svc_rdma_transport.c 37.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
/*
 * Copyright (c) 2005-2007 Network Appliance, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the BSD-type
 * license below:
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 *      Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *
 *      Redistributions in binary form must reproduce the above
 *      copyright notice, this list of conditions and the following
 *      disclaimer in the documentation and/or other materials provided
 *      with the distribution.
 *
 *      Neither the name of the Network Appliance, Inc. nor the names of
 *      its contributors may be used to endorse or promote products
 *      derived from this software without specific prior written
 *      permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * Author: Tom Tucker <tom@opengridcomputing.com>
 */

#include <linux/sunrpc/svc_xprt.h>
#include <linux/sunrpc/debug.h>
#include <linux/sunrpc/rpc_rdma.h>
#include <linux/spinlock.h>
#include <rdma/ib_verbs.h>
#include <rdma/rdma_cm.h>
#include <linux/sunrpc/svc_rdma.h>

#define RPCDBG_FACILITY	RPCDBG_SVCXPRT

static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
					struct sockaddr *sa, int salen,
					int flags);
static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt);
static void svc_rdma_release_rqst(struct svc_rqst *);
static void dto_tasklet_func(unsigned long data);
static void svc_rdma_detach(struct svc_xprt *xprt);
static void svc_rdma_free(struct svc_xprt *xprt);
static int svc_rdma_has_wspace(struct svc_xprt *xprt);
static void rq_cq_reap(struct svcxprt_rdma *xprt);
static void sq_cq_reap(struct svcxprt_rdma *xprt);

DECLARE_TASKLET(dto_tasklet, dto_tasklet_func, 0UL);
static DEFINE_SPINLOCK(dto_lock);
static LIST_HEAD(dto_xprt_q);

static struct svc_xprt_ops svc_rdma_ops = {
	.xpo_create = svc_rdma_create,
	.xpo_recvfrom = svc_rdma_recvfrom,
	.xpo_sendto = svc_rdma_sendto,
	.xpo_release_rqst = svc_rdma_release_rqst,
	.xpo_detach = svc_rdma_detach,
	.xpo_free = svc_rdma_free,
	.xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
	.xpo_has_wspace = svc_rdma_has_wspace,
	.xpo_accept = svc_rdma_accept,
};

struct svc_xprt_class svc_rdma_class = {
	.xcl_name = "rdma",
	.xcl_owner = THIS_MODULE,
	.xcl_ops = &svc_rdma_ops,
	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
};

87 88
/* WR context cache. Created in svc_rdma.c  */
extern struct kmem_cache *svc_rdma_ctxt_cachep;
89 90 91 92 93 94

struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
{
	struct svc_rdma_op_ctxt *ctxt;

	while (1) {
95 96 97 98
		ctxt = kmem_cache_alloc(svc_rdma_ctxt_cachep, GFP_KERNEL);
		if (ctxt)
			break;
		schedule_timeout_uninterruptible(msecs_to_jiffies(500));
99
	}
100 101 102
	ctxt->xprt = xprt;
	INIT_LIST_HEAD(&ctxt->dto_q);
	ctxt->count = 0;
T
Tom Tucker 已提交
103
	ctxt->frmr = NULL;
104
	atomic_inc(&xprt->sc_ctxt_used);
105 106 107
	return ctxt;
}

108
void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
109 110 111 112
{
	struct svcxprt_rdma *xprt = ctxt->xprt;
	int i;
	for (i = 0; i < ctxt->count && ctxt->sge[i].length; i++) {
T
Tom Tucker 已提交
113 114 115 116 117 118 119 120 121 122 123 124 125
		/*
		 * Unmap the DMA addr in the SGE if the lkey matches
		 * the sc_dma_lkey, otherwise, ignore it since it is
		 * an FRMR lkey and will be unmapped later when the
		 * last WR that uses it completes.
		 */
		if (ctxt->sge[i].lkey == xprt->sc_dma_lkey) {
			atomic_dec(&xprt->sc_dma_used);
			ib_dma_unmap_single(xprt->sc_cm_id->device,
					    ctxt->sge[i].addr,
					    ctxt->sge[i].length,
					    ctxt->direction);
		}
126 127 128
	}
}

129 130 131 132 133 134 135 136 137 138 139
void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
{
	struct svcxprt_rdma *xprt;
	int i;

	BUG_ON(!ctxt);
	xprt = ctxt->xprt;
	if (free_pages)
		for (i = 0; i < ctxt->count; i++)
			put_page(ctxt->pages[i]);

140
	kmem_cache_free(svc_rdma_ctxt_cachep, ctxt);
141
	atomic_dec(&xprt->sc_ctxt_used);
142 143
}

144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161
/* Temporary NFS request map cache. Created in svc_rdma.c  */
extern struct kmem_cache *svc_rdma_map_cachep;

/*
 * Temporary NFS req mappings are shared across all transport
 * instances. These are short lived and should be bounded by the number
 * of concurrent server threads * depth of the SQ.
 */
struct svc_rdma_req_map *svc_rdma_get_req_map(void)
{
	struct svc_rdma_req_map *map;
	while (1) {
		map = kmem_cache_alloc(svc_rdma_map_cachep, GFP_KERNEL);
		if (map)
			break;
		schedule_timeout_uninterruptible(msecs_to_jiffies(500));
	}
	map->count = 0;
T
Tom Tucker 已提交
162
	map->frmr = NULL;
163 164 165 166 167 168 169 170
	return map;
}

void svc_rdma_put_req_map(struct svc_rdma_req_map *map)
{
	kmem_cache_free(svc_rdma_map_cachep, map);
}

171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229
/* ib_cq event handler */
static void cq_event_handler(struct ib_event *event, void *context)
{
	struct svc_xprt *xprt = context;
	dprintk("svcrdma: received CQ event id=%d, context=%p\n",
		event->event, context);
	set_bit(XPT_CLOSE, &xprt->xpt_flags);
}

