xref: /openbmc/linux/net/sunrpc/xprtrdma/verbs.c (revision 3e26a691)
1 /*
2  * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
3  *
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
8  * license below:
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  *
14  *      Redistributions of source code must retain the above copyright
15  *      notice, this list of conditions and the following disclaimer.
16  *
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.
21  *
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
25  *      permission.
26  *
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.
38  */
39 
40 /*
41  * verbs.c
42  *
43  * Encapsulates the major functions managing:
44  *  o adapters
45  *  o endpoints
46  *  o connections
47  *  o buffer memory
48  */
49 
50 #include <linux/interrupt.h>
51 #include <linux/slab.h>
52 #include <linux/prefetch.h>
53 #include <linux/sunrpc/addr.h>
54 #include <asm/bitops.h>
55 #include <linux/module.h> /* try_module_get()/module_put() */
56 
57 #include "xprt_rdma.h"
58 
59 /*
60  * Globals/Macros
61  */
62 
63 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
64 # define RPCDBG_FACILITY	RPCDBG_TRANS
65 #endif
66 
67 /*
68  * internal functions
69  */
70 
71 static struct workqueue_struct *rpcrdma_receive_wq;
72 
73 int
74 rpcrdma_alloc_wq(void)
75 {
76 	struct workqueue_struct *recv_wq;
77 
78 	recv_wq = alloc_workqueue("xprtrdma_receive",
79 				  WQ_MEM_RECLAIM | WQ_UNBOUND | WQ_HIGHPRI,
80 				  0);
81 	if (!recv_wq)
82 		return -ENOMEM;
83 
84 	rpcrdma_receive_wq = recv_wq;
85 	return 0;
86 }
87 
88 void
89 rpcrdma_destroy_wq(void)
90 {
91 	struct workqueue_struct *wq;
92 
93 	if (rpcrdma_receive_wq) {
94 		wq = rpcrdma_receive_wq;
95 		rpcrdma_receive_wq = NULL;
96 		destroy_workqueue(wq);
97 	}
98 }
99 
100 static void
101 rpcrdma_qp_async_error_upcall(struct ib_event *event, void *context)
102 {
103 	struct rpcrdma_ep *ep = context;
104 
105 	pr_err("RPC:       %s: %s on device %s ep %p\n",
106 	       __func__, ib_event_msg(event->event),
107 		event->device->name, context);
108 	if (ep->rep_connected == 1) {
109 		ep->rep_connected = -EIO;
110 		rpcrdma_conn_func(ep);
111 		wake_up_all(&ep->rep_connect_wait);
112 	}
113 }
114 
115 /**
116  * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC
117  * @cq:	completion queue (ignored)
118  * @wc:	completed WR
119  *
120  */
121 static void
122 rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
123 {
124 	/* WARNING: Only wr_cqe and status are reliable at this point */
125 	if (wc->status != IB_WC_SUCCESS && wc->status != IB_WC_WR_FLUSH_ERR)
126 		pr_err("rpcrdma: Send: %s (%u/0x%x)\n",
127 		       ib_wc_status_msg(wc->status),
128 		       wc->status, wc->vendor_err);
129 }
130 
131 static void
132 rpcrdma_receive_worker(struct work_struct *work)
133 {
134 	struct rpcrdma_rep *rep =
135 			container_of(work, struct rpcrdma_rep, rr_work);
136 
137 	rpcrdma_reply_handler(rep);
138 }
139 
140 /* Perform basic sanity checking to avoid using garbage
141  * to update the credit grant value.
142  */
143 static void
144 rpcrdma_update_granted_credits(struct rpcrdma_rep *rep)
145 {
146 	struct rpcrdma_msg *rmsgp = rdmab_to_msg(rep->rr_rdmabuf);
147 	struct rpcrdma_buffer *buffer = &rep->rr_rxprt->rx_buf;
148 	u32 credits;
149 
150 	if (rep->rr_len < RPCRDMA_HDRLEN_ERR)
151 		return;
152 
153 	credits = be32_to_cpu(rmsgp->rm_credit);
154 	if (credits == 0)
155 		credits = 1;	/* don't deadlock */
156 	else if (credits > buffer->rb_max_requests)
157 		credits = buffer->rb_max_requests;
158 
159 	atomic_set(&buffer->rb_credits, credits);
160 }
161 
162 /**
163  * rpcrdma_receive_wc - Invoked by RDMA provider for each polled Receive WC
164  * @cq:	completion queue (ignored)
165  * @wc:	completed WR
166  *
167  */
168 static void
169 rpcrdma_receive_wc(struct ib_cq *cq, struct ib_wc *wc)
170 {
171 	struct ib_cqe *cqe = wc->wr_cqe;
172 	struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep,
173 					       rr_cqe);
174 
175 	/* WARNING: Only wr_id and status are reliable at this point */
176 	if (wc->status != IB_WC_SUCCESS)
177 		goto out_fail;
178 
179 	/* status == SUCCESS means all fields in wc are trustworthy */
180 	if (wc->opcode != IB_WC_RECV)
181 		return;
182 
183 	dprintk("RPC:       %s: rep %p opcode 'recv', length %u: success\n",
184 		__func__, rep, wc->byte_len);
185 
186 	rep->rr_len = wc->byte_len;
187 	ib_dma_sync_single_for_cpu(rep->rr_device,
188 				   rdmab_addr(rep->rr_rdmabuf),
189 				   rep->rr_len, DMA_FROM_DEVICE);
190 
191 	rpcrdma_update_granted_credits(rep);
192 
193 out_schedule:
194 	queue_work(rpcrdma_receive_wq, &rep->rr_work);
195 	return;
196 
