ib_srp.c 93.9 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
/*
 * Copyright (c) 2005 Cisco Systems.  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
 * OpenIB.org BSD 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.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

33
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34

35 36 37 38 39 40 41
#include <linux/module.h>
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/err.h>
#include <linux/string.h>
#include <linux/parser.h>
#include <linux/random.h>
42
#include <linux/jiffies.h>
43
#include <rdma/ib_cache.h>
44

A
Arun Sharma 已提交
45
#include <linux/atomic.h>
46 47 48 49

#include <scsi/scsi.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_dbg.h>
50
#include <scsi/scsi_tcq.h>
51
#include <scsi/srp.h>
52
#include <scsi/scsi_transport_srp.h>
53 54 55 56 57

#include "ib_srp.h"

#define DRV_NAME	"ib_srp"
#define PFX		DRV_NAME ": "
58 59
#define DRV_VERSION	"2.0"
#define DRV_RELDATE	"July 26, 2015"
60 61

MODULE_AUTHOR("Roland Dreier");
62
MODULE_DESCRIPTION("InfiniBand SCSI RDMA Protocol initiator");
63
MODULE_LICENSE("Dual BSD/GPL");
64 65
MODULE_VERSION(DRV_VERSION);
MODULE_INFO(release_date, DRV_RELDATE);
66

67 68
static unsigned int srp_sg_tablesize;
static unsigned int cmd_sg_entries;
69 70
static unsigned int indirect_sg_entries;
static bool allow_ext_sg;
71 72
static bool prefer_fr = true;
static bool register_always = true;
73
static int topspin_workarounds = 1;
74

75 76
module_param(srp_sg_tablesize, uint, 0444);
MODULE_PARM_DESC(srp_sg_tablesize, "Deprecated name for cmd_sg_entries");
77

78 79 80
module_param(cmd_sg_entries, uint, 0444);
MODULE_PARM_DESC(cmd_sg_entries,
		 "Default number of gather/scatter entries in the SRP command (default is 12, max 255)");
81

82 83 84 85 86 87 88 89
module_param(indirect_sg_entries, uint, 0444);
MODULE_PARM_DESC(indirect_sg_entries,
		 "Default max number of gather/scatter entries (default is 12, max is " __stringify(SCSI_MAX_SG_CHAIN_SEGMENTS) ")");

module_param(allow_ext_sg, bool, 0444);
MODULE_PARM_DESC(allow_ext_sg,
		  "Default behavior when there are more than cmd_sg_entries S/G entries after mapping; fails the request when false (default false)");

90 91 92 93
module_param(topspin_workarounds, int, 0444);
MODULE_PARM_DESC(topspin_workarounds,
		 "Enable workarounds for Topspin/Cisco SRP target bugs if != 0");

94 95 96 97
module_param(prefer_fr, bool, 0444);
MODULE_PARM_DESC(prefer_fr,
"Whether to use fast registration if both FMR and fast registration are supported");

98 99 100 101
module_param(register_always, bool, 0444);
MODULE_PARM_DESC(register_always,
		 "Use memory registration even for contiguous memory regions");

102
static const struct kernel_param_ops srp_tmo_ops;
103

104 105 106 107 108
static int srp_reconnect_delay = 10;
module_param_cb(reconnect_delay, &srp_tmo_ops, &srp_reconnect_delay,
		S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(reconnect_delay, "Time between successive reconnect attempts");

109 110 111 112 113 114 115 116
static int srp_fast_io_fail_tmo = 15;
module_param_cb(fast_io_fail_tmo, &srp_tmo_ops, &srp_fast_io_fail_tmo,
		S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(fast_io_fail_tmo,
		 "Number of seconds between the observation of a transport"
		 " layer error and failing all I/O. \"off\" means that this"
		 " functionality is disabled.");

117
static int srp_dev_loss_tmo = 600;
118 119 120 121 122 123 124 125 126 127
module_param_cb(dev_loss_tmo, &srp_tmo_ops, &srp_dev_loss_tmo,
		S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(dev_loss_tmo,
		 "Maximum number of seconds that the SRP transport should"
		 " insulate transport layer errors. After this time has been"
		 " exceeded the SCSI host is removed. Should be"
		 " between 1 and " __stringify(SCSI_DEVICE_BLOCK_MAX_TIMEOUT)
		 " if fast_io_fail_tmo has not been set. \"off\" means that"
		 " this functionality is disabled.");

B
Bart Van Assche 已提交
128 129 130 131 132
static unsigned ch_count;
module_param(ch_count, uint, 0444);
MODULE_PARM_DESC(ch_count,
		 "Number of RDMA channels to use for communication with an SRP target. Using more than one channel improves performance if the HCA supports multiple completion vectors. The default value is the minimum of four times the number of online CPU sockets and the number of completion vectors supported by the HCA.");

133
static void srp_add_one(struct ib_device *device);
134
static void srp_remove_one(struct ib_device *device, void *client_data);
C
Christoph Hellwig 已提交
135 136 137
static void srp_recv_done(struct ib_cq *cq, struct ib_wc *wc);
static void srp_handle_qp_err(struct ib_cq *cq, struct ib_wc *wc,
		const char *opname);
138 139
static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event);

140
static struct scsi_transport_template *ib_srp_transport_template;
141
static struct workqueue_struct *srp_remove_wq;
142

143 144 145 146 147 148
static struct ib_client srp_client = {
	.name   = "srp",
	.add    = srp_add_one,
	.remove = srp_remove_one
};

149 150
static struct ib_sa_client srp_sa_client;

151 152 153 154 155 156 157 158 159 160 161 162 163 164
static int srp_tmo_get(char *buffer, const struct kernel_param *kp)
{
	int tmo = *(int *)kp->arg;

	if (tmo >= 0)
		return sprintf(buffer, "%d", tmo);
	else
		return sprintf(buffer, "off");
}

static int srp_tmo_set(const char *val, const struct kernel_param *kp)
{
	int tmo, res;

165 166 167 168
	res = srp_parse_tmo(&tmo, val);
	if (res)
		goto out;

169 170 171 172 173
	if (kp->arg == &srp_reconnect_delay)
		res = srp_tmo_valid(tmo, srp_fast_io_fail_tmo,
				    srp_dev_loss_tmo);
	else if (kp->arg == &srp_fast_io_fail_tmo)
		res = srp_tmo_valid(srp_reconnect_delay, tmo, srp_dev_loss_tmo);
174
	else
175 176
		res = srp_tmo_valid(srp_reconnect_delay, srp_fast_io_fail_tmo,
				    tmo);
177 178 179 180 181 182 183 184
	if (res)
		goto out;
	*(int *)kp->arg = tmo;

out:
	return res;
}

185
static const struct kernel_param_ops srp_tmo_ops = {
186 187 188 189
	.get = srp_tmo_get,
	.set = srp_tmo_set,
};

190 191 192 193 194 195 196 197 198 199
static inline struct srp_target_port *host_to_target(struct Scsi_Host *host)
{
	return (struct srp_target_port *) host->hostdata;
}

static const char *srp_target_info(struct Scsi_Host *host)
{
	return host_to_target(host)->target_name;
}

200 201 202
static int srp_target_is_topspin(struct srp_target_port *target)
{
	static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad };
203
	static const u8 cisco_oui[3]   = { 0x00, 0x1b, 0x0d };
204 205

	return topspin_workarounds &&
206 207
		(!memcmp(&target->ioc_guid, topspin_oui, sizeof topspin_oui) ||
		 !memcmp(&target->ioc_guid, cisco_oui, sizeof cisco_oui));
208 209
}

210 211 212 213 214 215 216 217 218 219 220 221 222 223
static struct srp_iu *srp_alloc_iu(struct srp_host *host, size_t size,
				   gfp_t gfp_mask,
				   enum dma_data_direction direction)
{
	struct srp_iu *iu;

	iu = kmalloc(sizeof *iu, gfp_mask);
	if (!iu)
		goto out;

	iu->buf = kzalloc(size, gfp_mask);
	if (!iu->buf)
		goto out_free_iu;

224 225 226
	iu->dma = ib_dma_map_single(host->srp_dev->dev, iu->buf, size,
				    direction);
	if (ib_dma_mapping_error(host->srp_dev->dev, iu->dma))
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246
		goto out_free_buf;

	iu->size      = size;
	iu->direction = direction;

	return iu;

out_free_buf:
	kfree(iu->buf);
out_free_iu:
	kfree(iu);
out:
	return NULL;
}

static void srp_free_iu(struct srp_host *host, struct srp_iu *iu)
{
	if (!iu)
		return;

247 248
	ib_dma_unmap_single(host->srp_dev->dev, iu->dma, iu->size,
			    iu->direction);
249 250 251 252 253 254
	kfree(iu->buf);
	kfree(iu);
}

static void srp_qp_event(struct ib_event *event, void *context)
{
255 256
	pr_debug("QP event %s (%d)\n",
		 ib_event_msg(event->event), event->event);
257 258 259 260 261 262 263 264 265 266 267 268
}

static int srp_init_qp(struct srp_target_port *target,
		       struct ib_qp *qp)
{
	struct ib_qp_attr *attr;
	int ret;

	attr = kmalloc(sizeof *attr, GFP_KERNEL);
	if (!attr)
		return -ENOMEM;

269 270 271 272
	ret = ib_find_cached_pkey(target->srp_host->srp_dev->dev,
				  target->srp_host->port,
				  be16_to_cpu(target->pkey),
				  &attr->pkey_index);
273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291
	if (ret)
		goto out;

	attr->qp_state        = IB_QPS_INIT;
	attr->qp_access_flags = (IB_ACCESS_REMOTE_READ |
				    IB_ACCESS_REMOTE_WRITE);
	attr->port_num        = target->srp_host->port;

	ret = ib_modify_qp(qp, attr,
			   IB_QP_STATE		|
			   IB_QP_PKEY_INDEX	|
			   IB_QP_ACCESS_FLAGS	|
			   IB_QP_PORT);

out:
	kfree(attr);
	return ret;
}

292
static int srp_new_cm_id(struct srp_rdma_ch *ch)
D
David Dillow 已提交
293
{
294
	struct srp_target_port *target = ch->target;
D
David Dillow 已提交
295 296
	struct ib_cm_id *new_cm_id;

297
	new_cm_id = ib_create_cm_id(target->srp_host->srp_dev->dev,
298
				    srp_cm_handler, ch);
D
David Dillow 已提交
299 300 301
	if (IS_ERR(new_cm_id))
		return PTR_ERR(new_cm_id);

302 303 304 305 306 307 308
	if (ch->cm_id)
		ib_destroy_cm_id(ch->cm_id);
	ch->cm_id = new_cm_id;
	ch->path.sgid = target->sgid;
	ch->path.dgid = target->orig_dgid;
	ch->path.pkey = target->pkey;
	ch->path.service_id = target->service_id;
D
David Dillow 已提交
309 310 311 312

	return 0;
}

313 314 315 316 317 318
static struct ib_fmr_pool *srp_alloc_fmr_pool(struct srp_target_port *target)
{
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_fmr_pool_param fmr_param;

	memset(&fmr_param, 0, sizeof(fmr_param));
319
	fmr_param.pool_size	    = target->mr_pool_size;
320 321
	fmr_param.dirty_watermark   = fmr_param.pool_size / 4;
	fmr_param.cache		    = 1;
322 323
	fmr_param.max_pages_per_fmr = dev->max_pages_per_mr;
	fmr_param.page_shift	    = ilog2(dev->mr_page_size);
324 325 326 327 328 329 330
	fmr_param.access	    = (IB_ACCESS_LOCAL_WRITE |
				       IB_ACCESS_REMOTE_WRITE |
				       IB_ACCESS_REMOTE_READ);

	return ib_create_fmr_pool(dev->pd, &fmr_param);
}

331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
/**
 * srp_destroy_fr_pool() - free the resources owned by a pool
 * @pool: Fast registration pool to be destroyed.
 */
static void srp_destroy_fr_pool(struct srp_fr_pool *pool)
{
	int i;
	struct srp_fr_desc *d;

	if (!pool)
		return;

	for (i = 0, d = &pool->desc[0]; i < pool->size; i++, d++) {
		if (d->mr)
			ib_dereg_mr(d->mr);
	}
	kfree(pool);
}

/**
 * srp_create_fr_pool() - allocate and initialize a pool for fast registration
 * @device:            IB device to allocate fast registration descriptors for.
 * @pd:                Protection domain associated with the FR descriptors.
 * @pool_size:         Number of descriptors to allocate.
 * @max_page_list_len: Maximum fast registration work request page list length.
 */
static struct srp_fr_pool *srp_create_fr_pool(struct ib_device *device,
					      struct ib_pd *pd, int pool_size,
					      int max_page_list_len)
{
	struct srp_fr_pool *pool;
	struct srp_fr_desc *d;
	struct ib_mr *mr;
	int i, ret = -EINVAL;

	if (pool_size <= 0)
		goto err;
	ret = -ENOMEM;
	pool = kzalloc(sizeof(struct srp_fr_pool) +
		       pool_size * sizeof(struct srp_fr_desc), GFP_KERNEL);
	if (!pool)
		goto err;
	pool->size = pool_size;
	pool->max_page_list_len = max_page_list_len;
	spin_lock_init(&pool->lock);
	INIT_LIST_HEAD(&pool->free_list);

	for (i = 0, d = &pool->desc[0]; i < pool->size; i++, d++) {
S
Sagi Grimberg 已提交
379 380
		mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG,
				 max_page_list_len);
381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
		if (IS_ERR(mr)) {
			ret = PTR_ERR(mr);
			goto destroy_pool;
		}
		d->mr = mr;
		list_add_tail(&d->entry, &pool->free_list);
	}

out:
	return pool;

destroy_pool:
	srp_destroy_fr_pool(pool);

err:
	pool = ERR_PTR(ret);
	goto out;
}

/**
 * srp_fr_pool_get() - obtain a descriptor suitable for fast registration
 * @pool: Pool to obtain descriptor from.
 */
static struct srp_fr_desc *srp_fr_pool_get(struct srp_fr_pool *pool)
{
	struct srp_fr_desc *d = NULL;
	unsigned long flags;

	spin_lock_irqsave(&pool->lock, flags);
	if (!list_empty(&pool->free_list)) {
		d = list_first_entry(&pool->free_list, typeof(*d), entry);
		list_del(&d->entry);
	}
	spin_unlock_irqrestore(&pool->lock, flags);

	return d;
}

/**
 * srp_fr_pool_put() - put an FR descriptor back in the free list
 * @pool: Pool the descriptor was allocated from.
 * @desc: Pointer to an array of fast registration descriptor pointers.
 * @n:    Number of descriptors to put back.
 *
 * Note: The caller must already have queued an invalidation request for
 * desc->mr->rkey before calling this function.
 */
static void srp_fr_pool_put(struct srp_fr_pool *pool, struct srp_fr_desc **desc,
			    int n)
{
	unsigned long flags;
	int i;

	spin_lock_irqsave(&pool->lock, flags);
	for (i = 0; i < n; i++)
		list_add(&desc[i]->entry, &pool->free_list);
	spin_unlock_irqrestore(&pool->lock, flags);
}

static struct srp_fr_pool *srp_alloc_fr_pool(struct srp_target_port *target)
{
	struct srp_device *dev = target->srp_host->srp_dev;

444
	return srp_create_fr_pool(dev->dev, dev->pd, target->mr_pool_size,
445 446 447
				  dev->max_pages_per_mr);
}

448 449
/**
 * srp_destroy_qp() - destroy an RDMA queue pair
450
 * @qp: RDMA queue pair.
451
 *
S
Steve Wise 已提交
452 453
 * Drain the qp before destroying it.  This avoids that the receive
 * completion handler can access the queue pair while it is
454 455
 * being destroyed.
 */
456
static void srp_destroy_qp(struct ib_qp *qp)
457
{
458 459
	ib_drain_rq(qp);
	ib_destroy_qp(qp);
460 461
}

462
static int srp_create_ch_ib(struct srp_rdma_ch *ch)
463
{
464
	struct srp_target_port *target = ch->target;
465
	struct srp_device *dev = target->srp_host->srp_dev;
466
	struct ib_qp_init_attr *init_attr;
467 468
	struct ib_cq *recv_cq, *send_cq;
	struct ib_qp *qp;
469
	struct ib_fmr_pool *fmr_pool = NULL;
470
	struct srp_fr_pool *fr_pool = NULL;
471
	const int m = dev->use_fast_reg ? 3 : 1;
472 473 474 475 476 477
	int ret;

	init_attr = kzalloc(sizeof *init_attr, GFP_KERNEL);
	if (!init_attr)
		return -ENOMEM;

S
Steve Wise 已提交
478
	/* queue_size + 1 for ib_drain_rq() */
C
Christoph Hellwig 已提交
479 480
	recv_cq = ib_alloc_cq(dev->dev, ch, target->queue_size + 1,
				ch->comp_vector, IB_POLL_SOFTIRQ);
481 482
	if (IS_ERR(recv_cq)) {
		ret = PTR_ERR(recv_cq);
483
		goto err;
484 485
	}

C
Christoph Hellwig 已提交
486 487
	send_cq = ib_alloc_cq(dev->dev, ch, m * target->queue_size,
				ch->comp_vector, IB_POLL_DIRECT);
488 489
	if (IS_ERR(send_cq)) {
		ret = PTR_ERR(send_cq);
490
		goto err_recv_cq;
491 492
	}

493
	init_attr->event_handler       = srp_qp_event;
494
	init_attr->cap.max_send_wr     = m * target->queue_size;
495
	init_attr->cap.max_recv_wr     = target->queue_size + 1;
496 497
	init_attr->cap.max_recv_sge    = 1;
	init_attr->cap.max_send_sge    = 1;
498
	init_attr->sq_sig_type         = IB_SIGNAL_REQ_WR;
499
	init_attr->qp_type             = IB_QPT_RC;
500 501
	init_attr->send_cq             = send_cq;
	init_attr->recv_cq             = recv_cq;
502

