ib_srp.c 100.3 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 <linux/lockdep.h>
44
#include <rdma/ib_cache.h>
45

A
Arun Sharma 已提交
46
#include <linux/atomic.h>
47 48 49 50

#include <scsi/scsi.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_dbg.h>
51
#include <scsi/scsi_tcq.h>
52
#include <scsi/srp.h>
53
#include <scsi/scsi_transport_srp.h>
54 55 56 57 58

#include "ib_srp.h"

#define DRV_NAME	"ib_srp"
#define PFX		DRV_NAME ": "
59 60
#define DRV_VERSION	"2.0"
#define DRV_RELDATE	"July 26, 2015"
61 62

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

67 68 69 70 71
#if !defined(CONFIG_DYNAMIC_DEBUG)
#define DEFINE_DYNAMIC_DEBUG_METADATA(name, fmt)
#define DYNAMIC_DEBUG_BRANCH(descriptor) false
#endif

72 73
static unsigned int srp_sg_tablesize;
static unsigned int cmd_sg_entries;
74 75
static unsigned int indirect_sg_entries;
static bool allow_ext_sg;
76 77
static bool prefer_fr = true;
static bool register_always = true;
78
static bool never_register;
79
static int topspin_workarounds = 1;
80

81 82
module_param(srp_sg_tablesize, uint, 0444);
MODULE_PARM_DESC(srp_sg_tablesize, "Deprecated name for cmd_sg_entries");
83

84 85 86
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)");
87

88 89
module_param(indirect_sg_entries, uint, 0444);
MODULE_PARM_DESC(indirect_sg_entries,
90
		 "Default max number of gather/scatter entries (default is 12, max is " __stringify(SG_MAX_SEGMENTS) ")");
91 92 93 94 95

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)");

96 97 98 99
module_param(topspin_workarounds, int, 0444);
MODULE_PARM_DESC(topspin_workarounds,
		 "Enable workarounds for Topspin/Cisco SRP target bugs if != 0");

100 101 102 103
module_param(prefer_fr, bool, 0444);
MODULE_PARM_DESC(prefer_fr,
"Whether to use fast registration if both FMR and fast registration are supported");

104 105 106 107
module_param(register_always, bool, 0444);
MODULE_PARM_DESC(register_always,
		 "Use memory registration even for contiguous memory regions");

108 109 110
module_param(never_register, bool, 0444);
MODULE_PARM_DESC(never_register, "Never register memory");

111
static const struct kernel_param_ops srp_tmo_ops;
112

113 114 115 116 117
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");

118 119 120 121 122 123 124 125
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.");

126
static int srp_dev_loss_tmo = 600;
127 128 129 130 131 132 133 134 135 136
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
137 138 139 140 141
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.");

142
static void srp_add_one(struct ib_device *device);
143
static void srp_remove_one(struct ib_device *device, void *client_data);
C
Christoph Hellwig 已提交
144 145 146
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);
147 148
static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event);

149
static struct scsi_transport_template *ib_srp_transport_template;
150
static struct workqueue_struct *srp_remove_wq;
151

152 153 154 155 156 157
static struct ib_client srp_client = {
	.name   = "srp",
	.add    = srp_add_one,
	.remove = srp_remove_one
};

158 159
static struct ib_sa_client srp_sa_client;

160 161 162 163 164 165 166 167 168 169 170 171 172 173
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;

174 175 176 177
	res = srp_parse_tmo(&tmo, val);
	if (res)
		goto out;

178 179 180 181 182
	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);
183
	else
184 185
		res = srp_tmo_valid(srp_reconnect_delay, srp_fast_io_fail_tmo,
				    tmo);
186 187 188 189 190 191 192 193
	if (res)
		goto out;
	*(int *)kp->arg = tmo;

out:
	return res;
}

194
static const struct kernel_param_ops srp_tmo_ops = {
195 196 197 198
	.get = srp_tmo_get,
	.set = srp_tmo_set,
};

199 200 201 202 203 204 205 206 207 208
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;
}

209 210 211
static int srp_target_is_topspin(struct srp_target_port *target)
{
	static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad };
212
	static const u8 cisco_oui[3]   = { 0x00, 0x1b, 0x0d };
213 214

	return topspin_workarounds &&
215 216
		(!memcmp(&target->ioc_guid, topspin_oui, sizeof topspin_oui) ||
		 !memcmp(&target->ioc_guid, cisco_oui, sizeof cisco_oui));
217 218
}

219 220 221 222 223 224 225 226 227 228 229 230 231 232
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;

233 234 235
	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))
236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
		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;

256 257
	ib_dma_unmap_single(host->srp_dev->dev, iu->dma, iu->size,
			    iu->direction);
258 259 260 261 262 263
	kfree(iu->buf);
	kfree(iu);
}

static void srp_qp_event(struct ib_event *event, void *context)
{
264 265
	pr_debug("QP event %s (%d)\n",
		 ib_event_msg(event->event), event->event);
266 267 268 269 270 271 272 273 274 275 276 277
}

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;

278 279 280 281
	ret = ib_find_cached_pkey(target->srp_host->srp_dev->dev,
				  target->srp_host->port,
				  be16_to_cpu(target->pkey),
				  &attr->pkey_index);
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
	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;
}