197 out_fail:
198 	if (wc->status != IB_WC_WR_FLUSH_ERR)
199 		pr_err("rpcrdma: Recv: %s (%u/0x%x)\n",
200 		       ib_wc_status_msg(wc->status),
201 		       wc->status, wc->vendor_err);
202 	rep->rr_len = RPCRDMA_BAD_LEN;
203 	goto out_schedule;
204 }
205 
206 static void
207 rpcrdma_flush_cqs(struct rpcrdma_ep *ep)
208 {
209 	struct ib_wc wc;
210 
211 	while (ib_poll_cq(ep->rep_attr.recv_cq, 1, &wc) > 0)
212 		rpcrdma_receive_wc(NULL, &wc);
213 }
214 
215 static int
216 rpcrdma_conn_upcall(struct rdma_cm_id *id, struct rdma_cm_event *event)
217 {
218 	struct rpcrdma_xprt *xprt = id->context;
219 	struct rpcrdma_ia *ia = &xprt->rx_ia;
220 	struct rpcrdma_ep *ep = &xprt->rx_ep;
221 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
222 	struct sockaddr *sap = (struct sockaddr *)&ep->rep_remote_addr;
223 #endif
224 	struct ib_qp_attr *attr = &ia->ri_qp_attr;
225 	struct ib_qp_init_attr *iattr = &ia->ri_qp_init_attr;
226 	int connstate = 0;
227 
228 	switch (event->event) {
229 	case RDMA_CM_EVENT_ADDR_RESOLVED:
230 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
231 		ia->ri_async_rc = 0;
232 		complete(&ia->ri_done);
233 		break;
234 	case RDMA_CM_EVENT_ADDR_ERROR:
235 		ia->ri_async_rc = -EHOSTUNREACH;
236 		dprintk("RPC:       %s: CM address resolution error, ep 0x%p\n",
237 			__func__, ep);
238 		complete(&ia->ri_done);
239 		break;
240 	case RDMA_CM_EVENT_ROUTE_ERROR:
241 		ia->ri_async_rc = -ENETUNREACH;
242 		dprintk("RPC:       %s: CM route resolution error, ep 0x%p\n",
243 			__func__, ep);
244 		complete(&ia->ri_done);
245 		break;
246 	case RDMA_CM_EVENT_ESTABLISHED:
247 		connstate = 1;
248 		ib_query_qp(ia->ri_id->qp, attr,
249 			    IB_QP_MAX_QP_RD_ATOMIC | IB_QP_MAX_DEST_RD_ATOMIC,
250 			    iattr);
251 		dprintk("RPC:       %s: %d responder resources"
252 			" (%d initiator)\n",
253 			__func__, attr->max_dest_rd_atomic,
254 			attr->max_rd_atomic);
255 		goto connected;
256 	case RDMA_CM_EVENT_CONNECT_ERROR:
257 		connstate = -ENOTCONN;
258 		goto connected;
259 	case RDMA_CM_EVENT_UNREACHABLE:
260 		connstate = -ENETDOWN;
261 		goto connected;
262 	case RDMA_CM_EVENT_REJECTED:
263 		connstate = -ECONNREFUSED;
264 		goto connected;
265 	case RDMA_CM_EVENT_DISCONNECTED:
266 		connstate = -ECONNABORTED;
267 		goto connected;
268 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
269 		connstate = -ENODEV;
270 connected:
271 		dprintk("RPC:       %s: %sconnected\n",
272 					__func__, connstate > 0 ? "" : "dis");
273 		atomic_set(&xprt->rx_buf.rb_credits, 1);
274 		ep->rep_connected = connstate;
275 		rpcrdma_conn_func(ep);
276 		wake_up_all(&ep->rep_connect_wait);
277 		/*FALLTHROUGH*/
278 	default:
279 		dprintk("RPC:       %s: %pIS:%u (ep 0x%p): %s\n",
280 			__func__, sap, rpc_get_port(sap), ep,
281 			rdma_event_msg(event->event));
282 		break;
283 	}
284 
285 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
286 	if (connstate == 1) {
287 		int ird = attr->max_dest_rd_atomic;
288 		int tird = ep->rep_remote_cma.responder_resources;
289 
290 		pr_info("rpcrdma: connection to %pIS:%u on %s, memreg '%s', %d credits, %d responders%s\n",
291 			sap, rpc_get_port(sap),
292 			ia->ri_device->name,
293 			ia->ri_ops->ro_displayname,
294 			xprt->rx_buf.rb_max_requests,
295 			ird, ird < 4 && ird < tird / 2 ? " (low!)" : "");
296 	} else if (connstate < 0) {
297 		pr_info("rpcrdma: connection to %pIS:%u closed (%d)\n",
298 			sap, rpc_get_port(sap), connstate);
299 	}
300 #endif
301 
302 	return 0;
303 }
304 
305 static void rpcrdma_destroy_id(struct rdma_cm_id *id)
306 {
307 	if (id) {
308 		module_put(id->device->owner);
309 		rdma_destroy_id(id);
310 	}
311 }
312 
313 static struct rdma_cm_id *
314 rpcrdma_create_id(struct rpcrdma_xprt *xprt,
315 			struct rpcrdma_ia *ia, struct sockaddr *addr)
316 {
317 	struct rdma_cm_id *id;
318 	int rc;
319 
320 	init_completion(&ia->ri_done);
321 
322 	id = rdma_create_id(&init_net, rpcrdma_conn_upcall, xprt, RDMA_PS_TCP,
323 			    IB_QPT_RC);
324 	if (IS_ERR(id)) {
325 		rc = PTR_ERR(id);
326 		dprintk("RPC:       %s: rdma_create_id() failed %i\n",
327 			__func__, rc);
328 		return id;
329 	}
330 
331 	ia->ri_async_rc = -ETIMEDOUT;
332 	rc = rdma_resolve_addr(id, NULL, addr, RDMA_RESOLVE_TIMEOUT);
333 	if (rc) {
334 		dprintk("RPC:       %s: rdma_resolve_addr() failed %i\n",
335 			__func__, rc);
336 		goto out;
337 	}
338 	wait_for_completion_interruptible_timeout(&ia->ri_done,
339 				msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
340 
341 	/* FIXME:
342 	 * Until xprtrdma supports DEVICE_REMOVAL, the provider must
343 	 * be pinned while there are active NFS/RDMA mounts to prevent
344 	 * hangs and crashes at umount time.