/* QP event handler */
static void qp_event_handler(struct ib_event *event, void *context)
{
	struct svc_xprt *xprt = context;

	switch (event->event) {
	/* These are considered benign events */
	case IB_EVENT_PATH_MIG:
	case IB_EVENT_COMM_EST:
	case IB_EVENT_SQ_DRAINED:
	case IB_EVENT_QP_LAST_WQE_REACHED:
		dprintk("svcrdma: QP event %d received for QP=%p\n",
			event->event, event->element.qp);
		break;
	/* These are considered fatal events */
	case IB_EVENT_PATH_MIG_ERR:
	case IB_EVENT_QP_FATAL:
	case IB_EVENT_QP_REQ_ERR:
	case IB_EVENT_QP_ACCESS_ERR:
	case IB_EVENT_DEVICE_FATAL:
	default:
		dprintk("svcrdma: QP ERROR event %d received for QP=%p, "
			"closing transport\n",
			event->event, event->element.qp);
		set_bit(XPT_CLOSE, &xprt->xpt_flags);
		break;
	}
}

/*
 * Data Transfer Operation Tasklet
 *
 * Walks a list of transports with I/O pending, removing entries as
 * they are added to the server's I/O pending list. Two bits indicate
 * if SQ, RQ, or both have I/O pending. The dto_lock is an irqsave
 * spinlock that serializes access to the transport list with the RQ
 * and SQ interrupt handlers.
 */
static void dto_tasklet_func(unsigned long data)
{
	struct svcxprt_rdma *xprt;
	unsigned long flags;

	spin_lock_irqsave(&dto_lock, flags);
	while (!list_empty(&dto_xprt_q)) {
		xprt = list_entry(dto_xprt_q.next,
				  struct svcxprt_rdma, sc_dto_q);
		list_del_init(&xprt->sc_dto_q);
		spin_unlock_irqrestore(&dto_lock, flags);

230 231
		rq_cq_reap(xprt);
		sq_cq_reap(xprt);
232

233
		svc_xprt_put(&xprt->sc_xprt);
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
		spin_lock_irqsave(&dto_lock, flags);
	}
	spin_unlock_irqrestore(&dto_lock, flags);
}

/*
 * Receive Queue Completion Handler
 *
 * Since an RQ completion handler is called on interrupt context, we
 * need to defer the handling of the I/O to a tasklet
 */
static void rq_comp_handler(struct ib_cq *cq, void *cq_context)
{
	struct svcxprt_rdma *xprt = cq_context;
	unsigned long flags;

250 251 252 253
	/* Guard against unconditional flush call for destroyed QP */
	if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0)
		return;

254 255 256 257
	/*
	 * Set the bit regardless of whether or not it's on the list
	 * because it may be on the list already due to an SQ
	 * completion.
258
	 */
259 260 261 262 263 264 265
	set_bit(RDMAXPRT_RQ_PENDING, &xprt->sc_flags);

	/*
	 * If this transport is not already on the DTO transport queue,
	 * add it
	 */
	spin_lock_irqsave(&dto_lock, flags);
266 267
	if (list_empty(&xprt->sc_dto_q)) {
		svc_xprt_get(&xprt->sc_xprt);
268
		list_add_tail(&xprt->sc_dto_q, &dto_xprt_q);
269
	}
270 271 272 273 274 275 276 277 278 279 280
	spin_unlock_irqrestore(&dto_lock, flags);

	/* Tasklet does all the work to avoid irqsave locks. */
	tasklet_schedule(&dto_tasklet);
}

/*
 * rq_cq_reap - Process the RQ CQ.
 *
 * Take all completing WC off the CQE and enqueue the associated DTO
 * context on the dto_q for the transport.
281 282
 *
 * Note that caller must hold a transport reference.
283 284 285 286 287 288 289
 */
static void rq_cq_reap(struct svcxprt_rdma *xprt)
{
	int ret;
	struct ib_wc wc;
	struct svc_rdma_op_ctxt *ctxt = NULL;

290 291 292 293
	if (!test_and_clear_bit(RDMAXPRT_RQ_PENDING, &xprt->sc_flags))
		return;

	ib_req_notify_cq(xprt->sc_rq_cq, IB_CQ_NEXT_COMP);
294 295 296 297 298 299
	atomic_inc(&rdma_stat_rq_poll);

	while ((ret = ib_poll_cq(xprt->sc_rq_cq, 1, &wc)) > 0) {
		ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id;
		ctxt->wc_status = wc.status;
		ctxt->byte_len = wc.byte_len;
300
		svc_rdma_unmap_dma(ctxt);
301 302
		if (wc.status != IB_WC_SUCCESS) {
			/* Close the transport */
303
			dprintk("svcrdma: transport closing putting ctxt %p\n", ctxt);
304 305
			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
			svc_rdma_put_context(ctxt, 1);
306
			svc_xprt_put(&xprt->sc_xprt);
307 308
			continue;
		}
309
		spin_lock_bh(&xprt->sc_rq_dto_lock);
310
		list_add_tail(&ctxt->dto_q, &xprt->sc_rq_dto_q);
311
		spin_unlock_bh(&xprt->sc_rq_dto_lock);
312
		svc_xprt_put(&xprt->sc_xprt);
313 314 315 316
	}

	if (ctxt)
		atomic_inc(&rdma_stat_rq_prod);
317 318 319 320 321 322 323 324 325

	set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
	/*
	 * If data arrived before established event,
	 * don't enqueue. This defers RPC I/O until the
	 * RDMA connection is complete.
	 */
	if (!test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags))
		svc_xprt_enqueue(&xprt->sc_xprt);
326 327
}

328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345
/*
 * Processs a completion context
 */
static void process_context(struct svcxprt_rdma *xprt,
			    struct svc_rdma_op_ctxt *ctxt)
{
	svc_rdma_unmap_dma(ctxt);

	switch (ctxt->wr_op) {
	case IB_WR_SEND:
		svc_rdma_put_context(ctxt, 1);
		break;

	case IB_WR_RDMA_WRITE:
		svc_rdma_put_context(ctxt, 0);
		break;

	case IB_WR_RDMA_READ:
346
	case IB_WR_RDMA_READ_WITH_INV:
347 348 349
		if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
			struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;
			BUG_ON(!read_hdr);
350 351
			if (test_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags))
				svc_rdma_put_frmr(xprt, ctxt->frmr);
352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369
			spin_lock_bh(&xprt->sc_rq_dto_lock);
			set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
			list_add_tail(&read_hdr->dto_q,
				      &xprt->sc_read_complete_q);
			spin_unlock_bh(&xprt->sc_rq_dto_lock);
			svc_xprt_enqueue(&xprt->sc_xprt);
		}
		svc_rdma_put_context(ctxt, 0);
		break;

	default:
		printk(KERN_ERR "svcrdma: unexpected completion type, "
		       "opcode=%d\n",
		       ctxt->wr_op);
		break;
	}
}

