.recv_cq = NULL;
428 ib_free_cq(ep->rep_attr.send_cq);
429 ep->rep_attr.send_cq = NULL;
431 /* The ULP is responsible for ensuring all DMA
432 * mappings and MRs are gone.
434 rpcrdma_reps_destroy(buf);
435 list_for_each_entry(req, &buf->rb_allreqs, rl_all) {
436 rpcrdma_regbuf_dma_unmap(req->rl_rdmabuf);
437 rpcrdma_regbuf_dma_unmap(req->rl_sendbuf);
438 rpcrdma_regbuf_dma_unmap(req->rl_recvbuf);
440 rpcrdma_mrs_destroy(buf);
441 ib_dealloc_pd(ia->ri_pd);
444 /* Allow waiters to continue */
445 complete(&ia->ri_remove_done);
447 trace_xprtrdma_remove(r_xprt);
451 * rpcrdma_ia_close - Clean up/close an IA.
452 * @ia: interface adapter to close
456 rpcrdma_ia_close(struct rpcrdma_ia *ia)
458 if (ia->ri_id != NULL && !IS_ERR(ia->ri_id)) {
460 rdma_destroy_qp(ia->ri_id);
461 rdma_destroy_id(ia->ri_id);
465 /* If the pd is still busy, xprtrdma missed freeing a resource */
466 if (ia->ri_pd && !IS_ERR(ia->ri_pd))
467 ib_dealloc_pd(ia->ri_pd);
472 * rpcrdma_ep_create - Create unconnected endpoint
473 * @r_xprt: transport to instantiate
475 * Returns zero on success, or a negative errno.
477 int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
479 struct rpcrdma_ep *ep = &r_xprt->rx_ep;
480 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
481 struct rpcrdma_connect_private *pmsg = &ep->rep_cm_private;
482 struct ib_cq *sendcq, *recvcq;
483 unsigned int max_sge;
486 ep->rep_max_requests = xprt_rdma_slot_table_entries;
487 ep->rep_inline_send = xprt_rdma_max_inline_write;
488 ep->rep_inline_recv = xprt_rdma_max_inline_read;
490 max_sge = min_t(unsigned int, ia->ri_id->device->attrs.max_send_sge,
491 RPCRDMA_MAX_SEND_SGES);
492 if (max_sge < RPCRDMA_MIN_SEND_SGES) {
493 pr_warn("rpcrdma: HCA provides only %d send SGEs\n", max_sge);
496 ia->ri_max_send_sges = max_sge;
498 rc = frwr_open(ia, ep);
502 ep->rep_attr.event_handler = rpcrdma_qp_event_handler;
503 ep->rep_attr.qp_context = ep;
504 ep->rep_attr.srq = NULL;
505 ep->rep_attr.cap.max_send_sge = max_sge;
506 ep->rep_attr.cap.max_recv_sge = 1;
507 ep->rep_attr.cap.max_inline_data = 0;
508 ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
509 ep->rep_attr.qp_type = IB_QPT_RC;
510 ep->rep_attr.port_num = ~0;
512 dprintk("RPC: %s: requested max: dtos: send %d recv %d; "
513 "iovs: send %d recv %d\n",
515 ep->rep_attr.cap.max_send_wr,
516 ep->rep_attr.cap.max_recv_wr,
517 ep->rep_attr.cap.max_send_sge,
518 ep->rep_attr.cap.max_recv_sge);
520 ep->rep_send_batch = ep->rep_max_requests >> 3;
521 ep->rep_send_count = ep->rep_send_batch;
522 init_waitqueue_head(&ep->rep_connect_wait);
523 ep->rep_receive_count = 0;
525 sendcq = ib_alloc_cq_any(ia->ri_id->device, NULL,
526 ep->rep_attr.cap.max_send_wr + 1,
528 if (IS_ERR(sendcq)) {
529 rc = PTR_ERR(sendcq);
533 recvcq = ib_alloc_cq_any(ia->ri_id->device, NULL,
534 ep->rep_attr.cap.max_recv_wr + 1,
536 if (IS_ERR(recvcq)) {
537 rc = PTR_ERR(recvcq);
541 ep->rep_attr.send_cq = sendcq;
542 ep->rep_attr.recv_cq = recvcq;
544 /* Initialize cma parameters */
545 memset(&ep->rep_remote_cma, 0, sizeof(ep->rep_remote_cma));
