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/pci.h> /* for Tavor hack below */
52 #include <linux/slab.h>
54 #include "xprt_rdma.h"
61 # define RPCDBG_FACILITY RPCDBG_TRANS
69 * handle replies in tasklet context, using a single, global list
70 * rdma tasklet function -- just turn around and call the func
71 * for all replies on the list
74 static DEFINE_SPINLOCK(rpcrdma_tk_lock_g);
75 static LIST_HEAD(rpcrdma_tasklets_g);
78 rpcrdma_run_tasklet(unsigned long data)
80 struct rpcrdma_rep *rep;
81 void (*func)(struct rpcrdma_rep *);
85 spin_lock_irqsave(&rpcrdma_tk_lock_g, flags);
86 while (!list_empty(&rpcrdma_tasklets_g)) {
87 rep = list_entry(rpcrdma_tasklets_g.next,
88 struct rpcrdma_rep, rr_list);
89 list_del(&rep->rr_list);
92 spin_unlock_irqrestore(&rpcrdma_tk_lock_g, flags);
97 rpcrdma_recv_buffer_put(rep);
99 spin_lock_irqsave(&rpcrdma_tk_lock_g, flags);
101 spin_unlock_irqrestore(&rpcrdma_tk_lock_g, flags);
104 static DECLARE_TASKLET(rpcrdma_tasklet_g, rpcrdma_run_tasklet, 0UL);
107 rpcrdma_schedule_tasklet(struct rpcrdma_rep *rep)
111 spin_lock_irqsave(&rpcrdma_tk_lock_g, flags);
112 list_add_tail(&rep->rr_list, &rpcrdma_tasklets_g);
113 spin_unlock_irqrestore(&rpcrdma_tk_lock_g, flags);
114 tasklet_schedule(&rpcrdma_tasklet_g);
118 rpcrdma_qp_async_error_upcall(struct ib_event *event, void *context)
120 struct rpcrdma_ep *ep = context;
122 dprintk("RPC: %s: QP error %X on device %s ep %p\n",
123 __func__, event->event, event->device->name, context);
124 if (ep->rep_connected == 1) {
125 ep->rep_connected = -EIO;
127 wake_up_all(&ep->rep_connect_wait);
132 rpcrdma_cq_async_error_upcall(struct ib_event *event, void *context)
134 struct rpcrdma_ep *ep = context;
136 dprintk("RPC: %s: CQ error %X on device %s ep %p\n",
137 __func__, event->event, event->device->name, context);
138 if (ep->rep_connected == 1) {
139 ep->rep_connected = -EIO;
141 wake_up_all(&ep->rep_connect_wait);
146 void rpcrdma_event_process(struct ib_wc *wc)
148 struct rpcrdma_mw *frmr;
149 struct rpcrdma_rep *rep =
150 (struct rpcrdma_rep *)(unsigned long) wc->wr_id;
152 dprintk("RPC: %s: event rep %p status %X opcode %X length %u\n",
153 __func__, rep, wc->status, wc->opcode, wc->byte_len);
155 if (!rep) /* send or bind completion that we don't care about */
158 if (IB_WC_SUCCESS != wc->status) {
159 dprintk("RPC: %s: WC opcode %d status %X, connection lost\n",
160 __func__, wc->opcode, wc->status);
162 if (wc->opcode != IB_WC_FAST_REG_MR && wc->opcode != IB_WC_LOCAL_INV)
163 rpcrdma_schedule_tasklet(rep);
167 switch (wc->opcode) {
168 case IB_WC_FAST_REG_MR:
169 frmr = (struct rpcrdma_mw *)(unsigned long)wc->wr_id;
170 frmr->r.frmr.state = FRMR_IS_VALID;
172 case IB_WC_LOCAL_INV:
173 frmr = (struct rpcrdma_mw *)(unsigned long)wc->wr_id;
174 frmr->r.frmr.state = FRMR_IS_INVALID;
177 rep->rr_len = wc->byte_len;
178 ib_dma_sync_single_for_cpu(
179 rdmab_to_ia(rep->rr_buffer)->ri_id->device,
180 rep->rr_iov.addr, rep->rr_len, DMA_FROM_DEVICE);
181 /* Keep (only) the most recent credits, after check validity */
182 if (rep->rr_len >= 16) {
183 struct rpcrdma_msg *p =
184 (struct rpcrdma_msg *) rep->rr_base;
185 unsigned int credits = ntohl(p->rm_credit);
187 dprintk("RPC: %s: server"
188 " dropped credits to 0!\n", __func__);
191 } else if (credits > rep->rr_buffer->rb_max_requests) {
192 dprintk("RPC: %s: server"
193 " over-crediting: %d (%d)\n",
195 rep->rr_buffer->rb_max_requests);
196 credits = rep->rr_buffer->rb_max_requests;
198 atomic_set(&rep->rr_buffer->rb_credits, credits);
202 rpcrdma_schedule_tasklet(rep);
205 dprintk("RPC: %s: unexpected WC event %X\n",
206 __func__, wc->opcode);
212 rpcrdma_cq_poll(struct ib_cq *cq)
218 rc = ib_poll_cq(cq, 1, &wc);
220 dprintk("RPC: %s: ib_poll_cq failed %i\n",
227 rpcrdma_event_process(&wc);
234 * rpcrdma_cq_event_upcall
236 * This upcall handles recv, send, bind and unbind events.
237 * It is reentrant but processes single events in order to maintain
238 * ordering of receives to keep server credits.
240 * It is the responsibility of the scheduled tasklet to return
241 * recv buffers to the pool. NOTE: this affects synchronization of
242 * connection shutdown. That is, the structures required for
243 * the completion of the reply handler must remain intact until
244 * all memory has been reclaimed.
246 * Note that send events are suppressed and do not result in an upcall.