345 	 */
346 	if (!ia->ri_async_rc && !try_module_get(id->device->owner)) {
347 		dprintk("RPC:       %s: Failed to get device module\n",
348 			__func__);
349 		ia->ri_async_rc = -ENODEV;
350 	}
351 	rc = ia->ri_async_rc;
352 	if (rc)
353 		goto out;
354 
355 	ia->ri_async_rc = -ETIMEDOUT;
356 	rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
357 	if (rc) {
358 		dprintk("RPC:       %s: rdma_resolve_route() failed %i\n",
359 			__func__, rc);
360 		goto put;
361 	}
362 	wait_for_completion_interruptible_timeout(&ia->ri_done,
363 				msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
364 	rc = ia->ri_async_rc;
365 	if (rc)
366 		goto put;
367 
368 	return id;
369 put:
370 	module_put(id->device->owner);
371 out:
372 	rdma_destroy_id(id);
373 	return ERR_PTR(rc);
374 }
375 
376 /*
377  * Drain any cq, prior to teardown.
378  */
379 static void
380 rpcrdma_clean_cq(struct ib_cq *cq)
381 {
382 	struct ib_wc wc;
383 	int count = 0;
384 
385 	while (1 == ib_poll_cq(cq, 1, &wc))
386 		++count;
387 
388 	if (count)
389 		dprintk("RPC:       %s: flushed %d events (last 0x%x)\n",
390 			__func__, count, wc.opcode);
391 }
392 
393 /*
394  * Exported functions.
395  */
396 
397 /*
398  * Open and initialize an Interface Adapter.
399  *  o initializes fields of struct rpcrdma_ia, including
400  *    interface and provider attributes and protection zone.
401  */
402 int
403 rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
404 {
405 	struct rpcrdma_ia *ia = &xprt->rx_ia;
406 	int rc;
407 
408 	ia->ri_dma_mr = NULL;
409 
410 	ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
411 	if (IS_ERR(ia->ri_id)) {
412 		rc = PTR_ERR(ia->ri_id);
413 		goto out1;
414 	}
415 	ia->ri_device = ia->ri_id->device;
416 
417 	ia->ri_pd = ib_alloc_pd(ia->ri_device);
418 	if (IS_ERR(ia->ri_pd)) {
419 		rc = PTR_ERR(ia->ri_pd);
420 		dprintk("RPC:       %s: ib_alloc_pd() failed %i\n",
421 			__func__, rc);
422 		goto out2;
423 	}
424 
425 	if (memreg == RPCRDMA_FRMR) {
426 		if (!(ia->ri_device->attrs.device_cap_flags &
427 				IB_DEVICE_MEM_MGT_EXTENSIONS) ||
428 		    (ia->ri_device->attrs.max_fast_reg_page_list_len == 0)) {
429 			dprintk("RPC:       %s: FRMR registration "
430 				"not supported by HCA\n", __func__);
431 			memreg = RPCRDMA_MTHCAFMR;
432 		}
433 	}
434 	if (memreg == RPCRDMA_MTHCAFMR) {
435 		if (!ia->ri_device->alloc_fmr) {
436 			dprintk("RPC:       %s: MTHCAFMR registration "
437 				"not supported by HCA\n", __func__);
438 			rc = -EINVAL;
439 			goto out3;
440 		}
441 	}
442 
443 	switch (memreg) {
444 	case RPCRDMA_FRMR:
445 		ia->ri_ops = &rpcrdma_frwr_memreg_ops;
446 		break;
447 	case RPCRDMA_ALLPHYSICAL:
448 		ia->ri_ops = &rpcrdma_physical_memreg_ops;
449 		break;
450 	case RPCRDMA_MTHCAFMR:
451 		ia->ri_ops = &rpcrdma_fmr_memreg_ops;
452 		break;
453 	default:
454 		printk(KERN_ERR "RPC: Unsupported memory "
455 				"registration mode: %d\n", memreg);
456 		rc = -ENOMEM;
457 		goto out3;
458 	}
459 	dprintk("RPC:       %s: memory registration strategy is '%s'\n",
460 		__func__, ia->ri_ops->ro_displayname);
461 
462 	rwlock_init(&ia->ri_qplock);
463 	return 0;
464 
465 out3:
466 	ib_dealloc_pd(ia->ri_pd);
467 	ia->ri_pd = NULL;
468 out2:
469 	rpcrdma_destroy_id(ia->ri_id);
470 	ia->ri_id = NULL;
471 out1:
472 	return rc;
473 }
474 
475 /*
476  * Clean up/close an IA.
477  *   o if event handles and PD have been initialized, free them.
478  *   o close the IA
479  */
480 void
481 rpcrdma_ia_close(struct rpcrdma_ia *ia)
482 {
483 	dprintk("RPC:       %s: entering\n", __func__);
484 	if (ia->ri_id != NULL && !IS_ERR(ia->ri_id)) {
485 		if (ia->ri_id->qp)
486 			rdma_destroy_qp(ia->ri_id);
487 		rpcrdma_destroy_id(ia->ri_id);
488 		ia->ri_id = NULL;
489 	}
490 
491 	/* If the pd is still busy, xprtrdma missed freeing a resource */
492 	if (ia->ri_pd && !IS_ERR(ia->ri_pd))
493 		ib_dealloc_pd(ia->ri_pd);
494 }
495 
496 /*
497  * Create unconnected endpoint.