547 /* Prepare RDMA-CM private message */
548 pmsg->cp_magic = rpcrdma_cmp_magic;
549 pmsg->cp_version = RPCRDMA_CMP_VERSION;
550 pmsg->cp_flags |= RPCRDMA_CMP_F_SND_W_INV_OK;
551 pmsg->cp_send_size = rpcrdma_encode_buffer_size(ep->rep_inline_send);
552 pmsg->cp_recv_size = rpcrdma_encode_buffer_size(ep->rep_inline_recv);
553 ep->rep_remote_cma.private_data = pmsg;
554 ep->rep_remote_cma.private_data_len = sizeof(*pmsg);
556 /* Client offers RDMA Read but does not initiate */
557 ep->rep_remote_cma.initiator_depth = 0;
558 ep->rep_remote_cma.responder_resources =
559 min_t(int, U8_MAX, ia->ri_id->device->attrs.max_qp_rd_atom);
561 /* Limit transport retries so client can detect server
562 * GID changes quickly. RPC layer handles re-establishing
563 * transport connection and retransmission.
565 ep->rep_remote_cma.retry_count = 6;
567 /* RPC-over-RDMA handles its own flow control. In addition,
568 * make all RNR NAKs visible so we know that RPC-over-RDMA
569 * flow control is working correctly (no NAKs should be seen).
571 ep->rep_remote_cma.flow_control = 0;
572 ep->rep_remote_cma.rnr_retry_count = 0;
583 * rpcrdma_ep_destroy - Disconnect and destroy endpoint.
584 * @r_xprt: transport instance to shut down
587 void rpcrdma_ep_destroy(struct rpcrdma_xprt *r_xprt)
589 struct rpcrdma_ep *ep = &r_xprt->rx_ep;
590 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
592 if (ia->ri_id && ia->ri_id->qp) {
593 rpcrdma_ep_disconnect(ep, ia);
594 rdma_destroy_qp(ia->ri_id);
595 ia->ri_id->qp = NULL;
598 if (ep->rep_attr.recv_cq)
599 ib_free_cq(ep->rep_attr.recv_cq);
600 if (ep->rep_attr.send_cq)
601 ib_free_cq(ep->rep_attr.send_cq);
604 /* Re-establish a connection after a device removal event.
605 * Unlike a normal reconnection, a fresh PD and a new set
606 * of MRs and buffers is needed.
608 static int rpcrdma_ep_recreate_xprt(struct rpcrdma_xprt *r_xprt,
609 struct ib_qp_init_attr *qp_init_attr)
611 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
614 trace_xprtrdma_reinsert(r_xprt);
617 if (rpcrdma_ia_open(r_xprt))
621 err = rpcrdma_ep_create(r_xprt);
623 pr_err("rpcrdma: rpcrdma_ep_create returned %d\n", err);
628 err = rdma_create_qp(ia->ri_id, ia->ri_pd, qp_init_attr);
630 pr_err("rpcrdma: rdma_create_qp returned %d\n", err);
634 rpcrdma_mrs_create(r_xprt);
638 rpcrdma_ep_destroy(r_xprt);
640 rpcrdma_ia_close(ia);
645 static int rpcrdma_ep_reconnect(struct rpcrdma_xprt *r_xprt,
646 struct ib_qp_init_attr *qp_init_attr)
648 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
649 struct rdma_cm_id *id, *old;
652 trace_xprtrdma_reconnect(r_xprt);
654 rpcrdma_ep_disconnect(&r_xprt->rx_ep, ia);
657 id = rpcrdma_create_id(r_xprt, ia);
661 /* As long as the new ID points to the same device as the
662 * old ID, we can reuse the transport's existing PD and all
663 * previously allocated MRs. Also, the same device means
664 * the transport's previous DMA mappings are still valid.
666 * This is a sanity check only. There should be no way these
667 * point to two different devices here.