370 371
/*
 * Send Queue Completion Handler - potentially called on interrupt context.
372 373
 *
 * Note that caller must hold a transport reference.
374 375 376 377 378 379 380 381
 */
static void sq_cq_reap(struct svcxprt_rdma *xprt)
{
	struct svc_rdma_op_ctxt *ctxt = NULL;
	struct ib_wc wc;
	struct ib_cq *cq = xprt->sc_sq_cq;
	int ret;

382 383 384 385
	if (!test_and_clear_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags))
		return;

	ib_req_notify_cq(xprt->sc_sq_cq, IB_CQ_NEXT_COMP);
386 387 388 389 390 391 392 393 394 395
	atomic_inc(&rdma_stat_sq_poll);
	while ((ret = ib_poll_cq(cq, 1, &wc)) > 0) {
		if (wc.status != IB_WC_SUCCESS)
			/* Close the transport */
			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);

		/* Decrement used SQ WR count */
		atomic_dec(&xprt->sc_sq_count);
		wake_up(&xprt->sc_send_wait);

396 397 398
		ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id;
		if (ctxt)
			process_context(xprt, ctxt);
399

400
		svc_xprt_put(&xprt->sc_xprt);
401 402 403 404 405 406 407 408 409 410 411
	}

	if (ctxt)
		atomic_inc(&rdma_stat_sq_prod);
}

static void sq_comp_handler(struct ib_cq *cq, void *cq_context)
{
	struct svcxprt_rdma *xprt = cq_context;
	unsigned long flags;

412 413 414 415
	/* Guard against unconditional flush call for destroyed QP */
	if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0)
		return;

416 417 418 419
	/*
	 * Set the bit regardless of whether or not it's on the list
	 * because it may be on the list already due to an RQ
	 * completion.
420
	 */
421 422 423 424 425 426 427
	set_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags);

	/*
	 * If this transport is not already on the DTO transport queue,
	 * add it
	 */
	spin_lock_irqsave(&dto_lock, flags);
428 429
	if (list_empty(&xprt->sc_dto_q)) {
		svc_xprt_get(&xprt->sc_xprt);
430
		list_add_tail(&xprt->sc_dto_q, &dto_xprt_q);
431
	}
432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449
	spin_unlock_irqrestore(&dto_lock, flags);

	/* Tasklet does all the work to avoid irqsave locks. */
	tasklet_schedule(&dto_tasklet);
}

static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
					     int listener)
{
	struct svcxprt_rdma *cma_xprt = kzalloc(sizeof *cma_xprt, GFP_KERNEL);

	if (!cma_xprt)
		return NULL;
	svc_xprt_init(&svc_rdma_class, &cma_xprt->sc_xprt, serv);
	INIT_LIST_HEAD(&cma_xprt->sc_accept_q);
	INIT_LIST_HEAD(&cma_xprt->sc_dto_q);
	INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
	INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
T
Tom Tucker 已提交
450
	INIT_LIST_HEAD(&cma_xprt->sc_frmr_q);
451 452 453 454
	init_waitqueue_head(&cma_xprt->sc_send_wait);

	spin_lock_init(&cma_xprt->sc_lock);
	spin_lock_init(&cma_xprt->sc_rq_dto_lock);
T
Tom Tucker 已提交
455
	spin_lock_init(&cma_xprt->sc_frmr_q_lock);
456 457 458 459 460 461 462

	cma_xprt->sc_ord = svcrdma_ord;

	cma_xprt->sc_max_req_size = svcrdma_max_req_size;
	cma_xprt->sc_max_requests = svcrdma_max_requests;
	cma_xprt->sc_sq_depth = svcrdma_max_requests * RPCRDMA_SQ_DEPTH_MULT;
	atomic_set(&cma_xprt->sc_sq_count, 0);
463
	atomic_set(&cma_xprt->sc_ctxt_used, 0);
464

465
	if (listener)
466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488
		set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);

	return cma_xprt;
}

struct page *svc_rdma_get_page(void)
{
	struct page *page;

	while ((page = alloc_page(GFP_KERNEL)) == NULL) {
		/* If we can't get memory, wait a bit and try again */
		printk(KERN_INFO "svcrdma: out of memory...retrying in 1000 "
		       "jiffies.\n");
		schedule_timeout_uninterruptible(msecs_to_jiffies(1000));
	}
	return page;
}

int svc_rdma_post_recv(struct svcxprt_rdma *xprt)
{
	struct ib_recv_wr recv_wr, *bad_recv_wr;
	struct svc_rdma_op_ctxt *ctxt;
	struct page *page;
489
	dma_addr_t pa;
490 491 492 493 494 495 496 497 498 499 500 501 502 503
	int sge_no;
	int buflen;
	int ret;

	ctxt = svc_rdma_get_context(xprt);
	buflen = 0;
	ctxt->direction = DMA_FROM_DEVICE;
	for (sge_no = 0; buflen < xprt->sc_max_req_size; sge_no++) {
		BUG_ON(sge_no >= xprt->sc_max_sge);
		page = svc_rdma_get_page();
		ctxt->pages[sge_no] = page;
		pa = ib_dma_map_page(xprt->sc_cm_id->device,
				     page, 0, PAGE_SIZE,
				     DMA_FROM_DEVICE);
504 505 506
		if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa))
			goto err_put_ctxt;
		atomic_inc(&xprt->sc_dma_used);
507 508
		ctxt->sge[sge_no].addr = pa;
		ctxt->sge[sge_no].length = PAGE_SIZE;
509
		ctxt->sge[sge_no].lkey = xprt->sc_dma_lkey;
510 511 512 513 514 515 516 517
		buflen += PAGE_SIZE;
	}
	ctxt->count = sge_no;
	recv_wr.next = NULL;
	recv_wr.sg_list = &ctxt->sge[0];
	recv_wr.num_sge = ctxt->count;
	recv_wr.wr_id = (u64)(unsigned long)ctxt;