498  */
499 int
500 rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
501 				struct rpcrdma_create_data_internal *cdata)
502 {
503 	struct ib_cq *sendcq, *recvcq;
504 	unsigned int max_qp_wr;
505 	int rc;
506 
507 	if (ia->ri_device->attrs.max_sge < RPCRDMA_MAX_IOVS) {
508 		dprintk("RPC:       %s: insufficient sge's available\n",
509 			__func__);
510 		return -ENOMEM;
511 	}
512 
513 	if (ia->ri_device->attrs.max_qp_wr <= RPCRDMA_BACKWARD_WRS) {
514 		dprintk("RPC:       %s: insufficient wqe's available\n",
515 			__func__);
516 		return -ENOMEM;
517 	}
518 	max_qp_wr = ia->ri_device->attrs.max_qp_wr - RPCRDMA_BACKWARD_WRS;
519 
520 	/* check provider's send/recv wr limits */
521 	if (cdata->max_requests > max_qp_wr)
522 		cdata->max_requests = max_qp_wr;
523 
524 	ep->rep_attr.event_handler = rpcrdma_qp_async_error_upcall;
525 	ep->rep_attr.qp_context = ep;
526 	ep->rep_attr.srq = NULL;
527 	ep->rep_attr.cap.max_send_wr = cdata->max_requests;
528 	ep->rep_attr.cap.max_send_wr += RPCRDMA_BACKWARD_WRS;
529 	rc = ia->ri_ops->ro_open(ia, ep, cdata);
530 	if (rc)
531 		return rc;
532 	ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
533 	ep->rep_attr.cap.max_recv_wr += RPCRDMA_BACKWARD_WRS;
534 	ep->rep_attr.cap.max_send_sge = RPCRDMA_MAX_IOVS;
535 	ep->rep_attr.cap.max_recv_sge = 1;
536 	ep->rep_attr.cap.max_inline_data = 0;
537 	ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
538 	ep->rep_attr.qp_type = IB_QPT_RC;
539 	ep->rep_attr.port_num = ~0;
540 
541 	dprintk("RPC:       %s: requested max: dtos: send %d recv %d; "
542 		"iovs: send %d recv %d\n",
543 		__func__,
544 		ep->rep_attr.cap.max_send_wr,
545 		ep->rep_attr.cap.max_recv_wr,
546 		ep->rep_attr.cap.max_send_sge,
547 		ep->rep_attr.cap.max_recv_sge);
548 
549 	/* set trigger for requesting send completion */
550 	ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 - 1;
551 	if (ep->rep_cqinit <= 2)
552 		ep->rep_cqinit = 0;	/* always signal? */
553 	INIT_CQCOUNT(ep);
554 	init_waitqueue_head(&ep->rep_connect_wait);
555 	INIT_DELAYED_WORK(&ep->rep_connect_worker, rpcrdma_connect_worker);
556 
557 	sendcq = ib_alloc_cq(ia->ri_device, NULL,
558 			     ep->rep_attr.cap.max_send_wr + 1,
559 			     0, IB_POLL_SOFTIRQ);
560 	if (IS_ERR(sendcq)) {
561 		rc = PTR_ERR(sendcq);
562 		dprintk("RPC:       %s: failed to create send CQ: %i\n",
563 			__func__, rc);
564 		goto out1;
565 	}
566 
567 	recvcq = ib_alloc_cq(ia->ri_device, NULL,
568 			     ep->rep_attr.cap.max_recv_wr + 1,
569 			     0, IB_POLL_SOFTIRQ);
570 	if (IS_ERR(recvcq)) {
571 		rc = PTR_ERR(recvcq);
572 		dprintk("RPC:       %s: failed to create recv CQ: %i\n",
573 			__func__, rc);
574 		goto out2;
575 	}
576 
577 	ep->rep_attr.send_cq = sendcq;
578 	ep->rep_attr.recv_cq = recvcq;
579 
580 	/* Initialize cma parameters */
581 
582 	/* RPC/RDMA does not use private data */
583 	ep->rep_remote_cma.private_data = NULL;
584 	ep->rep_remote_cma.private_data_len = 0;
585 
586 	/* Client offers RDMA Read but does not initiate */
587 	ep->rep_remote_cma.initiator_depth = 0;
588 	if (ia->ri_device->attrs.max_qp_rd_atom > 32)	/* arbitrary but <= 255 */
589 		ep->rep_remote_cma.responder_resources = 32;
590 	else
591 		ep->rep_remote_cma.responder_resources =
592 						ia->ri_device->attrs.max_qp_rd_atom;
593 
594 	ep->rep_remote_cma.retry_count = 7;
595 	ep->rep_remote_cma.flow_control = 0;
596 	ep->rep_remote_cma.rnr_retry_count = 0;
597 
598 	return 0;
599 
600 out2:
601 	ib_free_cq(sendcq);
602 out1:
603 	if (ia->ri_dma_mr)
604 		ib_dereg_mr(ia->ri_dma_mr);
605 	return rc;
606 }
607 
608 /*
609  * rpcrdma_ep_destroy
610  *
611  * Disconnect and destroy endpoint. After this, the only
612  * valid operations on the ep are to free it (if dynamically
613  * allocated) or re-create it.
614  */
615 void
616 rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
617 {
618 	int rc;
619 
620 	dprintk("RPC:       %s: entering, connected is %d\n",
621 		__func__, ep->rep_connected);
622 
623 	cancel_delayed_work_sync(&ep->rep_connect_worker);
624 
625 	if (ia->ri_id->qp)
626 		rpcrdma_ep_disconnect(ep, ia);
627 
628 	rpcrdma_clean_cq(ep->rep_attr.recv_cq);
629 	rpcrdma_clean_cq(ep->rep_attr.send_cq);
630 
631 	if (ia->ri_id->qp) {
632 		rdma_destroy_qp(ia->ri_id);
633 		ia->ri_id->qp = NULL;
634 	}
635 
636 	ib_free_cq(ep->rep_attr.recv_cq);
637 	ib_free_cq(ep->rep_attr.send_cq);
638 
639 	if (ia->ri_dma_mr) {
640 		rc = ib_dereg_mr(ia->ri_dma_mr);
641 		dprintk("RPC:       %s: ib_dereg_mr returned %i\n",
642 			__func__, rc);
643 	}
644 }
645 
646 /*
647  * Connect unconnected endpoint.
648  */
649 int
650 rpcrdma_ep_connect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
651 {
652 	struct rdma_cm_id *id, *old;
653 	int rc = 0;
654 	int retry_count = 0;
655 
656 	if (ep->rep_connected != 0) {
657 		struct rpcrdma_xprt *xprt;
658 retry:
659 		dprintk("RPC:       %s: reconnecting...\n", __func__);
660 
661 		rpcrdma_ep_disconnect(ep, ia);
662 		rpcrdma_flush_cqs(ep);
663 
664 		xprt = container_of(ia, struct rpcrdma_xprt, rx_ia);
665 		id = rpcrdma_create_id(xprt, ia,
666 				(struct sockaddr *)&xprt->rx_data.addr);
667 		if (IS_ERR(id)) {
668 			rc = -EHOSTUNREACH;
669 			goto out;
670 		}
671 		/* TEMP TEMP TEMP - fail if new device:
672 		 * Deregister/remarshal *all* requests!
