Loading...
1/*
2 * Copyright (c) 2005-2007 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 * Author: Tom Tucker <tom@opengridcomputing.com>
40 */
41
42#include <linux/sunrpc/svc_xprt.h>
43#include <linux/sunrpc/debug.h>
44#include <linux/sunrpc/rpc_rdma.h>
45#include <linux/interrupt.h>
46#include <linux/sched.h>
47#include <linux/slab.h>
48#include <linux/spinlock.h>
49#include <linux/workqueue.h>
50#include <rdma/ib_verbs.h>
51#include <rdma/rdma_cm.h>
52#include <linux/sunrpc/svc_rdma.h>
53#include <linux/export.h>
54#include "xprt_rdma.h"
55
56#define RPCDBG_FACILITY RPCDBG_SVCXPRT
57
58static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
59 struct net *net,
60 struct sockaddr *sa, int salen,
61 int flags);
62static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt);
63static void svc_rdma_release_rqst(struct svc_rqst *);
64static void dto_tasklet_func(unsigned long data);
65static void svc_rdma_detach(struct svc_xprt *xprt);
66static void svc_rdma_free(struct svc_xprt *xprt);
67static int svc_rdma_has_wspace(struct svc_xprt *xprt);
68static void rq_cq_reap(struct svcxprt_rdma *xprt);
69static void sq_cq_reap(struct svcxprt_rdma *xprt);
70
71static DECLARE_TASKLET(dto_tasklet, dto_tasklet_func, 0UL);
72static DEFINE_SPINLOCK(dto_lock);
73static LIST_HEAD(dto_xprt_q);
74
75static struct svc_xprt_ops svc_rdma_ops = {
76 .xpo_create = svc_rdma_create,
77 .xpo_recvfrom = svc_rdma_recvfrom,
78 .xpo_sendto = svc_rdma_sendto,
79 .xpo_release_rqst = svc_rdma_release_rqst,
80 .xpo_detach = svc_rdma_detach,
81 .xpo_free = svc_rdma_free,
82 .xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
83 .xpo_has_wspace = svc_rdma_has_wspace,
84 .xpo_accept = svc_rdma_accept,
85};
86
87struct svc_xprt_class svc_rdma_class = {
88 .xcl_name = "rdma",
89 .xcl_owner = THIS_MODULE,
90 .xcl_ops = &svc_rdma_ops,
91 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
92};
93
94struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
95{
96 struct svc_rdma_op_ctxt *ctxt;
97
98 while (1) {
99 ctxt = kmem_cache_alloc(svc_rdma_ctxt_cachep, GFP_KERNEL);
100 if (ctxt)
101 break;
102 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
103 }
104 ctxt->xprt = xprt;
105 INIT_LIST_HEAD(&ctxt->dto_q);
106 ctxt->count = 0;
107 ctxt->frmr = NULL;
108 atomic_inc(&xprt->sc_ctxt_used);
109 return ctxt;
110}
111
112void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
113{
114 struct svcxprt_rdma *xprt = ctxt->xprt;
115 int i;
116 for (i = 0; i < ctxt->count && ctxt->sge[i].length; i++) {
117 /*
118 * Unmap the DMA addr in the SGE if the lkey matches
119 * the sc_dma_lkey, otherwise, ignore it since it is
120 * an FRMR lkey and will be unmapped later when the
121 * last WR that uses it completes.
122 */
123 if (ctxt->sge[i].lkey == xprt->sc_dma_lkey) {
124 atomic_dec(&xprt->sc_dma_used);
125 ib_dma_unmap_page(xprt->sc_cm_id->device,
126 ctxt->sge[i].addr,
127 ctxt->sge[i].length,
128 ctxt->direction);
129 }
130 }
131}
132
133void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
134{
135 struct svcxprt_rdma *xprt;
136 int i;
137
138 BUG_ON(!ctxt);
139 xprt = ctxt->xprt;
140 if (free_pages)
141 for (i = 0; i < ctxt->count; i++)
142 put_page(ctxt->pages[i]);
143
144 kmem_cache_free(svc_rdma_ctxt_cachep, ctxt);
145 atomic_dec(&xprt->sc_ctxt_used);
146}
147
148/*
149 * Temporary NFS req mappings are shared across all transport
150 * instances. These are short lived and should be bounded by the number
151 * of concurrent server threads * depth of the SQ.
152 */
153struct svc_rdma_req_map *svc_rdma_get_req_map(void)
154{
155 struct svc_rdma_req_map *map;
156 while (1) {
157 map = kmem_cache_alloc(svc_rdma_map_cachep, GFP_KERNEL);
158 if (map)
159 break;
160 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
161 }
162 map->count = 0;
163 map->frmr = NULL;
164 return map;
165}
166
167void svc_rdma_put_req_map(struct svc_rdma_req_map *map)
168{
169 kmem_cache_free(svc_rdma_map_cachep, map);
170}
171
172/* ib_cq event handler */
173static void cq_event_handler(struct ib_event *event, void *context)
174{
175 struct svc_xprt *xprt = context;
176 dprintk("svcrdma: received CQ event id=%d, context=%p\n",
177 event->event, context);
178 set_bit(XPT_CLOSE, &xprt->xpt_flags);
179}
180
181/* QP event handler */
182static void qp_event_handler(struct ib_event *event, void *context)
183{
184 struct svc_xprt *xprt = context;
185
186 switch (event->event) {
187 /* These are considered benign events */
188 case IB_EVENT_PATH_MIG:
189 case IB_EVENT_COMM_EST:
190 case IB_EVENT_SQ_DRAINED:
191 case IB_EVENT_QP_LAST_WQE_REACHED:
192 dprintk("svcrdma: QP event %d received for QP=%p\n",
193 event->event, event->element.qp);
194 break;
195 /* These are considered fatal events */
196 case IB_EVENT_PATH_MIG_ERR:
197 case IB_EVENT_QP_FATAL:
198 case IB_EVENT_QP_REQ_ERR:
199 case IB_EVENT_QP_ACCESS_ERR:
200 case IB_EVENT_DEVICE_FATAL:
201 default:
202 dprintk("svcrdma: QP ERROR event %d received for QP=%p, "
203 "closing transport\n",
204 event->event, event->element.qp);
205 set_bit(XPT_CLOSE, &xprt->xpt_flags);
206 break;
207 }
208}
209
210/*
211 * Data Transfer Operation Tasklet
212 *
213 * Walks a list of transports with I/O pending, removing entries as
214 * they are added to the server's I/O pending list. Two bits indicate
215 * if SQ, RQ, or both have I/O pending. The dto_lock is an irqsave
216 * spinlock that serializes access to the transport list with the RQ
217 * and SQ interrupt handlers.