671 if (ia->ri_id->device != id->device) {
672 pr_err("rpcrdma: can't reconnect on different device!\n");
676 err = rdma_create_qp(id, ia->ri_pd, qp_init_attr);
680 /* Atomically replace the transport's ID and QP. */
684 rdma_destroy_qp(old);
687 rdma_destroy_id(old);
693 * Connect unconnected endpoint.
696 rpcrdma_ep_connect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
698 struct rpcrdma_xprt *r_xprt = container_of(ia, struct rpcrdma_xprt,
700 struct rpc_xprt *xprt = &r_xprt->rx_xprt;
701 struct ib_qp_init_attr qp_init_attr;
705 memcpy(&qp_init_attr, &ep->rep_attr, sizeof(qp_init_attr));
706 switch (ep->rep_connected) {
708 dprintk("RPC: %s: connecting...\n", __func__);
709 rc = rdma_create_qp(ia->ri_id, ia->ri_pd, &qp_init_attr);
716 rc = rpcrdma_ep_recreate_xprt(r_xprt, &qp_init_attr);
721 rc = rpcrdma_ep_reconnect(r_xprt, &qp_init_attr);
726 ep->rep_connected = 0;
727 xprt_clear_connected(xprt);
729 rpcrdma_post_recvs(r_xprt, true);
731 rc = rdma_connect(ia->ri_id, &ep->rep_remote_cma);
735 if (xprt->reestablish_timeout < RPCRDMA_INIT_REEST_TO)
736 xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
737 wait_event_interruptible(ep->rep_connect_wait, ep->rep_connected != 0);
738 if (ep->rep_connected <= 0) {
739 if (ep->rep_connected == -EAGAIN)
741 rc = ep->rep_connected;
745 dprintk("RPC: %s: connected\n", __func__);
749 ep->rep_connected = rc;
756 * rpcrdma_ep_disconnect - Disconnect underlying transport
757 * @ep: endpoint to disconnect
758 * @ia: associated interface adapter
760 * This is separate from destroy to facilitate the ability
761 * to reconnect without recreating the endpoint.
763 * This call is not reentrant, and must not be made in parallel
764 * on the same endpoint.
767 rpcrdma_ep_disconnect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
769 struct rpcrdma_xprt *r_xprt = container_of(ep, struct rpcrdma_xprt,
773 /* returns without wait if ID is not connected */
774 rc = rdma_disconnect(ia->ri_id);
776 wait_event_interruptible(ep->rep_connect_wait,
777 ep->rep_connected != 1);
779 ep->rep_connected = rc;
780 trace_xprtrdma_disconnect(r_xprt, rc);
782 rpcrdma_xprt_drain(r_xprt);
785 /* Fixed-size circular FIFO queue. This implementation is wait-free and
788 * Consumer is the code path that posts Sends. This path dequeues a
789 * sendctx for use by a Send operation. Multiple consumer threads
790 * are serialized by the RPC transport lock, which allows only one
791 * ->send_request call at a time.
793 * Producer is the code path that handles Send completions. This path
794 * enqueues a sendctx that has been completed. Multiple producer
795 * threads are serialized by the ib_poll_cq() function.
798 /* rpcrdma_sendctxs_destroy() assumes caller has already quiesced
799 * queue activity, and rpcrdma_xprt_drain has flushed all remaining
802 static void rpcrdma_sendctxs_destroy(struct rpcrdma_buffer *buf)
806 for (i = 0; i <= buf->rb_sc_last; i++)
807 kfree(buf->rb_sc_ctxs[i]);
808 kfree(buf->rb_sc_ctxs);
811 static struct rpcrdma_sendctx *rpcrdma_sendctx_create(struct rpcrdma_ia *ia)
813 struct rpcrdma_sendctx *sc;
815 sc = kzalloc(struct_size(sc, sc_sges, ia->ri_max_send_sges),
820 sc->sc_wr.wr_cqe = &sc->sc_cqe;
821 sc->sc_wr.sg_list = sc->sc_sges;
822 sc->sc_wr.opcode = IB_WR_SEND;
823 sc->sc_cqe.done = rpcrdma_wc_send;
827 static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt)
829 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
830 struct rpcrdma_sendctx *sc;
833 /* Maximum number of concurrent outstanding Send WRs. Capping
834 * the circular queue size stops Send Queue overflow by causing
835 * the ->send_request call to fail temporarily before too many
838 i = buf->rb_max_requests + RPCRDMA_MAX_BC_REQUESTS;
839 dprintk("RPC: %s: allocating %lu send_ctxs\n", __func__, i);
840 buf->rb_sc_ctxs = kcalloc(i, sizeof(sc), GFP_KERNEL);
841 if (!buf->rb_sc_ctxs)
844 buf->rb_sc_last = i - 1;
845 for (i = 0; i <= buf->rb_sc_last; i++) {
846 sc = rpcrdma_sendctx_create(&r_xprt->rx_ia);
850 sc->sc_xprt = r_xprt;
851 buf->rb_sc_ctxs[i] = sc;
857 /* The sendctx queue is not guaranteed to have a size that is a
858 * power of two, thus the helpers in circ_buf.h cannot be used.