673 		 * Close and recreate adapter, pd, etc!
674 		 * Re-determine all attributes still sane!
675 		 * More stuff I haven't thought of!
676 		 * Rrrgh!
677 		 */
678 		if (ia->ri_device != id->device) {
679 			printk("RPC:       %s: can't reconnect on "
680 				"different device!\n", __func__);
681 			rpcrdma_destroy_id(id);
682 			rc = -ENETUNREACH;
683 			goto out;
684 		}
685 		/* END TEMP */
686 		rc = rdma_create_qp(id, ia->ri_pd, &ep->rep_attr);
687 		if (rc) {
688 			dprintk("RPC:       %s: rdma_create_qp failed %i\n",
689 				__func__, rc);
690 			rpcrdma_destroy_id(id);
691 			rc = -ENETUNREACH;
692 			goto out;
693 		}
694 
695 		write_lock(&ia->ri_qplock);
696 		old = ia->ri_id;
697 		ia->ri_id = id;
698 		write_unlock(&ia->ri_qplock);
699 
700 		rdma_destroy_qp(old);
701 		rpcrdma_destroy_id(old);
702 	} else {
703 		dprintk("RPC:       %s: connecting...\n", __func__);
704 		rc = rdma_create_qp(ia->ri_id, ia->ri_pd, &ep->rep_attr);
705 		if (rc) {
706 			dprintk("RPC:       %s: rdma_create_qp failed %i\n",
707 				__func__, rc);
708 			/* do not update ep->rep_connected */
709 			return -ENETUNREACH;
710 		}
711 	}
712 
713 	ep->rep_connected = 0;
714 
715 	rc = rdma_connect(ia->ri_id, &ep->rep_remote_cma);
716 	if (rc) {
717 		dprintk("RPC:       %s: rdma_connect() failed with %i\n",
718 				__func__, rc);
719 		goto out;
720 	}
721 
722 	wait_event_interruptible(ep->rep_connect_wait, ep->rep_connected != 0);
723 
724 	/*
725 	 * Check state. A non-peer reject indicates no listener
726 	 * (ECONNREFUSED), which may be a transient state. All
727 	 * others indicate a transport condition which has already
728 	 * undergone a best-effort.
729 	 */
730 	if (ep->rep_connected == -ECONNREFUSED &&
731 	    ++retry_count <= RDMA_CONNECT_RETRY_MAX) {
732 		dprintk("RPC:       %s: non-peer_reject, retry\n", __func__);
733 		goto retry;
734 	}
735 	if (ep->rep_connected <= 0) {
736 		/* Sometimes, the only way to reliably connect to remote
737 		 * CMs is to use same nonzero values for ORD and IRD. */
738 		if (retry_count++ <= RDMA_CONNECT_RETRY_MAX + 1 &&
739 		    (ep->rep_remote_cma.responder_resources == 0 ||
740 		     ep->rep_remote_cma.initiator_depth !=
741 				ep->rep_remote_cma.responder_resources)) {
742 			if (ep->rep_remote_cma.responder_resources == 0)
743 				ep->rep_remote_cma.responder_resources = 1;
744 			ep->rep_remote_cma.initiator_depth =
745 				ep->rep_remote_cma.responder_resources;
746 			goto retry;
747 		}
748 		rc = ep->rep_connected;
749 	} else {
750 		struct rpcrdma_xprt *r_xprt;
751 		unsigned int extras;
752 
753 		dprintk("RPC:       %s: connected\n", __func__);
754 
755 		r_xprt = container_of(ia, struct rpcrdma_xprt, rx_ia);
756 		extras = r_xprt->rx_buf.rb_bc_srv_max_requests;
757 
758 		if (extras) {
759 			rc = rpcrdma_ep_post_extra_recv(r_xprt, extras);
760 			if (rc) {
761 				pr_warn("%s: rpcrdma_ep_post_extra_recv: %i\n",
762 					__func__, rc);
763 				rc = 0;
764 			}
765 		}
766 	}
767 
768 out:
769 	if (rc)
770 		ep->rep_connected = rc;
771 	return rc;
772 }
773 
774 /*
775  * rpcrdma_ep_disconnect
776  *
777  * This is separate from destroy to facilitate the ability
778  * to reconnect without recreating the endpoint.
779  *
780  * This call is not reentrant, and must not be made in parallel
781  * on the same endpoint.