218 */
219static void dto_tasklet_func(unsigned long data)
220{
221 struct svcxprt_rdma *xprt;
222 unsigned long flags;
223
224 spin_lock_irqsave(&dto_lock, flags);
225 while (!list_empty(&dto_xprt_q)) {
226 xprt = list_entry(dto_xprt_q.next,
227 struct svcxprt_rdma, sc_dto_q);
228 list_del_init(&xprt->sc_dto_q);
229 spin_unlock_irqrestore(&dto_lock, flags);
230
231 rq_cq_reap(xprt);
232 sq_cq_reap(xprt);
233
234 svc_xprt_put(&xprt->sc_xprt);
235 spin_lock_irqsave(&dto_lock, flags);
236 }
237 spin_unlock_irqrestore(&dto_lock, flags);
238}
239
240/*
241 * Receive Queue Completion Handler
242 *
243 * Since an RQ completion handler is called on interrupt context, we
244 * need to defer the handling of the I/O to a tasklet
245 */
246static void rq_comp_handler(struct ib_cq *cq, void *cq_context)
247{
248 struct svcxprt_rdma *xprt = cq_context;
249 unsigned long flags;
250
251 /* Guard against unconditional flush call for destroyed QP */
252 if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0)
253 return;
254
255 /*
256 * Set the bit regardless of whether or not it's on the list
257 * because it may be on the list already due to an SQ
258 * completion.
259 */
260 set_bit(RDMAXPRT_RQ_PENDING, &xprt->sc_flags);
261
262 /*
263 * If this transport is not already on the DTO transport queue,
264 * add it
265 */
266 spin_lock_irqsave(&dto_lock, flags);
267 if (list_empty(&xprt->sc_dto_q)) {
268 svc_xprt_get(&xprt->sc_xprt);
269 list_add_tail(&xprt->sc_dto_q, &dto_xprt_q);
270 }
271 spin_unlock_irqrestore(&dto_lock, flags);
272
273 /* Tasklet does all the work to avoid irqsave locks. */
274 tasklet_schedule(&dto_tasklet);
275}
276
277/*
278 * rq_cq_reap - Process the RQ CQ.
279 *
280 * Take all completing WC off the CQE and enqueue the associated DTO
281 * context on the dto_q for the transport.
282 *
283 * Note that caller must hold a transport reference.
284 */
285static void rq_cq_reap(struct svcxprt_rdma *xprt)
286{
287 int ret;
288 struct ib_wc wc;
289 struct svc_rdma_op_ctxt *ctxt = NULL;
290
291 if (!test_and_clear_bit(RDMAXPRT_RQ_PENDING, &xprt->sc_flags))
292 return;
293
294 ib_req_notify_cq(xprt->sc_rq_cq, IB_CQ_NEXT_COMP);
295 atomic_inc(&rdma_stat_rq_poll);
296
297 while ((ret = ib_poll_cq(xprt->sc_rq_cq, 1, &wc)) > 0) {
298 ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id;
299 ctxt->wc_status = wc.status;
300 ctxt->byte_len = wc.byte_len;
301 svc_rdma_unmap_dma(ctxt);
302 if (wc.status != IB_WC_SUCCESS) {
303 /* Close the transport */
304 dprintk("svcrdma: transport closing putting ctxt %p\n", ctxt);
305 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
306 svc_rdma_put_context(ctxt, 1);
307 svc_xprt_put(&xprt->sc_xprt);
308 continue;
309 }
310 spin_lock_bh(&xprt->sc_rq_dto_lock);
311 list_add_tail(&ctxt->dto_q, &xprt->sc_rq_dto_q);
312 spin_unlock_bh(&xprt->sc_rq_dto_lock);
313 svc_xprt_put(&xprt->sc_xprt);
314 }
315
316 if (ctxt)
317 atomic_inc(&rdma_stat_rq_prod);
318
319 set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
320 /*
321 * If data arrived before established event,
322 * don't enqueue. This defers RPC I/O until the
323 * RDMA connection is complete.
324 */
325 if (!test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags))
326 svc_xprt_enqueue(&xprt->sc_xprt);
327}
328
329/*
330 * Process a completion context
331 */
332static void process_context(struct svcxprt_rdma *xprt,
333 struct svc_rdma_op_ctxt *ctxt)
334{
335 svc_rdma_unmap_dma(ctxt);
336
337 switch (ctxt->wr_op) {
338 case IB_WR_SEND:
339 if (test_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags))
340 svc_rdma_put_frmr(xprt, ctxt->frmr);
341 svc_rdma_put_context(ctxt, 1);
342 break;
343
344 case IB_WR_RDMA_WRITE:
345 svc_rdma_put_context(ctxt, 0);
346 break;
347
348 case IB_WR_RDMA_READ:
349 case IB_WR_RDMA_READ_WITH_INV:
350 if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
351 struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;
352 BUG_ON(!read_hdr);
353 if (test_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags))
354 svc_rdma_put_frmr(xprt, ctxt->frmr);
355 spin_lock_bh(&xprt->sc_rq_dto_lock);
356 set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
357 list_add_tail(&read_hdr->dto_q,
358 &xprt->sc_read_complete_q);
359 spin_unlock_bh(&xprt->sc_rq_dto_lock);
360 svc_xprt_enqueue(&xprt->sc_xprt);
361 }
362 svc_rdma_put_context(ctxt, 0);
363 break;
364
365 default:
366 printk(KERN_ERR "svcrdma: unexpected completion type, "
367 "opcode=%d\n",
368 ctxt->wr_op);
369 break;
370 }
371}
372
373/*
374 * Send Queue Completion Handler - potentially called on interrupt context.
375 *
376 * Note that caller must hold a transport reference.