503
	qp = ib_create_qp(dev->pd, init_attr);
504 505
	if (IS_ERR(qp)) {
		ret = PTR_ERR(qp);
506
		goto err_send_cq;
507 508
	}

509
	ret = srp_init_qp(target, qp);
510 511
	if (ret)
		goto err_qp;
512

513
	if (dev->use_fast_reg) {
514 515 516 517 518 519 520
		fr_pool = srp_alloc_fr_pool(target);
		if (IS_ERR(fr_pool)) {
			ret = PTR_ERR(fr_pool);
			shost_printk(KERN_WARNING, target->scsi_host, PFX
				     "FR pool allocation failed (%d)\n", ret);
			goto err_qp;
		}
521
	} else if (dev->use_fmr) {
522 523 524 525 526 527 528 529 530
		fmr_pool = srp_alloc_fmr_pool(target);
		if (IS_ERR(fmr_pool)) {
			ret = PTR_ERR(fmr_pool);
			shost_printk(KERN_WARNING, target->scsi_host, PFX
				     "FMR pool allocation failed (%d)\n", ret);
			goto err_qp;
		}
	}

531
	if (ch->qp)
532
		srp_destroy_qp(ch->qp);
533
	if (ch->recv_cq)
C
Christoph Hellwig 已提交
534
		ib_free_cq(ch->recv_cq);
535
	if (ch->send_cq)
C
Christoph Hellwig 已提交
536
		ib_free_cq(ch->send_cq);
537

538 539 540
	ch->qp = qp;
	ch->recv_cq = recv_cq;
	ch->send_cq = send_cq;
541

542 543 544 545 546 547 548 549 550 551
	if (dev->use_fast_reg) {
		if (ch->fr_pool)
			srp_destroy_fr_pool(ch->fr_pool);
		ch->fr_pool = fr_pool;
	} else if (dev->use_fmr) {
		if (ch->fmr_pool)
			ib_destroy_fmr_pool(ch->fmr_pool);
		ch->fmr_pool = fmr_pool;
	}

552 553 554 555
	kfree(init_attr);
	return 0;

err_qp:
556
	srp_destroy_qp(qp);
557 558

err_send_cq:
C
Christoph Hellwig 已提交
559
	ib_free_cq(send_cq);
560 561

err_recv_cq:
C
Christoph Hellwig 已提交
562
	ib_free_cq(recv_cq);
563 564

err:
565 566 567 568
	kfree(init_attr);
	return ret;
}

569 570
/*
 * Note: this function may be called without srp_alloc_iu_bufs() having been
571
 * invoked. Hence the ch->[rt]x_ring checks.
572
 */
573 574
static void srp_free_ch_ib(struct srp_target_port *target,
			   struct srp_rdma_ch *ch)
575
{
576
	struct srp_device *dev = target->srp_host->srp_dev;
577 578
	int i;

B
Bart Van Assche 已提交
579 580 581
	if (!ch->target)
		return;

582 583 584
	if (ch->cm_id) {
		ib_destroy_cm_id(ch->cm_id);
		ch->cm_id = NULL;
585 586
	}

B
Bart Van Assche 已提交
587 588 589 590
	/* If srp_new_cm_id() succeeded but srp_create_ch_ib() not, return. */
	if (!ch->qp)
		return;

591
	if (dev->use_fast_reg) {
592 593
		if (ch->fr_pool)
			srp_destroy_fr_pool(ch->fr_pool);
594
	} else if (dev->use_fmr) {
595 596
		if (ch->fmr_pool)
			ib_destroy_fmr_pool(ch->fmr_pool);
597
	}
C
Christoph Hellwig 已提交
598

599
	srp_destroy_qp(ch->qp);
C
Christoph Hellwig 已提交
600 601
	ib_free_cq(ch->send_cq);
	ib_free_cq(ch->recv_cq);
602

B
Bart Van Assche 已提交
603 604 605 606 607 608 609 610
	/*
	 * Avoid that the SCSI error handler tries to use this channel after
	 * it has been freed. The SCSI error handler can namely continue
	 * trying to perform recovery actions after scsi_remove_host()
	 * returned.
	 */
	ch->target = NULL;

611 612
	ch->qp = NULL;
	ch->send_cq = ch->recv_cq = NULL;
613

614
	if (ch->rx_ring) {
615
		for (i = 0; i < target->queue_size; ++i)
616 617 618
			srp_free_iu(target->srp_host, ch->rx_ring[i]);
		kfree(ch->rx_ring);
		ch->rx_ring = NULL;
619
	}
620
	if (ch->tx_ring) {
621
		for (i = 0; i < target->queue_size; ++i)
622 623 624
			srp_free_iu(target->srp_host, ch->tx_ring[i]);
		kfree(ch->tx_ring);
		ch->tx_ring = NULL;
625
	}
626 627 628 629
}

static void srp_path_rec_completion(int status,
				    struct ib_sa_path_rec *pathrec,
630
				    void *ch_ptr)
631
{
632 633
	struct srp_rdma_ch *ch = ch_ptr;
	struct srp_target_port *target = ch->target;
634

635
	ch->status = status;
636
	if (status)
637 638
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "Got failed path rec status %d\n", status);
639
	else
640 641
		ch->path = *pathrec;
	complete(&ch->done);
642 643
}

644
static int srp_lookup_path(struct srp_rdma_ch *ch)
645
{
646
	struct srp_target_port *target = ch->target;
647 648
	int ret;

649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669
	ch->path.numb_path = 1;

	init_completion(&ch->done);

	ch->path_query_id = ib_sa_path_rec_get(&srp_sa_client,
					       target->srp_host->srp_dev->dev,
					       target->srp_host->port,
					       &ch->path,
					       IB_SA_PATH_REC_SERVICE_ID |
					       IB_SA_PATH_REC_DGID	 |
					       IB_SA_PATH_REC_SGID	 |
					       IB_SA_PATH_REC_NUMB_PATH	 |
					       IB_SA_PATH_REC_PKEY,
					       SRP_PATH_REC_TIMEOUT_MS,
					       GFP_KERNEL,
					       srp_path_rec_completion,
					       ch, &ch->path_query);
	if (ch->path_query_id < 0)
		return ch->path_query_id;

	ret = wait_for_completion_interruptible(&ch->done);
670 671
	if (ret < 0)
		return ret;
672

673
	if (ch->status < 0)
674 675
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Path record query failed\n");
676

677
	return ch->status;
678 679
}

B
Bart Van Assche 已提交
680
static int srp_send_req(struct srp_rdma_ch *ch, bool multich)
681
{
682
	struct srp_target_port *target = ch->target;
683 684 685 686 687 688 689 690 691 692
	struct {
		struct ib_cm_req_param param;
		struct srp_login_req   priv;
	} *req = NULL;
	int status;

	req = kzalloc(sizeof *req, GFP_KERNEL);
	if (!req)
		return -ENOMEM;

693
	req->param.primary_path		      = &ch->path;
694 695
	req->param.alternate_path 	      = NULL;
	req->param.service_id 		      = target->service_id;
696 697
	req->param.qp_num		      = ch->qp->qp_num;
	req->param.qp_type		      = ch->qp->qp_type;
698 699 700 701 702 703 704 705 706 707 708 709 710 711
	req->param.private_data 	      = &req->priv;
	req->param.private_data_len 	      = sizeof req->priv;
	req->param.flow_control 	      = 1;

	get_random_bytes(&req->param.starting_psn, 4);
	req->param.starting_psn 	     &= 0xffffff;

	/*
	 * Pick some arbitrary defaults here; we could make these
	 * module parameters if anyone cared about setting them.
	 */
	req->param.responder_resources	      = 4;
	req->param.remote_cm_response_timeout = 20;
	req->param.local_cm_response_timeout  = 20;
712
	req->param.retry_count                = target->tl_retry_count;
713 714 715 716 717
	req->param.rnr_retry_count 	      = 7;
	req->param.max_cm_retries 	      = 15;

	req->priv.opcode     	= SRP_LOGIN_REQ;
	req->priv.tag        	= 0;
718
	req->priv.req_it_iu_len = cpu_to_be32(target->max_iu_len);
719 720
	req->priv.req_buf_fmt 	= cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
					      SRP_BUF_FORMAT_INDIRECT);
B
Bart Van Assche 已提交
721 722
	req->priv.req_flags	= (multich ? SRP_MULTICHAN_MULTI :
				   SRP_MULTICHAN_SINGLE);
723
	/*
R
Roland Dreier 已提交
724
	 * In the published SRP specification (draft rev. 16a), the
725 726 727 728 729 730 731 732 733
	 * port identifier format is 8 bytes of ID extension followed
	 * by 8 bytes of GUID.  Older drafts put the two halves in the
	 * opposite order, so that the GUID comes first.
	 *
	 * Targets conforming to these obsolete drafts can be
	 * recognized by the I/O Class they report.
	 */
	if (target->io_class == SRP_REV10_IB_IO_CLASS) {
		memcpy(req->priv.initiator_port_id,
734
		       &target->sgid.global.interface_id, 8);
735
		memcpy(req->priv.initiator_port_id + 8,
736
		       &target->initiator_ext, 8);
737 738 739 740
		memcpy(req->priv.target_port_id,     &target->ioc_guid, 8);
		memcpy(req->priv.target_port_id + 8, &target->id_ext, 8);
	} else {
		memcpy(req->priv.initiator_port_id,
741 742
		       &target->initiator_ext, 8);
		memcpy(req->priv.initiator_port_id + 8,
743
		       &target->sgid.global.interface_id, 8);
744 745 746 747
		memcpy(req->priv.target_port_id,     &target->id_ext, 8);
		memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8);
	}

748 749
	/*
	 * Topspin/Cisco SRP targets will reject our login unless we
750 751
	 * zero out the first 8 bytes of our initiator port ID and set
	 * the second 8 bytes to the local node GUID.
752
	 */
753
	if (srp_target_is_topspin(target)) {
754 755 756
		shost_printk(KERN_DEBUG, target->scsi_host,
			     PFX "Topspin/Cisco initiator port ID workaround "
			     "activated for target GUID %016llx\n",
757
			     be64_to_cpu(target->ioc_guid));
758
		memset(req->priv.initiator_port_id, 0, 8);
759
		memcpy(req->priv.initiator_port_id + 8,
760
		       &target->srp_host->srp_dev->dev->node_guid, 8);
761 762
	}

763
	status = ib_send_cm_req(ch->cm_id, &req->param);
764 765 766 767 768 769

	kfree(req);

	return status;
}

770 771 772 773 774 775 776 777 778 779 780 781
static bool srp_queue_remove_work(struct srp_target_port *target)
{
	bool changed = false;

	spin_lock_irq(&target->lock);
	if (target->state != SRP_TARGET_REMOVED) {
		target->state = SRP_TARGET_REMOVED;
		changed = true;
	}
	spin_unlock_irq(&target->lock);

	if (changed)
782
		queue_work(srp_remove_wq, &target->remove_work);
783 784 785 786

	return changed;
}

787 788
static void srp_disconnect_target(struct srp_target_port *target)
{
B
Bart Van Assche 已提交
789 790
	struct srp_rdma_ch *ch;
	int i;
791

792
	/* XXX should send SRP_I_LOGOUT request */
793

794 795 796 797 798 799
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		ch->connected = false;
		if (ch->cm_id && ib_send_cm_dreq(ch->cm_id, NULL, 0)) {
			shost_printk(KERN_DEBUG, target->scsi_host,
				     PFX "Sending CM DREQ failed\n");
800
		}
801
	}
802 803
}

804 805
static void srp_free_req_data(struct srp_target_port *target,
			      struct srp_rdma_ch *ch)
806
{
807 808
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = dev->dev;
809 810 811
	struct srp_request *req;
	int i;

812
	if (!ch->req_ring)
813 814 815
		return;

	for (i = 0; i < target->req_ring_size; ++i) {
816
		req = &ch->req_ring[i];
817
		if (dev->use_fast_reg) {
818
			kfree(req->fr_list);
819
		} else {
820
			kfree(req->fmr_list);
821 822
			kfree(req->map_page);
		}
823 824 825 826 827 828
		if (req->indirect_dma_addr) {
			ib_dma_unmap_single(ibdev, req->indirect_dma_addr,
					    target->indirect_size,
					    DMA_TO_DEVICE);
		}
		kfree(req->indirect_desc);
829
	}
830

831 832
	kfree(ch->req_ring);
	ch->req_ring = NULL;
833 834
}

835
static int srp_alloc_req_data(struct srp_rdma_ch *ch)
836
{
837
	struct srp_target_port *target = ch->target;
838 839 840
	struct srp_device *srp_dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = srp_dev->dev;
	struct srp_request *req;
841
	void *mr_list;
842 843 844
	dma_addr_t dma_addr;
	int i, ret = -ENOMEM;

845 846 847
	ch->req_ring = kcalloc(target->req_ring_size, sizeof(*ch->req_ring),
			       GFP_KERNEL);
	if (!ch->req_ring)
848 849 850
		goto out;

	for (i = 0; i < target->req_ring_size; ++i) {
851
		req = &ch->req_ring[i];
852 853 854 855
		mr_list = kmalloc(target->cmd_sg_cnt * sizeof(void *),
				  GFP_KERNEL);
		if (!mr_list)
			goto out;
856
		if (srp_dev->use_fast_reg) {
857
			req->fr_list = mr_list;
858
		} else {
859
			req->fmr_list = mr_list;
860 861 862 863 864
			req->map_page = kmalloc(srp_dev->max_pages_per_mr *
						sizeof(void *), GFP_KERNEL);
			if (!req->map_page)
				goto out;
		}
865
		req->indirect_desc = kmalloc(target->indirect_size, GFP_KERNEL);
866
		if (!req->indirect_desc)
867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882
			goto out;

		dma_addr = ib_dma_map_single(ibdev, req->indirect_desc,
					     target->indirect_size,
					     DMA_TO_DEVICE);
		if (ib_dma_mapping_error(ibdev, dma_addr))
			goto out;

		req->indirect_dma_addr = dma_addr;
	}
	ret = 0;

out:
	return ret;
}

883 884 885 886 887 888 889 890 891 892 893 894 895 896 897
/**
 * srp_del_scsi_host_attr() - Remove attributes defined in the host template.
 * @shost: SCSI host whose attributes to remove from sysfs.
 *
 * Note: Any attributes defined in the host template and that did not exist
 * before invocation of this function will be ignored.
 */
static void srp_del_scsi_host_attr(struct Scsi_Host *shost)
{
	struct device_attribute **attr;

	for (attr = shost->hostt->shost_attrs; attr && *attr; ++attr)
		device_remove_file(&shost->shost_dev, *attr);
}

898 899
static void srp_remove_target(struct srp_target_port *target)
{
B
Bart Van Assche 已提交
900 901
	struct srp_rdma_ch *ch;
	int i;
902

903 904
	WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);

905
	srp_del_scsi_host_attr(target->scsi_host);
906
	srp_rport_get(target->rport);
907 908
	srp_remove_host(target->scsi_host);
	scsi_remove_host(target->scsi_host);
909
	srp_stop_rport_timers(target->rport);
910
	srp_disconnect_target(target);
B
Bart Van Assche 已提交
911 912 913 914
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_ch_ib(target, ch);
	}
915
	cancel_work_sync(&target->tl_err_work);
916
	srp_rport_put(target->rport);
B
Bart Van Assche 已提交
917 918 919 920 921 922
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_req_data(target, ch);
	}
	kfree(target->ch);
	target->ch = NULL;
923 924 925 926 927

	spin_lock(&target->srp_host->target_lock);
	list_del(&target->list);
	spin_unlock(&target->srp_host->target_lock);

928 929 930
	scsi_host_put(target->scsi_host);
}

D
David Howells 已提交
931
static void srp_remove_work(struct work_struct *work)
932
{
D
David Howells 已提交
933
	struct srp_target_port *target =
934
		container_of(work, struct srp_target_port, remove_work);
935

936
	WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);
937

938
	srp_remove_target(target);
939 940
}

941 942 943 944 945 946 947
static void srp_rport_delete(struct srp_rport *rport)
{
	struct srp_target_port *target = rport->lld_data;

	srp_queue_remove_work(target);
}

948 949 950 951 952 953 954 955 956 957 958 959 960 961
/**
 * srp_connected_ch() - number of connected channels
 * @target: SRP target port.
 */
static int srp_connected_ch(struct srp_target_port *target)
{
	int i, c = 0;

	for (i = 0; i < target->ch_count; i++)
		c += target->ch[i].connected;

	return c;
}

B
Bart Van Assche 已提交
962
static int srp_connect_ch(struct srp_rdma_ch *ch, bool multich)
963
{
964
	struct srp_target_port *target = ch->target;
965 966
	int ret;

967
	WARN_ON_ONCE(!multich && srp_connected_ch(target) > 0);
968

969
	ret = srp_lookup_path(ch);
970
	if (ret)
B
Bart Van Assche 已提交
971
		goto out;
972 973

	while (1) {
974
		init_completion(&ch->done);
B
Bart Van Assche 已提交
975
		ret = srp_send_req(ch, multich);
976
		if (ret)
B
Bart Van Assche 已提交
977
			goto out;
978
		ret = wait_for_completion_interruptible(&ch->done);
979
		if (ret < 0)
B
Bart Van Assche 已提交
980
			goto out;
981 982 983 984 985 986 987

		/*
		 * The CM event handling code will set status to
		 * SRP_PORT_REDIRECT if we get a port redirect REJ
		 * back, or SRP_DLID_REDIRECT if we get a lid/qp
		 * redirect REJ back.
		 */
B
Bart Van Assche 已提交
988 989
		ret = ch->status;
		switch (ret) {
990
		case 0:
991
			ch->connected = true;
B
Bart Van Assche 已提交
992
			goto out;
993 994

		case SRP_PORT_REDIRECT:
995
			ret = srp_lookup_path(ch);
996
			if (ret)
B
Bart Van Assche 已提交
997
				goto out;
998 999 1000 1001 1002
			break;

		case SRP_DLID_REDIRECT:
			break;