301
static int srp_new_cm_id(struct srp_rdma_ch *ch)
D
David Dillow 已提交
302
{
303
	struct srp_target_port *target = ch->target;
D
David Dillow 已提交
304 305
	struct ib_cm_id *new_cm_id;

306
	new_cm_id = ib_create_cm_id(target->srp_host->srp_dev->dev,
307
				    srp_cm_handler, ch);
D
David Dillow 已提交
308 309 310
	if (IS_ERR(new_cm_id))
		return PTR_ERR(new_cm_id);

311 312 313
	if (ch->cm_id)
		ib_destroy_cm_id(ch->cm_id);
	ch->cm_id = new_cm_id;
314 315 316 317 318
	if (rdma_cap_opa_ah(target->srp_host->srp_dev->dev,
			    target->srp_host->port))
		ch->path.rec_type = SA_PATH_REC_TYPE_OPA;
	else
		ch->path.rec_type = SA_PATH_REC_TYPE_IB;
319 320 321
	ch->path.sgid = target->sgid;
	ch->path.dgid = target->orig_dgid;
	ch->path.pkey = target->pkey;
322
	ch->path.service_id = target->service_id;
D
David Dillow 已提交
323 324 325 326

	return 0;
}

327 328 329 330 331 332
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));
333
	fmr_param.pool_size	    = target->mr_pool_size;
334 335
	fmr_param.dirty_watermark   = fmr_param.pool_size / 4;
	fmr_param.cache		    = 1;
336 337
	fmr_param.max_pages_per_fmr = dev->max_pages_per_mr;
	fmr_param.page_shift	    = ilog2(dev->mr_page_size);
338 339 340 341 342 343 344
	fmr_param.access	    = (IB_ACCESS_LOCAL_WRITE |
				       IB_ACCESS_REMOTE_WRITE |
				       IB_ACCESS_REMOTE_READ);

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

345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
/**
 * 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++) {
393 394
		mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG,
				 max_page_list_len);
395 396
		if (IS_ERR(mr)) {
			ret = PTR_ERR(mr);
397 398 399
			if (ret == -ENOMEM)
				pr_info("%s: ib_alloc_mr() failed. Try to reduce max_cmd_per_lun, max_sect or ch_count\n",
					dev_name(&device->dev));
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460
			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;

461
	return srp_create_fr_pool(dev->dev, dev->pd, target->mr_pool_size,
462 463 464
				  dev->max_pages_per_mr);
}

465 466
/**
 * srp_destroy_qp() - destroy an RDMA queue pair
467
 * @ch: SRP RDMA channel.
468
 *
S
Steve Wise 已提交
469 470
 * Drain the qp before destroying it.  This avoids that the receive
 * completion handler can access the queue pair while it is
471 472
 * being destroyed.
 */
473
static void srp_destroy_qp(struct srp_rdma_ch *ch)
474
{
475 476 477 478
	spin_lock_irq(&ch->lock);
	ib_process_cq_direct(ch->send_cq, -1);
	spin_unlock_irq(&ch->lock);

479 480
	ib_drain_qp(ch->qp);
	ib_destroy_qp(ch->qp);
481 482
}

483
static int srp_create_ch_ib(struct srp_rdma_ch *ch)
484
{
485
	struct srp_target_port *target = ch->target;
486
	struct srp_device *dev = target->srp_host->srp_dev;
487
	struct ib_qp_init_attr *init_attr;
488 489
	struct ib_cq *recv_cq, *send_cq;
	struct ib_qp *qp;
490
	struct ib_fmr_pool *fmr_pool = NULL;
491
	struct srp_fr_pool *fr_pool = NULL;
492
	const int m = 1 + dev->use_fast_reg * target->mr_per_cmd * 2;
493 494 495 496 497 498
	int ret;

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

S
Steve Wise 已提交
499
	/* queue_size + 1 for ib_drain_rq() */
C
Christoph Hellwig 已提交
500 501
	recv_cq = ib_alloc_cq(dev->dev, ch, target->queue_size + 1,
				ch->comp_vector, IB_POLL_SOFTIRQ);
502 503
	if (IS_ERR(recv_cq)) {
		ret = PTR_ERR(recv_cq);
504
		goto err;
505 506
	}

C
Christoph Hellwig 已提交
507 508
	send_cq = ib_alloc_cq(dev->dev, ch, m * target->queue_size,
				ch->comp_vector, IB_POLL_DIRECT);
509 510
	if (IS_ERR(send_cq)) {
		ret = PTR_ERR(send_cq);
511
		goto err_recv_cq;
512 513
	}

514
	init_attr->event_handler       = srp_qp_event;
515
	init_attr->cap.max_send_wr     = m * target->queue_size;
516
	init_attr->cap.max_recv_wr     = target->queue_size + 1;
517 518
	init_attr->cap.max_recv_sge    = 1;
	init_attr->cap.max_send_sge    = 1;
519
	init_attr->sq_sig_type         = IB_SIGNAL_REQ_WR;
520
	init_attr->qp_type             = IB_QPT_RC;
521 522
	init_attr->send_cq             = send_cq;
	init_attr->recv_cq             = recv_cq;
523

524
	qp = ib_create_qp(dev->pd, init_attr);
525 526
	if (IS_ERR(qp)) {
		ret = PTR_ERR(qp);
527
		goto err_send_cq;
528 529
	}