377 */
378static void sq_cq_reap(struct svcxprt_rdma *xprt)
379{
380 struct svc_rdma_op_ctxt *ctxt = NULL;
381 struct ib_wc wc;
382 struct ib_cq *cq = xprt->sc_sq_cq;
383 int ret;
384
385 if (!test_and_clear_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags))
386 return;
387
388 ib_req_notify_cq(xprt->sc_sq_cq, IB_CQ_NEXT_COMP);
389 atomic_inc(&rdma_stat_sq_poll);
390 while ((ret = ib_poll_cq(cq, 1, &wc)) > 0) {
391 if (wc.status != IB_WC_SUCCESS)
392 /* Close the transport */
393 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
394
395 /* Decrement used SQ WR count */
396 atomic_dec(&xprt->sc_sq_count);
397 wake_up(&xprt->sc_send_wait);
398
399 ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id;
400 if (ctxt)
401 process_context(xprt, ctxt);
402
403 svc_xprt_put(&xprt->sc_xprt);
404 }
405
406 if (ctxt)
407 atomic_inc(&rdma_stat_sq_prod);
408}
409
410static void sq_comp_handler(struct ib_cq *cq, void *cq_context)
411{
412 struct svcxprt_rdma *xprt = cq_context;
413 unsigned long flags;
414
415 /* Guard against unconditional flush call for destroyed QP */
416 if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0)
417 return;
418
419 /*
420 * Set the bit regardless of whether or not it's on the list
421 * because it may be on the list already due to an RQ
422 * completion.
423 */
424 set_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags);
425
426 /*
427 * If this transport is not already on the DTO transport queue,
428 * add it
429 */
430 spin_lock_irqsave(&dto_lock, flags);
431 if (list_empty(&xprt->sc_dto_q)) {
432 svc_xprt_get(&xprt->sc_xprt);
433 list_add_tail(&xprt->sc_dto_q, &dto_xprt_q);
434 }
435 spin_unlock_irqrestore(&dto_lock, flags);
436
437 /* Tasklet does all the work to avoid irqsave locks. */
438 tasklet_schedule(&dto_tasklet);
439}
440
441static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
442 int listener)
443{
444 struct svcxprt_rdma *cma_xprt = kzalloc(sizeof *cma_xprt, GFP_KERNEL);
445
446 if (!cma_xprt)
447 return NULL;
448 svc_xprt_init(&init_net, &svc_rdma_class, &cma_xprt->sc_xprt, serv);
449 INIT_LIST_HEAD(&cma_xprt->sc_accept_q);
450 INIT_LIST_HEAD(&cma_xprt->sc_dto_q);
451 INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
452 INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
453 INIT_LIST_HEAD(&cma_xprt->sc_frmr_q);
454 init_waitqueue_head(&cma_xprt->sc_send_wait);
455
456 spin_lock_init(&cma_xprt->sc_lock);
457 spin_lock_init(&cma_xprt->sc_rq_dto_lock);
458 spin_lock_init(&cma_xprt->sc_frmr_q_lock);
459
460 cma_xprt->sc_ord = svcrdma_ord;
461
462 cma_xprt->sc_max_req_size = svcrdma_max_req_size;
463 cma_xprt->sc_max_requests = svcrdma_max_requests;
464 cma_xprt->sc_sq_depth = svcrdma_max_requests * RPCRDMA_SQ_DEPTH_MULT;
465 atomic_set(&cma_xprt->sc_sq_count, 0);
466 atomic_set(&cma_xprt->sc_ctxt_used, 0);
467
468 if (listener)
469 set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
470
471 return cma_xprt;
472}
473
474struct page *svc_rdma_get_page(void)
475{
476 struct page *page;
477
478 while ((page = alloc_page(GFP_KERNEL)) == NULL) {
479 /* If we can't get memory, wait a bit and try again */
480 printk(KERN_INFO "svcrdma: out of memory...retrying in 1000 "
481 "jiffies.\n");
482 schedule_timeout_uninterruptible(msecs_to_jiffies(1000));
483 }
484 return page;
485}
486
487int svc_rdma_post_recv(struct svcxprt_rdma *xprt)
488{
489 struct ib_recv_wr recv_wr, *bad_recv_wr;
490 struct svc_rdma_op_ctxt *ctxt;
491 struct page *page;
492 dma_addr_t pa;
493 int sge_no;
494 int buflen;
495 int ret;
496
497 ctxt = svc_rdma_get_context(xprt);
498 buflen = 0;
499 ctxt->direction = DMA_FROM_DEVICE;
500 for (sge_no = 0; buflen < xprt->sc_max_req_size; sge_no++) {
501 BUG_ON(sge_no >= xprt->sc_max_sge);
502 page = svc_rdma_get_page();
503 ctxt->pages[sge_no] = page;
504 pa = ib_dma_map_page(xprt->sc_cm_id->device,
505 page, 0, PAGE_SIZE,
506 DMA_FROM_DEVICE);
507 if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa))
508 goto err_put_ctxt;
509 atomic_inc(&xprt->sc_dma_used);
510 ctxt->sge[sge_no].addr = pa;
511 ctxt->sge[sge_no].length = PAGE_SIZE;
512 ctxt->sge[sge_no].lkey = xprt->sc_dma_lkey;
513 ctxt->count = sge_no + 1;
514 buflen += PAGE_SIZE;
515 }
516 recv_wr.next = NULL;
517 recv_wr.sg_list = &ctxt->sge[0];
518 recv_wr.num_sge = ctxt->count;
519 recv_wr.wr_id = (u64)(unsigned long)ctxt;
520
521 svc_xprt_get(&xprt->sc_xprt);
522 ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr);
523 if (ret) {
524 svc_rdma_unmap_dma(ctxt);
525 svc_rdma_put_context(ctxt, 1);
526 svc_xprt_put(&xprt->sc_xprt);
527 }
528 return ret;
529
530 err_put_ctxt:
531 svc_rdma_unmap_dma(ctxt);
532 svc_rdma_put_context(ctxt, 1);
533 return -ENOMEM;
534}
535
536/*
537 * This function handles the CONNECT_REQUEST event on a listening
538 * endpoint. It is passed the cma_id for the _new_ connection. The context in
539 * this cma_id is inherited from the listening cma_id and is the svc_xprt
540 * structure for the listening endpoint.
541 *
542 * This function creates a new xprt for the new connection and enqueues it on
543 * the accept queue for the listent xprt. When the listen thread is kicked, it
544 * will call the recvfrom method on the listen xprt which will accept the new
545 * connection.