D
David Dillow 已提交
1003 1004
		case SRP_STALE_CONN:
			shost_printk(KERN_ERR, target->scsi_host, PFX
1005
				     "giving up on stale connection\n");
B
Bart Van Assche 已提交
1006 1007
			ret = -ECONNRESET;
			goto out;
D
David Dillow 已提交
1008

1009
		default:
B
Bart Van Assche 已提交
1010
			goto out;
1011 1012
		}
	}
B
Bart Van Assche 已提交
1013 1014 1015

out:
	return ret <= 0 ? ret : -ENODEV;
1016 1017
}

C
Christoph Hellwig 已提交
1018 1019 1020 1021 1022 1023 1024
static void srp_inv_rkey_err_done(struct ib_cq *cq, struct ib_wc *wc)
{
	srp_handle_qp_err(cq, wc, "INV RKEY");
}

static int srp_inv_rkey(struct srp_request *req, struct srp_rdma_ch *ch,
		u32 rkey)
1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
{
	struct ib_send_wr *bad_wr;
	struct ib_send_wr wr = {
		.opcode		    = IB_WR_LOCAL_INV,
		.next		    = NULL,
		.num_sge	    = 0,
		.send_flags	    = 0,
		.ex.invalidate_rkey = rkey,
	};

C
Christoph Hellwig 已提交
1035 1036
	wr.wr_cqe = &req->reg_cqe;
	req->reg_cqe.done = srp_inv_rkey_err_done;
1037
	return ib_post_send(ch->qp, &wr, &bad_wr);
1038 1039
}

1040
static void srp_unmap_data(struct scsi_cmnd *scmnd,
1041
			   struct srp_rdma_ch *ch,
1042 1043
			   struct srp_request *req)
{
1044
	struct srp_target_port *target = ch->target;
1045 1046 1047
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = dev->dev;
	int i, res;
1048

1049
	if (!scsi_sglist(scmnd) ||
1050 1051 1052 1053
	    (scmnd->sc_data_direction != DMA_TO_DEVICE &&
	     scmnd->sc_data_direction != DMA_FROM_DEVICE))
		return;

1054 1055 1056 1057
	if (dev->use_fast_reg) {
		struct srp_fr_desc **pfr;

		for (i = req->nmdesc, pfr = req->fr_list; i > 0; i--, pfr++) {
C
Christoph Hellwig 已提交
1058
			res = srp_inv_rkey(req, ch, (*pfr)->mr->rkey);
1059 1060 1061 1062 1063 1064 1065 1066 1067
			if (res < 0) {
				shost_printk(KERN_ERR, target->scsi_host, PFX
				  "Queueing INV WR for rkey %#x failed (%d)\n",
				  (*pfr)->mr->rkey, res);
				queue_work(system_long_wq,
					   &target->tl_err_work);
			}
		}
		if (req->nmdesc)
1068
			srp_fr_pool_put(ch->fr_pool, req->fr_list,
1069
					req->nmdesc);
1070
	} else if (dev->use_fmr) {
1071 1072 1073 1074 1075
		struct ib_pool_fmr **pfmr;

		for (i = req->nmdesc, pfmr = req->fmr_list; i > 0; i--, pfmr++)
			ib_fmr_pool_unmap(*pfmr);
	}
1076

1077 1078
	ib_dma_unmap_sg(ibdev, scsi_sglist(scmnd), scsi_sg_count(scmnd),
			scmnd->sc_data_direction);
1079 1080
}

B
Bart Van Assche 已提交
1081 1082
/**
 * srp_claim_req - Take ownership of the scmnd associated with a request.
1083
 * @ch: SRP RDMA channel.
B
Bart Van Assche 已提交
1084
 * @req: SRP request.
1085
 * @sdev: If not NULL, only take ownership for this SCSI device.
B
Bart Van Assche 已提交
1086 1087 1088 1089 1090 1091
 * @scmnd: If NULL, take ownership of @req->scmnd. If not NULL, only take
 *         ownership of @req->scmnd if it equals @scmnd.
 *
 * Return value:
 * Either NULL or a pointer to the SCSI command the caller became owner of.
 */
1092
static struct scsi_cmnd *srp_claim_req(struct srp_rdma_ch *ch,
B
Bart Van Assche 已提交
1093
				       struct srp_request *req,
1094
				       struct scsi_device *sdev,
B
Bart Van Assche 已提交
1095 1096 1097 1098
				       struct scsi_cmnd *scmnd)
{
	unsigned long flags;

1099
	spin_lock_irqsave(&ch->lock, flags);
1100 1101 1102
	if (req->scmnd &&
	    (!sdev || req->scmnd->device == sdev) &&
	    (!scmnd || req->scmnd == scmnd)) {
B
Bart Van Assche 已提交
1103 1104 1105 1106 1107
		scmnd = req->scmnd;
		req->scmnd = NULL;
	} else {
		scmnd = NULL;
	}
1108
	spin_unlock_irqrestore(&ch->lock, flags);
B
Bart Van Assche 已提交
1109 1110 1111 1112 1113

	return scmnd;
}

/**
B
Bart Van Assche 已提交
1114
 * srp_free_req() - Unmap data and adjust ch->req_lim.
1115
 * @ch:     SRP RDMA channel.
1116 1117 1118
 * @req:    Request to be freed.
 * @scmnd:  SCSI command associated with @req.
 * @req_lim_delta: Amount to be added to @target->req_lim.
B
Bart Van Assche 已提交
1119
 */
1120 1121
static void srp_free_req(struct srp_rdma_ch *ch, struct srp_request *req,
			 struct scsi_cmnd *scmnd, s32 req_lim_delta)
1122
{
1123 1124
	unsigned long flags;

1125
	srp_unmap_data(scmnd, ch, req);
B
Bart Van Assche 已提交
1126

1127 1128 1129
	spin_lock_irqsave(&ch->lock, flags);
	ch->req_lim += req_lim_delta;
	spin_unlock_irqrestore(&ch->lock, flags);
1130 1131
}

1132 1133
static void srp_finish_req(struct srp_rdma_ch *ch, struct srp_request *req,
			   struct scsi_device *sdev, int result)
1134
{
1135
	struct scsi_cmnd *scmnd = srp_claim_req(ch, req, sdev, NULL);
B
Bart Van Assche 已提交
1136 1137

	if (scmnd) {
1138
		srp_free_req(ch, req, scmnd, 0);
1139
		scmnd->result = result;
B
Bart Van Assche 已提交
1140 1141
		scmnd->scsi_done(scmnd);
	}
1142 1143
}

1144
static void srp_terminate_io(struct srp_rport *rport)
1145
{
1146
	struct srp_target_port *target = rport->lld_data;
B
Bart Van Assche 已提交
1147
	struct srp_rdma_ch *ch;
1148 1149
	struct Scsi_Host *shost = target->scsi_host;
	struct scsi_device *sdev;
B
Bart Van Assche 已提交
1150
	int i, j;
1151

1152 1153 1154 1155 1156 1157 1158
	/*
	 * Invoking srp_terminate_io() while srp_queuecommand() is running
	 * is not safe. Hence the warning statement below.
	 */
	shost_for_each_device(sdev, shost)
		WARN_ON_ONCE(sdev->request_queue->request_fn_active);

B
Bart Van Assche 已提交
1159 1160
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
1161

B
Bart Van Assche 已提交
1162 1163 1164 1165 1166 1167
		for (j = 0; j < target->req_ring_size; ++j) {
			struct srp_request *req = &ch->req_ring[j];

			srp_finish_req(ch, req, NULL,
				       DID_TRANSPORT_FAILFAST << 16);
		}
1168 1169
	}
}
1170

1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182
/*
 * It is up to the caller to ensure that srp_rport_reconnect() calls are
 * serialized and that no concurrent srp_queuecommand(), srp_abort(),
 * srp_reset_device() or srp_reset_host() calls will occur while this function
 * is in progress. One way to realize that is not to call this function
 * directly but to call srp_reconnect_rport() instead since that last function
 * serializes calls of this function via rport->mutex and also blocks
 * srp_queuecommand() calls before invoking this function.
 */
static int srp_rport_reconnect(struct srp_rport *rport)
{
	struct srp_target_port *target = rport->lld_data;
B
Bart Van Assche 已提交
1183 1184 1185
	struct srp_rdma_ch *ch;
	int i, j, ret = 0;
	bool multich = false;
1186

1187
	srp_disconnect_target(target);
1188 1189 1190 1191

	if (target->state == SRP_TARGET_SCANNING)
		return -ENODEV;

1192
	/*
1193 1194 1195
	 * Now get a new local CM ID so that we avoid confusing the target in
	 * case things are really fouled up. Doing so also ensures that all CM
	 * callbacks will have finished before a new QP is allocated.
1196
	 */
B
Bart Van Assche 已提交
1197 1198 1199
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		ret += srp_new_cm_id(ch);
1200
	}
B
Bart Van Assche 已提交
1201 1202 1203 1204
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		for (j = 0; j < target->req_ring_size; ++j) {
			struct srp_request *req = &ch->req_ring[j];
1205

B
Bart Van Assche 已提交
1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216
			srp_finish_req(ch, req, NULL, DID_RESET << 16);
		}
	}
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		/*
		 * Whether or not creating a new CM ID succeeded, create a new
		 * QP. This guarantees that all completion callback function
		 * invocations have finished before request resetting starts.
		 */
		ret += srp_create_ch_ib(ch);
1217

B
Bart Van Assche 已提交
1218 1219 1220 1221
		INIT_LIST_HEAD(&ch->free_tx);
		for (j = 0; j < target->queue_size; ++j)
			list_add(&ch->tx_ring[j]->list, &ch->free_tx);
	}
1222 1223 1224

	target->qp_in_error = false;

B
Bart Van Assche 已提交
1225 1226
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
1227
		if (ret)
B
Bart Van Assche 已提交
1228 1229 1230 1231
			break;
		ret = srp_connect_ch(ch, multich);
		multich = true;
	}
1232

1233 1234 1235
	if (ret == 0)
		shost_printk(KERN_INFO, target->scsi_host,
			     PFX "reconnect succeeded\n");
1236 1237 1238 1239

	return ret;
}

1240 1241
static void srp_map_desc(struct srp_map_state *state, dma_addr_t dma_addr,
			 unsigned int dma_len, u32 rkey)
1242
{
1243
	struct srp_direct_buf *desc = state->desc;
1244

1245 1246
	WARN_ON_ONCE(!dma_len);

1247 1248 1249
	desc->va = cpu_to_be64(dma_addr);
	desc->key = cpu_to_be32(rkey);
	desc->len = cpu_to_be32(dma_len);
1250

1251 1252 1253 1254
	state->total_len += dma_len;
	state->desc++;
	state->ndesc++;
}
1255

1256
static int srp_map_finish_fmr(struct srp_map_state *state,
1257
			      struct srp_rdma_ch *ch)
1258
{
1259 1260
	struct srp_target_port *target = ch->target;
	struct srp_device *dev = target->srp_host->srp_dev;
1261 1262
	struct ib_pool_fmr *fmr;
	u64 io_addr = 0;
1263

1264 1265 1266
	if (state->fmr.next >= state->fmr.end)
		return -ENOMEM;

S
Sagi Grimberg 已提交
1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277
	WARN_ON_ONCE(!dev->use_fmr);

	if (state->npages == 0)
		return 0;

	if (state->npages == 1 && target->global_mr) {
		srp_map_desc(state, state->base_dma_addr, state->dma_len,
			     target->global_mr->rkey);
		goto reset_state;
	}

1278
	fmr = ib_fmr_pool_map_phys(ch->fmr_pool, state->pages,
1279 1280 1281
				   state->npages, io_addr);
	if (IS_ERR(fmr))
		return PTR_ERR(fmr);
1282

1283
	*state->fmr.next++ = fmr;
1284
	state->nmdesc++;
1285

1286 1287
	srp_map_desc(state, state->base_dma_addr & ~dev->mr_page_mask,
		     state->dma_len, fmr->fmr->rkey);
1288

S
Sagi Grimberg 已提交
1289 1290 1291 1292
reset_state:
	state->npages = 0;
	state->dma_len = 0;

1293 1294 1295
	return 0;
}

C
Christoph Hellwig 已提交
1296 1297 1298 1299 1300
static void srp_reg_mr_err_done(struct ib_cq *cq, struct ib_wc *wc)
{
	srp_handle_qp_err(cq, wc, "FAST REG");
}

1301
static int srp_map_finish_fr(struct srp_map_state *state,
C
Christoph Hellwig 已提交
1302
			     struct srp_request *req,
B
Bart Van Assche 已提交
1303
			     struct srp_rdma_ch *ch, int sg_nents)
1304
{
1305
	struct srp_target_port *target = ch->target;
1306 1307
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_send_wr *bad_wr;
1308
	struct ib_reg_wr wr;
1309 1310
	struct srp_fr_desc *desc;
	u32 rkey;
1311
	int n, err;
1312

1313 1314 1315
	if (state->fr.next >= state->fr.end)
		return -ENOMEM;

S
Sagi Grimberg 已提交
1316 1317
	WARN_ON_ONCE(!dev->use_fast_reg);

B
Bart Van Assche 已提交
1318
	if (sg_nents == 1 && target->global_mr) {
1319 1320
		srp_map_desc(state, sg_dma_address(state->sg),
			     sg_dma_len(state->sg),
S
Sagi Grimberg 已提交
1321
			     target->global_mr->rkey);
1322
		return 1;
S
Sagi Grimberg 已提交
1323 1324
	}

1325
	desc = srp_fr_pool_get(ch->fr_pool);
1326 1327 1328 1329 1330 1331
	if (!desc)
		return -ENOMEM;

	rkey = ib_inc_rkey(desc->mr->rkey);
	ib_update_fast_reg_key(desc->mr, rkey);

1332
	n = ib_map_mr_sg(desc->mr, state->sg, sg_nents, NULL, dev->mr_page_size);
1333 1334 1335 1336 1337
	if (unlikely(n < 0)) {
		srp_fr_pool_put(ch->fr_pool, &desc, 1);
		pr_debug("%s: ib_map_mr_sg(%d) returned %d.\n",
			 dev_name(&req->scmnd->device->sdev_gendev), sg_nents,
			 n);
1338
		return n;
1339
	}
1340

C
Christoph Hellwig 已提交
1341 1342
	req->reg_cqe.done = srp_reg_mr_err_done;

1343 1344
	wr.wr.next = NULL;
	wr.wr.opcode = IB_WR_REG_MR;
C
Christoph Hellwig 已提交
1345
	wr.wr.wr_cqe = &req->reg_cqe;
1346 1347 1348 1349 1350 1351 1352
	wr.wr.num_sge = 0;
	wr.wr.send_flags = 0;
	wr.mr = desc->mr;
	wr.key = desc->mr->rkey;
	wr.access = (IB_ACCESS_LOCAL_WRITE |
		     IB_ACCESS_REMOTE_READ |
		     IB_ACCESS_REMOTE_WRITE);
1353

1354
	*state->fr.next++ = desc;
1355 1356
	state->nmdesc++;

1357 1358
	srp_map_desc(state, desc->mr->iova,
		     desc->mr->length, desc->mr->rkey);
1359

S
Sagi Grimberg 已提交
1360
	err = ib_post_send(ch->qp, &wr.wr, &bad_wr);
1361
	if (unlikely(err))
S
Sagi Grimberg 已提交
1362 1363
		return err;

1364
	return n;
1365 1366
}

1367
static int srp_map_sg_entry(struct srp_map_state *state,
1368
			    struct srp_rdma_ch *ch,
1369
			    struct scatterlist *sg, int sg_index)
1370
{
1371
	struct srp_target_port *target = ch->target;
1372 1373 1374 1375
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = dev->dev;
	dma_addr_t dma_addr = ib_sg_dma_address(ibdev, sg);
	unsigned int dma_len = ib_sg_dma_len(ibdev, sg);
1376
	unsigned int len = 0;
1377 1378
	int ret;

1379
	WARN_ON_ONCE(!dma_len);
1380

1381
	while (dma_len) {
1382 1383
		unsigned offset = dma_addr & ~dev->mr_page_mask;
		if (state->npages == dev->max_pages_per_mr || offset != 0) {
1384
			ret = srp_map_finish_fmr(state, ch);
1385 1386 1387 1388
			if (ret)
				return ret;
		}

1389
		len = min_t(unsigned int, dma_len, dev->mr_page_size - offset);
1390

1391 1392
		if (!state->npages)
			state->base_dma_addr = dma_addr;
1393
		state->pages[state->npages++] = dma_addr & dev->mr_page_mask;
1394
		state->dma_len += len;
1395 1396 1397 1398
		dma_addr += len;
		dma_len -= len;
	}

1399 1400
	/*
	 * If the last entry of the MR wasn't a full page, then we need to
1401
	 * close it out and start a new one -- we can only merge at page
1402
	 * boundaries.
1403 1404
	 */
	ret = 0;
1405
	if (len != dev->mr_page_size)
1406
		ret = srp_map_finish_fmr(state, ch);
1407 1408 1409
	return ret;
}

S
Sagi Grimberg 已提交
1410 1411 1412
static int srp_map_sg_fmr(struct srp_map_state *state, struct srp_rdma_ch *ch,
			  struct srp_request *req, struct scatterlist *scat,
			  int count)
1413 1414
{
	struct scatterlist *sg;
1415
	int i, ret;
1416

S
Sagi Grimberg 已提交
1417 1418 1419 1420 1421 1422 1423 1424
	state->pages = req->map_page;
	state->fmr.next = req->fmr_list;
	state->fmr.end = req->fmr_list + ch->target->cmd_sg_cnt;

	for_each_sg(scat, sg, count, i) {
		ret = srp_map_sg_entry(state, ch, sg, i);
		if (ret)
			return ret;
1425
	}
1426

1427
	ret = srp_map_finish_fmr(state, ch);
S
Sagi Grimberg 已提交
1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438
	if (ret)
		return ret;

	return 0;
}

static int srp_map_sg_fr(struct srp_map_state *state, struct srp_rdma_ch *ch,
			 struct srp_request *req, struct scatterlist *scat,
			 int count)
{
	state->desc = req->indirect_desc;
1439 1440 1441
	state->fr.next = req->fr_list;
	state->fr.end = req->fr_list + ch->target->cmd_sg_cnt;
	state->sg = scat;
S
Sagi Grimberg 已提交
1442