530
	ret = srp_init_qp(target, qp);
531 532
	if (ret)
		goto err_qp;
533

534
	if (dev->use_fast_reg) {
535 536 537 538 539 540 541
		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;
		}
542
	} else if (dev->use_fmr) {
543 544 545 546 547 548 549 550 551
		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;
		}
	}

552
	if (ch->qp)
553
		srp_destroy_qp(ch);
554
	if (ch->recv_cq)
C
Christoph Hellwig 已提交
555
		ib_free_cq(ch->recv_cq);
556
	if (ch->send_cq)
C
Christoph Hellwig 已提交
557
		ib_free_cq(ch->send_cq);
558

559 560 561
	ch->qp = qp;
	ch->recv_cq = recv_cq;
	ch->send_cq = send_cq;
562

563 564 565 566 567 568 569 570 571 572
	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;
	}

573 574 575 576
	kfree(init_attr);
	return 0;

err_qp:
577
	ib_destroy_qp(qp);
578 579

err_send_cq:
C
Christoph Hellwig 已提交
580
	ib_free_cq(send_cq);
581 582

err_recv_cq:
C
Christoph Hellwig 已提交
583
	ib_free_cq(recv_cq);
584 585

err:
586 587 588 589
	kfree(init_attr);
	return ret;
}

590 591
/*
 * Note: this function may be called without srp_alloc_iu_bufs() having been
592
 * invoked. Hence the ch->[rt]x_ring checks.
593
 */
594 595
static void srp_free_ch_ib(struct srp_target_port *target,
			   struct srp_rdma_ch *ch)
596
{
597
	struct srp_device *dev = target->srp_host->srp_dev;
598 599
	int i;

B
Bart Van Assche 已提交
600 601 602
	if (!ch->target)
		return;

603 604 605
	if (ch->cm_id) {
		ib_destroy_cm_id(ch->cm_id);
		ch->cm_id = NULL;
606 607
	}

B
Bart Van Assche 已提交
608 609 610 611
	/* If srp_new_cm_id() succeeded but srp_create_ch_ib() not, return. */
	if (!ch->qp)
		return;

612
	if (dev->use_fast_reg) {
613 614
		if (ch->fr_pool)
			srp_destroy_fr_pool(ch->fr_pool);
615
	} else if (dev->use_fmr) {
616 617
		if (ch->fmr_pool)
			ib_destroy_fmr_pool(ch->fmr_pool);
618
	}
C
Christoph Hellwig 已提交
619

620
	srp_destroy_qp(ch);
C
Christoph Hellwig 已提交
621 622
	ib_free_cq(ch->send_cq);
	ib_free_cq(ch->recv_cq);
623

B
Bart Van Assche 已提交
624 625 626 627 628 629 630 631
	/*
	 * 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;

632 633
	ch->qp = NULL;
	ch->send_cq = ch->recv_cq = NULL;
634

635
	if (ch->rx_ring) {
636
		for (i = 0; i < target->queue_size; ++i)
637 638 639
			srp_free_iu(target->srp_host, ch->rx_ring[i]);
		kfree(ch->rx_ring);
		ch->rx_ring = NULL;
640
	}
641
	if (ch->tx_ring) {
642
		for (i = 0; i < target->queue_size; ++i)
643 644 645
			srp_free_iu(target->srp_host, ch->tx_ring[i]);
		kfree(ch->tx_ring);
		ch->tx_ring = NULL;
646
	}
647 648 649
}

static void srp_path_rec_completion(int status,
650
				    struct sa_path_rec *pathrec,
651
				    void *ch_ptr)
652
{
653 654
	struct srp_rdma_ch *ch = ch_ptr;
	struct srp_target_port *target = ch->target;
655

656
	ch->status = status;
657
	if (status)
658 659
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "Got failed path rec status %d\n", status);
660
	else
661 662
		ch->path = *pathrec;
	complete(&ch->done);
663 664
}

665
static int srp_lookup_path(struct srp_rdma_ch *ch)
666
{
667
	struct srp_target_port *target = ch->target;
668
	int ret = -ENODEV;
669

670 671 672 673
	ch->path.numb_path = 1;

	init_completion(&ch->done);

674 675 676 677 678 679 680
	/*
	 * Avoid that the SCSI host can be removed by srp_remove_target()
	 * before srp_path_rec_completion() is called.
	 */
	if (!scsi_host_get(target->scsi_host))
		goto out;

681 682 683 684 685 686 687 688 689 690 691 692 693
	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);
694 695 696
	ret = ch->path_query_id;
	if (ret < 0)
		goto put;
697 698

	ret = wait_for_completion_interruptible(&ch->done);
699
	if (ret < 0)
700
		goto put;
701

702 703
	ret = ch->status;
	if (ret < 0)
704
		shost_printk(KERN_WARNING, target->scsi_host,
705 706 707 708
			     PFX "Path record query failed: sgid %pI6, dgid %pI6, pkey %#04x, service_id %#16llx\n",
			     ch->path.sgid.raw, ch->path.dgid.raw,
			     be16_to_cpu(target->pkey),
			     be64_to_cpu(target->service_id));
709