546 */
547static void handle_connect_req(struct rdma_cm_id *new_cma_id, size_t client_ird)
548{
549 struct svcxprt_rdma *listen_xprt = new_cma_id->context;
550 struct svcxprt_rdma *newxprt;
551 struct sockaddr *sa;
552
553 /* Create a new transport */
554 newxprt = rdma_create_xprt(listen_xprt->sc_xprt.xpt_server, 0);
555 if (!newxprt) {
556 dprintk("svcrdma: failed to create new transport\n");
557 return;
558 }
559 newxprt->sc_cm_id = new_cma_id;
560 new_cma_id->context = newxprt;
561 dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n",
562 newxprt, newxprt->sc_cm_id, listen_xprt);
563
564 /* Save client advertised inbound read limit for use later in accept. */
565 newxprt->sc_ord = client_ird;
566
567 /* Set the local and remote addresses in the transport */
568 sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
569 svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
570 sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
571 svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));
572
573 /*
574 * Enqueue the new transport on the accept queue of the listening
575 * transport
576 */
577 spin_lock_bh(&listen_xprt->sc_lock);
578 list_add_tail(&newxprt->sc_accept_q, &listen_xprt->sc_accept_q);
579 spin_unlock_bh(&listen_xprt->sc_lock);
580
581 /*
582 * Can't use svc_xprt_received here because we are not on a
583 * rqstp thread
584 */
585 set_bit(XPT_CONN, &listen_xprt->sc_xprt.xpt_flags);
586 svc_xprt_enqueue(&listen_xprt->sc_xprt);
587}
588
589/*
590 * Handles events generated on the listening endpoint. These events will be
591 * either be incoming connect requests or adapter removal events.
592 */
593static int rdma_listen_handler(struct rdma_cm_id *cma_id,
594 struct rdma_cm_event *event)
595{
596 struct svcxprt_rdma *xprt = cma_id->context;
597 int ret = 0;
598
599 switch (event->event) {
600 case RDMA_CM_EVENT_CONNECT_REQUEST:
601 dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
602 "event=%d\n", cma_id, cma_id->context, event->event);
603 handle_connect_req(cma_id,
604 event->param.conn.initiator_depth);
605 break;
606
607 case RDMA_CM_EVENT_ESTABLISHED:
608 /* Accept complete */
609 dprintk("svcrdma: Connection completed on LISTEN xprt=%p, "
610 "cm_id=%p\n", xprt, cma_id);
611 break;
612
613 case RDMA_CM_EVENT_DEVICE_REMOVAL:
614 dprintk("svcrdma: Device removal xprt=%p, cm_id=%p\n",
615 xprt, cma_id);
616 if (xprt)
617 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
618 break;
619
620 default:
621 dprintk("svcrdma: Unexpected event on listening endpoint %p, "
622 "event=%d\n", cma_id, event->event);
623 break;
624 }
625
626 return ret;
627}
628
629static int rdma_cma_handler(struct rdma_cm_id *cma_id,
630 struct rdma_cm_event *event)
631{
632 struct svc_xprt *xprt = cma_id->context;
633 struct svcxprt_rdma *rdma =
634 container_of(xprt, struct svcxprt_rdma, sc_xprt);
635 switch (event->event) {
636 case RDMA_CM_EVENT_ESTABLISHED:
637 /* Accept complete */
638 svc_xprt_get(xprt);
639 dprintk("svcrdma: Connection completed on DTO xprt=%p, "
640 "cm_id=%p\n", xprt, cma_id);
641 clear_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags);
642 svc_xprt_enqueue(xprt);
643 break;
644 case RDMA_CM_EVENT_DISCONNECTED:
645 dprintk("svcrdma: Disconnect on DTO xprt=%p, cm_id=%p\n",
646 xprt, cma_id);
647 if (xprt) {
648 set_bit(XPT_CLOSE, &xprt->xpt_flags);
649 svc_xprt_enqueue(xprt);
650 svc_xprt_put(xprt);
651 }
652 break;
653 case RDMA_CM_EVENT_DEVICE_REMOVAL:
654 dprintk("svcrdma: Device removal cma_id=%p, xprt = %p, "
655 "event=%d\n", cma_id, xprt, event->event);
656 if (xprt) {
657 set_bit(XPT_CLOSE, &xprt->xpt_flags);
658 svc_xprt_enqueue(xprt);
659 }
660 break;
661 default:
662 dprintk("svcrdma: Unexpected event on DTO endpoint %p, "
663 "event=%d\n", cma_id, event->event);
664 break;
665 }
666 return 0;
667}
668
669/*
670 * Create a listening RDMA service endpoint.
671 */
672static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
673 struct net *net,
674 struct sockaddr *sa, int salen,
675 int flags)
676{
677 struct rdma_cm_id *listen_id;
678 struct svcxprt_rdma *cma_xprt;
679 struct svc_xprt *xprt;
680 int ret;
681
682 dprintk("svcrdma: Creating RDMA socket\n");
683 if (sa->sa_family != AF_INET) {
684 dprintk("svcrdma: Address family %d is not supported.\n", sa->sa_family);
685 return ERR_PTR(-EAFNOSUPPORT);
686 }
687 cma_xprt = rdma_create_xprt(serv, 1);
688 if (!cma_xprt)
689 return ERR_PTR(-ENOMEM);
690 xprt = &cma_xprt->sc_xprt;
691
692 listen_id = rdma_create_id(rdma_listen_handler, cma_xprt, RDMA_PS_TCP,
693 IB_QPT_RC);
694 if (IS_ERR(listen_id)) {
695 ret = PTR_ERR(listen_id);
696 dprintk("svcrdma: rdma_create_id failed = %d\n", ret);
697 goto err0;
698 }
699
700 ret = rdma_bind_addr(listen_id, sa);
701 if (ret) {
702 dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
703 goto err1;
704 }
705 cma_xprt->sc_cm_id = listen_id;
706
707 ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
708 if (ret) {
709 dprintk("svcrdma: rdma_listen failed = %d\n", ret);
710 goto err1;
711 }
712
713 /*
714 * We need to use the address from the cm_id in case the
715 * caller specified 0 for the port number.