782  */
783 void
784 rpcrdma_ep_disconnect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
785 {
786 	int rc;
787 
788 	rpcrdma_flush_cqs(ep);
789 	rc = rdma_disconnect(ia->ri_id);
790 	if (!rc) {
791 		/* returns without wait if not connected */
792 		wait_event_interruptible(ep->rep_connect_wait,
793 							ep->rep_connected != 1);
794 		dprintk("RPC:       %s: after wait, %sconnected\n", __func__,
795 			(ep->rep_connected == 1) ? "still " : "dis");
796 	} else {
797 		dprintk("RPC:       %s: rdma_disconnect %i\n", __func__, rc);
798 		ep->rep_connected = rc;
799 	}
800 }
801 
802 struct rpcrdma_req *
803 rpcrdma_create_req(struct rpcrdma_xprt *r_xprt)
804 {
805 	struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
806 	struct rpcrdma_req *req;
807 
808 	req = kzalloc(sizeof(*req), GFP_KERNEL);
809 	if (req == NULL)
810 		return ERR_PTR(-ENOMEM);
811 
812 	INIT_LIST_HEAD(&req->rl_free);
813 	spin_lock(&buffer->rb_reqslock);
814 	list_add(&req->rl_all, &buffer->rb_allreqs);
815 	spin_unlock(&buffer->rb_reqslock);
816 	req->rl_cqe.done = rpcrdma_wc_send;
817 	req->rl_buffer = &r_xprt->rx_buf;
818 	return req;
819 }
820 
821 struct rpcrdma_rep *
822 rpcrdma_create_rep(struct rpcrdma_xprt *r_xprt)
823 {
824 	struct rpcrdma_create_data_internal *cdata = &r_xprt->rx_data;
825 	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
826 	struct rpcrdma_rep *rep;
827 	int rc;
828 
829 	rc = -ENOMEM;
830 	rep = kzalloc(sizeof(*rep), GFP_KERNEL);
831 	if (rep == NULL)
832 		goto out;
833 
834 	rep->rr_rdmabuf = rpcrdma_alloc_regbuf(ia, cdata->inline_rsize,
835 					       GFP_KERNEL);
836 	if (IS_ERR(rep->rr_rdmabuf)) {
837 		rc = PTR_ERR(rep->rr_rdmabuf);
838 		goto out_free;
839 	}
840 
841 	rep->rr_device = ia->ri_device;
842 	rep->rr_cqe.done = rpcrdma_receive_wc;
843 	rep->rr_rxprt = r_xprt;
844 	INIT_WORK(&rep->rr_work, rpcrdma_receive_worker);
845 	return rep;
846 
847 out_free:
848 	kfree(rep);
849 out:
850 	return ERR_PTR(rc);
851 }
852 
853 int
854 rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt)
855 {
856 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
857 	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
858 	int i, rc;
859 
860 	buf->rb_max_requests = r_xprt->rx_data.max_requests;
861 	buf->rb_bc_srv_max_requests = 0;
862 	spin_lock_init(&buf->rb_lock);
863 	atomic_set(&buf->rb_credits, 1);
864 
865 	rc = ia->ri_ops->ro_init(r_xprt);
866 	if (rc)
867 		goto out;
868 
869 	INIT_LIST_HEAD(&buf->rb_send_bufs);
870 	INIT_LIST_HEAD(&buf->rb_allreqs);
871 	spin_lock_init(&buf->rb_reqslock);
872 	for (i = 0; i < buf->rb_max_requests; i++) {
873 		struct rpcrdma_req *req;
874 
875 		req = rpcrdma_create_req(r_xprt);
876 		if (IS_ERR(req)) {
877 			dprintk("RPC:       %s: request buffer %d alloc"
878 				" failed\n", __func__, i);
879 			rc = PTR_ERR(req);
880 			goto out;
881 		}
882 		req->rl_backchannel = false;
883 		list_add(&req->rl_free, &buf->rb_send_bufs);
884 	}
885 
886 	INIT_LIST_HEAD(&buf->rb_recv_bufs);
887 	for (i = 0; i < buf->rb_max_requests + 2; i++) {
888 		struct rpcrdma_rep *rep;
889 
890 		rep = rpcrdma_create_rep(r_xprt);
891 		if (IS_ERR(rep)) {
892 			dprintk("RPC:       %s: reply buffer %d alloc failed\n",
893 				__func__, i);
894 			rc = PTR_ERR(rep);
895 			goto out;
896 		}
897 		list_add(&rep->rr_list, &buf->rb_recv_bufs);
898 	}
899 
900 	return 0;
901 out:
902 	rpcrdma_buffer_destroy(buf);
903 	return rc;
904 }
905 
906 static struct rpcrdma_req *
907 rpcrdma_buffer_get_req_locked(struct rpcrdma_buffer *buf)
908 {
909 	struct rpcrdma_req *req;
910 
911 	req = list_first_entry(&buf->rb_send_bufs,
912 			       struct rpcrdma_req, rl_free);
913 	list_del(&req->rl_free);
914 	return req;
915 }
916 
917 static struct rpcrdma_rep *
918 rpcrdma_buffer_get_rep_locked(struct rpcrdma_buffer *buf)
919 {
920 	struct rpcrdma_rep *rep;
921 
922 	rep = list_first_entry(&buf->rb_recv_bufs,
923 			       struct rpcrdma_rep, rr_list);
924 	list_del(&rep->rr_list);
925 	return rep;
926 }
927 
928 static void
929 rpcrdma_destroy_rep(struct rpcrdma_ia *ia, struct rpcrdma_rep *rep)
930 {
931 	rpcrdma_free_regbuf(ia, rep->rr_rdmabuf);
932 	kfree(rep);
933 }
934 
935 void
936 rpcrdma_destroy_req(struct rpcrdma_ia *ia, struct rpcrdma_req *req)
937 {
938 	rpcrdma_free_regbuf(ia, req->rl_sendbuf);
939 	rpcrdma_free_regbuf(ia, req->rl_rdmabuf);
940 	kfree(req);
941 }
942 
943 void
944 rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf)
945 {
946 	struct rpcrdma_ia *ia = rdmab_to_ia(buf);
947 
948 	while (!list_empty(&buf->rb_recv_bufs)) {
949 		struct rpcrdma_rep *rep;
950 
951 		rep = rpcrdma_buffer_get_rep_locked(buf);
952 		rpcrdma_destroy_rep(ia, rep);
953 	}
954 
955 	spin_lock(&buf->rb_reqslock);
956 	while (!list_empty(&buf->rb_allreqs)) {
957 		struct rpcrdma_req *req;
958 
959 		req = list_first_entry(&buf->rb_allreqs,
960 				       struct rpcrdma_req, rl_all);
961 		list_del(&req->rl_all);
962 
963 		spin_unlock(&buf->rb_reqslock);
964 		rpcrdma_destroy_req(ia, req);
965 		spin_lock(&buf->rb_reqslock);
966 	}
967 	spin_unlock(&buf->rb_reqslock);
968 
969 	ia->ri_ops->ro_destroy(buf);
970 }
971 
972 struct rpcrdma_mw *
973 rpcrdma_get_mw(struct rpcrdma_xprt *r_xprt)
974 {
975 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
976 	struct rpcrdma_mw *mw = NULL;
977 
978 	spin_lock(&buf->rb_mwlock);
979 	if (!list_empty(&buf->rb_mws)) {
980 		mw = list_first_entry(&buf->rb_mws,
981 				      struct rpcrdma_mw, mw_list);
982 		list_del_init(&mw->mw_list);
983 	}
984 	spin_unlock(&buf->rb_mwlock);
985 
986 	if (!mw)
987 		pr_err("RPC:       %s: no MWs available\n", __func__);
988 	return mw;
989 }
990 
991 void
992 rpcrdma_put_mw(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mw *mw)
993 {
994 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
995 
996 	spin_lock(&buf->rb_mwlock);
997 	list_add_tail(&mw->mw_list, &buf->rb_mws);
998 	spin_unlock(&buf->rb_mwlock);
999 }
1000 
1001 /*
1002  * Get a set of request/reply buffers.