710 711 712 713 714
put:
	scsi_host_put(target->scsi_host);

out:
	return ret;
715 716
}

717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733
static u8 srp_get_subnet_timeout(struct srp_host *host)
{
	struct ib_port_attr attr;
	int ret;
	u8 subnet_timeout = 18;

	ret = ib_query_port(host->srp_dev->dev, host->port, &attr);
	if (ret == 0)
		subnet_timeout = attr.subnet_timeout;

	if (unlikely(subnet_timeout < 15))
		pr_warn("%s: subnet timeout %d may cause SRP login to fail.\n",
			dev_name(&host->srp_dev->dev->dev), subnet_timeout);

	return subnet_timeout;
}

B
Bart Van Assche 已提交
734
static int srp_send_req(struct srp_rdma_ch *ch, bool multich)
735
{
736
	struct srp_target_port *target = ch->target;
737 738 739 740 741
	struct {
		struct ib_cm_req_param param;
		struct srp_login_req   priv;
	} *req = NULL;
	int status;
742 743 744
	u8 subnet_timeout;

	subnet_timeout = srp_get_subnet_timeout(target->srp_host);
745 746 747 748 749

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

750
	req->param.primary_path		      = &ch->path;
751 752
	req->param.alternate_path 	      = NULL;
	req->param.service_id 		      = target->service_id;
753 754
	req->param.qp_num		      = ch->qp->qp_num;
	req->param.qp_type		      = ch->qp->qp_type;
755 756 757 758 759 760 761 762 763 764 765 766
	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;
767 768
	req->param.remote_cm_response_timeout = subnet_timeout + 2;
	req->param.local_cm_response_timeout  = subnet_timeout + 2;
769
	req->param.retry_count                = target->tl_retry_count;
770 771 772 773 774
	req->param.rnr_retry_count 	      = 7;
	req->param.max_cm_retries 	      = 15;

	req->priv.opcode     	= SRP_LOGIN_REQ;
	req->priv.tag        	= 0;
775
	req->priv.req_it_iu_len = cpu_to_be32(target->max_iu_len);
776 777
	req->priv.req_buf_fmt 	= cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
					      SRP_BUF_FORMAT_INDIRECT);
B
Bart Van Assche 已提交
778 779
	req->priv.req_flags	= (multich ? SRP_MULTICHAN_MULTI :
				   SRP_MULTICHAN_SINGLE);
780
	/*
R
Roland Dreier 已提交
781
	 * In the published SRP specification (draft rev. 16a), the
782 783 784 785 786 787 788 789 790
	 * 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,
791
		       &target->sgid.global.interface_id, 8);
792
		memcpy(req->priv.initiator_port_id + 8,
793
		       &target->initiator_ext, 8);
794 795 796 797
		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,
798 799
		       &target->initiator_ext, 8);
		memcpy(req->priv.initiator_port_id + 8,
800
		       &target->sgid.global.interface_id, 8);
801 802 803 804
		memcpy(req->priv.target_port_id,     &target->id_ext, 8);
		memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8);
	}

805 806
	/*
	 * Topspin/Cisco SRP targets will reject our login unless we
807 808
	 * zero out the first 8 bytes of our initiator port ID and set
	 * the second 8 bytes to the local node GUID.
809
	 */
810
	if (srp_target_is_topspin(target)) {
811 812 813
		shost_printk(KERN_DEBUG, target->scsi_host,
			     PFX "Topspin/Cisco initiator port ID workaround "
			     "activated for target GUID %016llx\n",
814
			     be64_to_cpu(target->ioc_guid));
815
		memset(req->priv.initiator_port_id, 0, 8);
816
		memcpy(req->priv.initiator_port_id + 8,
817
		       &target->srp_host->srp_dev->dev->node_guid, 8);
818 819
	}

820
	status = ib_send_cm_req(ch->cm_id, &req->param);
821 822 823 824 825 826

	kfree(req);

	return status;
}

827 828 829 830 831 832 833 834 835 836 837 838
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)
839
		queue_work(srp_remove_wq, &target->remove_work);
840 841 842 843

	return changed;
}

844 845
static void srp_disconnect_target(struct srp_target_port *target)
{
B
Bart Van Assche 已提交
846 847
	struct srp_rdma_ch *ch;
	int i;
848

849
	/* XXX should send SRP_I_LOGOUT request */
850

851 852 853 854 855 856
	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");
857
		}
858
	}
859 860
}

861 862
static void srp_free_req_data(struct srp_target_port *target,
			      struct srp_rdma_ch *ch)
863
{
864 865
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = dev->dev;
866 867 868
	struct srp_request *req;
	int i;

869
	if (!ch->req_ring)
870 871 872
		return;

	for (i = 0; i < target->req_ring_size; ++i) {
873
		req = &ch->req_ring[i];
874
		if (dev->use_fast_reg) {
875
			kfree(req->fr_list);
876
		} else {
877
			kfree(req->fmr_list);
878 879
			kfree(req->map_page);
		}
880 881 882 883 884 885
		if (req->indirect_dma_addr) {
			ib_dma_unmap_single(ibdev, req->indirect_dma_addr,
					    target->indirect_size,
					    DMA_TO_DEVICE);
		}
		kfree(req->indirect_desc);
886
	}
887

888 889
	kfree(ch->req_ring);
	ch->req_ring = NULL;
890 891
}

892
static int srp_alloc_req_data(struct srp_rdma_ch *ch)
893
{
894
	struct srp_target_port *target = ch->target;
895 896 897
	struct srp_device *srp_dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = srp_dev->dev;
	struct srp_request *req;
898
	void *mr_list;
899 900 901
	dma_addr_t dma_addr;
	int i, ret = -ENOMEM;