859 * The other option is to use modulus (%), which can be expensive.
861 static unsigned long rpcrdma_sendctx_next(struct rpcrdma_buffer *buf,
864 return likely(item < buf->rb_sc_last) ? item + 1 : 0;
868 * rpcrdma_sendctx_get_locked - Acquire a send context
869 * @r_xprt: controlling transport instance
871 * Returns pointer to a free send completion context; or NULL if
872 * the queue is empty.
874 * Usage: Called to acquire an SGE array before preparing a Send WR.
876 * The caller serializes calls to this function (per transport), and
877 * provides an effective memory barrier that flushes the new value
880 struct rpcrdma_sendctx *rpcrdma_sendctx_get_locked(struct rpcrdma_xprt *r_xprt)
882 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
883 struct rpcrdma_sendctx *sc;
884 unsigned long next_head;
886 next_head = rpcrdma_sendctx_next(buf, buf->rb_sc_head);
888 if (next_head == READ_ONCE(buf->rb_sc_tail))
891 /* ORDER: item must be accessed _before_ head is updated */
892 sc = buf->rb_sc_ctxs[next_head];
894 /* Releasing the lock in the caller acts as a memory
895 * barrier that flushes rb_sc_head.
897 buf->rb_sc_head = next_head;
902 /* The queue is "empty" if there have not been enough Send
903 * completions recently. This is a sign the Send Queue is
904 * backing up. Cause the caller to pause and try again.
906 xprt_wait_for_buffer_space(&r_xprt->rx_xprt);
907 r_xprt->rx_stats.empty_sendctx_q++;
912 * rpcrdma_sendctx_put_locked - Release a send context
913 * @sc: send context to release
915 * Usage: Called from Send completion to return a sendctxt
918 * The caller serializes calls to this function (per transport).
921 rpcrdma_sendctx_put_locked(struct rpcrdma_sendctx *sc)
923 struct rpcrdma_buffer *buf = &sc->sc_xprt->rx_buf;
924 unsigned long next_tail;
926 /* Unmap SGEs of previously completed but unsignaled
927 * Sends by walking up the queue until @sc is found.
929 next_tail = buf->rb_sc_tail;
931 next_tail = rpcrdma_sendctx_next(buf, next_tail);
933 /* ORDER: item must be accessed _before_ tail is updated */
934 rpcrdma_sendctx_unmap(buf->rb_sc_ctxs[next_tail]);
936 } while (buf->rb_sc_ctxs[next_tail] != sc);
938 /* Paired with READ_ONCE */
939 smp_store_release(&buf->rb_sc_tail, next_tail);
941 xprt_write_space(&sc->sc_xprt->rx_xprt);
945 rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt)
947 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
948 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
951 for (count = 0; count < ia->ri_max_segs; count++) {
952 struct rpcrdma_mr *mr;
955 mr = kzalloc(sizeof(*mr), GFP_NOFS);
959 rc = frwr_init_mr(ia, mr);
965 mr->mr_xprt = r_xprt;
967 spin_lock(&buf->rb_lock);
968 list_add(&mr->mr_list, &buf->rb_mrs);
969 list_add(&mr->mr_all, &buf->rb_all_mrs);
970 spin_unlock(&buf->rb_lock);
973 r_xprt->rx_stats.mrs_allocated += count;
974 trace_xprtrdma_createmrs(r_xprt, count);
978 rpcrdma_mr_refresh_worker(struct work_struct *work)
980 struct rpcrdma_buffer *buf = container_of(work, struct rpcrdma_buffer,
982 struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt,
985 rpcrdma_mrs_create(r_xprt);
986 xprt_write_space(&r_xprt->rx_xprt);
990 * rpcrdma_req_create - Allocate an rpcrdma_req object
991 * @r_xprt: controlling r_xprt
992 * @size: initial size, in bytes, of send and receive buffers
993 * @flags: GFP flags passed to memory allocators
995 * Returns an allocated and fully initialized rpcrdma_req or NULL.