249 rpcrdma_cq_event_upcall(struct ib_cq *cq, void *context)
253 rc = rpcrdma_cq_poll(cq);
257 rc = ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
259 dprintk("RPC: %s: ib_req_notify_cq failed %i\n",
268 static const char * const conn[] = {
285 rpcrdma_conn_upcall(struct rdma_cm_id *id, struct rdma_cm_event *event)
287 struct rpcrdma_xprt *xprt = id->context;
288 struct rpcrdma_ia *ia = &xprt->rx_ia;
289 struct rpcrdma_ep *ep = &xprt->rx_ep;
291 struct sockaddr_in *addr = (struct sockaddr_in *) &ep->rep_remote_addr;
293 struct ib_qp_attr attr;
294 struct ib_qp_init_attr iattr;
297 switch (event->event) {
298 case RDMA_CM_EVENT_ADDR_RESOLVED:
299 case RDMA_CM_EVENT_ROUTE_RESOLVED:
301 complete(&ia->ri_done);
303 case RDMA_CM_EVENT_ADDR_ERROR:
304 ia->ri_async_rc = -EHOSTUNREACH;
305 dprintk("RPC: %s: CM address resolution error, ep 0x%p\n",
307 complete(&ia->ri_done);
309 case RDMA_CM_EVENT_ROUTE_ERROR:
310 ia->ri_async_rc = -ENETUNREACH;
311 dprintk("RPC: %s: CM route resolution error, ep 0x%p\n",
313 complete(&ia->ri_done);
315 case RDMA_CM_EVENT_ESTABLISHED:
317 ib_query_qp(ia->ri_id->qp, &attr,
318 IB_QP_MAX_QP_RD_ATOMIC | IB_QP_MAX_DEST_RD_ATOMIC,
320 dprintk("RPC: %s: %d responder resources"
322 __func__, attr.max_dest_rd_atomic, attr.max_rd_atomic);
324 case RDMA_CM_EVENT_CONNECT_ERROR:
325 connstate = -ENOTCONN;
327 case RDMA_CM_EVENT_UNREACHABLE:
328 connstate = -ENETDOWN;
330 case RDMA_CM_EVENT_REJECTED:
331 connstate = -ECONNREFUSED;
333 case RDMA_CM_EVENT_DISCONNECTED:
334 connstate = -ECONNABORTED;
336 case RDMA_CM_EVENT_DEVICE_REMOVAL:
339 dprintk("RPC: %s: %s: %pI4:%u (ep 0x%p event 0x%x)\n",
341 (event->event <= 11) ? conn[event->event] :
342 "unknown connection error",
343 &addr->sin_addr.s_addr,
344 ntohs(addr->sin_port),
346 atomic_set(&rpcx_to_rdmax(ep->rep_xprt)->rx_buf.rb_credits, 1);
347 dprintk("RPC: %s: %sconnected\n",
348 __func__, connstate > 0 ? "" : "dis");
349 ep->rep_connected = connstate;
351 wake_up_all(&ep->rep_connect_wait);
354 dprintk("RPC: %s: unexpected CM event %d\n",
355 __func__, event->event);
360 if (connstate == 1) {
361 int ird = attr.max_dest_rd_atomic;
362 int tird = ep->rep_remote_cma.responder_resources;
363 printk(KERN_INFO "rpcrdma: connection to %pI4:%u "
364 "on %s, memreg %d slots %d ird %d%s\n",
365 &addr->sin_addr.s_addr,
366 ntohs(addr->sin_port),
367 ia->ri_id->device->name,
368 ia->ri_memreg_strategy,
369 xprt->rx_buf.rb_max_requests,
370 ird, ird < 4 && ird < tird / 2 ? " (low!)" : "");
371 } else if (connstate < 0) {
372 printk(KERN_INFO "rpcrdma: connection to %pI4:%u closed (%d)\n",
373 &addr->sin_addr.s_addr,
374 ntohs(addr->sin_port),
382 static struct rdma_cm_id *
383 rpcrdma_create_id(struct rpcrdma_xprt *xprt,
384 struct rpcrdma_ia *ia, struct sockaddr *addr)
386 struct rdma_cm_id *id;
389 init_completion(&ia->ri_done);
391 id = rdma_create_id(rpcrdma_conn_upcall, xprt, RDMA_PS_TCP, IB_QPT_RC);
394 dprintk("RPC: %s: rdma_create_id() failed %i\n",
399 ia->ri_async_rc = -ETIMEDOUT;
400 rc = rdma_resolve_addr(id, NULL, addr, RDMA_RESOLVE_TIMEOUT);
402 dprintk("RPC: %s: rdma_resolve_addr() failed %i\n",
406 wait_for_completion_interruptible_timeout(&ia->ri_done,
407 msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
408 rc = ia->ri_async_rc;
412 ia->ri_async_rc = -ETIMEDOUT;
413 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
415 dprintk("RPC: %s: rdma_resolve_route() failed %i\n",
419 wait_for_completion_interruptible_timeout(&ia->ri_done,
420 msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
421 rc = ia->ri_async_rc;
433 * Drain any cq, prior to teardown.
436 rpcrdma_clean_cq(struct ib_cq *cq)
441 while (1 == ib_poll_cq(cq, 1, &wc))
445 dprintk("RPC: %s: flushed %d events (last 0x%x)\n",
446 __func__, count, wc.opcode);
450 * Exported functions.
454 * Open and initialize an Interface Adapter.
455 * o initializes fields of struct rpcrdma_ia, including
456 * interface and provider attributes and protection zone.
459 rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
462 struct ib_device_attr devattr;
463 struct rpcrdma_ia *ia = &xprt->rx_ia;
465 ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
466 if (IS_ERR(ia->ri_id)) {
467 rc = PTR_ERR(ia->ri_id);
471 ia->ri_pd = ib_alloc_pd(ia->ri_id->device);
472 if (IS_ERR(ia->ri_pd)) {
473 rc = PTR_ERR(ia->ri_pd);
474 dprintk("RPC: %s: ib_alloc_pd() failed %i\n",
480 * Query the device to determine if the requested memory
481 * registration strategy is supported. If it isn't, set the
482 * strategy to a globally supported model.
484 rc = ib_query_device(ia->ri_id->device, &devattr);
486 dprintk("RPC: %s: ib_query_device failed %d\n",
491 if (devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY) {
492 ia->ri_have_dma_lkey = 1;
493 ia->ri_dma_lkey = ia->ri_id->device->local_dma_lkey;
497 case RPCRDMA_MEMWINDOWS:
498 case RPCRDMA_MEMWINDOWS_ASYNC:
499 if (!(devattr.device_cap_flags & IB_DEVICE_MEM_WINDOW)) {
500 dprintk("RPC: %s: MEMWINDOWS registration "
501 "specified but not supported by adapter, "
502 "using slower RPCRDMA_REGISTER\n",
504 memreg = RPCRDMA_REGISTER;
507 case RPCRDMA_MTHCAFMR:
508 if (!ia->ri_id->device->alloc_fmr) {
509 #if RPCRDMA_PERSISTENT_REGISTRATION
510 dprintk("RPC: %s: MTHCAFMR registration "
511 "specified but not supported by adapter, "
512 "using riskier RPCRDMA_ALLPHYSICAL\n",
514 memreg = RPCRDMA_ALLPHYSICAL;
516 dprintk("RPC: %s: MTHCAFMR registration "
517 "specified but not supported by adapter, "
518 "using slower RPCRDMA_REGISTER\n",
520 memreg = RPCRDMA_REGISTER;
525 /* Requires both frmr reg and local dma lkey */
526 if ((devattr.device_cap_flags &
527 (IB_DEVICE_MEM_MGT_EXTENSIONS|IB_DEVICE_LOCAL_DMA_LKEY)) !=
528 (IB_DEVICE_MEM_MGT_EXTENSIONS|IB_DEVICE_LOCAL_DMA_LKEY)) {
529 #if RPCRDMA_PERSISTENT_REGISTRATION
530 dprintk("RPC: %s: FRMR registration "
531 "specified but not supported by adapter, "
532 "using riskier RPCRDMA_ALLPHYSICAL\n",
534 memreg = RPCRDMA_ALLPHYSICAL;
536 dprintk("RPC: %s: FRMR registration "
537 "specified but not supported by adapter, "
538 "using slower RPCRDMA_REGISTER\n",
540 memreg = RPCRDMA_REGISTER;
547 * Optionally obtain an underlying physical identity mapping in
548 * order to do a memory window-based bind. This base registration
549 * is protected from remote access - that is enabled only by binding
550 * for the specific bytes targeted during each RPC operation, and
551 * revoked after the corresponding completion similar to a storage
555 case RPCRDMA_BOUNCEBUFFERS:
556 case RPCRDMA_REGISTER:
559 #if RPCRDMA_PERSISTENT_REGISTRATION
560 case RPCRDMA_ALLPHYSICAL:
561 mem_priv = IB_ACCESS_LOCAL_WRITE |
562 IB_ACCESS_REMOTE_WRITE |
563 IB_ACCESS_REMOTE_READ;
566 case RPCRDMA_MEMWINDOWS_ASYNC:
567 case RPCRDMA_MEMWINDOWS:
568 mem_priv = IB_ACCESS_LOCAL_WRITE |
571 case RPCRDMA_MTHCAFMR:
572 if (ia->ri_have_dma_lkey)
574 mem_priv = IB_ACCESS_LOCAL_WRITE;
576 ia->ri_bind_mem = ib_get_dma_mr(ia->ri_pd, mem_priv);
577 if (IS_ERR(ia->ri_bind_mem)) {
578 printk(KERN_ALERT "%s: ib_get_dma_mr for "
579 "phys register failed with %lX\n\t"
580 "Will continue with degraded performance\n",
581 __func__, PTR_ERR(ia->ri_bind_mem));
582 memreg = RPCRDMA_REGISTER;
583 ia->ri_bind_mem = NULL;
587 printk(KERN_ERR "%s: invalid memory registration mode %d\n",
592 dprintk("RPC: %s: memory registration strategy is %d\n",
595 /* Else will do memory reg/dereg for each chunk */
596 ia->ri_memreg_strategy = memreg;
600 rdma_destroy_id(ia->ri_id);
607 * Clean up/close an IA.