518
	svc_xprt_get(&xprt->sc_xprt);
519
	ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr);
520 521
	if (ret) {
		svc_xprt_put(&xprt->sc_xprt);
522
		svc_rdma_put_context(ctxt, 1);
523
	}
524
	return ret;
525 526 527 528

 err_put_ctxt:
	svc_rdma_put_context(ctxt, 1);
	return -ENOMEM;
529 530 531 532 533 534 535 536 537 538 539 540 541
}

/*
 * This function handles the CONNECT_REQUEST event on a listening
 * endpoint. It is passed the cma_id for the _new_ connection. The context in
 * this cma_id is inherited from the listening cma_id and is the svc_xprt
 * structure for the listening endpoint.
 *
 * This function creates a new xprt for the new connection and enqueues it on
 * the accept queue for the listent xprt. When the listen thread is kicked, it
 * will call the recvfrom method on the listen xprt which will accept the new
 * connection.
 */
542
static void handle_connect_req(struct rdma_cm_id *new_cma_id, size_t client_ird)
543 544 545
{
	struct svcxprt_rdma *listen_xprt = new_cma_id->context;
	struct svcxprt_rdma *newxprt;
546
	struct sockaddr *sa;
547 548 549 550 551 552 553 554 555 556 557 558

	/* Create a new transport */
	newxprt = rdma_create_xprt(listen_xprt->sc_xprt.xpt_server, 0);
	if (!newxprt) {
		dprintk("svcrdma: failed to create new transport\n");
		return;
	}
	newxprt->sc_cm_id = new_cma_id;
	new_cma_id->context = newxprt;
	dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n",
		newxprt, newxprt->sc_cm_id, listen_xprt);

559 560 561
	/* Save client advertised inbound read limit for use later in accept. */
	newxprt->sc_ord = client_ird;

562 563 564 565 566 567
	/* Set the local and remote addresses in the transport */
	sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
	svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
	sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
	svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));

568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597
	/*
	 * Enqueue the new transport on the accept queue of the listening
	 * transport
	 */
	spin_lock_bh(&listen_xprt->sc_lock);
	list_add_tail(&newxprt->sc_accept_q, &listen_xprt->sc_accept_q);
	spin_unlock_bh(&listen_xprt->sc_lock);

	/*
	 * Can't use svc_xprt_received here because we are not on a
	 * rqstp thread
	*/
	set_bit(XPT_CONN, &listen_xprt->sc_xprt.xpt_flags);
	svc_xprt_enqueue(&listen_xprt->sc_xprt);
}

/*
 * Handles events generated on the listening endpoint. These events will be
 * either be incoming connect requests or adapter removal  events.
 */
static int rdma_listen_handler(struct rdma_cm_id *cma_id,
			       struct rdma_cm_event *event)
{
	struct svcxprt_rdma *xprt = cma_id->context;
	int ret = 0;

	switch (event->event) {
	case RDMA_CM_EVENT_CONNECT_REQUEST:
		dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
			"event=%d\n", cma_id, cma_id->context, event->event);
598 599
		handle_connect_req(cma_id,
				   event->param.conn.responder_resources);
600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632
		break;

	case RDMA_CM_EVENT_ESTABLISHED:
		/* Accept complete */
		dprintk("svcrdma: Connection completed on LISTEN xprt=%p, "
			"cm_id=%p\n", xprt, cma_id);
		break;

	case RDMA_CM_EVENT_DEVICE_REMOVAL:
		dprintk("svcrdma: Device removal xprt=%p, cm_id=%p\n",
			xprt, cma_id);
		if (xprt)
			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
		break;

	default:
		dprintk("svcrdma: Unexpected event on listening endpoint %p, "
			"event=%d\n", cma_id, event->event);
		break;
	}

	return ret;
}

static int rdma_cma_handler(struct rdma_cm_id *cma_id,
			    struct rdma_cm_event *event)
{
	struct svc_xprt *xprt = cma_id->context;
	struct svcxprt_rdma *rdma =
		container_of(xprt, struct svcxprt_rdma, sc_xprt);
	switch (event->event) {
	case RDMA_CM_EVENT_ESTABLISHED:
		/* Accept complete */
633
		svc_xprt_get(xprt);
634 635 636 637 638 639 640 641 642 643 644
		dprintk("svcrdma: Connection completed on DTO xprt=%p, "
			"cm_id=%p\n", xprt, cma_id);
		clear_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags);
		svc_xprt_enqueue(xprt);
		break;
	case RDMA_CM_EVENT_DISCONNECTED:
		dprintk("svcrdma: Disconnect on DTO xprt=%p, cm_id=%p\n",
			xprt, cma_id);
		if (xprt) {
			set_bit(XPT_CLOSE, &xprt->xpt_flags);
			svc_xprt_enqueue(xprt);
645
			svc_xprt_put(xprt);
646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679
		}
		break;
	case RDMA_CM_EVENT_DEVICE_REMOVAL:
		dprintk("svcrdma: Device removal cma_id=%p, xprt = %p, "
			"event=%d\n", cma_id, xprt, event->event);
		if (xprt) {
			set_bit(XPT_CLOSE, &xprt->xpt_flags);
			svc_xprt_enqueue(xprt);
		}
		break;
	default:
		dprintk("svcrdma: Unexpected event on DTO endpoint %p, "
			"event=%d\n", cma_id, event->event);
		break;
	}
	return 0;
}

/*
 * Create a listening RDMA service endpoint.
 */
static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
					struct sockaddr *sa, int salen,
					int flags)
{
	struct rdma_cm_id *listen_id;
	struct svcxprt_rdma *cma_xprt;
	struct svc_xprt *xprt;
	int ret;

	dprintk("svcrdma: Creating RDMA socket\n");

	cma_xprt = rdma_create_xprt(serv, 1);
	if (!cma_xprt)
680
		return ERR_PTR(-ENOMEM);
681 682 683 684
	xprt = &cma_xprt->sc_xprt;

	listen_id = rdma_create_id(rdma_listen_handler, cma_xprt, RDMA_PS_TCP);
	if (IS_ERR(listen_id)) {
685 686 687
		ret = PTR_ERR(listen_id);
		dprintk("svcrdma: rdma_create_id failed = %d\n", ret);
		goto err0;
688
	}
689