716 */
717 sa = (struct sockaddr *)&cma_xprt->sc_cm_id->route.addr.src_addr;
718 svc_xprt_set_local(&cma_xprt->sc_xprt, sa, salen);
719
720 return &cma_xprt->sc_xprt;
721
722 err1:
723 rdma_destroy_id(listen_id);
724 err0:
725 kfree(cma_xprt);
726 return ERR_PTR(ret);
727}
728
729static struct svc_rdma_fastreg_mr *rdma_alloc_frmr(struct svcxprt_rdma *xprt)
730{
731 struct ib_mr *mr;
732 struct ib_fast_reg_page_list *pl;
733 struct svc_rdma_fastreg_mr *frmr;
734
735 frmr = kmalloc(sizeof(*frmr), GFP_KERNEL);
736 if (!frmr)
737 goto err;
738
739 mr = ib_alloc_fast_reg_mr(xprt->sc_pd, RPCSVC_MAXPAGES);
740 if (IS_ERR(mr))
741 goto err_free_frmr;
742
743 pl = ib_alloc_fast_reg_page_list(xprt->sc_cm_id->device,
744 RPCSVC_MAXPAGES);
745 if (IS_ERR(pl))
746 goto err_free_mr;
747
748 frmr->mr = mr;
749 frmr->page_list = pl;
750 INIT_LIST_HEAD(&frmr->frmr_list);
751 return frmr;
752
753 err_free_mr:
754 ib_dereg_mr(mr);
755 err_free_frmr:
756 kfree(frmr);
757 err:
758 return ERR_PTR(-ENOMEM);
759}
760
761static void rdma_dealloc_frmr_q(struct svcxprt_rdma *xprt)
762{
763 struct svc_rdma_fastreg_mr *frmr;
764
765 while (!list_empty(&xprt->sc_frmr_q)) {
766 frmr = list_entry(xprt->sc_frmr_q.next,
767 struct svc_rdma_fastreg_mr, frmr_list);
768 list_del_init(&frmr->frmr_list);
769 ib_dereg_mr(frmr->mr);
770 ib_free_fast_reg_page_list(frmr->page_list);
771 kfree(frmr);
772 }
773}
774
775struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *rdma)
776{
777 struct svc_rdma_fastreg_mr *frmr = NULL;
778
779 spin_lock_bh(&rdma->sc_frmr_q_lock);
780 if (!list_empty(&rdma->sc_frmr_q)) {
781 frmr = list_entry(rdma->sc_frmr_q.next,
782 struct svc_rdma_fastreg_mr, frmr_list);
783 list_del_init(&frmr->frmr_list);
784 frmr->map_len = 0;
785 frmr->page_list_len = 0;
786 }
787 spin_unlock_bh(&rdma->sc_frmr_q_lock);
788 if (frmr)
789 return frmr;
790
791 return rdma_alloc_frmr(rdma);
792}
793
794static void frmr_unmap_dma(struct svcxprt_rdma *xprt,
795 struct svc_rdma_fastreg_mr *frmr)
796{
797 int page_no;
798 for (page_no = 0; page_no < frmr->page_list_len; page_no++) {
799 dma_addr_t addr = frmr->page_list->page_list[page_no];
800 if (ib_dma_mapping_error(frmr->mr->device, addr))
801 continue;
802 atomic_dec(&xprt->sc_dma_used);
803 ib_dma_unmap_page(frmr->mr->device, addr, PAGE_SIZE,
804 frmr->direction);
805 }
806}
807
808void svc_rdma_put_frmr(struct svcxprt_rdma *rdma,
809 struct svc_rdma_fastreg_mr *frmr)
810{
811 if (frmr) {
812 frmr_unmap_dma(rdma, frmr);
813 spin_lock_bh(&rdma->sc_frmr_q_lock);
814 BUG_ON(!list_empty(&frmr->frmr_list));
815 list_add(&frmr->frmr_list, &rdma->sc_frmr_q);
816 spin_unlock_bh(&rdma->sc_frmr_q_lock);
817 }
818}
819
820/*
821 * This is the xpo_recvfrom function for listening endpoints. Its
822 * purpose is to accept incoming connections. The CMA callback handler
823 * has already created a new transport and attached it to the new CMA
824 * ID.
825 *
826 * There is a queue of pending connections hung on the listening
827 * transport. This queue contains the new svc_xprt structure. This
828 * function takes svc_xprt structures off the accept_q and completes
829 * the connection.
830 */
831static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
832{
833 struct svcxprt_rdma *listen_rdma;
834 struct svcxprt_rdma *newxprt = NULL;
835 struct rdma_conn_param conn_param;
836 struct ib_qp_init_attr qp_attr;
837 struct ib_device_attr devattr;
838 int uninitialized_var(dma_mr_acc);
839 int need_dma_mr;
840 int ret;
841 int i;
842
843 listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
844 clear_bit(XPT_CONN, &xprt->xpt_flags);
845 /* Get the next entry off the accept list */
846 spin_lock_bh(&listen_rdma->sc_lock);
847 if (!list_empty(&listen_rdma->sc_accept_q)) {
848 newxprt = list_entry(listen_rdma->sc_accept_q.next,
849 struct svcxprt_rdma, sc_accept_q);
850 list_del_init(&newxprt->sc_accept_q);
851 }
852 if (!list_empty(&listen_rdma->sc_accept_q))
853 set_bit(XPT_CONN, &listen_rdma->sc_xprt.xpt_flags);
854 spin_unlock_bh(&listen_rdma->sc_lock);
855 if (!newxprt)
856 return NULL;
857
858 dprintk("svcrdma: newxprt from accept queue = %p, cm_id=%p\n",
859 newxprt, newxprt->sc_cm_id);
860
861 ret = ib_query_device(newxprt->sc_cm_id->device, &devattr);
862 if (ret) {
863 dprintk("svcrdma: could not query device attributes on "
864 "device %p, rc=%d\n", newxprt->sc_cm_id->device, ret);
865 goto errout;
866 }
867
868 /* Qualify the transport resource defaults with the
869 * capabilities of this particular device */
870 newxprt->sc_max_sge = min((size_t)devattr.max_sge,
871 (size_t)RPCSVC_MAXPAGES);
872 newxprt->sc_max_requests = min((size_t)devattr.max_qp_wr,
873 (size_t)svcrdma_max_requests);
874 newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_max_requests;
875
876 /*
877 * Limit ORD based on client limit, local device limit, and
878 * configured svcrdma limit.