997 struct rpcrdma_req *rpcrdma_req_create(struct rpcrdma_xprt *r_xprt, size_t size,
1000 struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
1001 struct rpcrdma_regbuf *rb;
1002 struct rpcrdma_req *req;
1005 req = kzalloc(sizeof(*req), flags);
1009 /* Compute maximum header buffer size in bytes */
1010 maxhdrsize = rpcrdma_fixed_maxsz + 3 +
1011 r_xprt->rx_ia.ri_max_segs * rpcrdma_readchunk_maxsz;
1012 maxhdrsize *= sizeof(__be32);
1013 rb = rpcrdma_regbuf_alloc(__roundup_pow_of_two(maxhdrsize),
1014 DMA_TO_DEVICE, flags);
1017 req->rl_rdmabuf = rb;
1018 xdr_buf_init(&req->rl_hdrbuf, rdmab_data(rb), rdmab_length(rb));
1020 req->rl_sendbuf = rpcrdma_regbuf_alloc(size, DMA_TO_DEVICE, flags);
1021 if (!req->rl_sendbuf)
1024 req->rl_recvbuf = rpcrdma_regbuf_alloc(size, DMA_NONE, flags);
1025 if (!req->rl_recvbuf)
1028 INIT_LIST_HEAD(&req->rl_free_mrs);
1029 INIT_LIST_HEAD(&req->rl_registered);
1030 spin_lock(&buffer->rb_lock);
1031 list_add(&req->rl_all, &buffer->rb_allreqs);
1032 spin_unlock(&buffer->rb_lock);
1036 kfree(req->rl_sendbuf);
1038 kfree(req->rl_rdmabuf);
1045 static struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt,
1048 struct rpcrdma_rep *rep;
1050 rep = kzalloc(sizeof(*rep), GFP_KERNEL);
1054 rep->rr_rdmabuf = rpcrdma_regbuf_alloc(r_xprt->rx_ep.rep_inline_recv,
1055 DMA_FROM_DEVICE, GFP_KERNEL);
1056 if (!rep->rr_rdmabuf)
1059 xdr_buf_init(&rep->rr_hdrbuf, rdmab_data(rep->rr_rdmabuf),
1060 rdmab_length(rep->rr_rdmabuf));
1061 rep->rr_cqe.done = rpcrdma_wc_receive;
1062 rep->rr_rxprt = r_xprt;
1063 rep->rr_recv_wr.next = NULL;
1064 rep->rr_recv_wr.wr_cqe = &rep->rr_cqe;
1065 rep->rr_recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov;
1066 rep->rr_recv_wr.num_sge = 1;
1067 rep->rr_temp = temp;
1076 static void rpcrdma_rep_destroy(struct rpcrdma_rep *rep)
1078 rpcrdma_regbuf_free(rep->rr_rdmabuf);
1082 static struct rpcrdma_rep *rpcrdma_rep_get_locked(struct rpcrdma_buffer *buf)
1084 struct llist_node *node;
1086 /* Calls to llist_del_first are required to be serialized */
1087 node = llist_del_first(&buf->rb_free_reps);
1090 return llist_entry(node, struct rpcrdma_rep, rr_node);
1093 static void rpcrdma_rep_put(struct rpcrdma_buffer *buf,
1094 struct rpcrdma_rep *rep)
1097 llist_add(&rep->rr_node, &buf->rb_free_reps);
1099 rpcrdma_rep_destroy(rep);
1102 static void rpcrdma_reps_destroy(struct rpcrdma_buffer *buf)
1104 struct rpcrdma_rep *rep;
1106 while ((rep = rpcrdma_rep_get_locked(buf)) != NULL)
1107 rpcrdma_rep_destroy(rep);
1111 * rpcrdma_buffer_create - Create initial set of req/rep objects
1112 * @r_xprt: transport instance to (re)initialize
1114 * Returns zero on success, otherwise a negative errno.