1443 1444 1445
	if (count == 0)
		return 0;

B
Bart Van Assche 已提交
1446
	while (count) {
1447
		int i, n;
S
Sagi Grimberg 已提交
1448

1449
		n = srp_map_finish_fr(state, req, ch, count);
1450 1451 1452
		if (unlikely(n < 0))
			return n;

B
Bart Van Assche 已提交
1453
		count -= n;
1454 1455 1456
		for (i = 0; i < n; i++)
			state->sg = sg_next(state->sg);
	}
S
Sagi Grimberg 已提交
1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474

	return 0;
}

static int srp_map_sg_dma(struct srp_map_state *state, struct srp_rdma_ch *ch,
			  struct srp_request *req, struct scatterlist *scat,
			  int count)
{
	struct srp_target_port *target = ch->target;
	struct srp_device *dev = target->srp_host->srp_dev;
	struct scatterlist *sg;
	int i;

	state->desc = req->indirect_desc;
	for_each_sg(scat, sg, count, i) {
		srp_map_desc(state, ib_sg_dma_address(dev->dev, sg),
			     ib_sg_dma_len(dev->dev, sg),
			     target->global_mr->rkey);
1475
	}
1476

S
Sagi Grimberg 已提交
1477
	return 0;
1478 1479
}

1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495
/*
 * Register the indirect data buffer descriptor with the HCA.
 *
 * Note: since the indirect data buffer descriptor has been allocated with
 * kmalloc() it is guaranteed that this buffer is a physically contiguous
 * memory buffer.
 */
static int srp_map_idb(struct srp_rdma_ch *ch, struct srp_request *req,
		       void **next_mr, void **end_mr, u32 idb_len,
		       __be32 *idb_rkey)
{
	struct srp_target_port *target = ch->target;
	struct srp_device *dev = target->srp_host->srp_dev;
	struct srp_map_state state;
	struct srp_direct_buf idb_desc;
	u64 idb_pages[1];
1496
	struct scatterlist idb_sg[1];
1497 1498 1499 1500 1501 1502 1503 1504 1505
	int ret;

	memset(&state, 0, sizeof(state));
	memset(&idb_desc, 0, sizeof(idb_desc));
	state.gen.next = next_mr;
	state.gen.end = end_mr;
	state.desc = &idb_desc;
	state.base_dma_addr = req->indirect_dma_addr;
	state.dma_len = idb_len;
1506 1507 1508 1509 1510

	if (dev->use_fast_reg) {
		state.sg = idb_sg;
		sg_set_buf(idb_sg, req->indirect_desc, idb_len);
		idb_sg->dma_address = req->indirect_dma_addr; /* hack! */
1511 1512 1513
#ifdef CONFIG_NEED_SG_DMA_LENGTH
		idb_sg->dma_length = idb_sg->length;	      /* hack^2 */
#endif
1514
		ret = srp_map_finish_fr(&state, req, ch, 1);
1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527
		if (ret < 0)
			return ret;
	} else if (dev->use_fmr) {
		state.pages = idb_pages;
		state.pages[0] = (req->indirect_dma_addr &
				  dev->mr_page_mask);
		state.npages = 1;
		ret = srp_map_finish_fmr(&state, ch);
		if (ret < 0)
			return ret;
	} else {
		return -EINVAL;
	}
1528 1529 1530

	*idb_rkey = idb_desc.key;

1531
	return 0;
1532 1533
}

1534 1535 1536 1537 1538 1539 1540 1541 1542
/**
 * srp_map_data() - map SCSI data buffer onto an SRP request
 * @scmnd: SCSI command to map
 * @ch: SRP RDMA channel
 * @req: SRP request
 *
 * Returns the length in bytes of the SRP_CMD IU or a negative value if
 * mapping failed.
 */
1543
static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_rdma_ch *ch,
1544 1545
			struct srp_request *req)
{
1546
	struct srp_target_port *target = ch->target;
1547
	struct scatterlist *scat;
1548
	struct srp_cmd *cmd = req->cmd->buf;
1549
	int len, nents, count, ret;
1550 1551
	struct srp_device *dev;
	struct ib_device *ibdev;
1552 1553
	struct srp_map_state state;
	struct srp_indirect_buf *indirect_hdr;
1554 1555
	u32 idb_len, table_len;
	__be32 idb_rkey;
1556
	u8 fmt;
1557

1558
	if (!scsi_sglist(scmnd) || scmnd->sc_data_direction == DMA_NONE)
1559 1560 1561 1562
		return sizeof (struct srp_cmd);

	if (scmnd->sc_data_direction != DMA_FROM_DEVICE &&
	    scmnd->sc_data_direction != DMA_TO_DEVICE) {
1563 1564 1565
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled data direction %d\n",
			     scmnd->sc_data_direction);
1566 1567 1568
		return -EINVAL;
	}

1569 1570
	nents = scsi_sg_count(scmnd);
	scat  = scsi_sglist(scmnd);
1571

1572
	dev = target->srp_host->srp_dev;
1573 1574 1575
	ibdev = dev->dev;

	count = ib_dma_map_sg(ibdev, scat, nents, scmnd->sc_data_direction);
1576 1577
	if (unlikely(count == 0))
		return -EIO;
1578 1579 1580

	fmt = SRP_DATA_DESC_DIRECT;
	len = sizeof (struct srp_cmd) +	sizeof (struct srp_direct_buf);
1581

1582
	if (count == 1 && target->global_mr) {
1583 1584 1585 1586 1587 1588
		/*
		 * The midlayer only generated a single gather/scatter
		 * entry, or DMA mapping coalesced everything to a
		 * single entry.  So a direct descriptor along with
		 * the DMA MR suffices.
		 */
1589
		struct srp_direct_buf *buf = (void *) cmd->add_data;
1590

1591
		buf->va  = cpu_to_be64(ib_sg_dma_address(ibdev, scat));
1592
		buf->key = cpu_to_be32(target->global_mr->rkey);
1593
		buf->len = cpu_to_be32(ib_sg_dma_len(ibdev, scat));
1594

1595
		req->nmdesc = 0;
1596 1597 1598
		goto map_complete;
	}

1599 1600 1601
	/*
	 * We have more than one scatter/gather entry, so build our indirect
	 * descriptor table, trying to merge as many entries as we can.
1602 1603 1604
	 */
	indirect_hdr = (void *) cmd->add_data;

1605 1606 1607
	ib_dma_sync_single_for_cpu(ibdev, req->indirect_dma_addr,
				   target->indirect_size, DMA_TO_DEVICE);

1608
	memset(&state, 0, sizeof(state));
S
Sagi Grimberg 已提交
1609
	if (dev->use_fast_reg)
1610
		ret = srp_map_sg_fr(&state, ch, req, scat, count);
S
Sagi Grimberg 已提交
1611
	else if (dev->use_fmr)
1612
		ret = srp_map_sg_fmr(&state, ch, req, scat, count);
S
Sagi Grimberg 已提交
1613
	else
1614 1615 1616 1617
		ret = srp_map_sg_dma(&state, ch, req, scat, count);
	req->nmdesc = state.nmdesc;
	if (ret < 0)
		goto unmap;
1618

1619 1620 1621 1622 1623
	/* We've mapped the request, now pull as much of the indirect
	 * descriptor table as we can into the command buffer. If this
	 * target is not using an external indirect table, we are
	 * guaranteed to fit into the command, as the SCSI layer won't
	 * give us more S/G entries than we allow.
1624 1625
	 */
	if (state.ndesc == 1) {
1626 1627
		/*
		 * Memory registration collapsed the sg-list into one entry,
1628 1629 1630
		 * so use a direct descriptor.
		 */
		struct srp_direct_buf *buf = (void *) cmd->add_data;
1631

1632
		*buf = req->indirect_desc[0];
1633
		goto map_complete;
1634 1635
	}

1636 1637 1638 1639
	if (unlikely(target->cmd_sg_cnt < state.ndesc &&
						!target->allow_ext_sg)) {
		shost_printk(KERN_ERR, target->scsi_host,
			     "Could not fit S/G list into SRP_CMD\n");
1640 1641
		ret = -EIO;
		goto unmap;
1642 1643 1644
	}

	count = min(state.ndesc, target->cmd_sg_cnt);
1645
	table_len = state.ndesc * sizeof (struct srp_direct_buf);
1646
	idb_len = sizeof(struct srp_indirect_buf) + table_len;
1647 1648 1649

	fmt = SRP_DATA_DESC_INDIRECT;
	len = sizeof(struct srp_cmd) + sizeof (struct srp_indirect_buf);
1650
	len += count * sizeof (struct srp_direct_buf);
1651

1652 1653
	memcpy(indirect_hdr->desc_list, req->indirect_desc,
	       count * sizeof (struct srp_direct_buf));
1654

1655
	if (!target->global_mr) {
1656 1657 1658
		ret = srp_map_idb(ch, req, state.gen.next, state.gen.end,
				  idb_len, &idb_rkey);
		if (ret < 0)
1659
			goto unmap;
1660 1661
		req->nmdesc++;
	} else {
1662
		idb_rkey = cpu_to_be32(target->global_mr->rkey);
1663 1664
	}

1665
	indirect_hdr->table_desc.va = cpu_to_be64(req->indirect_dma_addr);
1666
	indirect_hdr->table_desc.key = idb_rkey;
1667 1668 1669 1670
	indirect_hdr->table_desc.len = cpu_to_be32(table_len);
	indirect_hdr->len = cpu_to_be32(state.total_len);

	if (scmnd->sc_data_direction == DMA_TO_DEVICE)
1671
		cmd->data_out_desc_cnt = count;
1672
	else
1673 1674 1675 1676
		cmd->data_in_desc_cnt = count;

	ib_dma_sync_single_for_device(ibdev, req->indirect_dma_addr, table_len,
				      DMA_TO_DEVICE);
1677 1678

map_complete:
1679 1680 1681 1682 1683 1684
	if (scmnd->sc_data_direction == DMA_TO_DEVICE)
		cmd->buf_fmt = fmt << 4;
	else
		cmd->buf_fmt = fmt;

	return len;
1685 1686 1687

unmap:
	srp_unmap_data(scmnd, ch, req);
1688 1689
	if (ret == -ENOMEM && req->nmdesc >= target->mr_pool_size)
		ret = -E2BIG;
1690
	return ret;
1691 1692
}

1693 1694 1695
/*
 * Return an IU and possible credit to the free pool
 */
1696
static void srp_put_tx_iu(struct srp_rdma_ch *ch, struct srp_iu *iu,
1697 1698 1699 1700
			  enum srp_iu_type iu_type)
{
	unsigned long flags;

1701 1702
	spin_lock_irqsave(&ch->lock, flags);
	list_add(&iu->list, &ch->free_tx);
1703
	if (iu_type != SRP_IU_RSP)
1704 1705
		++ch->req_lim;
	spin_unlock_irqrestore(&ch->lock, flags);
1706 1707
}

1708
/*
1709
 * Must be called with ch->lock held to protect req_lim and free_tx.
1710
 * If IU is not sent, it must be returned using srp_put_tx_iu().
1711 1712 1713 1714 1715 1716 1717 1718 1719 1720
 *
 * Note:
 * An upper limit for the number of allocated information units for each
 * request type is:
 * - SRP_IU_CMD: SRP_CMD_SQ_SIZE, since the SCSI mid-layer never queues
 *   more than Scsi_Host.can_queue requests.
 * - SRP_IU_TSK_MGMT: SRP_TSK_MGMT_SQ_SIZE.
 * - SRP_IU_RSP: 1, since a conforming SRP target never sends more than
 *   one unanswered SRP request to an initiator.
 */
1721
static struct srp_iu *__srp_get_tx_iu(struct srp_rdma_ch *ch,
1722 1723
				      enum srp_iu_type iu_type)
{
1724
	struct srp_target_port *target = ch->target;
1725 1726 1727
	s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE;
	struct srp_iu *iu;

C
Christoph Hellwig 已提交
1728
	ib_process_cq_direct(ch->send_cq, -1);
1729

1730
	if (list_empty(&ch->free_tx))
1731 1732 1733
		return NULL;

	/* Initiator responses to target requests do not consume credits */
1734
	if (iu_type != SRP_IU_RSP) {
1735
		if (ch->req_lim <= rsv) {
1736 1737 1738 1739
			++target->zero_req_lim;
			return NULL;
		}

1740
		--ch->req_lim;
1741 1742
	}

1743
	iu = list_first_entry(&ch->free_tx, struct srp_iu, list);
1744
	list_del(&iu->list);
1745 1746 1747
	return iu;
}

C
Christoph Hellwig 已提交
1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760
static void srp_send_done(struct ib_cq *cq, struct ib_wc *wc)
{
	struct srp_iu *iu = container_of(wc->wr_cqe, struct srp_iu, cqe);
	struct srp_rdma_ch *ch = cq->cq_context;

	if (unlikely(wc->status != IB_WC_SUCCESS)) {
		srp_handle_qp_err(cq, wc, "SEND");
		return;
	}

	list_add(&iu->list, &ch->free_tx);
}

1761
static int srp_post_send(struct srp_rdma_ch *ch, struct srp_iu *iu, int len)
1762
{
1763
	struct srp_target_port *target = ch->target;
1764 1765 1766 1767 1768
	struct ib_sge list;
	struct ib_send_wr wr, *bad_wr;

	list.addr   = iu->dma;
	list.length = len;
1769
	list.lkey   = target->lkey;
1770

C
Christoph Hellwig 已提交
1771 1772
	iu->cqe.done = srp_send_done;

1773
	wr.next       = NULL;
C
Christoph Hellwig 已提交
1774
	wr.wr_cqe     = &iu->cqe;
1775 1776 1777 1778 1779
	wr.sg_list    = &list;
	wr.num_sge    = 1;
	wr.opcode     = IB_WR_SEND;
	wr.send_flags = IB_SEND_SIGNALED;

1780
	return ib_post_send(ch->qp, &wr, &bad_wr);
1781 1782
}

1783
static int srp_post_recv(struct srp_rdma_ch *ch, struct srp_iu *iu)
1784
{
1785
	struct srp_target_port *target = ch->target;
1786
	struct ib_recv_wr wr, *bad_wr;
1787
	struct ib_sge list;
1788 1789 1790

	list.addr   = iu->dma;
	list.length = iu->size;
1791
	list.lkey   = target->lkey;
1792

C
Christoph Hellwig 已提交
1793 1794
	iu->cqe.done = srp_recv_done;

1795
	wr.next     = NULL;
C
Christoph Hellwig 已提交
1796
	wr.wr_cqe   = &iu->cqe;
1797 1798 1799
	wr.sg_list  = &list;
	wr.num_sge  = 1;

1800
	return ib_post_recv(ch->qp, &wr, &bad_wr);
1801 1802
}

1803
static void srp_process_rsp(struct srp_rdma_ch *ch, struct srp_rsp *rsp)
1804
{
1805
	struct srp_target_port *target = ch->target;
1806 1807 1808 1809 1810
	struct srp_request *req;
	struct scsi_cmnd *scmnd;
	unsigned long flags;

	if (unlikely(rsp->tag & SRP_TAG_TSK_MGMT)) {
1811 1812 1813
		spin_lock_irqsave(&ch->lock, flags);
		ch->req_lim += be32_to_cpu(rsp->req_lim_delta);
		spin_unlock_irqrestore(&ch->lock, flags);
1814

1815
		ch->tsk_mgmt_status = -1;
1816
		if (be32_to_cpu(rsp->resp_data_len) >= 4)
1817 1818
			ch->tsk_mgmt_status = rsp->data[3];
		complete(&ch->tsk_mgmt_done);
1819
	} else {
B
Bart Van Assche 已提交
1820 1821 1822 1823 1824
		scmnd = scsi_host_find_tag(target->scsi_host, rsp->tag);
		if (scmnd) {
			req = (void *)scmnd->host_scribble;
			scmnd = srp_claim_req(ch, req, NULL, scmnd);
		}
B
Bart Van Assche 已提交
1825
		if (!scmnd) {
1826
			shost_printk(KERN_ERR, target->scsi_host,
B
Bart Van Assche 已提交
1827 1828
				     "Null scmnd for RSP w/tag %#016llx received on ch %td / QP %#x\n",
				     rsp->tag, ch - target->ch, ch->qp->qp_num);
B
Bart Van Assche 已提交
1829

1830 1831 1832
			spin_lock_irqsave(&ch->lock, flags);
			ch->req_lim += be32_to_cpu(rsp->req_lim_delta);
			spin_unlock_irqrestore(&ch->lock, flags);
B
Bart Van Assche 已提交
1833 1834 1835

			return;
		}
1836 1837 1838 1839 1840 1841 1842 1843 1844
		scmnd->result = rsp->status;

		if (rsp->flags & SRP_RSP_FLAG_SNSVALID) {
			memcpy(scmnd->sense_buffer, rsp->data +
			       be32_to_cpu(rsp->resp_data_len),
			       min_t(int, be32_to_cpu(rsp->sense_data_len),
				     SCSI_SENSE_BUFFERSIZE));
		}

B
Bart Van Assche 已提交
1845
		if (unlikely(rsp->flags & SRP_RSP_FLAG_DIUNDER))
1846
			scsi_set_resid(scmnd, be32_to_cpu(rsp->data_in_res_cnt));
B
Bart Van Assche 已提交
1847 1848 1849 1850 1851 1852
		else if (unlikely(rsp->flags & SRP_RSP_FLAG_DIOVER))
			scsi_set_resid(scmnd, -be32_to_cpu(rsp->data_in_res_cnt));
		else if (unlikely(rsp->flags & SRP_RSP_FLAG_DOUNDER))
			scsi_set_resid(scmnd, be32_to_cpu(rsp->data_out_res_cnt));
		else if (unlikely(rsp->flags & SRP_RSP_FLAG_DOOVER))
			scsi_set_resid(scmnd, -be32_to_cpu(rsp->data_out_res_cnt));
1853