690 691 692
	ret = rdma_bind_addr(listen_id, sa);
	if (ret) {
		dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
693
		goto err1;
694 695 696 697 698 699
	}
	cma_xprt->sc_cm_id = listen_id;

	ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
	if (ret) {
		dprintk("svcrdma: rdma_listen failed = %d\n", ret);
700
		goto err1;
701 702 703 704 705 706 707 708 709 710
	}

	/*
	 * We need to use the address from the cm_id in case the
	 * caller specified 0 for the port number.
	 */
	sa = (struct sockaddr *)&cma_xprt->sc_cm_id->route.addr.src_addr;
	svc_xprt_set_local(&cma_xprt->sc_xprt, sa, salen);

	return &cma_xprt->sc_xprt;
711 712 713 714 715 716

 err1:
	rdma_destroy_id(listen_id);
 err0:
	kfree(cma_xprt);
	return ERR_PTR(ret);
717 718
}

T
Tom Tucker 已提交
719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809
static struct svc_rdma_fastreg_mr *rdma_alloc_frmr(struct svcxprt_rdma *xprt)
{
	struct ib_mr *mr;
	struct ib_fast_reg_page_list *pl;
	struct svc_rdma_fastreg_mr *frmr;

	frmr = kmalloc(sizeof(*frmr), GFP_KERNEL);
	if (!frmr)
		goto err;

	mr = ib_alloc_fast_reg_mr(xprt->sc_pd, RPCSVC_MAXPAGES);
	if (!mr)
		goto err_free_frmr;

	pl = ib_alloc_fast_reg_page_list(xprt->sc_cm_id->device,
					 RPCSVC_MAXPAGES);
	if (!pl)
		goto err_free_mr;

	frmr->mr = mr;
	frmr->page_list = pl;
	INIT_LIST_HEAD(&frmr->frmr_list);
	return frmr;

 err_free_mr:
	ib_dereg_mr(mr);
 err_free_frmr:
	kfree(frmr);
 err:
	return ERR_PTR(-ENOMEM);
}

static void rdma_dealloc_frmr_q(struct svcxprt_rdma *xprt)
{
	struct svc_rdma_fastreg_mr *frmr;

	while (!list_empty(&xprt->sc_frmr_q)) {
		frmr = list_entry(xprt->sc_frmr_q.next,
				  struct svc_rdma_fastreg_mr, frmr_list);
		list_del_init(&frmr->frmr_list);
		ib_dereg_mr(frmr->mr);
		ib_free_fast_reg_page_list(frmr->page_list);
		kfree(frmr);
	}
}

struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *rdma)
{
	struct svc_rdma_fastreg_mr *frmr = NULL;

	spin_lock_bh(&rdma->sc_frmr_q_lock);
	if (!list_empty(&rdma->sc_frmr_q)) {
		frmr = list_entry(rdma->sc_frmr_q.next,
				  struct svc_rdma_fastreg_mr, frmr_list);
		list_del_init(&frmr->frmr_list);
		frmr->map_len = 0;
		frmr->page_list_len = 0;
	}
	spin_unlock_bh(&rdma->sc_frmr_q_lock);
	if (frmr)
		return frmr;

	return rdma_alloc_frmr(rdma);
}

static void frmr_unmap_dma(struct svcxprt_rdma *xprt,
			   struct svc_rdma_fastreg_mr *frmr)
{
	int page_no;
	for (page_no = 0; page_no < frmr->page_list_len; page_no++) {
		dma_addr_t addr = frmr->page_list->page_list[page_no];
		if (ib_dma_mapping_error(frmr->mr->device, addr))
			continue;
		atomic_dec(&xprt->sc_dma_used);
		ib_dma_unmap_single(frmr->mr->device, addr, PAGE_SIZE,
				    frmr->direction);
	}
}

void svc_rdma_put_frmr(struct svcxprt_rdma *rdma,
		       struct svc_rdma_fastreg_mr *frmr)
{
	if (frmr) {
		frmr_unmap_dma(rdma, frmr);
		spin_lock_bh(&rdma->sc_frmr_q_lock);
		BUG_ON(!list_empty(&frmr->frmr_list));
		list_add(&frmr->frmr_list, &rdma->sc_frmr_q);
		spin_unlock_bh(&rdma->sc_frmr_q_lock);
	}
}

810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827
/*
 * This is the xpo_recvfrom function for listening endpoints. Its
 * purpose is to accept incoming connections. The CMA callback handler
 * has already created a new transport and attached it to the new CMA
 * ID.
 *
 * There is a queue of pending connections hung on the listening
 * transport. This queue contains the new svc_xprt structure. This
 * function takes svc_xprt structures off the accept_q and completes
 * the connection.
 */
static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
{
	struct svcxprt_rdma *listen_rdma;
	struct svcxprt_rdma *newxprt = NULL;
	struct rdma_conn_param conn_param;
	struct ib_qp_init_attr qp_attr;
	struct ib_device_attr devattr;
828 829
	int dma_mr_acc;
	int need_dma_mr;
830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
	int ret;
	int i;

	listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
	clear_bit(XPT_CONN, &xprt->xpt_flags);
	/* Get the next entry off the accept list */
	spin_lock_bh(&listen_rdma->sc_lock);
	if (!list_empty(&listen_rdma->sc_accept_q)) {
		newxprt = list_entry(listen_rdma->sc_accept_q.next,
				     struct svcxprt_rdma, sc_accept_q);
		list_del_init(&newxprt->sc_accept_q);
	}
	if (!list_empty(&listen_rdma->sc_accept_q))
		set_bit(XPT_CONN, &listen_rdma->sc_xprt.xpt_flags);
	spin_unlock_bh(&listen_rdma->sc_lock);
	if (!newxprt)
		return NULL;

	dprintk("svcrdma: newxprt from accept queue = %p, cm_id=%p\n",
		newxprt, newxprt->sc_cm_id);

	ret = ib_query_device(newxprt->sc_cm_id->device, &devattr);
	if (ret) {
		dprintk("svcrdma: could not query device attributes on "
			"device %p, rc=%d\n", newxprt->sc_cm_id->device, ret);
		goto errout;
	}