1116 int rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt)
1118 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1121 buf->rb_max_requests = r_xprt->rx_ep.rep_max_requests;
1122 buf->rb_bc_srv_max_requests = 0;
1123 spin_lock_init(&buf->rb_lock);
1124 INIT_LIST_HEAD(&buf->rb_mrs);
1125 INIT_LIST_HEAD(&buf->rb_all_mrs);
1126 INIT_WORK(&buf->rb_refresh_worker, rpcrdma_mr_refresh_worker);
1128 rpcrdma_mrs_create(r_xprt);
1130 INIT_LIST_HEAD(&buf->rb_send_bufs);
1131 INIT_LIST_HEAD(&buf->rb_allreqs);
1134 for (i = 0; i < buf->rb_max_requests; i++) {
1135 struct rpcrdma_req *req;
1137 req = rpcrdma_req_create(r_xprt, RPCRDMA_V1_DEF_INLINE_SIZE,
1141 list_add(&req->rl_list, &buf->rb_send_bufs);
1144 buf->rb_credits = 1;
1145 init_llist_head(&buf->rb_free_reps);
1147 rc = rpcrdma_sendctxs_create(r_xprt);
1153 rpcrdma_buffer_destroy(buf);
1158 * rpcrdma_req_destroy - Destroy an rpcrdma_req object
1159 * @req: unused object to be destroyed
1161 * This function assumes that the caller prevents concurrent device
1162 * unload and transport tear-down.
1164 void rpcrdma_req_destroy(struct rpcrdma_req *req)
1166 list_del(&req->rl_all);
1168 while (!list_empty(&req->rl_free_mrs))
1169 rpcrdma_mr_free(rpcrdma_mr_pop(&req->rl_free_mrs));
1171 rpcrdma_regbuf_free(req->rl_recvbuf);
1172 rpcrdma_regbuf_free(req->rl_sendbuf);
1173 rpcrdma_regbuf_free(req->rl_rdmabuf);
1178 rpcrdma_mrs_destroy(struct rpcrdma_buffer *buf)
1180 struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt,
1182 struct rpcrdma_mr *mr;
1186 spin_lock(&buf->rb_lock);
1187 while ((mr = list_first_entry_or_null(&buf->rb_all_mrs,
1190 list_del(&mr->mr_all);
1191 spin_unlock(&buf->rb_lock);
1193 frwr_release_mr(mr);
1195 spin_lock(&buf->rb_lock);
1197 spin_unlock(&buf->rb_lock);
1198 r_xprt->rx_stats.mrs_allocated = 0;
1202 * rpcrdma_buffer_destroy - Release all hw resources
1203 * @buf: root control block for resources
1205 * ORDERING: relies on a prior rpcrdma_xprt_drain :
1206 * - No more Send or Receive completions can occur
1207 * - All MRs, reps, and reqs are returned to their free lists
1210 rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf)
1212 cancel_work_sync(&buf->rb_refresh_worker);
1214 rpcrdma_sendctxs_destroy(buf);
1215 rpcrdma_reps_destroy(buf);
1217 while (!list_empty(&buf->rb_send_bufs)) {
1218 struct rpcrdma_req *req;
1220 req = list_first_entry(&buf->rb_send_bufs,
1221 struct rpcrdma_req, rl_list);
1222 list_del(&req->rl_list);
1223 rpcrdma_req_destroy(req);
1226 rpcrdma_mrs_destroy(buf);
1230 * rpcrdma_mr_get - Allocate an rpcrdma_mr object
1231 * @r_xprt: controlling transport
1233 * Returns an initialized rpcrdma_mr or NULL if no free
1234 * rpcrdma_mr objects are available.