1854
		srp_free_req(ch, req, scmnd,
B
Bart Van Assche 已提交
1855 1856
			     be32_to_cpu(rsp->req_lim_delta));

1857 1858
		scmnd->host_scribble = NULL;
		scmnd->scsi_done(scmnd);
1859 1860 1861
	}
}

1862
static int srp_response_common(struct srp_rdma_ch *ch, s32 req_delta,
1863 1864
			       void *rsp, int len)
{
1865
	struct srp_target_port *target = ch->target;
1866
	struct ib_device *dev = target->srp_host->srp_dev->dev;
1867 1868
	unsigned long flags;
	struct srp_iu *iu;
1869
	int err;
1870

1871 1872 1873 1874
	spin_lock_irqsave(&ch->lock, flags);
	ch->req_lim += req_delta;
	iu = __srp_get_tx_iu(ch, SRP_IU_RSP);
	spin_unlock_irqrestore(&ch->lock, flags);
1875

1876 1877 1878
	if (!iu) {
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "no IU available to send response\n");
1879
		return 1;
1880 1881 1882 1883 1884 1885
	}

	ib_dma_sync_single_for_cpu(dev, iu->dma, len, DMA_TO_DEVICE);
	memcpy(iu->buf, rsp, len);
	ib_dma_sync_single_for_device(dev, iu->dma, len, DMA_TO_DEVICE);

1886
	err = srp_post_send(ch, iu, len);
1887
	if (err) {
1888 1889
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "unable to post response: %d\n", err);
1890
		srp_put_tx_iu(ch, iu, SRP_IU_RSP);
1891
	}
1892 1893 1894 1895

	return err;
}

1896
static void srp_process_cred_req(struct srp_rdma_ch *ch,
1897 1898 1899 1900 1901 1902 1903 1904
				 struct srp_cred_req *req)
{
	struct srp_cred_rsp rsp = {
		.opcode = SRP_CRED_RSP,
		.tag = req->tag,
	};
	s32 delta = be32_to_cpu(req->req_lim_delta);

1905 1906
	if (srp_response_common(ch, delta, &rsp, sizeof(rsp)))
		shost_printk(KERN_ERR, ch->target->scsi_host, PFX
1907 1908 1909
			     "problems processing SRP_CRED_REQ\n");
}

1910
static void srp_process_aer_req(struct srp_rdma_ch *ch,
1911 1912
				struct srp_aer_req *req)
{
1913
	struct srp_target_port *target = ch->target;
1914 1915 1916 1917 1918 1919 1920
	struct srp_aer_rsp rsp = {
		.opcode = SRP_AER_RSP,
		.tag = req->tag,
	};
	s32 delta = be32_to_cpu(req->req_lim_delta);

	shost_printk(KERN_ERR, target->scsi_host, PFX
B
Bart Van Assche 已提交
1921
		     "ignoring AER for LUN %llu\n", scsilun_to_int(&req->lun));
1922

1923
	if (srp_response_common(ch, delta, &rsp, sizeof(rsp)))
1924 1925 1926 1927
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "problems processing SRP_AER_REQ\n");
}

C
Christoph Hellwig 已提交
1928
static void srp_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1929
{
C
Christoph Hellwig 已提交
1930 1931
	struct srp_iu *iu = container_of(wc->wr_cqe, struct srp_iu, cqe);
	struct srp_rdma_ch *ch = cq->cq_context;
1932
	struct srp_target_port *target = ch->target;
1933
	struct ib_device *dev = target->srp_host->srp_dev->dev;
1934
	int res;
1935 1936
	u8 opcode;

C
Christoph Hellwig 已提交
1937 1938 1939 1940 1941
	if (unlikely(wc->status != IB_WC_SUCCESS)) {
		srp_handle_qp_err(cq, wc, "RECV");
		return;
	}

1942
	ib_dma_sync_single_for_cpu(dev, iu->dma, ch->max_ti_iu_len,
1943
				   DMA_FROM_DEVICE);
1944 1945 1946 1947

	opcode = *(u8 *) iu->buf;

	if (0) {
1948 1949
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "recv completion, opcode 0x%02x\n", opcode);
B
Bart Van Assche 已提交
1950 1951
		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 8, 1,
			       iu->buf, wc->byte_len, true);
1952 1953 1954 1955
	}

	switch (opcode) {
	case SRP_RSP:
1956
		srp_process_rsp(ch, iu->buf);
1957 1958
		break;

1959
	case SRP_CRED_REQ:
1960
		srp_process_cred_req(ch, iu->buf);
1961 1962 1963
		break;

	case SRP_AER_REQ:
1964
		srp_process_aer_req(ch, iu->buf);
1965 1966
		break;

1967 1968
	case SRP_T_LOGOUT:
		/* XXX Handle target logout */
1969 1970
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Got target logout request\n");
1971 1972 1973
		break;

	default:
1974 1975
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled SRP opcode 0x%02x\n", opcode);
1976 1977 1978
		break;
	}

1979
	ib_dma_sync_single_for_device(dev, iu->dma, ch->max_ti_iu_len,
1980
				      DMA_FROM_DEVICE);
1981

1982
	res = srp_post_recv(ch, iu);
1983 1984 1985
	if (res != 0)
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "Recv failed with error code %d\n", res);
1986 1987
}

1988 1989
/**
 * srp_tl_err_work() - handle a transport layer error
1990
 * @work: Work structure embedded in an SRP target port.
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
 *
 * Note: This function may get invoked before the rport has been created,
 * hence the target->rport test.
 */
static void srp_tl_err_work(struct work_struct *work)
{
	struct srp_target_port *target;

	target = container_of(work, struct srp_target_port, tl_err_work);
	if (target->rport)
		srp_start_tl_fail_timers(target->rport);
}

C
Christoph Hellwig 已提交
2004 2005
static void srp_handle_qp_err(struct ib_cq *cq, struct ib_wc *wc,
		const char *opname)
2006
{
C
Christoph Hellwig 已提交
2007
	struct srp_rdma_ch *ch = cq->cq_context;
2008 2009
	struct srp_target_port *target = ch->target;

2010
	if (ch->connected && !target->qp_in_error) {
C
Christoph Hellwig 已提交
2011 2012 2013 2014
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "failed %s status %s (%d) for CQE %p\n",
			     opname, ib_wc_status_msg(wc->status), wc->status,
			     wc->wr_cqe);
2015
		queue_work(system_long_wq, &target->tl_err_work);
2016
	}
2017 2018 2019
	target->qp_in_error = true;
}

2020
static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
2021
{
2022
	struct srp_target_port *target = host_to_target(shost);
2023
	struct srp_rport *rport = target->rport;
2024
	struct srp_rdma_ch *ch;
2025 2026 2027
	struct srp_request *req;
	struct srp_iu *iu;
	struct srp_cmd *cmd;
2028
	struct ib_device *dev;
2029
	unsigned long flags;
B
Bart Van Assche 已提交
2030 2031
	u32 tag;
	u16 idx;
2032
	int len, ret;
2033 2034 2035 2036 2037 2038 2039 2040 2041 2042
	const bool in_scsi_eh = !in_interrupt() && current == shost->ehandler;

	/*
	 * The SCSI EH thread is the only context from which srp_queuecommand()
	 * can get invoked for blocked devices (SDEV_BLOCK /
	 * SDEV_CREATED_BLOCK). Avoid racing with srp_reconnect_rport() by
	 * locking the rport mutex if invoked from inside the SCSI EH.
	 */
	if (in_scsi_eh)
		mutex_lock(&rport->mutex);
2043

2044 2045 2046
	scmnd->result = srp_chkready(target->rport);
	if (unlikely(scmnd->result))
		goto err;
2047

B
Bart Van Assche 已提交
2048 2049
	WARN_ON_ONCE(scmnd->request->tag < 0);
	tag = blk_mq_unique_tag(scmnd->request);
B
Bart Van Assche 已提交
2050
	ch = &target->ch[blk_mq_unique_tag_to_hwq(tag)];
B
Bart Van Assche 已提交
2051 2052 2053 2054
	idx = blk_mq_unique_tag_to_tag(tag);
	WARN_ONCE(idx >= target->req_ring_size, "%s: tag %#x: idx %d >= %d\n",
		  dev_name(&shost->shost_gendev), tag, idx,
		  target->req_ring_size);
2055 2056 2057 2058

	spin_lock_irqsave(&ch->lock, flags);
	iu = __srp_get_tx_iu(ch, SRP_IU_CMD);
	spin_unlock_irqrestore(&ch->lock, flags);
2059

B
Bart Van Assche 已提交
2060 2061 2062 2063
	if (!iu)
		goto err;

	req = &ch->req_ring[idx];
2064
	dev = target->srp_host->srp_dev->dev;
2065
	ib_dma_sync_single_for_cpu(dev, iu->dma, target->max_iu_len,
2066
				   DMA_TO_DEVICE);
2067

2068
	scmnd->host_scribble = (void *) req;
2069 2070 2071 2072 2073

	cmd = iu->buf;
	memset(cmd, 0, sizeof *cmd);

	cmd->opcode = SRP_CMD;
B
Bart Van Assche 已提交
2074
	int_to_scsilun(scmnd->device->lun, &cmd->lun);
B
Bart Van Assche 已提交
2075
	cmd->tag    = tag;
2076 2077 2078 2079 2080
	memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);

	req->scmnd    = scmnd;
	req->cmd      = iu;

2081
	len = srp_map_data(scmnd, ch, req);
2082
	if (len < 0) {
2083
		shost_printk(KERN_ERR, target->scsi_host,
2084 2085 2086 2087
			     PFX "Failed to map data (%d)\n", len);
		/*
		 * If we ran out of memory descriptors (-ENOMEM) because an
		 * application is queuing many requests with more than
2088
		 * max_pages_per_mr sg-list elements, tell the SCSI mid-layer
2089 2090 2091 2092
		 * to reduce queue depth temporarily.
		 */
		scmnd->result = len == -ENOMEM ?
			DID_OK << 16 | QUEUE_FULL << 1 : DID_ERROR << 16;
2093
		goto err_iu;
2094 2095
	}

2096
	ib_dma_sync_single_for_device(dev, iu->dma, target->max_iu_len,
2097
				      DMA_TO_DEVICE);
2098

2099
	if (srp_post_send(ch, iu, len)) {
2100
		shost_printk(KERN_ERR, target->scsi_host, PFX "Send failed\n");
2101 2102 2103
		goto err_unmap;
	}

2104 2105
	ret = 0;

2106 2107 2108 2109
unlock_rport:
	if (in_scsi_eh)
		mutex_unlock(&rport->mutex);

2110
	return ret;
2111 2112

err_unmap:
2113
	srp_unmap_data(scmnd, ch, req);
2114

2115
err_iu:
2116
	srp_put_tx_iu(ch, iu, SRP_IU_CMD);
2117

2118 2119 2120 2121 2122 2123
	/*
	 * Avoid that the loops that iterate over the request ring can
	 * encounter a dangling SCSI command pointer.
	 */
	req->scmnd = NULL;

2124 2125 2126 2127 2128 2129 2130
err:
	if (scmnd->result) {
		scmnd->scsi_done(scmnd);
		ret = 0;
	} else {
		ret = SCSI_MLQUEUE_HOST_BUSY;
	}
2131

2132
	goto unlock_rport;
2133 2134
}

2135 2136
/*
 * Note: the resources allocated in this function are freed in
2137
 * srp_free_ch_ib().
2138
 */
2139
static int srp_alloc_iu_bufs(struct srp_rdma_ch *ch)
2140
{
2141
	struct srp_target_port *target = ch->target;
2142 2143
	int i;

2144 2145 2146
	ch->rx_ring = kcalloc(target->queue_size, sizeof(*ch->rx_ring),
			      GFP_KERNEL);
	if (!ch->rx_ring)
2147
		goto err_no_ring;
2148 2149 2150
	ch->tx_ring = kcalloc(target->queue_size, sizeof(*ch->tx_ring),
			      GFP_KERNEL);
	if (!ch->tx_ring)
2151 2152 2153
		goto err_no_ring;

	for (i = 0; i < target->queue_size; ++i) {
2154 2155 2156 2157
		ch->rx_ring[i] = srp_alloc_iu(target->srp_host,
					      ch->max_ti_iu_len,
					      GFP_KERNEL, DMA_FROM_DEVICE);
		if (!ch->rx_ring[i])
2158 2159 2160
			goto err;
	}

2161
	for (i = 0; i < target->queue_size; ++i) {
2162 2163 2164 2165
		ch->tx_ring[i] = srp_alloc_iu(target->srp_host,
					      target->max_iu_len,
					      GFP_KERNEL, DMA_TO_DEVICE);
		if (!ch->tx_ring[i])
2166
			goto err;
2167

2168
		list_add(&ch->tx_ring[i]->list, &ch->free_tx);
2169 2170 2171 2172 2173
	}

	return 0;

err:
2174
	for (i = 0; i < target->queue_size; ++i) {
2175 2176
		srp_free_iu(target->srp_host, ch->rx_ring[i]);
		srp_free_iu(target->srp_host, ch->tx_ring[i]);
2177 2178
	}

2179 2180

err_no_ring:
2181 2182 2183 2184
	kfree(ch->tx_ring);
	ch->tx_ring = NULL;
	kfree(ch->rx_ring);
	ch->rx_ring = NULL;
2185

2186 2187 2188
	return -ENOMEM;
}

2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215
static uint32_t srp_compute_rq_tmo(struct ib_qp_attr *qp_attr, int attr_mask)
{
	uint64_t T_tr_ns, max_compl_time_ms;
	uint32_t rq_tmo_jiffies;

	/*
	 * According to section 11.2.4.2 in the IBTA spec (Modify Queue Pair,
	 * table 91), both the QP timeout and the retry count have to be set
	 * for RC QP's during the RTR to RTS transition.
	 */
	WARN_ON_ONCE((attr_mask & (IB_QP_TIMEOUT | IB_QP_RETRY_CNT)) !=
		     (IB_QP_TIMEOUT | IB_QP_RETRY_CNT));

	/*
	 * Set target->rq_tmo_jiffies to one second more than the largest time
	 * it can take before an error completion is generated. See also
	 * C9-140..142 in the IBTA spec for more information about how to
	 * convert the QP Local ACK Timeout value to nanoseconds.
	 */
	T_tr_ns = 4096 * (1ULL << qp_attr->timeout);
	max_compl_time_ms = qp_attr->retry_cnt * 4 * T_tr_ns;
	do_div(max_compl_time_ms, NSEC_PER_MSEC);
	rq_tmo_jiffies = msecs_to_jiffies(max_compl_time_ms + 1000);

	return rq_tmo_jiffies;
}

2216
static void srp_cm_rep_handler(struct ib_cm_id *cm_id,
2217
			       const struct srp_login_rsp *lrsp,
2218
			       struct srp_rdma_ch *ch)
2219
{
2220
	struct srp_target_port *target = ch->target;
2221 2222 2223 2224 2225 2226
	struct ib_qp_attr *qp_attr = NULL;
	int attr_mask = 0;
	int ret;
	int i;

	if (lrsp->opcode == SRP_LOGIN_RSP) {
2227 2228
		ch->max_ti_iu_len = be32_to_cpu(lrsp->max_ti_iu_len);
		ch->req_lim       = be32_to_cpu(lrsp->req_lim_delta);
2229 2230 2231 2232 2233 2234

		/*
		 * Reserve credits for task management so we don't
		 * bounce requests back to the SCSI mid-layer.
		 */
		target->scsi_host->can_queue
2235
			= min(ch->req_lim - SRP_TSK_MGMT_SQ_SIZE,
2236
			      target->scsi_host->can_queue);
2237 2238 2239
		target->scsi_host->cmd_per_lun
			= min_t(int, target->scsi_host->can_queue,
				target->scsi_host->cmd_per_lun);
2240 2241 2242 2243 2244 2245 2246
	} else {
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled RSP opcode %#x\n", lrsp->opcode);
		ret = -ECONNRESET;
		goto error;
	}

2247 2248
	if (!ch->rx_ring) {
		ret = srp_alloc_iu_bufs(ch);
2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262
		if (ret)
			goto error;
	}

	ret = -ENOMEM;
	qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL);
	if (!qp_attr)
		goto error;

	qp_attr->qp_state = IB_QPS_RTR;
	ret = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
	if (ret)
		goto error_free;

2263
	ret = ib_modify_qp(ch->qp, qp_attr, attr_mask);
2264 2265 2266
	if (ret)
		goto error_free;

2267
	for (i = 0; i < target->queue_size; i++) {
2268 2269 2270
		struct srp_iu *iu = ch->rx_ring[i];

		ret = srp_post_recv(ch, iu);
2271 2272 2273 2274 2275 2276 2277 2278 2279
		if (ret)
			goto error_free;
	}

	qp_attr->qp_state = IB_QPS_RTS;
	ret = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
	if (ret)
		goto error_free;

2280 2281
	target->rq_tmo_jiffies = srp_compute_rq_tmo(qp_attr, attr_mask);

2282
	ret = ib_modify_qp(ch->qp, qp_attr, attr_mask);
2283 2284 2285 2286 2287 2288 2289 2290 2291
	if (ret)
		goto error_free;

	ret = ib_send_cm_rtu(cm_id, NULL, 0);

error_free:
	kfree(qp_attr);

error:
2292
	ch->status = ret;
2293 2294
}

2295 2296
static void srp_cm_rej_handler(struct ib_cm_id *cm_id,
			       struct ib_cm_event *event,
2297
			       struct srp_rdma_ch *ch)
2298
{
2299
	struct srp_target_port *target = ch->target;
2300
	struct Scsi_Host *shost = target->scsi_host;
2301 2302 2303 2304 2305 2306
	struct ib_class_port_info *cpi;
	int opcode;

	switch (event->param.rej_rcvd.reason) {
	case IB_CM_REJ_PORT_CM_REDIRECT:
		cpi = event->param.rej_rcvd.ari;
2307 2308
		ch->path.dlid = cpi->redirect_lid;
		ch->path.pkey = cpi->redirect_pkey;
2309
		cm_id->remote_cm_qpn = be32_to_cpu(cpi->redirect_qp) & 0x00ffffff;
2310
		memcpy(ch->path.dgid.raw, cpi->redirect_gid, 16);
2311