	/* Qualify the transport resource defaults with the
	 * capabilities of this particular device */
	newxprt->sc_max_sge = min((size_t)devattr.max_sge,
				  (size_t)RPCSVC_MAXPAGES);
	newxprt->sc_max_requests = min((size_t)devattr.max_qp_wr,
				   (size_t)svcrdma_max_requests);
	newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_max_requests;

866 867 868 869 870 871
	/*
	 * Limit ORD based on client limit, local device limit, and
	 * configured svcrdma limit.
	 */
	newxprt->sc_ord = min_t(size_t, devattr.max_qp_rd_atom, newxprt->sc_ord);
	newxprt->sc_ord = min_t(size_t,	svcrdma_ord, newxprt->sc_ord);
872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944

	newxprt->sc_pd = ib_alloc_pd(newxprt->sc_cm_id->device);
	if (IS_ERR(newxprt->sc_pd)) {
		dprintk("svcrdma: error creating PD for connect request\n");
		goto errout;
	}
	newxprt->sc_sq_cq = ib_create_cq(newxprt->sc_cm_id->device,
					 sq_comp_handler,
					 cq_event_handler,
					 newxprt,
					 newxprt->sc_sq_depth,
					 0);
	if (IS_ERR(newxprt->sc_sq_cq)) {
		dprintk("svcrdma: error creating SQ CQ for connect request\n");
		goto errout;
	}
	newxprt->sc_rq_cq = ib_create_cq(newxprt->sc_cm_id->device,
					 rq_comp_handler,
					 cq_event_handler,
					 newxprt,
					 newxprt->sc_max_requests,
					 0);
	if (IS_ERR(newxprt->sc_rq_cq)) {
		dprintk("svcrdma: error creating RQ CQ for connect request\n");
		goto errout;
	}

	memset(&qp_attr, 0, sizeof qp_attr);
	qp_attr.event_handler = qp_event_handler;
	qp_attr.qp_context = &newxprt->sc_xprt;
	qp_attr.cap.max_send_wr = newxprt->sc_sq_depth;
	qp_attr.cap.max_recv_wr = newxprt->sc_max_requests;
	qp_attr.cap.max_send_sge = newxprt->sc_max_sge;
	qp_attr.cap.max_recv_sge = newxprt->sc_max_sge;
	qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
	qp_attr.qp_type = IB_QPT_RC;
	qp_attr.send_cq = newxprt->sc_sq_cq;
	qp_attr.recv_cq = newxprt->sc_rq_cq;
	dprintk("svcrdma: newxprt->sc_cm_id=%p, newxprt->sc_pd=%p\n"
		"    cm_id->device=%p, sc_pd->device=%p\n"
		"    cap.max_send_wr = %d\n"
		"    cap.max_recv_wr = %d\n"
		"    cap.max_send_sge = %d\n"
		"    cap.max_recv_sge = %d\n",
		newxprt->sc_cm_id, newxprt->sc_pd,
		newxprt->sc_cm_id->device, newxprt->sc_pd->device,
		qp_attr.cap.max_send_wr,
		qp_attr.cap.max_recv_wr,
		qp_attr.cap.max_send_sge,
		qp_attr.cap.max_recv_sge);

	ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd, &qp_attr);
	if (ret) {
		/*
		 * XXX: This is a hack. We need a xx_request_qp interface
		 * that will adjust the qp_attr's with a best-effort
		 * number
		 */
		qp_attr.cap.max_send_sge -= 2;
		qp_attr.cap.max_recv_sge -= 2;
		ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd,
				     &qp_attr);
		if (ret) {
			dprintk("svcrdma: failed to create QP, ret=%d\n", ret);
			goto errout;
		}
		newxprt->sc_max_sge = qp_attr.cap.max_send_sge;
		newxprt->sc_max_sge = qp_attr.cap.max_recv_sge;
		newxprt->sc_sq_depth = qp_attr.cap.max_send_wr;
		newxprt->sc_max_requests = qp_attr.cap.max_recv_wr;
	}
	newxprt->sc_qp = newxprt->sc_cm_id->qp;

945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
	/*
	 * Use the most secure set of MR resources based on the
	 * transport type and available memory management features in
	 * the device. Here's the table implemented below:
	 *
	 *		Fast	Global	DMA	Remote WR
	 *		Reg	LKEY	MR	Access
	 *		Sup'd	Sup'd	Needed	Needed
	 *
	 * IWARP	N	N	Y	Y
	 *		N	Y	Y	Y
	 *		Y	N	Y	N
	 *		Y	Y	N	-
	 *
	 * IB		N	N	Y	N
	 *		N	Y	N	-
	 *		Y	N	Y	N
	 *		Y	Y	N	-
	 *
	 * NB:	iWARP requires remote write access for the data sink
	 *	of an RDMA_READ. IB does not.
	 */
	if (devattr.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) {
		newxprt->sc_frmr_pg_list_len =
			devattr.max_fast_reg_page_list_len;
		newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_FAST_REG;
	}

	/*
	 * Determine if a DMA MR is required and if so, what privs are required
	 */
	switch (rdma_node_get_transport(newxprt->sc_cm_id->device->node_type)) {
	case RDMA_TRANSPORT_IWARP:
		newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_READ_W_INV;
		if (!(newxprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG)) {
			need_dma_mr = 1;
			dma_mr_acc =
				(IB_ACCESS_LOCAL_WRITE |
				 IB_ACCESS_REMOTE_WRITE);
		} else if (!(devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
			need_dma_mr = 1;
			dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
		} else
			need_dma_mr = 0;
		break;
	case RDMA_TRANSPORT_IB:
		if (!(devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
			need_dma_mr = 1;
			dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
		} else
			need_dma_mr = 0;
		break;
	default:
998 999 1000
		goto errout;
	}

1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015
	/* Create the DMA MR if needed, otherwise, use the DMA LKEY */
	if (need_dma_mr) {
		/* Register all of physical memory */
		newxprt->sc_phys_mr =
			ib_get_dma_mr(newxprt->sc_pd, dma_mr_acc);
		if (IS_ERR(newxprt->sc_phys_mr)) {
			dprintk("svcrdma: Failed to create DMA MR ret=%d\n",
				ret);
			goto errout;
		}
		newxprt->sc_dma_lkey = newxprt->sc_phys_mr->lkey;
	} else
		newxprt->sc_dma_lkey =
			newxprt->sc_cm_id->device->local_dma_lkey;