1237 rpcrdma_mr_get(struct rpcrdma_xprt *r_xprt)
1239 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1240 struct rpcrdma_mr *mr;
1242 spin_lock(&buf->rb_lock);
1243 mr = rpcrdma_mr_pop(&buf->rb_mrs);
1244 spin_unlock(&buf->rb_lock);
1249 * rpcrdma_mr_put - DMA unmap an MR and release it
1250 * @mr: MR to release
1253 void rpcrdma_mr_put(struct rpcrdma_mr *mr)
1255 struct rpcrdma_xprt *r_xprt = mr->mr_xprt;
1257 if (mr->mr_dir != DMA_NONE) {
1258 trace_xprtrdma_mr_unmap(mr);
1259 ib_dma_unmap_sg(r_xprt->rx_ia.ri_id->device,
1260 mr->mr_sg, mr->mr_nents, mr->mr_dir);
1261 mr->mr_dir = DMA_NONE;
1264 rpcrdma_mr_push(mr, &mr->mr_req->rl_free_mrs);
1267 static void rpcrdma_mr_free(struct rpcrdma_mr *mr)
1269 struct rpcrdma_xprt *r_xprt = mr->mr_xprt;
1270 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1273 spin_lock(&buf->rb_lock);
1274 rpcrdma_mr_push(mr, &buf->rb_mrs);
1275 spin_unlock(&buf->rb_lock);
1279 * rpcrdma_buffer_get - Get a request buffer
1280 * @buffers: Buffer pool from which to obtain a buffer
1282 * Returns a fresh rpcrdma_req, or NULL if none are available.
1284 struct rpcrdma_req *
1285 rpcrdma_buffer_get(struct rpcrdma_buffer *buffers)
1287 struct rpcrdma_req *req;
1289 spin_lock(&buffers->rb_lock);
1290 req = list_first_entry_or_null(&buffers->rb_send_bufs,
1291 struct rpcrdma_req, rl_list);
1293 list_del_init(&req->rl_list);
1294 spin_unlock(&buffers->rb_lock);
1299 * rpcrdma_buffer_put - Put request/reply buffers back into pool
1300 * @buffers: buffer pool
1301 * @req: object to return
1304 void rpcrdma_buffer_put(struct rpcrdma_buffer *buffers, struct rpcrdma_req *req)
1307 rpcrdma_rep_put(buffers, req->rl_reply);
1308 req->rl_reply = NULL;
1310 spin_lock(&buffers->rb_lock);
1311 list_add(&req->rl_list, &buffers->rb_send_bufs);
1312 spin_unlock(&buffers->rb_lock);
1316 * rpcrdma_recv_buffer_put - Release rpcrdma_rep back to free list
1317 * @rep: rep to release
1319 * Used after error conditions.
1321 void rpcrdma_recv_buffer_put(struct rpcrdma_rep *rep)
1323 rpcrdma_rep_put(&rep->rr_rxprt->rx_buf, rep);
1326 /* Returns a pointer to a rpcrdma_regbuf object, or NULL.
1328 * xprtrdma uses a regbuf for posting an outgoing RDMA SEND, or for
1329 * receiving the payload of RDMA RECV operations. During Long Calls
1330 * or Replies they may be registered externally via frwr_map.
1332 static struct rpcrdma_regbuf *
1333 rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction,
1336 struct rpcrdma_regbuf *rb;
1338 rb = kmalloc(sizeof(*rb), flags);
1341 rb->rg_data = kmalloc(size, flags);
1347 rb->rg_device = NULL;
1348 rb->rg_direction = direction;
1349 rb->rg_iov.length = size;
1354 * rpcrdma_regbuf_realloc - re-allocate a SEND/RECV buffer
1355 * @rb: regbuf to reallocate
1356 * @size: size of buffer to be allocated, in bytes
1359 * Returns true if reallocation was successful. If false is
1360 * returned, @rb is left untouched.