2312
		ch->status = ch->path.dlid ?
2313 2314 2315 2316
			SRP_DLID_REDIRECT : SRP_PORT_REDIRECT;
		break;

	case IB_CM_REJ_PORT_REDIRECT:
2317
		if (srp_target_is_topspin(target)) {
2318 2319 2320 2321 2322
			/*
			 * Topspin/Cisco SRP gateways incorrectly send
			 * reject reason code 25 when they mean 24
			 * (port redirect).
			 */
2323
			memcpy(ch->path.dgid.raw,
2324 2325
			       event->param.rej_rcvd.ari, 16);

2326 2327
			shost_printk(KERN_DEBUG, shost,
				     PFX "Topspin/Cisco redirect to target port GID %016llx%016llx\n",
2328 2329
				     be64_to_cpu(ch->path.dgid.global.subnet_prefix),
				     be64_to_cpu(ch->path.dgid.global.interface_id));
2330

2331
			ch->status = SRP_PORT_REDIRECT;
2332
		} else {
2333 2334
			shost_printk(KERN_WARNING, shost,
				     "  REJ reason: IB_CM_REJ_PORT_REDIRECT\n");
2335
			ch->status = -ECONNRESET;
2336 2337 2338 2339
		}
		break;

	case IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID:
2340 2341
		shost_printk(KERN_WARNING, shost,
			    "  REJ reason: IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID\n");
2342
		ch->status = -ECONNRESET;
2343 2344 2345 2346 2347 2348 2349 2350 2351
		break;

	case IB_CM_REJ_CONSUMER_DEFINED:
		opcode = *(u8 *) event->private_data;
		if (opcode == SRP_LOGIN_REJ) {
			struct srp_login_rej *rej = event->private_data;
			u32 reason = be32_to_cpu(rej->reason);

			if (reason == SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE)
2352 2353
				shost_printk(KERN_WARNING, shost,
					     PFX "SRP_LOGIN_REJ: requested max_it_iu_len too large\n");
2354
			else
B
Bart Van Assche 已提交
2355 2356
				shost_printk(KERN_WARNING, shost, PFX
					     "SRP LOGIN from %pI6 to %pI6 REJECTED, reason 0x%08x\n",
2357 2358
					     target->sgid.raw,
					     target->orig_dgid.raw, reason);
2359
		} else
2360 2361 2362
			shost_printk(KERN_WARNING, shost,
				     "  REJ reason: IB_CM_REJ_CONSUMER_DEFINED,"
				     " opcode 0x%02x\n", opcode);
2363
		ch->status = -ECONNRESET;
2364 2365
		break;

D
David Dillow 已提交
2366 2367
	case IB_CM_REJ_STALE_CONN:
		shost_printk(KERN_WARNING, shost, "  REJ reason: stale connection\n");
2368
		ch->status = SRP_STALE_CONN;
D
David Dillow 已提交
2369 2370
		break;

2371
	default:
2372 2373
		shost_printk(KERN_WARNING, shost, "  REJ reason 0x%x\n",
			     event->param.rej_rcvd.reason);
2374
		ch->status = -ECONNRESET;
2375 2376 2377 2378 2379
	}
}

static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
{
2380 2381
	struct srp_rdma_ch *ch = cm_id->context;
	struct srp_target_port *target = ch->target;
2382 2383 2384 2385
	int comp = 0;

	switch (event->event) {
	case IB_CM_REQ_ERROR:
2386 2387
		shost_printk(KERN_DEBUG, target->scsi_host,
			     PFX "Sending CM REQ failed\n");
2388
		comp = 1;
2389
		ch->status = -ECONNRESET;
2390 2391 2392 2393
		break;

	case IB_CM_REP_RECEIVED:
		comp = 1;
2394
		srp_cm_rep_handler(cm_id, event->private_data, ch);
2395 2396 2397
		break;

	case IB_CM_REJ_RECEIVED:
2398
		shost_printk(KERN_DEBUG, target->scsi_host, PFX "REJ received\n");
2399 2400
		comp = 1;

2401
		srp_cm_rej_handler(cm_id, event, ch);
2402 2403
		break;

2404
	case IB_CM_DREQ_RECEIVED:
2405 2406
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "DREQ received - connection closed\n");
2407
		ch->connected = false;
2408
		if (ib_send_cm_drep(cm_id, NULL, 0))
2409 2410
			shost_printk(KERN_ERR, target->scsi_host,
				     PFX "Sending CM DREP failed\n");
2411
		queue_work(system_long_wq, &target->tl_err_work);
2412 2413 2414
		break;

	case IB_CM_TIMEWAIT_EXIT:
2415 2416
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "connection closed\n");
2417
		comp = 1;
2418

2419
		ch->status = 0;
2420 2421
		break;

2422 2423 2424 2425 2426
	case IB_CM_MRA_RECEIVED:
	case IB_CM_DREQ_ERROR:
	case IB_CM_DREP_RECEIVED:
		break;

2427
	default:
2428 2429
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled CM event %d\n", event->event);
2430 2431 2432 2433
		break;
	}

	if (comp)
2434
		complete(&ch->done);
2435 2436 2437 2438

	return 0;
}

2439 2440 2441 2442 2443 2444 2445 2446
/**
 * srp_change_queue_depth - setting device queue depth
 * @sdev: scsi device struct
 * @qdepth: requested queue depth
 *
 * Returns queue depth.
 */
static int
2447
srp_change_queue_depth(struct scsi_device *sdev, int qdepth)
2448
{
2449
	if (!sdev->tagged_supported)
2450
		qdepth = 1;
2451
	return scsi_change_queue_depth(sdev, qdepth);
2452 2453
}

B
Bart Van Assche 已提交
2454 2455
static int srp_send_tsk_mgmt(struct srp_rdma_ch *ch, u64 req_tag, u64 lun,
			     u8 func)
2456
{
2457
	struct srp_target_port *target = ch->target;
2458
	struct srp_rport *rport = target->rport;
2459
	struct ib_device *dev = target->srp_host->srp_dev->dev;
2460 2461 2462
	struct srp_iu *iu;
	struct srp_tsk_mgmt *tsk_mgmt;

2463
	if (!ch->connected || target->qp_in_error)
2464 2465
		return -1;

2466
	init_completion(&ch->tsk_mgmt_done);
2467

2468
	/*
2469
	 * Lock the rport mutex to avoid that srp_create_ch_ib() is
2470 2471 2472
	 * invoked while a task management function is being sent.
	 */
	mutex_lock(&rport->mutex);
2473 2474 2475
	spin_lock_irq(&ch->lock);
	iu = __srp_get_tx_iu(ch, SRP_IU_TSK_MGMT);
	spin_unlock_irq(&ch->lock);
2476

2477 2478 2479
	if (!iu) {
		mutex_unlock(&rport->mutex);

2480
		return -1;
2481
	}
2482

2483 2484
	ib_dma_sync_single_for_cpu(dev, iu->dma, sizeof *tsk_mgmt,
				   DMA_TO_DEVICE);
2485 2486 2487 2488
	tsk_mgmt = iu->buf;
	memset(tsk_mgmt, 0, sizeof *tsk_mgmt);

	tsk_mgmt->opcode 	= SRP_TSK_MGMT;
B
Bart Van Assche 已提交
2489
	int_to_scsilun(lun, &tsk_mgmt->lun);
2490
	tsk_mgmt->tag		= req_tag | SRP_TAG_TSK_MGMT;
2491
	tsk_mgmt->tsk_mgmt_func = func;
2492
	tsk_mgmt->task_tag	= req_tag;
2493

2494 2495
	ib_dma_sync_single_for_device(dev, iu->dma, sizeof *tsk_mgmt,
				      DMA_TO_DEVICE);
2496 2497
	if (srp_post_send(ch, iu, sizeof(*tsk_mgmt))) {
		srp_put_tx_iu(ch, iu, SRP_IU_TSK_MGMT);
2498 2499
		mutex_unlock(&rport->mutex);

2500 2501
		return -1;
	}
2502
	mutex_unlock(&rport->mutex);
2503

2504
	if (!wait_for_completion_timeout(&ch->tsk_mgmt_done,
2505
					 msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS)))
2506
		return -1;
2507

2508 2509 2510
	return 0;
}

2511 2512
static int srp_abort(struct scsi_cmnd *scmnd)
{
2513
	struct srp_target_port *target = host_to_target(scmnd->device->host);
2514
	struct srp_request *req = (struct srp_request *) scmnd->host_scribble;
B
Bart Van Assche 已提交
2515
	u32 tag;
B
Bart Van Assche 已提交
2516
	u16 ch_idx;
2517
	struct srp_rdma_ch *ch;
2518
	int ret;
2519

2520
	shost_printk(KERN_ERR, target->scsi_host, "SRP abort called\n");
2521

B
Bart Van Assche 已提交
2522
	if (!req)
2523
		return SUCCESS;
B
Bart Van Assche 已提交
2524
	tag = blk_mq_unique_tag(scmnd->request);
B
Bart Van Assche 已提交
2525 2526 2527 2528 2529 2530 2531 2532
	ch_idx = blk_mq_unique_tag_to_hwq(tag);
	if (WARN_ON_ONCE(ch_idx >= target->ch_count))
		return SUCCESS;
	ch = &target->ch[ch_idx];
	if (!srp_claim_req(ch, req, NULL, scmnd))
		return SUCCESS;
	shost_printk(KERN_ERR, target->scsi_host,
		     "Sending SRP abort for tag %#x\n", tag);
B
Bart Van Assche 已提交
2533
	if (srp_send_tsk_mgmt(ch, tag, scmnd->device->lun,
2534
			      SRP_TSK_ABORT_TASK) == 0)
2535
		ret = SUCCESS;
2536
	else if (target->rport->state == SRP_RPORT_LOST)
2537
		ret = FAST_IO_FAIL;
2538 2539
	else
		ret = FAILED;
2540
	srp_free_req(ch, req, scmnd, 0);
B
Bart Van Assche 已提交
2541
	scmnd->result = DID_ABORT << 16;
2542
	scmnd->scsi_done(scmnd);
2543

2544
	return ret;
2545 2546 2547 2548
}

static int srp_reset_device(struct scsi_cmnd *scmnd)
{
2549
	struct srp_target_port *target = host_to_target(scmnd->device->host);
B
Bart Van Assche 已提交
2550
	struct srp_rdma_ch *ch;
2551
	int i;
2552

2553
	shost_printk(KERN_ERR, target->scsi_host, "SRP reset_device called\n");
2554

B
Bart Van Assche 已提交
2555
	ch = &target->ch[0];
2556
	if (srp_send_tsk_mgmt(ch, SRP_TAG_NO_REQ, scmnd->device->lun,
2557
			      SRP_TSK_LUN_RESET))
2558
		return FAILED;
2559
	if (ch->tsk_mgmt_status)
2560 2561
		return FAILED;

B
Bart Van Assche 已提交
2562 2563 2564 2565
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		for (i = 0; i < target->req_ring_size; ++i) {
			struct srp_request *req = &ch->req_ring[i];
2566

B
Bart Van Assche 已提交
2567 2568
			srp_finish_req(ch, req, scmnd->device, DID_RESET << 16);
		}
2569
	}
2570 2571

	return SUCCESS;
2572 2573 2574 2575 2576 2577
}

static int srp_reset_host(struct scsi_cmnd *scmnd)
{
	struct srp_target_port *target = host_to_target(scmnd->device->host);

2578
	shost_printk(KERN_ERR, target->scsi_host, PFX "SRP reset_host called\n");
2579

2580
	return srp_reconnect_rport(target->rport) == 0 ? SUCCESS : FAILED;
2581 2582
}

2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597
static int srp_slave_configure(struct scsi_device *sdev)
{
	struct Scsi_Host *shost = sdev->host;
	struct srp_target_port *target = host_to_target(shost);
	struct request_queue *q = sdev->request_queue;
	unsigned long timeout;

	if (sdev->type == TYPE_DISK) {
		timeout = max_t(unsigned, 30 * HZ, target->rq_tmo_jiffies);
		blk_queue_rq_timeout(q, timeout);
	}

	return 0;
}

2598 2599
static ssize_t show_id_ext(struct device *dev, struct device_attribute *attr,
			   char *buf)
2600
{
2601
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2602

2603
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->id_ext));
2604 2605
}

2606 2607
static ssize_t show_ioc_guid(struct device *dev, struct device_attribute *attr,
			     char *buf)
2608
{
2609
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2610

2611
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->ioc_guid));
2612 2613
}

2614 2615
static ssize_t show_service_id(struct device *dev,
			       struct device_attribute *attr, char *buf)
2616
{
2617
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2618

2619
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->service_id));
2620 2621
}

2622 2623
static ssize_t show_pkey(struct device *dev, struct device_attribute *attr,
			 char *buf)
2624
{
2625
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2626

2627
	return sprintf(buf, "0x%04x\n", be16_to_cpu(target->pkey));
2628 2629
}

B
Bart Van Assche 已提交
2630 2631 2632 2633 2634
static ssize_t show_sgid(struct device *dev, struct device_attribute *attr,
			 char *buf)
{
	struct srp_target_port *target = host_to_target(class_to_shost(dev));

2635
	return sprintf(buf, "%pI6\n", target->sgid.raw);
B
Bart Van Assche 已提交
2636 2637
}

2638 2639
static ssize_t show_dgid(struct device *dev, struct device_attribute *attr,
			 char *buf)
2640
{
2641
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
B
Bart Van Assche 已提交
2642
	struct srp_rdma_ch *ch = &target->ch[0];
2643

2644
	return sprintf(buf, "%pI6\n", ch->path.dgid.raw);
2645 2646
}

2647 2648
static ssize_t show_orig_dgid(struct device *dev,
			      struct device_attribute *attr, char *buf)
2649
{
2650
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2651

2652
	return sprintf(buf, "%pI6\n", target->orig_dgid.raw);
2653 2654
}

2655 2656 2657 2658
static ssize_t show_req_lim(struct device *dev,
			    struct device_attribute *attr, char *buf)
{
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
B
Bart Van Assche 已提交
2659 2660
	struct srp_rdma_ch *ch;
	int i, req_lim = INT_MAX;
2661

B
Bart Van Assche 已提交
2662 2663 2664 2665 2666
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		req_lim = min(req_lim, ch->req_lim);
	}
	return sprintf(buf, "%d\n", req_lim);
2667 2668
}

2669 2670
static ssize_t show_zero_req_lim(struct device *dev,
				 struct device_attribute *attr, char *buf)
2671
{
2672
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2673 2674 2675 2676

	return sprintf(buf, "%d\n", target->zero_req_lim);
}

2677 2678
static ssize_t show_local_ib_port(struct device *dev,
				  struct device_attribute *attr, char *buf)
2679
{
2680
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2681 2682 2683 2684

	return sprintf(buf, "%d\n", target->srp_host->port);
}

2685 2686
static ssize_t show_local_ib_device(struct device *dev,
				    struct device_attribute *attr, char *buf)
2687
{
2688
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2689

2690
	return sprintf(buf, "%s\n", target->srp_host->srp_dev->dev->name);
2691 2692
}

B
Bart Van Assche 已提交
2693 2694 2695 2696 2697 2698 2699 2700
static ssize_t show_ch_count(struct device *dev, struct device_attribute *attr,
			     char *buf)
{
	struct srp_target_port *target = host_to_target(class_to_shost(dev));

	return sprintf(buf, "%d\n", target->ch_count);
}

2701 2702 2703 2704 2705 2706 2707 2708
static ssize_t show_comp_vector(struct device *dev,
				struct device_attribute *attr, char *buf)
{
	struct srp_target_port *target = host_to_target(class_to_shost(dev));

	return sprintf(buf, "%d\n", target->comp_vector);
}

2709 2710 2711 2712 2713 2714 2715 2716
static ssize_t show_tl_retry_count(struct device *dev,
				   struct device_attribute *attr, char *buf)
{
	struct srp_target_port *target = host_to_target(class_to_shost(dev));

	return sprintf(buf, "%d\n", target->tl_retry_count);
}

2717 2718 2719 2720 2721 2722 2723 2724
static ssize_t show_cmd_sg_entries(struct device *dev,
				   struct device_attribute *attr, char *buf)
{
	struct srp_target_port *target = host_to_target(class_to_shost(dev));

	return sprintf(buf, "%u\n", target->cmd_sg_cnt);
}

2725 2726 2727 2728 2729 2730 2731 2732
static ssize_t show_allow_ext_sg(struct device *dev,
				 struct device_attribute *attr, char *buf)
{
	struct srp_target_port *target = host_to_target(class_to_shost(dev));

	return sprintf(buf, "%s\n", target->allow_ext_sg ? "true" : "false");
}

2733 2734 2735 2736
static DEVICE_ATTR(id_ext,	    S_IRUGO, show_id_ext,	   NULL);
static DEVICE_ATTR(ioc_guid,	    S_IRUGO, show_ioc_guid,	   NULL);
static DEVICE_ATTR(service_id,	    S_IRUGO, show_service_id,	   NULL);
static DEVICE_ATTR(pkey,	    S_IRUGO, show_pkey,		   NULL);
B
Bart Van Assche 已提交
2737
static DEVICE_ATTR(sgid,	    S_IRUGO, show_sgid,		   NULL);
2738 2739
static DEVICE_ATTR(dgid,	    S_IRUGO, show_dgid,		   NULL);
static DEVICE_ATTR(orig_dgid,	    S_IRUGO, show_orig_dgid,	   NULL);
2740
static DEVICE_ATTR(req_lim,         S_IRUGO, show_req_lim,         NULL);
2741 2742 2743
static DEVICE_ATTR(zero_req_lim,    S_IRUGO, show_zero_req_lim,	   NULL);
static DEVICE_ATTR(local_ib_port,   S_IRUGO, show_local_ib_port,   NULL);
static DEVICE_ATTR(local_ib_device, S_IRUGO, show_local_ib_device, NULL);
B
Bart Van Assche 已提交
2744
static DEVICE_ATTR(ch_count,        S_IRUGO, show_ch_count,        NULL);
2745
static DEVICE_ATTR(comp_vector,     S_IRUGO, show_comp_vector,     NULL);
2746
static DEVICE_ATTR(tl_retry_count,  S_IRUGO, show_tl_retry_count,  NULL);
2747
static DEVICE_ATTR(cmd_sg_entries,  S_IRUGO, show_cmd_sg_entries,  NULL);
2748
static DEVICE_ATTR(allow_ext_sg,    S_IRUGO, show_allow_ext_sg,    NULL);
2749 2750 2751 2752 2753 2754

static struct device_attribute *srp_host_attrs[] = {
	&dev_attr_id_ext,
	&dev_attr_ioc_guid,
	&dev_attr_service_id,
	&dev_attr_pkey,
B
Bart Van Assche 已提交
2755
	&dev_attr_sgid,
2756 2757
	&dev_attr_dgid,
	&dev_attr_orig_dgid,
2758
	&dev_attr_req_lim,
2759 2760 2761
	&dev_attr_zero_req_lim,
	&dev_attr_local_ib_port,
	&dev_attr_local_ib_device,
B
Bart Van Assche 已提交
2762
	&dev_attr_ch_count,
2763
	&dev_attr_comp_vector,
2764
	&dev_attr_tl_retry_count,
2765
	&dev_attr_cmd_sg_entries,
2766
	&dev_attr_allow_ext_sg,
2767 2768 2769
	NULL
};