608 * o if event handles and PD have been initialized, free them.
612 rpcrdma_ia_close(struct rpcrdma_ia *ia)
616 dprintk("RPC: %s: entering\n", __func__);
617 if (ia->ri_bind_mem != NULL) {
618 rc = ib_dereg_mr(ia->ri_bind_mem);
619 dprintk("RPC: %s: ib_dereg_mr returned %i\n",
622 if (ia->ri_id != NULL && !IS_ERR(ia->ri_id)) {
624 rdma_destroy_qp(ia->ri_id);
625 rdma_destroy_id(ia->ri_id);
628 if (ia->ri_pd != NULL && !IS_ERR(ia->ri_pd)) {
629 rc = ib_dealloc_pd(ia->ri_pd);
630 dprintk("RPC: %s: ib_dealloc_pd returned %i\n",
636 * Create unconnected endpoint.
639 rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
640 struct rpcrdma_create_data_internal *cdata)
642 struct ib_device_attr devattr;
645 rc = ib_query_device(ia->ri_id->device, &devattr);
647 dprintk("RPC: %s: ib_query_device failed %d\n",
652 /* check provider's send/recv wr limits */
653 if (cdata->max_requests > devattr.max_qp_wr)
654 cdata->max_requests = devattr.max_qp_wr;
656 ep->rep_attr.event_handler = rpcrdma_qp_async_error_upcall;
657 ep->rep_attr.qp_context = ep;
658 /* send_cq and recv_cq initialized below */
659 ep->rep_attr.srq = NULL;
660 ep->rep_attr.cap.max_send_wr = cdata->max_requests;
661 switch (ia->ri_memreg_strategy) {
663 /* Add room for frmr register and invalidate WRs.
664 * 1. FRMR reg WR for head
665 * 2. FRMR invalidate WR for head
666 * 3. FRMR reg WR for pagelist
667 * 4. FRMR invalidate WR for pagelist
668 * 5. FRMR reg WR for tail
669 * 6. FRMR invalidate WR for tail
670 * 7. The RDMA_SEND WR
672 ep->rep_attr.cap.max_send_wr *= 7;
673 if (ep->rep_attr.cap.max_send_wr > devattr.max_qp_wr) {
674 cdata->max_requests = devattr.max_qp_wr / 7;
675 if (!cdata->max_requests)
677 ep->rep_attr.cap.max_send_wr = cdata->max_requests * 7;
680 case RPCRDMA_MEMWINDOWS_ASYNC:
681 case RPCRDMA_MEMWINDOWS:
682 /* Add room for mw_binds+unbinds - overkill! */
683 ep->rep_attr.cap.max_send_wr++;
684 ep->rep_attr.cap.max_send_wr *= (2 * RPCRDMA_MAX_SEGS);
685 if (ep->rep_attr.cap.max_send_wr > devattr.max_qp_wr)
691 ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
692 ep->rep_attr.cap.max_send_sge = (cdata->padding ? 4 : 2);
693 ep->rep_attr.cap.max_recv_sge = 1;
694 ep->rep_attr.cap.max_inline_data = 0;
695 ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
696 ep->rep_attr.qp_type = IB_QPT_RC;
697 ep->rep_attr.port_num = ~0;
699 dprintk("RPC: %s: requested max: dtos: send %d recv %d; "
700 "iovs: send %d recv %d\n",
702 ep->rep_attr.cap.max_send_wr,
703 ep->rep_attr.cap.max_recv_wr,
704 ep->rep_attr.cap.max_send_sge,
705 ep->rep_attr.cap.max_recv_sge);
707 /* set trigger for requesting send completion */
708 ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 /* - 1*/;
709 switch (ia->ri_memreg_strategy) {
710 case RPCRDMA_MEMWINDOWS_ASYNC:
711 case RPCRDMA_MEMWINDOWS:
712 ep->rep_cqinit -= RPCRDMA_MAX_SEGS;
717 if (ep->rep_cqinit <= 2)
721 init_waitqueue_head(&ep->rep_connect_wait);
724 * Create a single cq for receive dto and mw_bind (only ever
725 * care about unbind, really). Send completions are suppressed.
726 * Use single threaded tasklet upcalls to maintain ordering.
728 ep->rep_cq = ib_create_cq(ia->ri_id->device, rpcrdma_cq_event_upcall,
729 rpcrdma_cq_async_error_upcall, NULL,
730 ep->rep_attr.cap.max_recv_wr +
731 ep->rep_attr.cap.max_send_wr + 1, 0);
732 if (IS_ERR(ep->rep_cq)) {
733 rc = PTR_ERR(ep->rep_cq);
734 dprintk("RPC: %s: ib_create_cq failed: %i\n",
739 rc = ib_req_notify_cq(ep->rep_cq, IB_CQ_NEXT_COMP);
741 dprintk("RPC: %s: ib_req_notify_cq failed: %i\n",
746 ep->rep_attr.send_cq = ep->rep_cq;
747 ep->rep_attr.recv_cq = ep->rep_cq;
749 /* Initialize cma parameters */
751 /* RPC/RDMA does not use private data */
752 ep->rep_remote_cma.private_data = NULL;
753 ep->rep_remote_cma.private_data_len = 0;
755 /* Client offers RDMA Read but does not initiate */
756 ep->rep_remote_cma.initiator_depth = 0;
757 if (ia->ri_memreg_strategy == RPCRDMA_BOUNCEBUFFERS)
758 ep->rep_remote_cma.responder_resources = 0;
759 else if (devattr.max_qp_rd_atom > 32) /* arbitrary but <= 255 */
760 ep->rep_remote_cma.responder_resources = 32;
762 ep->rep_remote_cma.responder_resources = devattr.max_qp_rd_atom;
764 ep->rep_remote_cma.retry_count = 7;
765 ep->rep_remote_cma.flow_control = 0;
766 ep->rep_remote_cma.rnr_retry_count = 0;
771 err = ib_destroy_cq(ep->rep_cq);
773 dprintk("RPC: %s: ib_destroy_cq returned %i\n",
782 * Disconnect and destroy endpoint. After this, the only
783 * valid operations on the ep are to free it (if dynamically
784 * allocated) or re-create it.
786 * The caller's error handling must be sure to not leak the endpoint
787 * if this function fails.
790 rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
794 dprintk("RPC: %s: entering, connected is %d\n",
795 __func__, ep->rep_connected);
798 rc = rpcrdma_ep_disconnect(ep, ia);
800 dprintk("RPC: %s: rpcrdma_ep_disconnect"
801 " returned %i\n", __func__, rc);
802 rdma_destroy_qp(ia->ri_id);
803 ia->ri_id->qp = NULL;
806 /* padding - could be done in rpcrdma_buffer_destroy... */
807 if (ep->rep_pad_mr) {
808 rpcrdma_deregister_internal(ia, ep->rep_pad_mr, &ep->rep_pad);
809 ep->rep_pad_mr = NULL;
812 rpcrdma_clean_cq(ep->rep_cq);
813 rc = ib_destroy_cq(ep->rep_cq);
815 dprintk("RPC: %s: ib_destroy_cq returned %i\n",
822 * Connect unconnected endpoint.