902 903 904
	ch->req_ring = kcalloc(target->req_ring_size, sizeof(*ch->req_ring),
			       GFP_KERNEL);
	if (!ch->req_ring)
905 906 907
		goto out;

	for (i = 0; i < target->req_ring_size; ++i) {
908
		req = &ch->req_ring[i];
909
		mr_list = kmalloc(target->mr_per_cmd * sizeof(void *),
910 911 912
				  GFP_KERNEL);
		if (!mr_list)
			goto out;
913
		if (srp_dev->use_fast_reg) {
914
			req->fr_list = mr_list;
915
		} else {
916
			req->fmr_list = mr_list;
917 918 919 920 921
			req->map_page = kmalloc(srp_dev->max_pages_per_mr *
						sizeof(void *), GFP_KERNEL);
			if (!req->map_page)
				goto out;
		}
922
		req->indirect_desc = kmalloc(target->indirect_size, GFP_KERNEL);
923
		if (!req->indirect_desc)
924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939
			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;
}

940 941 942 943 944 945 946 947 948 949 950 951 952 953 954
/**
 * 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);
}

955 956
static void srp_remove_target(struct srp_target_port *target)
{
B
Bart Van Assche 已提交
957 958
	struct srp_rdma_ch *ch;
	int i;
959

960 961
	WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);

962
	srp_del_scsi_host_attr(target->scsi_host);
963
	srp_rport_get(target->rport);
964 965
	srp_remove_host(target->scsi_host);
	scsi_remove_host(target->scsi_host);
966
	srp_stop_rport_timers(target->rport);
967
	srp_disconnect_target(target);
B
Bart Van Assche 已提交
968 969 970 971
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_ch_ib(target, ch);
	}
972
	cancel_work_sync(&target->tl_err_work);
973
	srp_rport_put(target->rport);
B
Bart Van Assche 已提交
974 975 976 977 978 979
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_req_data(target, ch);
	}
	kfree(target->ch);
	target->ch = NULL;
980 981 982 983 984

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

985 986 987
	scsi_host_put(target->scsi_host);
}

D
David Howells 已提交
988
static void srp_remove_work(struct work_struct *work)
989
{
D
David Howells 已提交
990
	struct srp_target_port *target =
991
		container_of(work, struct srp_target_port, remove_work);
992

993
	WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);
994

995
	srp_remove_target(target);
996 997
}

998 999 1000 1001 1002 1003 1004
static void srp_rport_delete(struct srp_rport *rport)
{
	struct srp_target_port *target = rport->lld_data;

	srp_queue_remove_work(target);
}

1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
/**
 * 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
1019
static int srp_connect_ch(struct srp_rdma_ch *ch, bool multich)
1020
{
1021
	struct srp_target_port *target = ch->target;
1022 1023
	int ret;

1024
	WARN_ON_ONCE(!multich && srp_connected_ch(target) > 0);
1025

1026
	ret = srp_lookup_path(ch);
1027
	if (ret)
B
Bart Van Assche 已提交
1028
		goto out;
1029 1030

	while (1) {
1031
		init_completion(&ch->done);
B
Bart Van Assche 已提交
1032
		ret = srp_send_req(ch, multich);
1033
		if (ret)
B
Bart Van Assche 已提交
1034
			goto out;
1035
		ret = wait_for_completion_interruptible(&ch->done);
1036
		if (ret < 0)
B
Bart Van Assche 已提交
1037
			goto out;
1038 1039 1040 1041 1042 1043 1044

		/*
		 * 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
1045 1046
		ret = ch->status;
		switch (ret) {
1047
		case 0:
1048
			ch->connected = true;
B
Bart Van Assche 已提交
1049
			goto out;
1050 1051

		case SRP_PORT_REDIRECT:
1052
			ret = srp_lookup_path(ch);
1053
			if (ret)
B
Bart Van Assche 已提交
1054
				goto out;
1055 1056 1057 1058 1059
			break;

		case SRP_DLID_REDIRECT:
			break;

D
David Dillow 已提交
1060 1061
		case SRP_STALE_CONN:
			shost_printk(KERN_ERR, target->scsi_host, PFX
1062
				     "giving up on stale connection\n");
B
Bart Van Assche 已提交
1063 1064
			ret = -ECONNRESET;
			goto out;
D
David Dillow 已提交
1065

1066
		default:
B
Bart Van Assche 已提交
1067
			goto out;
1068 1069
		}
	}
B
Bart Van Assche 已提交
1070 1071 1072

out:
	return ret <= 0 ? ret : -ENODEV;
1073 1074
}

C
Christoph Hellwig 已提交
1075 1076 1077 1078 1079 1080 1081
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)
1082 1083 1084 1085 1086 1087 1088 1089 1090 1091
{
	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
1092 1093
	wr.wr_cqe = &req->reg_cqe;
	req->reg_cqe.done = srp_inv_rkey_err_done;
1094
	return ib_post_send(ch->qp, &wr, &bad_wr);
1095 1096
}

1097
static void srp_unmap_data(struct scsi_cmnd *scmnd,
1098
			   struct srp_rdma_ch *ch,
1099 1100
			   struct srp_request *req)
{
1101
	struct srp_target_port *target = ch->target;
1102 1103 1104
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = dev->dev;
	int i, res;
1105