1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027
	/* Post receive buffers */
	for (i = 0; i < newxprt->sc_max_requests; i++) {
		ret = svc_rdma_post_recv(newxprt);
		if (ret) {
			dprintk("svcrdma: failure posting receive buffers\n");
			goto errout;
		}
	}

	/* Swap out the handler */
	newxprt->sc_cm_id->event_handler = rdma_cma_handler;

1028 1029 1030 1031 1032 1033 1034
	/*
	 * Arm the CQs for the SQ and RQ before accepting so we can't
	 * miss the first message
	 */
	ib_req_notify_cq(newxprt->sc_sq_cq, IB_CQ_NEXT_COMP);
	ib_req_notify_cq(newxprt->sc_rq_cq, IB_CQ_NEXT_COMP);

1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
	/* Accept Connection */
	set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
	memset(&conn_param, 0, sizeof conn_param);
	conn_param.responder_resources = 0;
	conn_param.initiator_depth = newxprt->sc_ord;
	ret = rdma_accept(newxprt->sc_cm_id, &conn_param);
	if (ret) {
		dprintk("svcrdma: failed to accept new connection, ret=%d\n",
		       ret);
		goto errout;
	}

	dprintk("svcrdma: new connection %p accepted with the following "
		"attributes:\n"
		"    local_ip        : %d.%d.%d.%d\n"
		"    local_port	     : %d\n"
		"    remote_ip       : %d.%d.%d.%d\n"
		"    remote_port     : %d\n"
		"    max_sge         : %d\n"
		"    sq_depth        : %d\n"
		"    max_requests    : %d\n"
		"    ord             : %d\n",
		newxprt,
		NIPQUAD(((struct sockaddr_in *)&newxprt->sc_cm_id->
			 route.addr.src_addr)->sin_addr.s_addr),
		ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
		       route.addr.src_addr)->sin_port),
		NIPQUAD(((struct sockaddr_in *)&newxprt->sc_cm_id->
			 route.addr.dst_addr)->sin_addr.s_addr),
		ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
		       route.addr.dst_addr)->sin_port),
		newxprt->sc_max_sge,
		newxprt->sc_sq_depth,
		newxprt->sc_max_requests,
		newxprt->sc_ord);

	return &newxprt->sc_xprt;

 errout:
	dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
1075 1076
	/* Take a reference in case the DTO handler runs */
	svc_xprt_get(&newxprt->sc_xprt);
1077
	if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp))
1078
		ib_destroy_qp(newxprt->sc_qp);
1079
	rdma_destroy_id(newxprt->sc_cm_id);
1080 1081
	/* This call to put will destroy the transport */
	svc_xprt_put(&newxprt->sc_xprt);
1082 1083 1084 1085 1086 1087 1088
	return NULL;
}

static void svc_rdma_release_rqst(struct svc_rqst *rqstp)
{
}

1089
/*
1090
 * When connected, an svc_xprt has at least two references:
1091 1092 1093 1094 1095 1096 1097 1098
 *
 * - A reference held by the cm_id between the ESTABLISHED and
 *   DISCONNECTED events. If the remote peer disconnected first, this
 *   reference could be gone.
 *
 * - A reference held by the svc_recv code that called this function
 *   as part of close processing.
 *
1099
 * At a minimum one references should still be held.
1100
 */
1101 1102 1103 1104 1105
static void svc_rdma_detach(struct svc_xprt *xprt)
{
	struct svcxprt_rdma *rdma =
		container_of(xprt, struct svcxprt_rdma, sc_xprt);
	dprintk("svc: svc_rdma_detach(%p)\n", xprt);
1106 1107

	/* Disconnect and flush posted WQE */
1108 1109 1110
	rdma_disconnect(rdma->sc_cm_id);
}

1111
static void __svc_rdma_free(struct work_struct *work)
1112
{
1113 1114
	struct svcxprt_rdma *rdma =
		container_of(work, struct svcxprt_rdma, sc_work);
1115
	dprintk("svcrdma: svc_rdma_free(%p)\n", rdma);
1116

1117
	/* We should only be called from kref_put */
1118 1119
	BUG_ON(atomic_read(&rdma->sc_xprt.xpt_ref.refcount) != 0);

1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146
	/*
	 * Destroy queued, but not processed read completions. Note
	 * that this cleanup has to be done before destroying the
	 * cm_id because the device ptr is needed to unmap the dma in
	 * svc_rdma_put_context.
	 */
	while (!list_empty(&rdma->sc_read_complete_q)) {
		struct svc_rdma_op_ctxt *ctxt;
		ctxt = list_entry(rdma->sc_read_complete_q.next,
				  struct svc_rdma_op_ctxt,
				  dto_q);
		list_del_init(&ctxt->dto_q);
		svc_rdma_put_context(ctxt, 1);
	}

	/* Destroy queued, but not processed recv completions */
	while (!list_empty(&rdma->sc_rq_dto_q)) {
		struct svc_rdma_op_ctxt *ctxt;
		ctxt = list_entry(rdma->sc_rq_dto_q.next,
				  struct svc_rdma_op_ctxt,
				  dto_q);
		list_del_init(&ctxt->dto_q);
		svc_rdma_put_context(ctxt, 1);
	}

	/* Warn if we leaked a resource or under-referenced */
	WARN_ON(atomic_read(&rdma->sc_ctxt_used) != 0);
1147
	WARN_ON(atomic_read(&rdma->sc_dma_used) != 0);
1148

T
Tom Tucker 已提交
1149 1150 1151
	/* De-allocate fastreg mr */
	rdma_dealloc_frmr_q(rdma);

1152 1153 1154 1155
	/* Destroy the QP if present (not a listener) */
	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
		ib_destroy_qp(rdma->sc_qp);

1156 1157
	if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
		ib_destroy_cq(rdma->sc_sq_cq);
1158

1159 1160
	if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
		ib_destroy_cq(rdma->sc_rq_cq);
1161

1162 1163
	if (rdma->sc_phys_mr && !IS_ERR(rdma->sc_phys_mr))
		ib_dereg_mr(rdma->sc_phys_mr);
1164

1165 1166
	if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
		ib_dealloc_pd(rdma->sc_pd);
1167