825 rpcrdma_ep_connect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
827 struct rdma_cm_id *id;
831 if (ep->rep_connected != 0) {
832 struct rpcrdma_xprt *xprt;
834 rc = rpcrdma_ep_disconnect(ep, ia);
835 if (rc && rc != -ENOTCONN)
836 dprintk("RPC: %s: rpcrdma_ep_disconnect"
837 " status %i\n", __func__, rc);
838 rpcrdma_clean_cq(ep->rep_cq);
840 xprt = container_of(ia, struct rpcrdma_xprt, rx_ia);
841 id = rpcrdma_create_id(xprt, ia,
842 (struct sockaddr *)&xprt->rx_data.addr);
847 /* TEMP TEMP TEMP - fail if new device:
848 * Deregister/remarshal *all* requests!
849 * Close and recreate adapter, pd, etc!
850 * Re-determine all attributes still sane!
851 * More stuff I haven't thought of!
854 if (ia->ri_id->device != id->device) {
855 printk("RPC: %s: can't reconnect on "
856 "different device!\n", __func__);
862 rdma_destroy_qp(ia->ri_id);
863 rdma_destroy_id(ia->ri_id);
867 rc = rdma_create_qp(ia->ri_id, ia->ri_pd, &ep->rep_attr);
869 dprintk("RPC: %s: rdma_create_qp failed %i\n",
874 /* XXX Tavor device performs badly with 2K MTU! */
875 if (strnicmp(ia->ri_id->device->dma_device->bus->name, "pci", 3) == 0) {
876 struct pci_dev *pcid = to_pci_dev(ia->ri_id->device->dma_device);
877 if (pcid->device == PCI_DEVICE_ID_MELLANOX_TAVOR &&
878 (pcid->vendor == PCI_VENDOR_ID_MELLANOX ||
879 pcid->vendor == PCI_VENDOR_ID_TOPSPIN)) {
880 struct ib_qp_attr attr = {
881 .path_mtu = IB_MTU_1024
883 rc = ib_modify_qp(ia->ri_id->qp, &attr, IB_QP_PATH_MTU);
887 ep->rep_connected = 0;
889 rc = rdma_connect(ia->ri_id, &ep->rep_remote_cma);
891 dprintk("RPC: %s: rdma_connect() failed with %i\n",
896 wait_event_interruptible(ep->rep_connect_wait, ep->rep_connected != 0);
899 * Check state. A non-peer reject indicates no listener
900 * (ECONNREFUSED), which may be a transient state. All
901 * others indicate a transport condition which has already
902 * undergone a best-effort.
904 if (ep->rep_connected == -ECONNREFUSED &&
905 ++retry_count <= RDMA_CONNECT_RETRY_MAX) {
906 dprintk("RPC: %s: non-peer_reject, retry\n", __func__);
909 if (ep->rep_connected <= 0) {
910 /* Sometimes, the only way to reliably connect to remote
911 * CMs is to use same nonzero values for ORD and IRD. */
912 if (retry_count++ <= RDMA_CONNECT_RETRY_MAX + 1 &&
913 (ep->rep_remote_cma.responder_resources == 0 ||
914 ep->rep_remote_cma.initiator_depth !=
915 ep->rep_remote_cma.responder_resources)) {
916 if (ep->rep_remote_cma.responder_resources == 0)
917 ep->rep_remote_cma.responder_resources = 1;
918 ep->rep_remote_cma.initiator_depth =
919 ep->rep_remote_cma.responder_resources;
922 rc = ep->rep_connected;
924 dprintk("RPC: %s: connected\n", __func__);
929 ep->rep_connected = rc;
934 * rpcrdma_ep_disconnect
936 * This is separate from destroy to facilitate the ability
937 * to reconnect without recreating the endpoint.
939 * This call is not reentrant, and must not be made in parallel
940 * on the same endpoint.
943 rpcrdma_ep_disconnect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
947 rpcrdma_clean_cq(ep->rep_cq);
948 rc = rdma_disconnect(ia->ri_id);
950 /* returns without wait if not connected */
951 wait_event_interruptible(ep->rep_connect_wait,
952 ep->rep_connected != 1);
953 dprintk("RPC: %s: after wait, %sconnected\n", __func__,
954 (ep->rep_connected == 1) ? "still " : "dis");
956 dprintk("RPC: %s: rdma_disconnect %i\n", __func__, rc);
957 ep->rep_connected = rc;
963 * Initialize buffer memory
966 rpcrdma_buffer_create(struct rpcrdma_buffer *buf, struct rpcrdma_ep *ep,
967 struct rpcrdma_ia *ia, struct rpcrdma_create_data_internal *cdata)
972 struct rpcrdma_mw *r;
974 buf->rb_max_requests = cdata->max_requests;
975 spin_lock_init(&buf->rb_lock);
976 atomic_set(&buf->rb_credits, 1);
979 * 1. arrays for send and recv pointers
980 * 2. arrays of struct rpcrdma_req to fill in pointers
981 * 3. array of struct rpcrdma_rep for replies
983 * 5. mw's, fmr's or frmr's, if any
984 * Send/recv buffers in req/rep need to be registered
987 len = buf->rb_max_requests *
988 (sizeof(struct rpcrdma_req *) + sizeof(struct rpcrdma_rep *));
989 len += cdata->padding;
990 switch (ia->ri_memreg_strategy) {
992 len += buf->rb_max_requests * RPCRDMA_MAX_SEGS *
993 sizeof(struct rpcrdma_mw);
995 case RPCRDMA_MTHCAFMR:
996 /* TBD we are perhaps overallocating here */
997 len += (buf->rb_max_requests + 1) * RPCRDMA_MAX_SEGS *
998 sizeof(struct rpcrdma_mw);
1000 case RPCRDMA_MEMWINDOWS_ASYNC:
1001 case RPCRDMA_MEMWINDOWS:
1002 len += (buf->rb_max_requests + 1) * RPCRDMA_MAX_SEGS *
1003 sizeof(struct rpcrdma_mw);
1009 /* allocate 1, 4 and 5 in one shot */
1010 p = kzalloc(len, GFP_KERNEL);
1012 dprintk("RPC: %s: req_t/rep_t/pad kzalloc(%zd) failed\n",
1017 buf->rb_pool = p; /* for freeing it later */
1019 buf->rb_send_bufs = (struct rpcrdma_req **) p;
1020 p = (char *) &buf->rb_send_bufs[buf->rb_max_requests];
1021 buf->rb_recv_bufs = (struct rpcrdma_rep **) p;
1022 p = (char *) &buf->rb_recv_bufs[buf->rb_max_requests];
1025 * Register the zeroed pad buffer, if any.
1027 if (cdata->padding) {
1028 rc = rpcrdma_register_internal(ia, p, cdata->padding,
1029 &ep->rep_pad_mr, &ep->rep_pad);
1033 p += cdata->padding;
1036 * Allocate the fmr's, or mw's for mw_bind chunk registration.
1037 * We "cycle" the mw's in order to minimize rkey reuse,
1038 * and also reduce unbind-to-bind collision.