1106
	if (!scsi_sglist(scmnd) ||
1107 1108 1109 1110
	    (scmnd->sc_data_direction != DMA_TO_DEVICE &&
	     scmnd->sc_data_direction != DMA_FROM_DEVICE))
		return;

1111 1112 1113 1114
	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
1115
			res = srp_inv_rkey(req, ch, (*pfr)->mr->rkey);
1116 1117 1118 1119 1120 1121 1122 1123 1124
			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)
1125
			srp_fr_pool_put(ch->fr_pool, req->fr_list,
1126
					req->nmdesc);
1127
	} else if (dev->use_fmr) {
1128 1129 1130 1131 1132
		struct ib_pool_fmr **pfmr;

		for (i = req->nmdesc, pfmr = req->fmr_list; i > 0; i--, pfmr++)
			ib_fmr_pool_unmap(*pfmr);
	}
1133

1134 1135
	ib_dma_unmap_sg(ibdev, scsi_sglist(scmnd), scsi_sg_count(scmnd),
			scmnd->sc_data_direction);
1136 1137
}

B
Bart Van Assche 已提交
1138 1139
/**
 * srp_claim_req - Take ownership of the scmnd associated with a request.
1140
 * @ch: SRP RDMA channel.
B
Bart Van Assche 已提交
1141
 * @req: SRP request.
1142
 * @sdev: If not NULL, only take ownership for this SCSI device.
B
Bart Van Assche 已提交
1143 1144 1145 1146 1147 1148
 * @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.
 */
1149
static struct scsi_cmnd *srp_claim_req(struct srp_rdma_ch *ch,
B
Bart Van Assche 已提交
1150
				       struct srp_request *req,
1151
				       struct scsi_device *sdev,
B
Bart Van Assche 已提交
1152 1153 1154 1155
				       struct scsi_cmnd *scmnd)
{
	unsigned long flags;

1156
	spin_lock_irqsave(&ch->lock, flags);
1157 1158 1159
	if (req->scmnd &&
	    (!sdev || req->scmnd->device == sdev) &&
	    (!scmnd || req->scmnd == scmnd)) {
B
Bart Van Assche 已提交
1160 1161 1162 1163 1164
		scmnd = req->scmnd;
		req->scmnd = NULL;
	} else {
		scmnd = NULL;
	}
1165
	spin_unlock_irqrestore(&ch->lock, flags);
B
Bart Van Assche 已提交
1166 1167 1168 1169 1170

	return scmnd;
}

/**
B
Bart Van Assche 已提交
1171
 * srp_free_req() - Unmap data and adjust ch->req_lim.
1172
 * @ch:     SRP RDMA channel.
1173 1174 1175
 * @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
1176
 */
1177 1178
static void srp_free_req(struct srp_rdma_ch *ch, struct srp_request *req,
			 struct scsi_cmnd *scmnd, s32 req_lim_delta)
1179
{
1180 1181
	unsigned long flags;

1182
	srp_unmap_data(scmnd, ch, req);
B
Bart Van Assche 已提交
1183

1184 1185 1186
	spin_lock_irqsave(&ch->lock, flags);
	ch->req_lim += req_lim_delta;
	spin_unlock_irqrestore(&ch->lock, flags);
1187 1188
}

1189 1190
static void srp_finish_req(struct srp_rdma_ch *ch, struct srp_request *req,
			   struct scsi_device *sdev, int result)
1191
{
1192
	struct scsi_cmnd *scmnd = srp_claim_req(ch, req, sdev, NULL);
B
Bart Van Assche 已提交
1193 1194

	if (scmnd) {
1195
		srp_free_req(ch, req, scmnd, 0);
1196
		scmnd->result = result;
B
Bart Van Assche 已提交
1197 1198
		scmnd->scsi_done(scmnd);
	}
1199 1200
}

1201
static void srp_terminate_io(struct srp_rport *rport)
1202
{
1203
	struct srp_target_port *target = rport->lld_data;
B
Bart Van Assche 已提交
1204
	struct srp_rdma_ch *ch;
1205 1206
	struct Scsi_Host *shost = target->scsi_host;
	struct scsi_device *sdev;
B
Bart Van Assche 已提交
1207
	int i, j;
1208

1209 1210 1211 1212 1213 1214 1215
	/*
	 * 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
1216 1217
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
1218

B
Bart Van Assche 已提交
1219 1220 1221 1222 1223 1224
		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);
		}
1225 1226
	}
}
1227

1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239
/*
 * 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
1240 1241 1242
	struct srp_rdma_ch *ch;
	int i, j, ret = 0;
	bool multich = false;
1243

1244
	srp_disconnect_target(target);
1245 1246 1247 1248

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

1249
	/*
1250 1251 1252
	 * 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.
1253
	 */
B
Bart Van Assche 已提交
1254 1255 1256
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		ret += srp_new_cm_id(ch);
1257
	}
B
Bart Van Assche 已提交
1258 1259 1260 1261
	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];
1262

B
Bart Van Assche 已提交
1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273
			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);
1274

B
Bart Van Assche 已提交
1275 1276 1277 1278
		INIT_LIST_HEAD(&ch->free_tx);
		for (j = 0; j < target->queue_size; ++j)
			list_add(&ch->tx_ring[j]->list, &ch->free_tx);
	}
1279 1280 1281

	target->qp_in_error = false;