1362 bool rpcrdma_regbuf_realloc(struct rpcrdma_regbuf *rb, size_t size, gfp_t flags)
1366 buf = kmalloc(size, flags);
1370 rpcrdma_regbuf_dma_unmap(rb);
1374 rb->rg_iov.length = size;
1379 * __rpcrdma_regbuf_dma_map - DMA-map a regbuf
1380 * @r_xprt: controlling transport instance
1381 * @rb: regbuf to be mapped
1383 * Returns true if the buffer is now DMA mapped to @r_xprt's device
1385 bool __rpcrdma_regbuf_dma_map(struct rpcrdma_xprt *r_xprt,
1386 struct rpcrdma_regbuf *rb)
1388 struct ib_device *device = r_xprt->rx_ia.ri_id->device;
1390 if (rb->rg_direction == DMA_NONE)
1393 rb->rg_iov.addr = ib_dma_map_single(device, rdmab_data(rb),
1394 rdmab_length(rb), rb->rg_direction);
1395 if (ib_dma_mapping_error(device, rdmab_addr(rb))) {
1396 trace_xprtrdma_dma_maperr(rdmab_addr(rb));
1400 rb->rg_device = device;
1401 rb->rg_iov.lkey = r_xprt->rx_ia.ri_pd->local_dma_lkey;
1405 static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb)
1410 if (!rpcrdma_regbuf_is_mapped(rb))
1413 ib_dma_unmap_single(rb->rg_device, rdmab_addr(rb), rdmab_length(rb),
1415 rb->rg_device = NULL;
1418 static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb)
1420 rpcrdma_regbuf_dma_unmap(rb);
1427 * rpcrdma_ep_post - Post WRs to a transport's Send Queue
1428 * @ia: transport's device information
1429 * @ep: transport's RDMA endpoint information
1430 * @req: rpcrdma_req containing the Send WR to post
1432 * Returns 0 if the post was successful, otherwise -ENOTCONN
1436 rpcrdma_ep_post(struct rpcrdma_ia *ia,
1437 struct rpcrdma_ep *ep,
1438 struct rpcrdma_req *req)
1440 struct ib_send_wr *send_wr = &req->rl_sendctx->sc_wr;
1443 if (!ep->rep_send_count || kref_read(&req->rl_kref) > 1) {
1444 send_wr->send_flags |= IB_SEND_SIGNALED;
1445 ep->rep_send_count = ep->rep_send_batch;
1447 send_wr->send_flags &= ~IB_SEND_SIGNALED;
1448 --ep->rep_send_count;
1451 rc = frwr_send(ia, req);
1452 trace_xprtrdma_post_send(req, rc);
1459 rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp)
1461 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1462 struct rpcrdma_ep *ep = &r_xprt->rx_ep;
1463 struct ib_recv_wr *i, *wr, *bad_wr;
1464 struct rpcrdma_rep *rep;
1465 int needed, count, rc;
1470 needed = buf->rb_credits + (buf->rb_bc_srv_max_requests << 1);
1471 if (likely(ep->rep_receive_count > needed))
1473 needed -= ep->rep_receive_count;
1475 needed += RPCRDMA_MAX_RECV_BATCH;
1477 /* fast path: all needed reps can be found on the free list */
1480 rep = rpcrdma_rep_get_locked(buf);
1482 rep = rpcrdma_rep_create(r_xprt, temp);
1486 rep->rr_recv_wr.next = wr;
1487 wr = &rep->rr_recv_wr;
1493 for (i = wr; i; i = i->next) {
1494 rep = container_of(i, struct rpcrdma_rep, rr_recv_wr);
1496 if (!rpcrdma_regbuf_dma_map(r_xprt, rep->rr_rdmabuf))
1499 trace_xprtrdma_post_recv(rep);
1503 rc = ib_post_recv(r_xprt->rx_ia.ri_id->qp, wr,
1504 (const struct ib_recv_wr **)&bad_wr);
1506 trace_xprtrdma_post_recvs(r_xprt, count, rc);
1508 for (wr = bad_wr; wr;) {
1509 struct rpcrdma_rep *rep;
1511 rep = container_of(wr, struct rpcrdma_rep, rr_recv_wr);
1513 rpcrdma_recv_buffer_put(rep);
1517 ep->rep_receive_count += count;
1522 rep = container_of(i, struct rpcrdma_rep, rr_recv_wr);
1524 rpcrdma_recv_buffer_put(rep);