1040 INIT_LIST_HEAD(&buf->rb_mws);
1041 r = (struct rpcrdma_mw *)p;
1042 switch (ia->ri_memreg_strategy) {
1044 for (i = buf->rb_max_requests * RPCRDMA_MAX_SEGS; i; i--) {
1045 r->r.frmr.fr_mr = ib_alloc_fast_reg_mr(ia->ri_pd,
1047 if (IS_ERR(r->r.frmr.fr_mr)) {
1048 rc = PTR_ERR(r->r.frmr.fr_mr);
1049 dprintk("RPC: %s: ib_alloc_fast_reg_mr"
1050 " failed %i\n", __func__, rc);
1054 ib_alloc_fast_reg_page_list(ia->ri_id->device,
1056 if (IS_ERR(r->r.frmr.fr_pgl)) {
1057 rc = PTR_ERR(r->r.frmr.fr_pgl);
1059 "ib_alloc_fast_reg_page_list "
1060 "failed %i\n", __func__, rc);
1063 list_add(&r->mw_list, &buf->rb_mws);
1067 case RPCRDMA_MTHCAFMR:
1068 /* TBD we are perhaps overallocating here */
1069 for (i = (buf->rb_max_requests+1) * RPCRDMA_MAX_SEGS; i; i--) {
1070 static struct ib_fmr_attr fa =
1071 { RPCRDMA_MAX_DATA_SEGS, 1, PAGE_SHIFT };
1072 r->r.fmr = ib_alloc_fmr(ia->ri_pd,
1073 IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ,
1075 if (IS_ERR(r->r.fmr)) {
1076 rc = PTR_ERR(r->r.fmr);
1077 dprintk("RPC: %s: ib_alloc_fmr"
1078 " failed %i\n", __func__, rc);
1081 list_add(&r->mw_list, &buf->rb_mws);
1085 case RPCRDMA_MEMWINDOWS_ASYNC:
1086 case RPCRDMA_MEMWINDOWS:
1087 /* Allocate one extra request's worth, for full cycling */
1088 for (i = (buf->rb_max_requests+1) * RPCRDMA_MAX_SEGS; i; i--) {
1089 r->r.mw = ib_alloc_mw(ia->ri_pd, IB_MW_TYPE_1);
1090 if (IS_ERR(r->r.mw)) {
1091 rc = PTR_ERR(r->r.mw);
1092 dprintk("RPC: %s: ib_alloc_mw"
1093 " failed %i\n", __func__, rc);
1096 list_add(&r->mw_list, &buf->rb_mws);
1105 * Allocate/init the request/reply buffers. Doing this
1106 * using kmalloc for now -- one for each buf.
1108 for (i = 0; i < buf->rb_max_requests; i++) {
1109 struct rpcrdma_req *req;
1110 struct rpcrdma_rep *rep;
1112 len = cdata->inline_wsize + sizeof(struct rpcrdma_req);
1113 /* RPC layer requests *double* size + 1K RPC_SLACK_SPACE! */
1114 /* Typical ~2400b, so rounding up saves work later */
1117 req = kmalloc(len, GFP_KERNEL);
1119 dprintk("RPC: %s: request buffer %d alloc"
1120 " failed\n", __func__, i);
1124 memset(req, 0, sizeof(struct rpcrdma_req));
1125 buf->rb_send_bufs[i] = req;
1126 buf->rb_send_bufs[i]->rl_buffer = buf;
1128 rc = rpcrdma_register_internal(ia, req->rl_base,
1129 len - offsetof(struct rpcrdma_req, rl_base),
1130 &buf->rb_send_bufs[i]->rl_handle,
1131 &buf->rb_send_bufs[i]->rl_iov);
1135 buf->rb_send_bufs[i]->rl_size = len-sizeof(struct rpcrdma_req);
1137 len = cdata->inline_rsize + sizeof(struct rpcrdma_rep);
1138 rep = kmalloc(len, GFP_KERNEL);
1140 dprintk("RPC: %s: reply buffer %d alloc failed\n",
1145 memset(rep, 0, sizeof(struct rpcrdma_rep));
1146 buf->rb_recv_bufs[i] = rep;
1147 buf->rb_recv_bufs[i]->rr_buffer = buf;
1148 init_waitqueue_head(&rep->rr_unbind);
1150 rc = rpcrdma_register_internal(ia, rep->rr_base,
1151 len - offsetof(struct rpcrdma_rep, rr_base),
1152 &buf->rb_recv_bufs[i]->rr_handle,
1153 &buf->rb_recv_bufs[i]->rr_iov);
1158 dprintk("RPC: %s: max_requests %d\n",
1159 __func__, buf->rb_max_requests);
1163 rpcrdma_buffer_destroy(buf);
1168 * Unregister and destroy buffer memory. Need to deal with
1169 * partial initialization, so it's callable from failed create.
1170 * Must be called before destroying endpoint, as registrations
1174 rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf)
1177 struct rpcrdma_ia *ia = rdmab_to_ia(buf);
1178 struct rpcrdma_mw *r;
1180 /* clean up in reverse order from create
1181 * 1. recv mr memory (mr free, then kfree)
1182 * 1a. bind mw memory
1183 * 2. send mr memory (mr free, then kfree)
1184 * 3. padding (if any) [moved to rpcrdma_ep_destroy]
1187 dprintk("RPC: %s: entering\n", __func__);
1189 for (i = 0; i < buf->rb_max_requests; i++) {
1190 if (buf->rb_recv_bufs && buf->rb_recv_bufs[i]) {
1191 rpcrdma_deregister_internal(ia,
1192 buf->rb_recv_bufs[i]->rr_handle,
1193 &buf->rb_recv_bufs[i]->rr_iov);
1194 kfree(buf->rb_recv_bufs[i]);
1196 if (buf->rb_send_bufs && buf->rb_send_bufs[i]) {
1197 while (!list_empty(&buf->rb_mws)) {
1198 r = list_entry(buf->rb_mws.next,
1199 struct rpcrdma_mw, mw_list);
1200 list_del(&r->mw_list);
1201 switch (ia->ri_memreg_strategy) {
1203 rc = ib_dereg_mr(r->r.frmr.fr_mr);
1209 ib_free_fast_reg_page_list(r->r.frmr.fr_pgl);
1211 case RPCRDMA_MTHCAFMR:
1212 rc = ib_dealloc_fmr(r->r.fmr);
1219 case RPCRDMA_MEMWINDOWS_ASYNC:
1220 case RPCRDMA_MEMWINDOWS:
1221 rc = ib_dealloc_mw(r->r.mw);
1232 rpcrdma_deregister_internal(ia,
1233 buf->rb_send_bufs[i]->rl_handle,
1234 &buf->rb_send_bufs[i]->rl_iov);
1235 kfree(buf->rb_send_bufs[i]);
1239 kfree(buf->rb_pool);
1243 * Get a set of request/reply buffers.
1245 * Reply buffer (if needed) is attached to send buffer upon return.
1247 * rb_send_index and rb_recv_index MUST always be pointing to the
1248 * *next* available buffer (non-NULL). They are incremented after
1249 * removing buffers, and decremented *before* returning them.
1251 struct rpcrdma_req *
1252 rpcrdma_buffer_get(struct rpcrdma_buffer *buffers)
1254 struct rpcrdma_req *req;
1255 unsigned long flags;
1257 struct rpcrdma_mw *r;
1259 spin_lock_irqsave(&buffers->rb_lock, flags);
1260 if (buffers->rb_send_index == buffers->rb_max_requests) {
1261 spin_unlock_irqrestore(&buffers->rb_lock, flags);
1262 dprintk("RPC: %s: out of request buffers\n", __func__);
1263 return ((struct rpcrdma_req *)NULL);
1266 req = buffers->rb_send_bufs[buffers->rb_send_index];
1267 if (buffers->rb_send_index < buffers->rb_recv_index) {
1268 dprintk("RPC: %s: %d extra receives outstanding (ok)\n",
1270 buffers->rb_recv_index - buffers->rb_send_index);
1271 req->rl_reply = NULL;
1273 req->rl_reply = buffers->rb_recv_bufs[buffers->rb_recv_index];
1274 buffers->rb_recv_bufs[buffers->rb_recv_index++] = NULL;
1276 buffers->rb_send_bufs[buffers->rb_send_index++] = NULL;
1277 if (!list_empty(&buffers->rb_mws)) {
1278 i = RPCRDMA_MAX_SEGS - 1;
1280 r = list_entry(buffers->rb_mws.next,
1281 struct rpcrdma_mw, mw_list);
1282 list_del(&r->mw_list);
1283 req->rl_segments[i].mr_chunk.rl_mw = r;
1286 spin_unlock_irqrestore(&buffers->rb_lock, flags);
1291 * Put request/reply buffers back into pool.
1292 * Pre-decrement counter/array index.