1003  *
1004  * Reply buffer (if available) is attached to send buffer upon return.
1005  */
1006 struct rpcrdma_req *
1007 rpcrdma_buffer_get(struct rpcrdma_buffer *buffers)
1008 {
1009 	struct rpcrdma_req *req;
1010 
1011 	spin_lock(&buffers->rb_lock);
1012 	if (list_empty(&buffers->rb_send_bufs))
1013 		goto out_reqbuf;
1014 	req = rpcrdma_buffer_get_req_locked(buffers);
1015 	if (list_empty(&buffers->rb_recv_bufs))
1016 		goto out_repbuf;
1017 	req->rl_reply = rpcrdma_buffer_get_rep_locked(buffers);
1018 	spin_unlock(&buffers->rb_lock);
1019 	return req;
1020 
1021 out_reqbuf:
1022 	spin_unlock(&buffers->rb_lock);
1023 	pr_warn("RPC:       %s: out of request buffers\n", __func__);
1024 	return NULL;
1025 out_repbuf:
1026 	spin_unlock(&buffers->rb_lock);
1027 	pr_warn("RPC:       %s: out of reply buffers\n", __func__);
1028 	req->rl_reply = NULL;
1029 	return req;
1030 }
1031 
1032 /*
1033  * Put request/reply buffers back into pool.
1034  * Pre-decrement counter/array index.
1035  */
1036 void
1037 rpcrdma_buffer_put(struct rpcrdma_req *req)
1038 {
1039 	struct rpcrdma_buffer *buffers = req->rl_buffer;
1040 	struct rpcrdma_rep *rep = req->rl_reply;
1041 
1042 	req->rl_niovs = 0;
1043 	req->rl_reply = NULL;
1044 
1045 	spin_lock(&buffers->rb_lock);
1046 	list_add_tail(&req->rl_free, &buffers->rb_send_bufs);
1047 	if (rep)
1048 		list_add_tail(&rep->rr_list, &buffers->rb_recv_bufs);
1049 	spin_unlock(&buffers->rb_lock);
1050 }
1051 
1052 /*
1053  * Recover reply buffers from pool.
1054  * This happens when recovering from disconnect.
1055  */
1056 void
1057 rpcrdma_recv_buffer_get(struct rpcrdma_req *req)
1058 {
1059 	struct rpcrdma_buffer *buffers = req->rl_buffer;
1060 
1061 	spin_lock(&buffers->rb_lock);
1062 	if (!list_empty(&buffers->rb_recv_bufs))
1063 		req->rl_reply = rpcrdma_buffer_get_rep_locked(buffers);
1064 	spin_unlock(&buffers->rb_lock);
1065 }
1066 
1067 /*
1068  * Put reply buffers back into pool when not attached to
1069  * request. This happens in error conditions.
1070  */
1071 void
1072 rpcrdma_recv_buffer_put(struct rpcrdma_rep *rep)
1073 {
1074 	struct rpcrdma_buffer *buffers = &rep->rr_rxprt->rx_buf;
1075 
1076 	spin_lock(&buffers->rb_lock);
1077 	list_add_tail(&rep->rr_list, &buffers->rb_recv_bufs);
1078 	spin_unlock(&buffers->rb_lock);
1079 }
1080 
1081 /*
1082  * Wrappers for internal-use kmalloc memory registration, used by buffer code.
1083  */
1084 
1085 void
1086 rpcrdma_mapping_error(struct rpcrdma_mr_seg *seg)
1087 {
1088 	dprintk("RPC:       map_one: offset %p iova %llx len %zu\n",
1089 		seg->mr_offset,
1090 		(unsigned long long)seg->mr_dma, seg->mr_dmalen);
1091 }
1092 
1093 /**
1094  * rpcrdma_alloc_regbuf - kmalloc and register memory for SEND/RECV buffers
1095  * @ia: controlling rpcrdma_ia
1096  * @size: size of buffer to be allocated, in bytes
1097  * @flags: GFP flags
1098  *
1099  * Returns pointer to private header of an area of internally
1100  * registered memory, or an ERR_PTR. The registered buffer follows
1101  * the end of the private header.
1102  *
1103  * xprtrdma uses a regbuf for posting an outgoing RDMA SEND, or for
1104  * receiving the payload of RDMA RECV operations. regbufs are not
1105  * used for RDMA READ/WRITE operations, thus are registered only for
1106  * LOCAL access.
1107  */
1108 struct rpcrdma_regbuf *
1109 rpcrdma_alloc_regbuf(struct rpcrdma_ia *ia, size_t size, gfp_t flags)
1110 {
1111 	struct rpcrdma_regbuf *rb;
1112 	struct ib_sge *iov;
1113 
1114 	rb = kmalloc(sizeof(*rb) + size, flags);
1115 	if (rb == NULL)
1116 		goto out;
1117 
1118 	iov = &rb->rg_iov;
1119 	iov->addr = ib_dma_map_single(ia->ri_device,
1120 				      (void *)rb->rg_base, size,
1121 				      DMA_BIDIRECTIONAL);
1122 	if (ib_dma_mapping_error(ia->ri_device, iov->addr))
1123 		goto out_free;
1124 
1125 	iov->length = size;
1126 	iov->lkey = ia->ri_pd->local_dma_lkey;
1127 	rb->rg_size = size;
1128 	rb->rg_owner = NULL;
1129 	return rb;
1130 
1131 out_free:
1132 	kfree(rb);
1133 out:
1134 	return ERR_PTR(-ENOMEM);
1135 }
1136 
1137 /**
1138  * rpcrdma_free_regbuf - deregister and free registered buffer
1139  * @ia: controlling rpcrdma_ia
1140  * @rb: regbuf to be deregistered and freed
1141  */
1142 void
1143 rpcrdma_free_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
1144 {
1145 	struct ib_sge *iov;
1146 
1147 	if (!rb)
1148 		return;
1149 
1150 	iov = &rb->rg_iov;
1151 	ib_dma_unmap_single(ia->ri_device,
1152 			    iov->addr, iov->length, DMA_BIDIRECTIONAL);
1153 	kfree(rb);
1154 }
1155 
1156 /*
1157  * Prepost any receive buffer, then post send.