879 */
880 newxprt->sc_ord = min_t(size_t, devattr.max_qp_rd_atom, newxprt->sc_ord);
881 newxprt->sc_ord = min_t(size_t, svcrdma_ord, newxprt->sc_ord);
882
883 newxprt->sc_pd = ib_alloc_pd(newxprt->sc_cm_id->device);
884 if (IS_ERR(newxprt->sc_pd)) {
885 dprintk("svcrdma: error creating PD for connect request\n");
886 goto errout;
887 }
888 newxprt->sc_sq_cq = ib_create_cq(newxprt->sc_cm_id->device,
889 sq_comp_handler,
890 cq_event_handler,
891 newxprt,
892 newxprt->sc_sq_depth,
893 0);
894 if (IS_ERR(newxprt->sc_sq_cq)) {
895 dprintk("svcrdma: error creating SQ CQ for connect request\n");
896 goto errout;
897 }
898 newxprt->sc_rq_cq = ib_create_cq(newxprt->sc_cm_id->device,
899 rq_comp_handler,
900 cq_event_handler,
901 newxprt,
902 newxprt->sc_max_requests,
903 0);
904 if (IS_ERR(newxprt->sc_rq_cq)) {
905 dprintk("svcrdma: error creating RQ CQ for connect request\n");
906 goto errout;
907 }
908
909 memset(&qp_attr, 0, sizeof qp_attr);
910 qp_attr.event_handler = qp_event_handler;
911 qp_attr.qp_context = &newxprt->sc_xprt;
912 qp_attr.cap.max_send_wr = newxprt->sc_sq_depth;
913 qp_attr.cap.max_recv_wr = newxprt->sc_max_requests;
914 qp_attr.cap.max_send_sge = newxprt->sc_max_sge;
915 qp_attr.cap.max_recv_sge = newxprt->sc_max_sge;
916 qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
917 qp_attr.qp_type = IB_QPT_RC;
918 qp_attr.send_cq = newxprt->sc_sq_cq;
919 qp_attr.recv_cq = newxprt->sc_rq_cq;
920 dprintk("svcrdma: newxprt->sc_cm_id=%p, newxprt->sc_pd=%p\n"
921 " cm_id->device=%p, sc_pd->device=%p\n"
922 " cap.max_send_wr = %d\n"
923 " cap.max_recv_wr = %d\n"
924 " cap.max_send_sge = %d\n"
925 " cap.max_recv_sge = %d\n",
926 newxprt->sc_cm_id, newxprt->sc_pd,
927 newxprt->sc_cm_id->device, newxprt->sc_pd->device,
928 qp_attr.cap.max_send_wr,
929 qp_attr.cap.max_recv_wr,
930 qp_attr.cap.max_send_sge,
931 qp_attr.cap.max_recv_sge);
932
933 ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd, &qp_attr);
934 if (ret) {
935 /*
936 * XXX: This is a hack. We need a xx_request_qp interface
937 * that will adjust the qp_attr's with a best-effort
938 * number
939 */
940 qp_attr.cap.max_send_sge -= 2;
941 qp_attr.cap.max_recv_sge -= 2;
942 ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd,
943 &qp_attr);
944 if (ret) {
945 dprintk("svcrdma: failed to create QP, ret=%d\n", ret);
946 goto errout;
947 }
948 newxprt->sc_max_sge = qp_attr.cap.max_send_sge;
949 newxprt->sc_max_sge = qp_attr.cap.max_recv_sge;
950 newxprt->sc_sq_depth = qp_attr.cap.max_send_wr;
951 newxprt->sc_max_requests = qp_attr.cap.max_recv_wr;
952 }
953 newxprt->sc_qp = newxprt->sc_cm_id->qp;
954
955 /*
956 * Use the most secure set of MR resources based on the
957 * transport type and available memory management features in
958 * the device. Here's the table implemented below:
959 *
960 * Fast Global DMA Remote WR
961 * Reg LKEY MR Access
962 * Sup'd Sup'd Needed Needed
963 *
964 * IWARP N N Y Y
965 * N Y Y Y
966 * Y N Y N
967 * Y Y N -
968 *
969 * IB N N Y N
970 * N Y N -
971 * Y N Y N
972 * Y Y N -
973 *
974 * NB: iWARP requires remote write access for the data sink
975 * of an RDMA_READ. IB does not.
976 */
977 if (devattr.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) {
978 newxprt->sc_frmr_pg_list_len =
979 devattr.max_fast_reg_page_list_len;
980 newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_FAST_REG;
981 }
982
983 /*
984 * Determine if a DMA MR is required and if so, what privs are required
985 */
986 switch (rdma_node_get_transport(newxprt->sc_cm_id->device->node_type)) {
987 case RDMA_TRANSPORT_IWARP:
988 newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_READ_W_INV;
989 if (!(newxprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG)) {
990 need_dma_mr = 1;
991 dma_mr_acc =
992 (IB_ACCESS_LOCAL_WRITE |
993 IB_ACCESS_REMOTE_WRITE);
994 } else if (!(devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
995 need_dma_mr = 1;
996 dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
997 } else
998 need_dma_mr = 0;
999 break;
1000 case RDMA_TRANSPORT_IB:
1001 if (!(devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
1002 need_dma_mr = 1;
1003 dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
1004 } else
1005 need_dma_mr = 0;
1006 break;
1007 default:
1008 goto errout;
1009 }
1010
1011 /* Create the DMA MR if needed, otherwise, use the DMA LKEY */
1012 if (need_dma_mr) {
1013 /* Register all of physical memory */
1014 newxprt->sc_phys_mr =
1015 ib_get_dma_mr(newxprt->sc_pd, dma_mr_acc);
1016 if (IS_ERR(newxprt->sc_phys_mr)) {
1017 dprintk("svcrdma: Failed to create DMA MR ret=%d\n",
1018 ret);
1019 goto errout;
1020 }
1021 newxprt->sc_dma_lkey = newxprt->sc_phys_mr->lkey;
1022 } else
1023 newxprt->sc_dma_lkey =
1024 newxprt->sc_cm_id->device->local_dma_lkey;
1025
1026 /* Post receive buffers */
1027 for (i = 0; i < newxprt->sc_max_requests; i++) {
1028 ret = svc_rdma_post_recv(newxprt);
1029 if (ret) {
1030 dprintk("svcrdma: failure posting receive buffers\n");
1031 goto errout;
1032 }
1033 }
1034
1035 /* Swap out the handler */
1036 newxprt->sc_cm_id->event_handler = rdma_cma_handler;
1037
1038 /*
1039 * Arm the CQs for the SQ and RQ before accepting so we can't
1040 * miss the first message