1295 rpcrdma_buffer_put(struct rpcrdma_req *req)
1297 struct rpcrdma_buffer *buffers = req->rl_buffer;
1298 struct rpcrdma_ia *ia = rdmab_to_ia(buffers);
1300 unsigned long flags;
1302 BUG_ON(req->rl_nchunks != 0);
1303 spin_lock_irqsave(&buffers->rb_lock, flags);
1304 buffers->rb_send_bufs[--buffers->rb_send_index] = req;
1306 if (req->rl_reply) {
1307 buffers->rb_recv_bufs[--buffers->rb_recv_index] = req->rl_reply;
1308 init_waitqueue_head(&req->rl_reply->rr_unbind);
1309 req->rl_reply->rr_func = NULL;
1310 req->rl_reply = NULL;
1312 switch (ia->ri_memreg_strategy) {
1314 case RPCRDMA_MTHCAFMR:
1315 case RPCRDMA_MEMWINDOWS_ASYNC:
1316 case RPCRDMA_MEMWINDOWS:
1318 * Cycle mw's back in reverse order, and "spin" them.
1319 * This delays and scrambles reuse as much as possible.
1323 struct rpcrdma_mw **mw;
1324 mw = &req->rl_segments[i].mr_chunk.rl_mw;
1325 list_add_tail(&(*mw)->mw_list, &buffers->rb_mws);
1327 } while (++i < RPCRDMA_MAX_SEGS);
1328 list_add_tail(&req->rl_segments[0].mr_chunk.rl_mw->mw_list,
1330 req->rl_segments[0].mr_chunk.rl_mw = NULL;
1335 spin_unlock_irqrestore(&buffers->rb_lock, flags);
1339 * Recover reply buffers from pool.
1340 * This happens when recovering from error conditions.
1341 * Post-increment counter/array index.
1344 rpcrdma_recv_buffer_get(struct rpcrdma_req *req)
1346 struct rpcrdma_buffer *buffers = req->rl_buffer;
1347 unsigned long flags;
1349 if (req->rl_iov.length == 0) /* special case xprt_rdma_allocate() */
1350 buffers = ((struct rpcrdma_req *) buffers)->rl_buffer;
1351 spin_lock_irqsave(&buffers->rb_lock, flags);
1352 if (buffers->rb_recv_index < buffers->rb_max_requests) {
1353 req->rl_reply = buffers->rb_recv_bufs[buffers->rb_recv_index];
1354 buffers->rb_recv_bufs[buffers->rb_recv_index++] = NULL;
1356 spin_unlock_irqrestore(&buffers->rb_lock, flags);
1360 * Put reply buffers back into pool when not attached to
1361 * request. This happens in error conditions, and when
1362 * aborting unbinds. Pre-decrement counter/array index.
1365 rpcrdma_recv_buffer_put(struct rpcrdma_rep *rep)
1367 struct rpcrdma_buffer *buffers = rep->rr_buffer;
1368 unsigned long flags;
1370 rep->rr_func = NULL;
1371 spin_lock_irqsave(&buffers->rb_lock, flags);
1372 buffers->rb_recv_bufs[--buffers->rb_recv_index] = rep;
1373 spin_unlock_irqrestore(&buffers->rb_lock, flags);
1377 * Wrappers for internal-use kmalloc memory registration, used by buffer code.
1381 rpcrdma_register_internal(struct rpcrdma_ia *ia, void *va, int len,
1382 struct ib_mr **mrp, struct ib_sge *iov)
1384 struct ib_phys_buf ipb;
1389 * All memory passed here was kmalloc'ed, therefore phys-contiguous.
1391 iov->addr = ib_dma_map_single(ia->ri_id->device,
1392 va, len, DMA_BIDIRECTIONAL);
1395 if (ia->ri_have_dma_lkey) {
1397 iov->lkey = ia->ri_dma_lkey;
1399 } else if (ia->ri_bind_mem != NULL) {
1401 iov->lkey = ia->ri_bind_mem->lkey;
1405 ipb.addr = iov->addr;
1406 ipb.size = iov->length;
1407 mr = ib_reg_phys_mr(ia->ri_pd, &ipb, 1,
1408 IB_ACCESS_LOCAL_WRITE, &iov->addr);
1410 dprintk("RPC: %s: phys convert: 0x%llx "
1411 "registered 0x%llx length %d\n",
1412 __func__, (unsigned long long)ipb.addr,
1413 (unsigned long long)iov->addr, len);
1418 dprintk("RPC: %s: failed with %i\n", __func__, rc);
1421 iov->lkey = mr->lkey;
1429 rpcrdma_deregister_internal(struct rpcrdma_ia *ia,
1430 struct ib_mr *mr, struct ib_sge *iov)
1434 ib_dma_unmap_single(ia->ri_id->device,
1435 iov->addr, iov->length, DMA_BIDIRECTIONAL);
1440 rc = ib_dereg_mr(mr);
1442 dprintk("RPC: %s: ib_dereg_mr failed %i\n", __func__, rc);
1447 * Wrappers for chunk registration, shared by read/write chunk code.
1451 rpcrdma_map_one(struct rpcrdma_ia *ia, struct rpcrdma_mr_seg *seg, int writing)
1453 seg->mr_dir = writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
1454 seg->mr_dmalen = seg->mr_len;
1456 seg->mr_dma = ib_dma_map_page(ia->ri_id->device,
1457 seg->mr_page, offset_in_page(seg->mr_offset),
1458 seg->mr_dmalen, seg->mr_dir);
1460 seg->mr_dma = ib_dma_map_single(ia->ri_id->device,
1462 seg->mr_dmalen, seg->mr_dir);
1463 if (ib_dma_mapping_error(ia->ri_id->device, seg->mr_dma)) {
1464 dprintk("RPC: %s: mr_dma %llx mr_offset %p mr_dma_len %zu\n",
1466 (unsigned long long)seg->mr_dma,
1467 seg->mr_offset, seg->mr_dmalen);
1472 rpcrdma_unmap_one(struct rpcrdma_ia *ia, struct rpcrdma_mr_seg *seg)
1475 ib_dma_unmap_page(ia->ri_id->device,
1476 seg->mr_dma, seg->mr_dmalen, seg->mr_dir);
1478 ib_dma_unmap_single(ia->ri_id->device,
1479 seg->mr_dma, seg->mr_dmalen, seg->mr_dir);
1483 rpcrdma_register_frmr_external(struct rpcrdma_mr_seg *seg,
1484 int *nsegs, int writing, struct rpcrdma_ia *ia,
1485 struct rpcrdma_xprt *r_xprt)
1487 struct rpcrdma_mr_seg *seg1 = seg;
1488 struct ib_send_wr invalidate_wr, frmr_wr, *bad_wr, *post_wr;
1497 pageoff = offset_in_page(seg1->mr_offset);
1498 seg1->mr_offset -= pageoff; /* start of page */
1499 seg1->mr_len += pageoff;
1501 if (*nsegs > RPCRDMA_MAX_DATA_SEGS)
1502 *nsegs = RPCRDMA_MAX_DATA_SEGS;
1503 for (page_no = i = 0; i < *nsegs;) {
1504 rpcrdma_map_one(ia, seg, writing);
1506 for (seg_len = seg->mr_len; seg_len > 0; seg_len -= PAGE_SIZE) {
1507 seg1->mr_chunk.rl_mw->r.frmr.fr_pgl->
1508 page_list[page_no++] = pa;
1514 /* Check for holes */
1515 if ((i < *nsegs && offset_in_page(seg->mr_offset)) ||
1516 offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
1519 dprintk("RPC: %s: Using frmr %p to map %d segments\n",
1520 __func__, seg1->mr_chunk.rl_mw, i);
1522 if (unlikely(seg1->mr_chunk.rl_mw->r.frmr.state == FRMR_IS_VALID)) {
1523 dprintk("RPC: %s: frmr %x left valid, posting invalidate.\n",
1525 seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey);
1526 /* Invalidate before using. */
1527 memset(&invalidate_wr, 0, sizeof invalidate_wr);
1528 invalidate_wr.wr_id = (unsigned long)(void *)seg1->mr_chunk.rl_mw;
1529 invalidate_wr.next = &frmr_wr;
1530 invalidate_wr.opcode = IB_WR_LOCAL_INV;
1531 invalidate_wr.send_flags = IB_SEND_SIGNALED;
1532 invalidate_wr.ex.invalidate_rkey =
1533 seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
1534 DECR_CQCOUNT(&r_xprt->rx_ep);
1535 post_wr = &invalidate_wr;
1540 key = (u8)(seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey & 0x000000FF);
1541 ib_update_fast_reg_key(seg1->mr_chunk.rl_mw->r.frmr.fr_mr, ++key);
1543 /* Prepare FRMR WR */
1544 memset(&frmr_wr, 0, sizeof frmr_wr);
1545 frmr_wr.wr_id = (unsigned long)(void *)seg1->mr_chunk.rl_mw;
1546 frmr_wr.opcode = IB_WR_FAST_REG_MR;
1547 frmr_wr.send_flags = IB_SEND_SIGNALED;
1548 frmr_wr.wr.fast_reg.iova_start = seg1->mr_dma;
1549 frmr_wr.wr.fast_reg.page_list = seg1->mr_chunk.rl_mw->r.frmr.fr_pgl;
1550 frmr_wr.wr.fast_reg.page_list_len = page_no;
1551 frmr_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
1552 frmr_wr.wr.fast_reg.length = page_no << PAGE_SHIFT;
1553 BUG_ON(frmr_wr.wr.fast_reg.length < len);
1554 frmr_wr.wr.fast_reg.access_flags = (writing ?