2770 2771
static struct scsi_host_template srp_template = {
	.module				= THIS_MODULE,
R
Roland Dreier 已提交
2772 2773
	.name				= "InfiniBand SRP initiator",
	.proc_name			= DRV_NAME,
2774
	.slave_configure		= srp_slave_configure,
2775 2776
	.info				= srp_target_info,
	.queuecommand			= srp_queuecommand,
2777
	.change_queue_depth             = srp_change_queue_depth,
2778 2779 2780
	.eh_abort_handler		= srp_abort,
	.eh_device_reset_handler	= srp_reset_device,
	.eh_host_reset_handler		= srp_reset_host,
B
Bart Van Assche 已提交
2781
	.skip_settle_delay		= true,
2782
	.sg_tablesize			= SRP_DEF_SG_TABLESIZE,
2783
	.can_queue			= SRP_DEFAULT_CMD_SQ_SIZE,
2784
	.this_id			= -1,
2785
	.cmd_per_lun			= SRP_DEFAULT_CMD_SQ_SIZE,
2786
	.use_clustering			= ENABLE_CLUSTERING,
B
Bart Van Assche 已提交
2787
	.shost_attrs			= srp_host_attrs,
2788
	.track_queue_depth		= 1,
2789 2790
};

2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801
static int srp_sdev_count(struct Scsi_Host *host)
{
	struct scsi_device *sdev;
	int c = 0;

	shost_for_each_device(sdev, host)
		c++;

	return c;
}

2802 2803 2804 2805 2806 2807 2808
/*
 * Return values:
 * < 0 upon failure. Caller is responsible for SRP target port cleanup.
 * 0 and target->state == SRP_TARGET_REMOVED if asynchronous target port
 *    removal has been scheduled.
 * 0 and target->state != SRP_TARGET_REMOVED upon success.
 */
2809 2810
static int srp_add_target(struct srp_host *host, struct srp_target_port *target)
{
2811 2812 2813
	struct srp_rport_identifiers ids;
	struct srp_rport *rport;

2814
	target->state = SRP_TARGET_SCANNING;
2815
	sprintf(target->target_name, "SRP.T10:%016llX",
2816
		be64_to_cpu(target->id_ext));
2817

2818
	if (scsi_add_host(target->scsi_host, host->srp_dev->dev->dma_device))
2819 2820
		return -ENODEV;

2821 2822
	memcpy(ids.port_id, &target->id_ext, 8);
	memcpy(ids.port_id + 8, &target->ioc_guid, 8);
2823
	ids.roles = SRP_RPORT_ROLE_TARGET;
2824 2825 2826 2827 2828 2829
	rport = srp_rport_add(target->scsi_host, &ids);
	if (IS_ERR(rport)) {
		scsi_remove_host(target->scsi_host);
		return PTR_ERR(rport);
	}

2830
	rport->lld_data = target;
2831
	target->rport = rport;
2832

2833
	spin_lock(&host->target_lock);
2834
	list_add_tail(&target->list, &host->target_list);
2835
	spin_unlock(&host->target_lock);
2836 2837

	scsi_scan_target(&target->scsi_host->shost_gendev,
2838
			 0, target->scsi_id, SCAN_WILD_CARD, 0);
2839

2840 2841
	if (srp_connected_ch(target) < target->ch_count ||
	    target->qp_in_error) {
2842 2843 2844 2845 2846 2847
		shost_printk(KERN_INFO, target->scsi_host,
			     PFX "SCSI scan failed - removing SCSI host\n");
		srp_queue_remove_work(target);
		goto out;
	}

2848
	pr_debug("%s: SCSI scan succeeded - detected %d LUNs\n",
2849 2850 2851 2852 2853 2854 2855 2856 2857
		 dev_name(&target->scsi_host->shost_gendev),
		 srp_sdev_count(target->scsi_host));

	spin_lock_irq(&target->lock);
	if (target->state == SRP_TARGET_SCANNING)
		target->state = SRP_TARGET_LIVE;
	spin_unlock_irq(&target->lock);

out:
2858 2859 2860
	return 0;
}

2861
static void srp_release_dev(struct device *dev)
2862 2863
{
	struct srp_host *host =
2864
		container_of(dev, struct srp_host, dev);
2865 2866 2867 2868 2869 2870

	complete(&host->released);
}

static struct class srp_class = {
	.name    = "infiniband_srp",
2871
	.dev_release = srp_release_dev
2872 2873
};

2874 2875
/**
 * srp_conn_unique() - check whether the connection to a target is unique
2876 2877
 * @host:   SRP host.
 * @target: SRP target port.
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905
 */
static bool srp_conn_unique(struct srp_host *host,
			    struct srp_target_port *target)
{
	struct srp_target_port *t;
	bool ret = false;

	if (target->state == SRP_TARGET_REMOVED)
		goto out;

	ret = true;

	spin_lock(&host->target_lock);
	list_for_each_entry(t, &host->target_list, list) {
		if (t != target &&
		    target->id_ext == t->id_ext &&
		    target->ioc_guid == t->ioc_guid &&
		    target->initiator_ext == t->initiator_ext) {
			ret = false;
			break;
		}
	}
	spin_unlock(&host->target_lock);

out:
	return ret;
}

2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921
/*
 * Target ports are added by writing
 *
 *     id_ext=<SRP ID ext>,ioc_guid=<SRP IOC GUID>,dgid=<dest GID>,
 *     pkey=<P_Key>,service_id=<service ID>
 *
 * to the add_target sysfs attribute.
 */
enum {
	SRP_OPT_ERR		= 0,
	SRP_OPT_ID_EXT		= 1 << 0,
	SRP_OPT_IOC_GUID	= 1 << 1,
	SRP_OPT_DGID		= 1 << 2,
	SRP_OPT_PKEY		= 1 << 3,
	SRP_OPT_SERVICE_ID	= 1 << 4,
	SRP_OPT_MAX_SECT	= 1 << 5,
2922
	SRP_OPT_MAX_CMD_PER_LUN	= 1 << 6,
2923
	SRP_OPT_IO_CLASS	= 1 << 7,
2924
	SRP_OPT_INITIATOR_EXT	= 1 << 8,
2925
	SRP_OPT_CMD_SG_ENTRIES	= 1 << 9,
2926 2927
	SRP_OPT_ALLOW_EXT_SG	= 1 << 10,
	SRP_OPT_SG_TABLESIZE	= 1 << 11,
2928
	SRP_OPT_COMP_VECTOR	= 1 << 12,
2929
	SRP_OPT_TL_RETRY_COUNT	= 1 << 13,
2930
	SRP_OPT_QUEUE_SIZE	= 1 << 14,
2931 2932 2933 2934 2935 2936 2937
	SRP_OPT_ALL		= (SRP_OPT_ID_EXT	|
				   SRP_OPT_IOC_GUID	|
				   SRP_OPT_DGID		|
				   SRP_OPT_PKEY		|
				   SRP_OPT_SERVICE_ID),
};

2938
static const match_table_t srp_opt_tokens = {
2939 2940 2941 2942 2943 2944 2945
	{ SRP_OPT_ID_EXT,		"id_ext=%s" 		},
	{ SRP_OPT_IOC_GUID,		"ioc_guid=%s" 		},
	{ SRP_OPT_DGID,			"dgid=%s" 		},
	{ SRP_OPT_PKEY,			"pkey=%x" 		},
	{ SRP_OPT_SERVICE_ID,		"service_id=%s"		},
	{ SRP_OPT_MAX_SECT,		"max_sect=%d" 		},
	{ SRP_OPT_MAX_CMD_PER_LUN,	"max_cmd_per_lun=%d" 	},
2946
	{ SRP_OPT_IO_CLASS,		"io_class=%x"		},
2947
	{ SRP_OPT_INITIATOR_EXT,	"initiator_ext=%s"	},
2948
	{ SRP_OPT_CMD_SG_ENTRIES,	"cmd_sg_entries=%u"	},
2949 2950
	{ SRP_OPT_ALLOW_EXT_SG,		"allow_ext_sg=%u"	},
	{ SRP_OPT_SG_TABLESIZE,		"sg_tablesize=%u"	},
2951
	{ SRP_OPT_COMP_VECTOR,		"comp_vector=%u"	},
2952
	{ SRP_OPT_TL_RETRY_COUNT,	"tl_retry_count=%u"	},
2953
	{ SRP_OPT_QUEUE_SIZE,		"queue_size=%d"		},
2954
	{ SRP_OPT_ERR,			NULL 			}
2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972
};

static int srp_parse_options(const char *buf, struct srp_target_port *target)
{
	char *options, *sep_opt;
	char *p;
	char dgid[3];
	substring_t args[MAX_OPT_ARGS];
	int opt_mask = 0;
	int token;
	int ret = -EINVAL;
	int i;

	options = kstrdup(buf, GFP_KERNEL);
	if (!options)
		return -ENOMEM;

	sep_opt = options;
2973
	while ((p = strsep(&sep_opt, ",\n")) != NULL) {
2974 2975 2976 2977 2978 2979 2980 2981 2982
		if (!*p)
			continue;

		token = match_token(p, srp_opt_tokens, args);
		opt_mask |= token;

		switch (token) {
		case SRP_OPT_ID_EXT:
			p = match_strdup(args);
2983 2984 2985 2986
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
2987 2988 2989 2990 2991 2992
			target->id_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

		case SRP_OPT_IOC_GUID:
			p = match_strdup(args);
2993 2994 2995 2996
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
2997 2998 2999 3000 3001 3002
			target->ioc_guid = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

		case SRP_OPT_DGID:
			p = match_strdup(args);
3003 3004 3005 3006
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3007
			if (strlen(p) != 32) {
3008
				pr_warn("bad dest GID parameter '%s'\n", p);
3009
				kfree(p);
3010 3011 3012 3013
				goto out;
			}

			for (i = 0; i < 16; ++i) {
3014 3015 3016 3017 3018 3019 3020
				strlcpy(dgid, p + i * 2, sizeof(dgid));
				if (sscanf(dgid, "%hhx",
					   &target->orig_dgid.raw[i]) < 1) {
					ret = -EINVAL;
					kfree(p);
					goto out;
				}
3021
			}
3022
			kfree(p);
3023 3024 3025 3026
			break;

		case SRP_OPT_PKEY:
			if (match_hex(args, &token)) {
3027
				pr_warn("bad P_Key parameter '%s'\n", p);
3028 3029
				goto out;
			}
3030
			target->pkey = cpu_to_be16(token);
3031 3032 3033 3034
			break;

		case SRP_OPT_SERVICE_ID:
			p = match_strdup(args);
3035 3036 3037 3038
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3039 3040 3041 3042 3043 3044
			target->service_id = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

		case SRP_OPT_MAX_SECT:
			if (match_int(args, &token)) {
3045
				pr_warn("bad max sect parameter '%s'\n", p);
3046 3047 3048 3049 3050
				goto out;
			}
			target->scsi_host->max_sectors = token;
			break;

3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062
		case SRP_OPT_QUEUE_SIZE:
			if (match_int(args, &token) || token < 1) {
				pr_warn("bad queue_size parameter '%s'\n", p);
				goto out;
			}
			target->scsi_host->can_queue = token;
			target->queue_size = token + SRP_RSP_SQ_SIZE +
					     SRP_TSK_MGMT_SQ_SIZE;
			if (!(opt_mask & SRP_OPT_MAX_CMD_PER_LUN))
				target->scsi_host->cmd_per_lun = token;
			break;

3063
		case SRP_OPT_MAX_CMD_PER_LUN:
3064
			if (match_int(args, &token) || token < 1) {
3065 3066
				pr_warn("bad max cmd_per_lun parameter '%s'\n",
					p);
3067 3068
				goto out;
			}
3069
			target->scsi_host->cmd_per_lun = token;
3070 3071
			break;

3072 3073
		case SRP_OPT_IO_CLASS:
			if (match_hex(args, &token)) {
3074
				pr_warn("bad IO class parameter '%s'\n", p);
3075 3076 3077 3078
				goto out;
			}
			if (token != SRP_REV10_IB_IO_CLASS &&
			    token != SRP_REV16A_IB_IO_CLASS) {
3079 3080 3081
				pr_warn("unknown IO class parameter value %x specified (use %x or %x).\n",
					token, SRP_REV10_IB_IO_CLASS,
					SRP_REV16A_IB_IO_CLASS);
3082 3083 3084 3085 3086
				goto out;
			}
			target->io_class = token;
			break;

3087 3088
		case SRP_OPT_INITIATOR_EXT:
			p = match_strdup(args);
3089 3090 3091 3092
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3093 3094 3095 3096
			target->initiator_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

3097 3098
		case SRP_OPT_CMD_SG_ENTRIES:
			if (match_int(args, &token) || token < 1 || token > 255) {
3099 3100
				pr_warn("bad max cmd_sg_entries parameter '%s'\n",
					p);
3101 3102 3103 3104 3105
				goto out;
			}
			target->cmd_sg_cnt = token;
			break;

3106 3107
		case SRP_OPT_ALLOW_EXT_SG:
			if (match_int(args, &token)) {
3108
				pr_warn("bad allow_ext_sg parameter '%s'\n", p);
3109 3110 3111 3112 3113 3114 3115 3116
				goto out;
			}
			target->allow_ext_sg = !!token;
			break;

		case SRP_OPT_SG_TABLESIZE:
			if (match_int(args, &token) || token < 1 ||
					token > SCSI_MAX_SG_CHAIN_SEGMENTS) {
3117 3118
				pr_warn("bad max sg_tablesize parameter '%s'\n",
					p);
3119 3120 3121 3122 3123
				goto out;
			}
			target->sg_tablesize = token;
			break;

3124 3125 3126 3127 3128 3129 3130 3131
		case SRP_OPT_COMP_VECTOR:
			if (match_int(args, &token) || token < 0) {
				pr_warn("bad comp_vector parameter '%s'\n", p);
				goto out;
			}
			target->comp_vector = token;
			break;

3132 3133 3134 3135 3136 3137 3138 3139 3140
		case SRP_OPT_TL_RETRY_COUNT:
			if (match_int(args, &token) || token < 2 || token > 7) {
				pr_warn("bad tl_retry_count parameter '%s' (must be a number between 2 and 7)\n",
					p);
				goto out;
			}
			target->tl_retry_count = token;
			break;

3141
		default:
3142 3143
			pr_warn("unknown parameter or missing value '%s' in target creation request\n",
				p);
3144 3145 3146 3147 3148 3149 3150 3151 3152 3153
			goto out;
		}
	}

	if ((opt_mask & SRP_OPT_ALL) == SRP_OPT_ALL)
		ret = 0;
	else
		for (i = 0; i < ARRAY_SIZE(srp_opt_tokens); ++i)
			if ((srp_opt_tokens[i].token & SRP_OPT_ALL) &&
			    !(srp_opt_tokens[i].token & opt_mask))
3154 3155
				pr_warn("target creation request is missing parameter '%s'\n",
					srp_opt_tokens[i].pattern);
3156

3157 3158 3159 3160 3161 3162
	if (target->scsi_host->cmd_per_lun > target->scsi_host->can_queue
	    && (opt_mask & SRP_OPT_MAX_CMD_PER_LUN))
		pr_warn("cmd_per_lun = %d > queue_size = %d\n",
			target->scsi_host->cmd_per_lun,
			target->scsi_host->can_queue);

3163 3164 3165 3166 3167
out:
	kfree(options);
	return ret;
}

3168 3169
static ssize_t srp_create_target(struct device *dev,
				 struct device_attribute *attr,
3170 3171 3172
				 const char *buf, size_t count)
{
	struct srp_host *host =
3173
		container_of(dev, struct srp_host, dev);
3174 3175
	struct Scsi_Host *target_host;
	struct srp_target_port *target;
3176
	struct srp_rdma_ch *ch;
3177 3178
	struct srp_device *srp_dev = host->srp_dev;
	struct ib_device *ibdev = srp_dev->dev;
B
Bart Van Assche 已提交
3179 3180
	int ret, node_idx, node, cpu, i;
	bool multich = false;
3181 3182 3183 3184 3185 3186

	target_host = scsi_host_alloc(&srp_template,
				      sizeof (struct srp_target_port));
	if (!target_host)
		return -ENOMEM;