B
Bart Van Assche 已提交
1282 1283
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
1284
		if (ret)
B
Bart Van Assche 已提交
1285 1286 1287 1288
			break;
		ret = srp_connect_ch(ch, multich);
		multich = true;
	}
1289

1290 1291 1292
	if (ret == 0)
		shost_printk(KERN_INFO, target->scsi_host,
			     PFX "reconnect succeeded\n");
1293 1294 1295 1296

	return ret;
}

1297 1298
static void srp_map_desc(struct srp_map_state *state, dma_addr_t dma_addr,
			 unsigned int dma_len, u32 rkey)
1299
{
1300
	struct srp_direct_buf *desc = state->desc;
1301

1302 1303
	WARN_ON_ONCE(!dma_len);

1304 1305 1306
	desc->va = cpu_to_be64(dma_addr);
	desc->key = cpu_to_be32(rkey);
	desc->len = cpu_to_be32(dma_len);
1307

1308 1309 1310 1311
	state->total_len += dma_len;
	state->desc++;
	state->ndesc++;
}
1312

1313
static int srp_map_finish_fmr(struct srp_map_state *state,
1314
			      struct srp_rdma_ch *ch)
1315
{
1316 1317
	struct srp_target_port *target = ch->target;
	struct srp_device *dev = target->srp_host->srp_dev;
1318 1319
	struct ib_pool_fmr *fmr;
	u64 io_addr = 0;
1320

1321 1322 1323 1324
	if (state->fmr.next >= state->fmr.end) {
		shost_printk(KERN_ERR, ch->target->scsi_host,
			     PFX "Out of MRs (mr_per_cmd = %d)\n",
			     ch->target->mr_per_cmd);
1325
		return -ENOMEM;
1326
	}
1327

S
Sagi Grimberg 已提交
1328 1329 1330 1331 1332
	WARN_ON_ONCE(!dev->use_fmr);

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

B
Bart Van Assche 已提交
1333
	if (state->npages == 1 && target->global_rkey) {
S
Sagi Grimberg 已提交
1334
		srp_map_desc(state, state->base_dma_addr, state->dma_len,
B
Bart Van Assche 已提交
1335
			     target->global_rkey);
S
Sagi Grimberg 已提交
1336 1337 1338
		goto reset_state;
	}

1339
	fmr = ib_fmr_pool_map_phys(ch->fmr_pool, state->pages,
1340 1341 1342
				   state->npages, io_addr);
	if (IS_ERR(fmr))
		return PTR_ERR(fmr);
1343

1344
	*state->fmr.next++ = fmr;
1345
	state->nmdesc++;
1346

1347 1348
	srp_map_desc(state, state->base_dma_addr & ~dev->mr_page_mask,
		     state->dma_len, fmr->fmr->rkey);
1349

S
Sagi Grimberg 已提交
1350 1351 1352 1353
reset_state:
	state->npages = 0;
	state->dma_len = 0;

1354 1355 1356
	return 0;
}

C
Christoph Hellwig 已提交
1357 1358 1359 1360 1361
static void srp_reg_mr_err_done(struct ib_cq *cq, struct ib_wc *wc)
{
	srp_handle_qp_err(cq, wc, "FAST REG");
}

1362 1363 1364 1365 1366 1367
/*
 * Map up to sg_nents elements of state->sg where *sg_offset_p is the offset
 * where to start in the first element. If sg_offset_p != NULL then
 * *sg_offset_p is updated to the offset in state->sg[retval] of the first
 * byte that has not yet been mapped.
 */
1368
static int srp_map_finish_fr(struct srp_map_state *state,
C
Christoph Hellwig 已提交
1369
			     struct srp_request *req,
1370 1371
			     struct srp_rdma_ch *ch, int sg_nents,
			     unsigned int *sg_offset_p)
1372
{
1373
	struct srp_target_port *target = ch->target;
1374 1375
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_send_wr *bad_wr;
1376
	struct ib_reg_wr wr;
1377 1378
	struct srp_fr_desc *desc;
	u32 rkey;
1379
	int n, err;
1380

1381 1382 1383 1384
	if (state->fr.next >= state->fr.end) {
		shost_printk(KERN_ERR, ch->target->scsi_host,
			     PFX "Out of MRs (mr_per_cmd = %d)\n",
			     ch->target->mr_per_cmd);
1385
		return -ENOMEM;
1386
	}
1387

S
Sagi Grimberg 已提交
1388 1389
	WARN_ON_ONCE(!dev->use_fast_reg);

B
Bart Van Assche 已提交
1390
	if (sg_nents == 1 && target->global_rkey) {
1391 1392 1393 1394
		unsigned int sg_offset = sg_offset_p ? *sg_offset_p : 0;

		srp_map_desc(state, sg_dma_address(state->sg) + sg_offset,
			     sg_dma_len(state->sg) - sg_offset,
B
Bart Van Assche 已提交
1395
			     target->global_rkey);
1396 1397
		if (sg_offset_p)
			*sg_offset_p = 0;
1398
		return 1;
S
Sagi Grimberg 已提交
1399 1400
	}

1401
	desc = srp_fr_pool_get(ch->fr_pool);
1402 1403 1404 1405 1406 1407
	if (!desc)
		return -ENOMEM;