1555 IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
1556 IB_ACCESS_REMOTE_READ);
1557 frmr_wr.wr.fast_reg.rkey = seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
1558 DECR_CQCOUNT(&r_xprt->rx_ep);
1560 rc = ib_post_send(ia->ri_id->qp, post_wr, &bad_wr);
1563 dprintk("RPC: %s: failed ib_post_send for register,"
1564 " status %i\n", __func__, rc);
1566 rpcrdma_unmap_one(ia, --seg);
1568 seg1->mr_rkey = seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
1569 seg1->mr_base = seg1->mr_dma + pageoff;
1578 rpcrdma_deregister_frmr_external(struct rpcrdma_mr_seg *seg,
1579 struct rpcrdma_ia *ia, struct rpcrdma_xprt *r_xprt)
1581 struct rpcrdma_mr_seg *seg1 = seg;
1582 struct ib_send_wr invalidate_wr, *bad_wr;
1585 while (seg1->mr_nsegs--)
1586 rpcrdma_unmap_one(ia, seg++);
1588 memset(&invalidate_wr, 0, sizeof invalidate_wr);
1589 invalidate_wr.wr_id = (unsigned long)(void *)seg1->mr_chunk.rl_mw;
1590 invalidate_wr.opcode = IB_WR_LOCAL_INV;
1591 invalidate_wr.send_flags = IB_SEND_SIGNALED;
1592 invalidate_wr.ex.invalidate_rkey = seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
1593 DECR_CQCOUNT(&r_xprt->rx_ep);
1595 rc = ib_post_send(ia->ri_id->qp, &invalidate_wr, &bad_wr);
1597 dprintk("RPC: %s: failed ib_post_send for invalidate,"
1598 " status %i\n", __func__, rc);
1603 rpcrdma_register_fmr_external(struct rpcrdma_mr_seg *seg,
1604 int *nsegs, int writing, struct rpcrdma_ia *ia)
1606 struct rpcrdma_mr_seg *seg1 = seg;
1607 u64 physaddrs[RPCRDMA_MAX_DATA_SEGS];
1608 int len, pageoff, i, rc;
1610 pageoff = offset_in_page(seg1->mr_offset);
1611 seg1->mr_offset -= pageoff; /* start of page */
1612 seg1->mr_len += pageoff;
1614 if (*nsegs > RPCRDMA_MAX_DATA_SEGS)
1615 *nsegs = RPCRDMA_MAX_DATA_SEGS;
1616 for (i = 0; i < *nsegs;) {
1617 rpcrdma_map_one(ia, seg, writing);
1618 physaddrs[i] = seg->mr_dma;
1622 /* Check for holes */
1623 if ((i < *nsegs && offset_in_page(seg->mr_offset)) ||
1624 offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
1627 rc = ib_map_phys_fmr(seg1->mr_chunk.rl_mw->r.fmr,
1628 physaddrs, i, seg1->mr_dma);
1630 dprintk("RPC: %s: failed ib_map_phys_fmr "
1631 "%u@0x%llx+%i (%d)... status %i\n", __func__,
1632 len, (unsigned long long)seg1->mr_dma,
1635 rpcrdma_unmap_one(ia, --seg);
1637 seg1->mr_rkey = seg1->mr_chunk.rl_mw->r.fmr->rkey;
1638 seg1->mr_base = seg1->mr_dma + pageoff;
1647 rpcrdma_deregister_fmr_external(struct rpcrdma_mr_seg *seg,
1648 struct rpcrdma_ia *ia)
1650 struct rpcrdma_mr_seg *seg1 = seg;
1654 list_add(&seg1->mr_chunk.rl_mw->r.fmr->list, &l);
1655 rc = ib_unmap_fmr(&l);
1656 while (seg1->mr_nsegs--)
1657 rpcrdma_unmap_one(ia, seg++);
1659 dprintk("RPC: %s: failed ib_unmap_fmr,"
1660 " status %i\n", __func__, rc);
1665 rpcrdma_register_memwin_external(struct rpcrdma_mr_seg *seg,
1666 int *nsegs, int writing, struct rpcrdma_ia *ia,
1667 struct rpcrdma_xprt *r_xprt)
1669 int mem_priv = (writing ? IB_ACCESS_REMOTE_WRITE :
1670 IB_ACCESS_REMOTE_READ);
1671 struct ib_mw_bind param;
1675 rpcrdma_map_one(ia, seg, writing);
1676 param.bind_info.mr = ia->ri_bind_mem;
1677 param.wr_id = 0ULL; /* no send cookie */
1678 param.bind_info.addr = seg->mr_dma;
1679 param.bind_info.length = seg->mr_len;
1680 param.send_flags = 0;
1681 param.bind_info.mw_access_flags = mem_priv;
1683 DECR_CQCOUNT(&r_xprt->rx_ep);
1684 rc = ib_bind_mw(ia->ri_id->qp, seg->mr_chunk.rl_mw->r.mw, ¶m);
1686 dprintk("RPC: %s: failed ib_bind_mw "
1687 "%u@0x%llx status %i\n",
1688 __func__, seg->mr_len,
1689 (unsigned long long)seg->mr_dma, rc);
1690 rpcrdma_unmap_one(ia, seg);
1692 seg->mr_rkey = seg->mr_chunk.rl_mw->r.mw->rkey;
1693 seg->mr_base = param.bind_info.addr;
1700 rpcrdma_deregister_memwin_external(struct rpcrdma_mr_seg *seg,
1701 struct rpcrdma_ia *ia,
1702 struct rpcrdma_xprt *r_xprt, void **r)
1704 struct ib_mw_bind param;
1708 BUG_ON(seg->mr_nsegs != 1);
1709 param.bind_info.mr = ia->ri_bind_mem;
1710 param.bind_info.addr = 0ULL; /* unbind */
1711 param.bind_info.length = 0;
1712 param.bind_info.mw_access_flags = 0;
1714 param.wr_id = (u64) (unsigned long) *r;
1715 param.send_flags = IB_SEND_SIGNALED;
1716 INIT_CQCOUNT(&r_xprt->rx_ep);
1719 param.send_flags = 0;
1720 DECR_CQCOUNT(&r_xprt->rx_ep);
1722 rc = ib_bind_mw(ia->ri_id->qp, seg->mr_chunk.rl_mw->r.mw, ¶m);
1723 rpcrdma_unmap_one(ia, seg);
1725 dprintk("RPC: %s: failed ib_(un)bind_mw,"
1726 " status %i\n", __func__, rc);
1728 *r = NULL; /* will upcall on completion */
1733 rpcrdma_register_default_external(struct rpcrdma_mr_seg *seg,
1734 int *nsegs, int writing, struct rpcrdma_ia *ia)
1736 int mem_priv = (writing ? IB_ACCESS_REMOTE_WRITE :
1737 IB_ACCESS_REMOTE_READ);
1738 struct rpcrdma_mr_seg *seg1 = seg;
1739 struct ib_phys_buf ipb[RPCRDMA_MAX_DATA_SEGS];
1742 if (*nsegs > RPCRDMA_MAX_DATA_SEGS)