1158  *
1159  * Receive buffer is donated to hardware, reclaimed upon recv completion.
1160  */
1161 int
1162 rpcrdma_ep_post(struct rpcrdma_ia *ia,
1163 		struct rpcrdma_ep *ep,
1164 		struct rpcrdma_req *req)
1165 {
1166 	struct ib_device *device = ia->ri_device;
1167 	struct ib_send_wr send_wr, *send_wr_fail;
1168 	struct rpcrdma_rep *rep = req->rl_reply;
1169 	struct ib_sge *iov = req->rl_send_iov;
1170 	int i, rc;
1171 
1172 	if (rep) {
1173 		rc = rpcrdma_ep_post_recv(ia, ep, rep);
1174 		if (rc)
1175 			goto out;
1176 		req->rl_reply = NULL;
1177 	}
1178 
1179 	send_wr.next = NULL;
1180 	send_wr.wr_cqe = &req->rl_cqe;
1181 	send_wr.sg_list = iov;
1182 	send_wr.num_sge = req->rl_niovs;
1183 	send_wr.opcode = IB_WR_SEND;
1184 
1185 	for (i = 0; i < send_wr.num_sge; i++)
1186 		ib_dma_sync_single_for_device(device, iov[i].addr,
1187 					      iov[i].length, DMA_TO_DEVICE);
1188 	dprintk("RPC:       %s: posting %d s/g entries\n",
1189 		__func__, send_wr.num_sge);
1190 
1191 	if (DECR_CQCOUNT(ep) > 0)
1192 		send_wr.send_flags = 0;
1193 	else { /* Provider must take a send completion every now and then */
1194 		INIT_CQCOUNT(ep);
1195 		send_wr.send_flags = IB_SEND_SIGNALED;
1196 	}
1197 
1198 	rc = ib_post_send(ia->ri_id->qp, &send_wr, &send_wr_fail);
1199 	if (rc)
1200 		dprintk("RPC:       %s: ib_post_send returned %i\n", __func__,
1201 			rc);
1202 out:
1203 	return rc;
1204 }
1205 
1206 /*
1207  * (Re)post a receive buffer.
1208  */
1209 int
1210 rpcrdma_ep_post_recv(struct rpcrdma_ia *ia,
1211 		     struct rpcrdma_ep *ep,
1212 		     struct rpcrdma_rep *rep)
1213 {
1214 	struct ib_recv_wr recv_wr, *recv_wr_fail;
1215 	int rc;
1216 
1217 	recv_wr.next = NULL;
1218 	recv_wr.wr_cqe = &rep->rr_cqe;
1219 	recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov;
1220 	recv_wr.num_sge = 1;
1221 
1222 	ib_dma_sync_single_for_cpu(ia->ri_device,
1223 				   rdmab_addr(rep->rr_rdmabuf),
1224 				   rdmab_length(rep->rr_rdmabuf),
1225 				   DMA_BIDIRECTIONAL);
1226 
1227 	rc = ib_post_recv(ia->ri_id->qp, &recv_wr, &recv_wr_fail);
1228 
1229 	if (rc)
1230 		dprintk("RPC:       %s: ib_post_recv returned %i\n", __func__,
1231 			rc);
1232 	return rc;
1233 }
1234 
1235 /**
1236  * rpcrdma_ep_post_extra_recv - Post buffers for incoming backchannel requests
1237  * @r_xprt: transport associated with these backchannel resources
1238  * @min_reqs: minimum number of incoming requests expected
1239  *
1240  * Returns zero if all requested buffers were posted, or a negative errno.
1241  */
1242 int
1243 rpcrdma_ep_post_extra_recv(struct rpcrdma_xprt *r_xprt, unsigned int count)
1244 {
1245 	struct rpcrdma_buffer *buffers = &r_xprt->rx_buf;
1246 	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
1247 	struct rpcrdma_ep *ep = &r_xprt->rx_ep;
1248 	struct rpcrdma_rep *rep;
1249 	int rc;
1250 
1251 	while (count--) {
1252 		spin_lock(&buffers->rb_lock);
1253 		if (list_empty(&buffers->rb_recv_bufs))
1254 			goto out_reqbuf;
1255 		rep = rpcrdma_buffer_get_rep_locked(buffers);
1256 		spin_unlock(&buffers->rb_lock);
1257 
1258 		rc = rpcrdma_ep_post_recv(ia, ep, rep);
1259 		if (rc)
1260 			goto out_rc;
1261 	}
1262 
1263 	return 0;
1264 
1265 out_reqbuf:
1266 	spin_unlock(&buffers->rb_lock);
1267 	pr_warn("%s: no extra receive buffers\n", __func__);
1268 	return -ENOMEM;
1269 
1270 out_rc:
1271 	rpcrdma_recv_buffer_put(rep);
1272 	return rc;
1273 }
1274 
1275 /* How many chunk list items fit within our inline buffers?
1276  */
1277 unsigned int
1278 rpcrdma_max_segments(struct rpcrdma_xprt *r_xprt)
1279 {
1280 	struct rpcrdma_create_data_internal *cdata = &r_xprt->rx_data;
1281 	int bytes, segments;
1282 
1283 	bytes = min_t(unsigned int, cdata->inline_wsize, cdata->inline_rsize);
1284 	bytes -= RPCRDMA_HDRLEN_MIN;
1285 	if (bytes < sizeof(struct rpcrdma_segment) * 2) {
1286 		pr_warn("RPC:       %s: inline threshold too small\n",
1287 			__func__);
1288 		return 0;
1289 	}
1290 
1291 	segments = 1 << (fls(bytes / sizeof(struct rpcrdma_segment)) - 1);
1292 	dprintk("RPC:       %s: max chunk list size = %d segments\n",
1293 		__func__, segments);
1294 	return segments;
1295 }
1296