1041 */
1042 ib_req_notify_cq(newxprt->sc_sq_cq, IB_CQ_NEXT_COMP);
1043 ib_req_notify_cq(newxprt->sc_rq_cq, IB_CQ_NEXT_COMP);
1044
1045 /* Accept Connection */
1046 set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
1047 memset(&conn_param, 0, sizeof conn_param);
1048 conn_param.responder_resources = 0;
1049 conn_param.initiator_depth = newxprt->sc_ord;
1050 ret = rdma_accept(newxprt->sc_cm_id, &conn_param);
1051 if (ret) {
1052 dprintk("svcrdma: failed to accept new connection, ret=%d\n",
1053 ret);
1054 goto errout;
1055 }
1056
1057 dprintk("svcrdma: new connection %p accepted with the following "
1058 "attributes:\n"
1059 " local_ip : %pI4\n"
1060 " local_port : %d\n"
1061 " remote_ip : %pI4\n"
1062 " remote_port : %d\n"
1063 " max_sge : %d\n"
1064 " sq_depth : %d\n"
1065 " max_requests : %d\n"
1066 " ord : %d\n",
1067 newxprt,
1068 &((struct sockaddr_in *)&newxprt->sc_cm_id->
1069 route.addr.src_addr)->sin_addr.s_addr,
1070 ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
1071 route.addr.src_addr)->sin_port),
1072 &((struct sockaddr_in *)&newxprt->sc_cm_id->
1073 route.addr.dst_addr)->sin_addr.s_addr,
1074 ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
1075 route.addr.dst_addr)->sin_port),
1076 newxprt->sc_max_sge,
1077 newxprt->sc_sq_depth,
1078 newxprt->sc_max_requests,
1079 newxprt->sc_ord);
1080
1081 return &newxprt->sc_xprt;
1082
1083 errout:
1084 dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
1085 /* Take a reference in case the DTO handler runs */
1086 svc_xprt_get(&newxprt->sc_xprt);
1087 if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp))
1088 ib_destroy_qp(newxprt->sc_qp);
1089 rdma_destroy_id(newxprt->sc_cm_id);
1090 /* This call to put will destroy the transport */
1091 svc_xprt_put(&newxprt->sc_xprt);
1092 return NULL;
1093}
1094
1095static void svc_rdma_release_rqst(struct svc_rqst *rqstp)
1096{
1097}
1098
1099/*
1100 * When connected, an svc_xprt has at least two references:
1101 *
1102 * - A reference held by the cm_id between the ESTABLISHED and
1103 * DISCONNECTED events. If the remote peer disconnected first, this
1104 * reference could be gone.
1105 *
1106 * - A reference held by the svc_recv code that called this function
1107 * as part of close processing.
1108 *
1109 * At a minimum one references should still be held.
1110 */
1111static void svc_rdma_detach(struct svc_xprt *xprt)
1112{
1113 struct svcxprt_rdma *rdma =
1114 container_of(xprt, struct svcxprt_rdma, sc_xprt);
1115 dprintk("svc: svc_rdma_detach(%p)\n", xprt);
1116
1117 /* Disconnect and flush posted WQE */
1118 rdma_disconnect(rdma->sc_cm_id);
1119}
1120
1121static void __svc_rdma_free(struct work_struct *work)
1122{
1123 struct svcxprt_rdma *rdma =
1124 container_of(work, struct svcxprt_rdma, sc_work);
1125 dprintk("svcrdma: svc_rdma_free(%p)\n", rdma);
1126
1127 /* We should only be called from kref_put */
1128 BUG_ON(atomic_read(&rdma->sc_xprt.xpt_ref.refcount) != 0);
1129
1130 /*
1131 * Destroy queued, but not processed read completions. Note
1132 * that this cleanup has to be done before destroying the
1133 * cm_id because the device ptr is needed to unmap the dma in
1134 * svc_rdma_put_context.
1135 */
1136 while (!list_empty(&rdma->sc_read_complete_q)) {
1137 struct svc_rdma_op_ctxt *ctxt;
1138 ctxt = list_entry(rdma->sc_read_complete_q.next,
1139 struct svc_rdma_op_ctxt,
1140 dto_q);
1141 list_del_init(&ctxt->dto_q);
1142 svc_rdma_put_context(ctxt, 1);
1143 }
1144
1145 /* Destroy queued, but not processed recv completions */
1146 while (!list_empty(&rdma->sc_rq_dto_q)) {
1147 struct svc_rdma_op_ctxt *ctxt;
1148 ctxt = list_entry(rdma->sc_rq_dto_q.next,
1149 struct svc_rdma_op_ctxt,
1150 dto_q);
1151 list_del_init(&ctxt->dto_q);
1152 svc_rdma_put_context(ctxt, 1);
1153 }
1154
1155 /* Warn if we leaked a resource or under-referenced */
1156 WARN_ON(atomic_read(&rdma->sc_ctxt_used) != 0);
1157 WARN_ON(atomic_read(&rdma->sc_dma_used) != 0);
1158
1159 /* De-allocate fastreg mr */
1160 rdma_dealloc_frmr_q(rdma);
1161
1162 /* Destroy the QP if present (not a listener) */
1163 if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
1164 ib_destroy_qp(rdma->sc_qp);
1165
1166 if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
1167 ib_destroy_cq(rdma->sc_sq_cq);
1168
1169 if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
1170 ib_destroy_cq(rdma->sc_rq_cq);
1171
1172 if (rdma->sc_phys_mr && !IS_ERR(rdma->sc_phys_mr))
1173 ib_dereg_mr(rdma->sc_phys_mr);
1174
1175 if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
1176 ib_dealloc_pd(rdma->sc_pd);
1177
1178 /* Destroy the CM ID */
1179 rdma_destroy_id(rdma->sc_cm_id);
1180
1181 kfree(rdma);
1182}
1183
1184static void svc_rdma_free(struct svc_xprt *xprt)
1185{
1186 struct svcxprt_rdma *rdma =
1187 container_of(xprt, struct svcxprt_rdma, sc_xprt);
1188 INIT_WORK(&rdma->sc_work, __svc_rdma_free);
1189 queue_work(svc_rdma_wq, &rdma->sc_work);
1190}
1191
1192static int svc_rdma_has_wspace(struct svc_xprt *xprt)
1193{
1194 struct svcxprt_rdma *rdma =
1195 container_of(xprt, struct svcxprt_rdma, sc_xprt);
1196
1197 /*
1198 * If there are fewer SQ WR available than required to send a
1199 * simple response, return false.