1168 1169 1170
	/* Destroy the CM ID */
	rdma_destroy_id(rdma->sc_cm_id);

1171
	kfree(rdma);
1172 1173
}

1174 1175 1176 1177 1178 1179 1180 1181
static void svc_rdma_free(struct svc_xprt *xprt)
{
	struct svcxprt_rdma *rdma =
		container_of(xprt, struct svcxprt_rdma, sc_xprt);
	INIT_WORK(&rdma->sc_work, __svc_rdma_free);
	schedule_work(&rdma->sc_work);
}

1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
static int svc_rdma_has_wspace(struct svc_xprt *xprt)
{
	struct svcxprt_rdma *rdma =
		container_of(xprt, struct svcxprt_rdma, sc_xprt);

	/*
	 * If there are fewer SQ WR available than required to send a
	 * simple response, return false.
	 */
	if ((rdma->sc_sq_depth - atomic_read(&rdma->sc_sq_count) < 3))
		return 0;

	/*
	 * ...or there are already waiters on the SQ,
	 * return false.
	 */
	if (waitqueue_active(&rdma->sc_send_wait))
		return 0;

	/* Otherwise return true. */
	return 1;
}

1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
/*
 * Attempt to register the kvec representing the RPC memory with the
 * device.
 *
 * Returns:
 *  NULL : The device does not support fastreg or there were no more
 *         fastreg mr.
 *  frmr : The kvec register request was successfully posted.
 *    <0 : An error was encountered attempting to register the kvec.
 */
int svc_rdma_fastreg(struct svcxprt_rdma *xprt,
		     struct svc_rdma_fastreg_mr *frmr)
{
	struct ib_send_wr fastreg_wr;
	u8 key;

	/* Bump the key */
	key = (u8)(frmr->mr->lkey & 0x000000FF);
	ib_update_fast_reg_key(frmr->mr, ++key);

	/* Prepare FASTREG WR */
	memset(&fastreg_wr, 0, sizeof fastreg_wr);
	fastreg_wr.opcode = IB_WR_FAST_REG_MR;
	fastreg_wr.send_flags = IB_SEND_SIGNALED;
	fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
	fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
	fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
	fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
	fastreg_wr.wr.fast_reg.length = frmr->map_len;
	fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
	fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
	return svc_rdma_send(xprt, &fastreg_wr);
}

1239 1240
int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr)
{
1241 1242 1243
	struct ib_send_wr *bad_wr, *n_wr;
	int wr_count;
	int i;
1244 1245 1246
	int ret;

	if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1247
		return -ENOTCONN;
1248 1249

	BUG_ON(wr->send_flags != IB_SEND_SIGNALED);
1250 1251 1252 1253
	wr_count = 1;
	for (n_wr = wr->next; n_wr; n_wr = n_wr->next)
		wr_count++;

1254 1255 1256
	/* If the SQ is full, wait until an SQ entry is available */
	while (1) {
		spin_lock_bh(&xprt->sc_lock);
1257
		if (xprt->sc_sq_depth < atomic_read(&xprt->sc_sq_count) + wr_count) {
1258 1259
			spin_unlock_bh(&xprt->sc_lock);
			atomic_inc(&rdma_stat_sq_starve);
1260 1261

			/* See if we can opportunistically reap SQ WR to make room */
1262 1263 1264 1265 1266 1267
			sq_cq_reap(xprt);

			/* Wait until SQ WR available if SQ still full */
			wait_event(xprt->sc_send_wait,
				   atomic_read(&xprt->sc_sq_count) <
				   xprt->sc_sq_depth);
1268 1269
			if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
				return 0;
1270 1271
			continue;
		}
1272 1273 1274 1275 1276 1277
		/* Take a transport ref for each WR posted */
		for (i = 0; i < wr_count; i++)
			svc_xprt_get(&xprt->sc_xprt);

		/* Bump used SQ WR count and post */
		atomic_add(wr_count, &xprt->sc_sq_count);
1278
		ret = ib_post_send(xprt->sc_qp, wr, &bad_wr);
1279 1280 1281 1282 1283
		if (ret) {
			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
			atomic_sub(wr_count, &xprt->sc_sq_count);
			for (i = 0; i < wr_count; i ++)
				svc_xprt_put(&xprt->sc_xprt);
1284 1285 1286 1287
			dprintk("svcrdma: failed to post SQ WR rc=%d, "
			       "sc_sq_count=%d, sc_sq_depth=%d\n",
			       ret, atomic_read(&xprt->sc_sq_count),
			       xprt->sc_sq_depth);
1288
		}
1289
		spin_unlock_bh(&xprt->sc_lock);
1290 1291
		if (ret)
			wake_up(&xprt->sc_send_wait);
1292 1293 1294 1295 1296
		break;
	}
	return ret;
}

1297 1298
void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
			 enum rpcrdma_errcode err)
1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314
{
	struct ib_send_wr err_wr;
	struct ib_sge sge;
	struct page *p;
	struct svc_rdma_op_ctxt *ctxt;
	u32 *va;
	int length;
	int ret;

	p = svc_rdma_get_page();
	va = page_address(p);

	/* XDR encode error */
	length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va);

	/* Prepare SGE for local address */
1315
	atomic_inc(&xprt->sc_dma_used);
1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336
	sge.addr = ib_dma_map_page(xprt->sc_cm_id->device,
				   p, 0, PAGE_SIZE, DMA_FROM_DEVICE);
	sge.lkey = xprt->sc_phys_mr->lkey;
	sge.length = length;

	ctxt = svc_rdma_get_context(xprt);
	ctxt->count = 1;
	ctxt->pages[0] = p;

	/* Prepare SEND WR */
	memset(&err_wr, 0, sizeof err_wr);
	ctxt->wr_op = IB_WR_SEND;
	err_wr.wr_id = (unsigned long)ctxt;
	err_wr.sg_list = &sge;
	err_wr.num_sge = 1;
	err_wr.opcode = IB_WR_SEND;
	err_wr.send_flags = IB_SEND_SIGNALED;

	/* Post It */
	ret = svc_rdma_send(xprt, &err_wr);
	if (ret) {
1337 1338
		dprintk("svcrdma: Error %d posting send for protocol error\n",
			ret);
1339 1340 1341
		svc_rdma_put_context(ctxt, 1);
	}
}