3187
	target_host->transportt  = ib_srp_transport_template;
3188 3189
	target_host->max_channel = 0;
	target_host->max_id      = 1;
B
Bart Van Assche 已提交
3190
	target_host->max_lun     = -1LL;
A
Arne Redlich 已提交
3191
	target_host->max_cmd_len = sizeof ((struct srp_cmd *) (void *) 0L)->cdb;
R
Roland Dreier 已提交
3192

3193 3194
	target = host_to_target(target_host);

3195 3196 3197
	target->io_class	= SRP_REV16A_IB_IO_CLASS;
	target->scsi_host	= target_host;
	target->srp_host	= host;
J
Jason Gunthorpe 已提交
3198
	target->lkey		= host->srp_dev->pd->local_dma_lkey;
3199
	target->global_mr	= host->srp_dev->global_mr;
3200
	target->cmd_sg_cnt	= cmd_sg_entries;
3201 3202
	target->sg_tablesize	= indirect_sg_entries ? : cmd_sg_entries;
	target->allow_ext_sg	= allow_ext_sg;
3203
	target->tl_retry_count	= 7;
3204
	target->queue_size	= SRP_DEFAULT_QUEUE_SIZE;
3205

3206 3207 3208 3209 3210 3211
	/*
	 * Avoid that the SCSI host can be removed by srp_remove_target()
	 * before this function returns.
	 */
	scsi_host_get(target->scsi_host);

3212 3213
	mutex_lock(&host->add_target_mutex);

3214 3215
	ret = srp_parse_options(buf, target);
	if (ret)
3216
		goto out;
3217

3218 3219
	target->req_ring_size = target->queue_size - SRP_TSK_MGMT_SQ_SIZE;

3220 3221 3222 3223 3224 3225 3226
	if (!srp_conn_unique(target->srp_host, target)) {
		shost_printk(KERN_INFO, target->scsi_host,
			     PFX "Already connected to target port with id_ext=%016llx;ioc_guid=%016llx;initiator_ext=%016llx\n",
			     be64_to_cpu(target->id_ext),
			     be64_to_cpu(target->ioc_guid),
			     be64_to_cpu(target->initiator_ext));
		ret = -EEXIST;
3227
		goto out;
3228 3229
	}

3230
	if (!srp_dev->has_fmr && !srp_dev->has_fr && !target->allow_ext_sg &&
3231
	    target->cmd_sg_cnt < target->sg_tablesize) {
3232
		pr_warn("No MR pool and no external indirect descriptors, limiting sg_tablesize to cmd_sg_cnt\n");
3233 3234 3235 3236
		target->sg_tablesize = target->cmd_sg_cnt;
	}

	target_host->sg_tablesize = target->sg_tablesize;
3237
	target->mr_pool_size = target->scsi_host->can_queue;
3238 3239
	target->indirect_size = target->sg_tablesize *
				sizeof (struct srp_direct_buf);
3240 3241 3242 3243
	target->max_iu_len = sizeof (struct srp_cmd) +
			     sizeof (struct srp_indirect_buf) +
			     target->cmd_sg_cnt * sizeof (struct srp_direct_buf);

3244
	INIT_WORK(&target->tl_err_work, srp_tl_err_work);
3245
	INIT_WORK(&target->remove_work, srp_remove_work);
3246
	spin_lock_init(&target->lock);
3247
	ret = ib_query_gid(ibdev, host->port, 0, &target->sgid, NULL);
3248
	if (ret)
3249
		goto out;
3250

B
Bart Van Assche 已提交
3251 3252 3253 3254 3255 3256 3257 3258 3259
	ret = -ENOMEM;
	target->ch_count = max_t(unsigned, num_online_nodes(),
				 min(ch_count ? :
				     min(4 * num_online_nodes(),
					 ibdev->num_comp_vectors),
				     num_online_cpus()));
	target->ch = kcalloc(target->ch_count, sizeof(*target->ch),
			     GFP_KERNEL);
	if (!target->ch)
3260
		goto out;
3261

B
Bart Van Assche 已提交
3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289
	node_idx = 0;
	for_each_online_node(node) {
		const int ch_start = (node_idx * target->ch_count /
				      num_online_nodes());
		const int ch_end = ((node_idx + 1) * target->ch_count /
				    num_online_nodes());
		const int cv_start = (node_idx * ibdev->num_comp_vectors /
				      num_online_nodes() + target->comp_vector)
				     % ibdev->num_comp_vectors;
		const int cv_end = ((node_idx + 1) * ibdev->num_comp_vectors /
				    num_online_nodes() + target->comp_vector)
				   % ibdev->num_comp_vectors;
		int cpu_idx = 0;

		for_each_online_cpu(cpu) {
			if (cpu_to_node(cpu) != node)
				continue;
			if (ch_start + cpu_idx >= ch_end)
				continue;
			ch = &target->ch[ch_start + cpu_idx];
			ch->target = target;
			ch->comp_vector = cv_start == cv_end ? cv_start :
				cv_start + cpu_idx % (cv_end - cv_start);
			spin_lock_init(&ch->lock);
			INIT_LIST_HEAD(&ch->free_tx);
			ret = srp_new_cm_id(ch);
			if (ret)
				goto err_disconnect;
3290

B
Bart Van Assche 已提交
3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310
			ret = srp_create_ch_ib(ch);
			if (ret)
				goto err_disconnect;

			ret = srp_alloc_req_data(ch);
			if (ret)
				goto err_disconnect;

			ret = srp_connect_ch(ch, multich);
			if (ret) {
				shost_printk(KERN_ERR, target->scsi_host,
					     PFX "Connection %d/%d failed\n",
					     ch_start + cpu_idx,
					     target->ch_count);
				if (node_idx == 0 && cpu_idx == 0) {
					goto err_disconnect;
				} else {
					srp_free_ch_ib(target, ch);
					srp_free_req_data(target, ch);
					target->ch_count = ch - target->ch;
3311
					goto connected;
B
Bart Van Assche 已提交
3312 3313 3314 3315 3316 3317 3318
				}
			}

			multich = true;
			cpu_idx++;
		}
		node_idx++;
3319 3320
	}

3321
connected:
B
Bart Van Assche 已提交
3322 3323
	target->scsi_host->nr_hw_queues = target->ch_count;

3324 3325 3326 3327
	ret = srp_add_target(host, target);
	if (ret)
		goto err_disconnect;

3328 3329 3330 3331 3332
	if (target->state != SRP_TARGET_REMOVED) {
		shost_printk(KERN_DEBUG, target->scsi_host, PFX
			     "new target: id_ext %016llx ioc_guid %016llx pkey %04x service_id %016llx sgid %pI6 dgid %pI6\n",
			     be64_to_cpu(target->id_ext),
			     be64_to_cpu(target->ioc_guid),
3333
			     be16_to_cpu(target->pkey),
3334
			     be64_to_cpu(target->service_id),
3335
			     target->sgid.raw, target->orig_dgid.raw);
3336
	}
B
Bart Van Assche 已提交
3337

3338 3339 3340 3341
	ret = count;

out:
	mutex_unlock(&host->add_target_mutex);
3342 3343

	scsi_host_put(target->scsi_host);
3344 3345
	if (ret < 0)
		scsi_host_put(target->scsi_host);
3346

3347
	return ret;
3348 3349 3350 3351

err_disconnect:
	srp_disconnect_target(target);

B
Bart Van Assche 已提交
3352 3353 3354 3355 3356
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_ch_ib(target, ch);
		srp_free_req_data(target, ch);
	}
3357

B
Bart Van Assche 已提交
3358
	kfree(target->ch);
3359
	goto out;
3360 3361
}

3362
static DEVICE_ATTR(add_target, S_IWUSR, NULL, srp_create_target);
3363

3364 3365
static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
			  char *buf)
3366
{
3367
	struct srp_host *host = container_of(dev, struct srp_host, dev);
3368

3369
	return sprintf(buf, "%s\n", host->srp_dev->dev->name);
3370 3371
}

3372
static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
3373

3374 3375
static ssize_t show_port(struct device *dev, struct device_attribute *attr,
			 char *buf)
3376
{
3377
	struct srp_host *host = container_of(dev, struct srp_host, dev);
3378 3379 3380 3381

	return sprintf(buf, "%d\n", host->port);
}

3382
static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
3383

3384
static struct srp_host *srp_add_port(struct srp_device *device, u8 port)
3385 3386 3387 3388 3389 3390 3391 3392
{
	struct srp_host *host;

	host = kzalloc(sizeof *host, GFP_KERNEL);
	if (!host)
		return NULL;

	INIT_LIST_HEAD(&host->target_list);
3393
	spin_lock_init(&host->target_lock);
3394
	init_completion(&host->released);
3395
	mutex_init(&host->add_target_mutex);
3396
	host->srp_dev = device;
3397 3398
	host->port = port;

3399 3400
	host->dev.class = &srp_class;
	host->dev.parent = device->dev->dma_device;
3401
	dev_set_name(&host->dev, "srp-%s-%d", device->dev->name, port);
3402

3403
	if (device_register(&host->dev))
3404
		goto free_host;
3405
	if (device_create_file(&host->dev, &dev_attr_add_target))
3406
		goto err_class;
3407
	if (device_create_file(&host->dev, &dev_attr_ibdev))
3408
		goto err_class;
3409
	if (device_create_file(&host->dev, &dev_attr_port))
3410 3411 3412 3413 3414
		goto err_class;

	return host;

err_class:
3415
	device_unregister(&host->dev);
3416

3417
free_host:
3418 3419 3420 3421 3422 3423 3424
	kfree(host);

	return NULL;
}

static void srp_add_one(struct ib_device *device)
{
3425
	struct srp_device *srp_dev;
3426
	struct srp_host *host;
3427
	int mr_page_shift, p;
3428
	u64 max_pages_per_mr;
3429

3430 3431
	srp_dev = kmalloc(sizeof *srp_dev, GFP_KERNEL);
	if (!srp_dev)
3432
		return;
3433

3434 3435
	srp_dev->has_fmr = (device->alloc_fmr && device->dealloc_fmr &&
			    device->map_phys_fmr && device->unmap_fmr);
3436
	srp_dev->has_fr = (device->attrs.device_cap_flags &
3437 3438 3439 3440 3441 3442
			   IB_DEVICE_MEM_MGT_EXTENSIONS);
	if (!srp_dev->has_fmr && !srp_dev->has_fr)
		dev_warn(&device->dev, "neither FMR nor FR is supported\n");

	srp_dev->use_fast_reg = (srp_dev->has_fr &&
				 (!srp_dev->has_fmr || prefer_fr));
3443
	srp_dev->use_fmr = !srp_dev->use_fast_reg && srp_dev->has_fmr;
3444

3445 3446
	/*
	 * Use the smallest page size supported by the HCA, down to a
3447 3448
	 * minimum of 4096 bytes. We're unlikely to build large sglists
	 * out of smaller entries.
3449
	 */
3450
	mr_page_shift		= max(12, ffs(device->attrs.page_size_cap) - 1);
3451 3452
	srp_dev->mr_page_size	= 1 << mr_page_shift;
	srp_dev->mr_page_mask	= ~((u64) srp_dev->mr_page_size - 1);
3453
	max_pages_per_mr	= device->attrs.max_mr_size;
3454 3455 3456
	do_div(max_pages_per_mr, srp_dev->mr_page_size);
	srp_dev->max_pages_per_mr = min_t(u64, SRP_MAX_PAGES_PER_MR,
					  max_pages_per_mr);
3457 3458 3459
	if (srp_dev->use_fast_reg) {
		srp_dev->max_pages_per_mr =
			min_t(u32, srp_dev->max_pages_per_mr,
3460
			      device->attrs.max_fast_reg_page_list_len);
3461
	}
3462 3463
	srp_dev->mr_max_size	= srp_dev->mr_page_size *
				   srp_dev->max_pages_per_mr;
3464 3465 3466
	pr_debug("%s: mr_page_shift = %d, device->max_mr_size = %#llx, device->max_fast_reg_page_list_len = %u, max_pages_per_mr = %d, mr_max_size = %#x\n",
		 device->name, mr_page_shift, device->attrs.max_mr_size,
		 device->attrs.max_fast_reg_page_list_len,
3467
		 srp_dev->max_pages_per_mr, srp_dev->mr_max_size);
3468 3469 3470 3471 3472 3473 3474 3475

	INIT_LIST_HEAD(&srp_dev->dev_list);

	srp_dev->dev = device;
	srp_dev->pd  = ib_alloc_pd(device);
	if (IS_ERR(srp_dev->pd))
		goto free_dev;

3476 3477 3478 3479 3480 3481 3482 3483 3484 3485
	if (!register_always || (!srp_dev->has_fmr && !srp_dev->has_fr)) {
		srp_dev->global_mr = ib_get_dma_mr(srp_dev->pd,
						   IB_ACCESS_LOCAL_WRITE |
						   IB_ACCESS_REMOTE_READ |
						   IB_ACCESS_REMOTE_WRITE);
		if (IS_ERR(srp_dev->global_mr))
			goto err_pd;
	} else {
		srp_dev->global_mr = NULL;
	}
3486

3487
	for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) {
3488
		host = srp_add_port(srp_dev, p);
3489
		if (host)
3490
			list_add_tail(&host->list, &srp_dev->dev_list);
3491 3492
	}

3493
	ib_set_client_data(device, &srp_client, srp_dev);
3494
	return;
3495 3496 3497 3498 3499 3500

err_pd:
	ib_dealloc_pd(srp_dev->pd);

free_dev:
	kfree(srp_dev);
3501 3502
}

3503
static void srp_remove_one(struct ib_device *device, void *client_data)
3504
{
3505
	struct srp_device *srp_dev;
3506
	struct srp_host *host, *tmp_host;
3507
	struct srp_target_port *target;
3508

3509
	srp_dev = client_data;
3510 3511
	if (!srp_dev)
		return;
3512

3513
	list_for_each_entry_safe(host, tmp_host, &srp_dev->dev_list, list) {
3514
		device_unregister(&host->dev);
3515 3516 3517 3518 3519 3520 3521
		/*
		 * Wait for the sysfs entry to go away, so that no new
		 * target ports can be created.
		 */
		wait_for_completion(&host->released);

		/*
3522
		 * Remove all target ports.
3523
		 */
3524
		spin_lock(&host->target_lock);
3525 3526
		list_for_each_entry(target, &host->target_list, list)
			srp_queue_remove_work(target);
3527
		spin_unlock(&host->target_lock);
3528 3529

		/*
3530
		 * Wait for tl_err and target port removal tasks.
3531
		 */
3532
		flush_workqueue(system_long_wq);
3533
		flush_workqueue(srp_remove_wq);
3534 3535 3536 3537

		kfree(host);
	}

3538 3539
	if (srp_dev->global_mr)
		ib_dereg_mr(srp_dev->global_mr);
3540 3541 3542
	ib_dealloc_pd(srp_dev->pd);

	kfree(srp_dev);
3543 3544
}

3545
static struct srp_function_template ib_srp_transport_functions = {
3546 3547
	.has_rport_state	 = true,
	.reset_timer_if_blocked	 = true,
3548
	.reconnect_delay	 = &srp_reconnect_delay,
3549 3550 3551
	.fast_io_fail_tmo	 = &srp_fast_io_fail_tmo,
	.dev_loss_tmo		 = &srp_dev_loss_tmo,
	.reconnect		 = srp_rport_reconnect,
3552
	.rport_delete		 = srp_rport_delete,
3553
	.terminate_rport_io	 = srp_terminate_io,
3554 3555
};

3556 3557 3558 3559
static int __init srp_init_module(void)
{
	int ret;

3560
	if (srp_sg_tablesize) {
3561
		pr_warn("srp_sg_tablesize is deprecated, please use cmd_sg_entries\n");
3562 3563 3564 3565 3566 3567 3568 3569
		if (!cmd_sg_entries)
			cmd_sg_entries = srp_sg_tablesize;
	}

	if (!cmd_sg_entries)
		cmd_sg_entries = SRP_DEF_SG_TABLESIZE;

	if (cmd_sg_entries > 255) {
3570
		pr_warn("Clamping cmd_sg_entries to 255\n");
3571
		cmd_sg_entries = 255;
3572 3573
	}

3574 3575 3576
	if (!indirect_sg_entries)
		indirect_sg_entries = cmd_sg_entries;
	else if (indirect_sg_entries < cmd_sg_entries) {
3577 3578
		pr_warn("Bumping up indirect_sg_entries to match cmd_sg_entries (%u)\n",
			cmd_sg_entries);
3579 3580 3581
		indirect_sg_entries = cmd_sg_entries;
	}

3582
	srp_remove_wq = create_workqueue("srp_remove");
3583 3584
	if (!srp_remove_wq) {
		ret = -ENOMEM;
3585 3586 3587 3588
		goto out;
	}

	ret = -ENOMEM;
3589 3590 3591
	ib_srp_transport_template =
		srp_attach_transport(&ib_srp_transport_functions);
	if (!ib_srp_transport_template)
3592
		goto destroy_wq;
3593

3594 3595
	ret = class_register(&srp_class);
	if (ret) {
3596
		pr_err("couldn't register class infiniband_srp\n");
3597
		goto release_tr;
3598 3599
	}

3600 3601
	ib_sa_register_client(&srp_sa_client);

3602 3603
	ret = ib_register_client(&srp_client);
	if (ret) {
3604
		pr_err("couldn't register IB client\n");
3605
		goto unreg_sa;
3606 3607
	}

3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620
out:
	return ret;

unreg_sa:
	ib_sa_unregister_client(&srp_sa_client);
	class_unregister(&srp_class);

release_tr:
	srp_release_transport(ib_srp_transport_template);

destroy_wq:
	destroy_workqueue(srp_remove_wq);
	goto out;
3621 3622 3623 3624 3625
}

static void __exit srp_cleanup_module(void)
{
	ib_unregister_client(&srp_client);
3626
	ib_sa_unregister_client(&srp_sa_client);
3627
	class_unregister(&srp_class);
3628
	srp_release_transport(ib_srp_transport_template);
3629
	destroy_workqueue(srp_remove_wq);
3630 3631 3632 3633
}

module_init(srp_init_module);
module_exit(srp_cleanup_module);