1743 *nsegs = RPCRDMA_MAX_DATA_SEGS;
1744 for (len = 0, i = 0; i < *nsegs;) {
1745 rpcrdma_map_one(ia, seg, writing);
1746 ipb[i].addr = seg->mr_dma;
1747 ipb[i].size = seg->mr_len;
1751 /* Check for holes */
1752 if ((i < *nsegs && offset_in_page(seg->mr_offset)) ||
1753 offset_in_page((seg-1)->mr_offset+(seg-1)->mr_len))
1756 seg1->mr_base = seg1->mr_dma;
1757 seg1->mr_chunk.rl_mr = ib_reg_phys_mr(ia->ri_pd,
1758 ipb, i, mem_priv, &seg1->mr_base);
1759 if (IS_ERR(seg1->mr_chunk.rl_mr)) {
1760 rc = PTR_ERR(seg1->mr_chunk.rl_mr);
1761 dprintk("RPC: %s: failed ib_reg_phys_mr "
1762 "%u@0x%llx (%d)... status %i\n",
1764 (unsigned long long)seg1->mr_dma, i, rc);
1766 rpcrdma_unmap_one(ia, --seg);
1768 seg1->mr_rkey = seg1->mr_chunk.rl_mr->rkey;
1777 rpcrdma_deregister_default_external(struct rpcrdma_mr_seg *seg,
1778 struct rpcrdma_ia *ia)
1780 struct rpcrdma_mr_seg *seg1 = seg;
1783 rc = ib_dereg_mr(seg1->mr_chunk.rl_mr);
1784 seg1->mr_chunk.rl_mr = NULL;
1785 while (seg1->mr_nsegs--)
1786 rpcrdma_unmap_one(ia, seg++);
1788 dprintk("RPC: %s: failed ib_dereg_mr,"
1789 " status %i\n", __func__, rc);
1794 rpcrdma_register_external(struct rpcrdma_mr_seg *seg,
1795 int nsegs, int writing, struct rpcrdma_xprt *r_xprt)
1797 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
1800 switch (ia->ri_memreg_strategy) {
1802 #if RPCRDMA_PERSISTENT_REGISTRATION
1803 case RPCRDMA_ALLPHYSICAL:
1804 rpcrdma_map_one(ia, seg, writing);
1805 seg->mr_rkey = ia->ri_bind_mem->rkey;
1806 seg->mr_base = seg->mr_dma;
1812 /* Registration using frmr registration */
1814 rc = rpcrdma_register_frmr_external(seg, &nsegs, writing, ia, r_xprt);
1817 /* Registration using fmr memory registration */
1818 case RPCRDMA_MTHCAFMR:
1819 rc = rpcrdma_register_fmr_external(seg, &nsegs, writing, ia);
1822 /* Registration using memory windows */
1823 case RPCRDMA_MEMWINDOWS_ASYNC:
1824 case RPCRDMA_MEMWINDOWS:
1825 rc = rpcrdma_register_memwin_external(seg, &nsegs, writing, ia, r_xprt);
1828 /* Default registration each time */
1830 rc = rpcrdma_register_default_external(seg, &nsegs, writing, ia);
1840 rpcrdma_deregister_external(struct rpcrdma_mr_seg *seg,
1841 struct rpcrdma_xprt *r_xprt, void *r)
1843 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
1844 int nsegs = seg->mr_nsegs, rc;
1846 switch (ia->ri_memreg_strategy) {
1848 #if RPCRDMA_PERSISTENT_REGISTRATION
1849 case RPCRDMA_ALLPHYSICAL:
1851 rpcrdma_unmap_one(ia, seg);
1857 rc = rpcrdma_deregister_frmr_external(seg, ia, r_xprt);
1860 case RPCRDMA_MTHCAFMR:
1861 rc = rpcrdma_deregister_fmr_external(seg, ia);
1864 case RPCRDMA_MEMWINDOWS_ASYNC:
1865 case RPCRDMA_MEMWINDOWS:
1866 rc = rpcrdma_deregister_memwin_external(seg, ia, r_xprt, &r);
1870 rc = rpcrdma_deregister_default_external(seg, ia);
1874 struct rpcrdma_rep *rep = r;
1875 void (*func)(struct rpcrdma_rep *) = rep->rr_func;
1876 rep->rr_func = NULL;
1877 func(rep); /* dereg done, callback now */
1883 * Prepost any receive buffer, then post send.
1885 * Receive buffer is donated to hardware, reclaimed upon recv completion.
1888 rpcrdma_ep_post(struct rpcrdma_ia *ia,
1889 struct rpcrdma_ep *ep,
1890 struct rpcrdma_req *req)
1892 struct ib_send_wr send_wr, *send_wr_fail;
1893 struct rpcrdma_rep *rep = req->rl_reply;
1897 rc = rpcrdma_ep_post_recv(ia, ep, rep);
1900 req->rl_reply = NULL;
1903 send_wr.next = NULL;
1904 send_wr.wr_id = 0ULL; /* no send cookie */
1905 send_wr.sg_list = req->rl_send_iov;
1906 send_wr.num_sge = req->rl_niovs;
1907 send_wr.opcode = IB_WR_SEND;
1908 if (send_wr.num_sge == 4) /* no need to sync any pad (constant) */
1909 ib_dma_sync_single_for_device(ia->ri_id->device,
1910 req->rl_send_iov[3].addr, req->rl_send_iov[3].length,
1912 ib_dma_sync_single_for_device(ia->ri_id->device,
1913 req->rl_send_iov[1].addr, req->rl_send_iov[1].length,
1915 ib_dma_sync_single_for_device(ia->ri_id->device,
1916 req->rl_send_iov[0].addr, req->rl_send_iov[0].length,
1919 if (DECR_CQCOUNT(ep) > 0)
1920 send_wr.send_flags = 0;
1921 else { /* Provider must take a send completion every now and then */
1923 send_wr.send_flags = IB_SEND_SIGNALED;
1926 rc = ib_post_send(ia->ri_id->qp, &send_wr, &send_wr_fail);
1928 dprintk("RPC: %s: ib_post_send returned %i\n", __func__,
1935 * (Re)post a receive buffer.
1938 rpcrdma_ep_post_recv(struct rpcrdma_ia *ia,
1939 struct rpcrdma_ep *ep,
1940 struct rpcrdma_rep *rep)
1942 struct ib_recv_wr recv_wr, *recv_wr_fail;
1945 recv_wr.next = NULL;
1946 recv_wr.wr_id = (u64) (unsigned long) rep;
1947 recv_wr.sg_list = &rep->rr_iov;
1948 recv_wr.num_sge = 1;
1950 ib_dma_sync_single_for_cpu(ia->ri_id->device,
1951 rep->rr_iov.addr, rep->rr_iov.length, DMA_BIDIRECTIONAL);
1954 rc = ib_post_recv(ia->ri_id->qp, &recv_wr, &recv_wr_fail);
1957 dprintk("RPC: %s: ib_post_recv returned %i\n", __func__,