1200 */
1201 if ((rdma->sc_sq_depth - atomic_read(&rdma->sc_sq_count) < 3))
1202 return 0;
1203
1204 /*
1205 * ...or there are already waiters on the SQ,
1206 * return false.
1207 */
1208 if (waitqueue_active(&rdma->sc_send_wait))
1209 return 0;
1210
1211 /* Otherwise return true. */
1212 return 1;
1213}
1214
1215/*
1216 * Attempt to register the kvec representing the RPC memory with the
1217 * device.
1218 *
1219 * Returns:
1220 * NULL : The device does not support fastreg or there were no more
1221 * fastreg mr.
1222 * frmr : The kvec register request was successfully posted.
1223 * <0 : An error was encountered attempting to register the kvec.
1224 */
1225int svc_rdma_fastreg(struct svcxprt_rdma *xprt,
1226 struct svc_rdma_fastreg_mr *frmr)
1227{
1228 struct ib_send_wr fastreg_wr;
1229 u8 key;
1230
1231 /* Bump the key */
1232 key = (u8)(frmr->mr->lkey & 0x000000FF);
1233 ib_update_fast_reg_key(frmr->mr, ++key);
1234
1235 /* Prepare FASTREG WR */
1236 memset(&fastreg_wr, 0, sizeof fastreg_wr);
1237 fastreg_wr.opcode = IB_WR_FAST_REG_MR;
1238 fastreg_wr.send_flags = IB_SEND_SIGNALED;
1239 fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
1240 fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
1241 fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
1242 fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
1243 fastreg_wr.wr.fast_reg.length = frmr->map_len;
1244 fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
1245 fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
1246 return svc_rdma_send(xprt, &fastreg_wr);
1247}
1248
1249int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr)
1250{
1251 struct ib_send_wr *bad_wr, *n_wr;
1252 int wr_count;
1253 int i;
1254 int ret;
1255
1256 if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1257 return -ENOTCONN;
1258
1259 BUG_ON(wr->send_flags != IB_SEND_SIGNALED);
1260 wr_count = 1;
1261 for (n_wr = wr->next; n_wr; n_wr = n_wr->next)
1262 wr_count++;
1263
1264 /* If the SQ is full, wait until an SQ entry is available */
1265 while (1) {
1266 spin_lock_bh(&xprt->sc_lock);
1267 if (xprt->sc_sq_depth < atomic_read(&xprt->sc_sq_count) + wr_count) {
1268 spin_unlock_bh(&xprt->sc_lock);
1269 atomic_inc(&rdma_stat_sq_starve);
1270
1271 /* See if we can opportunistically reap SQ WR to make room */
1272 sq_cq_reap(xprt);
1273
1274 /* Wait until SQ WR available if SQ still full */
1275 wait_event(xprt->sc_send_wait,
1276 atomic_read(&xprt->sc_sq_count) <
1277 xprt->sc_sq_depth);
1278 if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1279 return -ENOTCONN;
1280 continue;
1281 }
1282 /* Take a transport ref for each WR posted */
1283 for (i = 0; i < wr_count; i++)
1284 svc_xprt_get(&xprt->sc_xprt);
1285
1286 /* Bump used SQ WR count and post */
1287 atomic_add(wr_count, &xprt->sc_sq_count);
1288 ret = ib_post_send(xprt->sc_qp, wr, &bad_wr);
1289 if (ret) {
1290 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
1291 atomic_sub(wr_count, &xprt->sc_sq_count);
1292 for (i = 0; i < wr_count; i ++)
1293 svc_xprt_put(&xprt->sc_xprt);
1294 dprintk("svcrdma: failed to post SQ WR rc=%d, "
1295 "sc_sq_count=%d, sc_sq_depth=%d\n",
1296 ret, atomic_read(&xprt->sc_sq_count),
1297 xprt->sc_sq_depth);
1298 }
1299 spin_unlock_bh(&xprt->sc_lock);
1300 if (ret)
1301 wake_up(&xprt->sc_send_wait);
1302 break;
1303 }
1304 return ret;
1305}
1306
1307void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
1308 enum rpcrdma_errcode err)
1309{
1310 struct ib_send_wr err_wr;
1311 struct page *p;
1312 struct svc_rdma_op_ctxt *ctxt;
1313 u32 *va;
1314 int length;
1315 int ret;
1316
1317 p = svc_rdma_get_page();
1318 va = page_address(p);
1319
1320 /* XDR encode error */
1321 length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va);
1322
1323 ctxt = svc_rdma_get_context(xprt);
1324 ctxt->direction = DMA_FROM_DEVICE;
1325 ctxt->count = 1;
1326 ctxt->pages[0] = p;
1327
1328 /* Prepare SGE for local address */
1329 ctxt->sge[0].addr = ib_dma_map_page(xprt->sc_cm_id->device,
1330 p, 0, length, DMA_FROM_DEVICE);
1331 if (ib_dma_mapping_error(xprt->sc_cm_id->device, ctxt->sge[0].addr)) {
1332 put_page(p);
1333 svc_rdma_put_context(ctxt, 1);
1334 return;
1335 }
1336 atomic_inc(&xprt->sc_dma_used);
1337 ctxt->sge[0].lkey = xprt->sc_dma_lkey;
1338 ctxt->sge[0].length = length;
1339
1340 /* Prepare SEND WR */
1341 memset(&err_wr, 0, sizeof err_wr);
1342 ctxt->wr_op = IB_WR_SEND;
1343 err_wr.wr_id = (unsigned long)ctxt;
1344 err_wr.sg_list = ctxt->sge;
1345 err_wr.num_sge = 1;
1346 err_wr.opcode = IB_WR_SEND;
1347 err_wr.send_flags = IB_SEND_SIGNALED;
1348
1349 /* Post It */
1350 ret = svc_rdma_send(xprt, &err_wr);
1351 if (ret) {
1352 dprintk("svcrdma: Error %d posting send for protocol error\n",
1353 ret);
1354 svc_rdma_unmap_dma(ctxt);
1355 svc_rdma_put_context(ctxt, 1);
1356 }
1357}