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

1408 1409
	n = ib_map_mr_sg(desc->mr, state->sg, sg_nents, sg_offset_p,
			 dev->mr_page_size);
1410 1411
	if (unlikely(n < 0)) {
		srp_fr_pool_put(ch->fr_pool, &desc, 1);
1412
		pr_debug("%s: ib_map_mr_sg(%d, %d) returned %d.\n",
1413
			 dev_name(&req->scmnd->device->sdev_gendev), sg_nents,
1414
			 sg_offset_p ? *sg_offset_p : -1, n);
1415
		return n;
1416
	}
1417

1418
	WARN_ON_ONCE(desc->mr->length == 0);
1419

C
Christoph Hellwig 已提交
1420 1421
	req->reg_cqe.done = srp_reg_mr_err_done;

1422 1423
	wr.wr.next = NULL;
	wr.wr.opcode = IB_WR_REG_MR;
C
Christoph Hellwig 已提交
1424
	wr.wr.wr_cqe = &req->reg_cqe;
1425 1426 1427 1428 1429 1430 1431
	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);
1432

1433
	*state->fr.next++ = desc;
1434 1435
	state->nmdesc++;

1436 1437
	srp_map_desc(state, desc->mr->iova,
		     desc->mr->length, desc->mr->rkey);
1438

S
Sagi Grimberg 已提交
1439
	err = ib_post_send(ch->qp, &wr.wr, &bad_wr);
1440 1441
	if (unlikely(err)) {
		WARN_ON_ONCE(err == -ENOMEM);
S
Sagi Grimberg 已提交
1442
		return err;
1443
	}
S
Sagi Grimberg 已提交
1444

1445
	return n;
1446 1447
}

1448
static int srp_map_sg_entry(struct srp_map_state *state,
1449
			    struct srp_rdma_ch *ch,
1450
			    struct scatterlist *sg)
1451
{
1452
	struct srp_target_port *target = ch->target;
1453 1454 1455 1456
	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);
1457
	unsigned int len = 0;
1458 1459
	int ret;

1460
	WARN_ON_ONCE(!dma_len);
1461

1462
	while (dma_len) {
1463
		unsigned offset = dma_addr & ~dev->mr_page_mask;
1464 1465 1466

		if (state->npages == dev->max_pages_per_mr ||
		    (state->npages > 0 && offset != 0)) {
1467
			ret = srp_map_finish_fmr(state, ch);
1468 1469 1470 1471
			if (ret)
				return ret;
		}

1472
		len = min_t(unsigned int, dma_len, dev->mr_page_size - offset);
1473

1474 1475
		if (!state->npages)
			state->base_dma_addr = dma_addr;
1476
		state->pages[state->npages++] = dma_addr & dev->mr_page_mask;
1477
		state->dma_len += len;
1478 1479 1480 1481
		dma_addr += len;
		dma_len -= len;
	}

1482
	/*
1483
	 * If the end of the MR is not on a page boundary then we need to
1484
	 * close it out and start a new one -- we can only merge at page
1485
	 * boundaries.
1486 1487
	 */
	ret = 0;
1488
	if ((dma_addr & ~dev->mr_page_mask) != 0)
1489
		ret = srp_map_finish_fmr(state, ch);
1490 1491 1492
	return ret;
}

S
Sagi Grimberg 已提交
1493 1494 1495
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)
1496 1497
{
	struct scatterlist *sg;
1498
	int i, ret;
1499

S
Sagi Grimberg 已提交
1500 1501
	state->pages = req->map_page;
	state->fmr.next = req->fmr_list;
1502
	state->fmr.end = req->fmr_list + ch->target->mr_per_cmd;
S
Sagi Grimberg 已提交
1503 1504

	for_each_sg(scat, sg, count, i) {
1505
		ret = srp_map_sg_entry(state, ch, sg);
S
Sagi Grimberg 已提交
1506 1507
		if (ret)
			return ret;
1508
	}
1509

1510
	ret = srp_map_finish_fmr(state, ch);
S
Sagi Grimberg 已提交
1511 1512 1513 1514 1515 1516 1517 1518 1519 1520
	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)
{
1521 1522
	unsigned int sg_offset = 0;

1523
	state->fr.next = req->fr_list;
1524
	state->fr.end = req->fr_list + ch->target->mr_per_cmd;
1525
	state->sg = scat;
S
Sagi Grimberg 已提交
1526

1527 1528 1529
	if (count == 0)
		return 0;

B
Bart Van Assche 已提交
1530
	while (count) {
1531
		int i, n;
S
Sagi Grimberg 已提交
1532

1533
		n = srp_map_finish_fr(state, req, ch, count, &sg_offset);
1534 1535 1536
		if (unlikely(n < 0))
			return n;

B
Bart Van Assche 已提交
1537
		count -= n;
1538 1539 1540
		for (i = 0; i < n; i++)
			state->sg = sg_next(state->sg);
	}
S
Sagi Grimberg 已提交
1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556

	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;

	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),
B
Bart Van Assche 已提交
1557
			     target->global_rkey);
1558
	}
1559

S
Sagi Grimberg 已提交
1560
	return 0;
1561 1562
}

1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578
/*
 * 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];
1579
	struct scatterlist idb_sg[1];
1580 1581 1582 1583 1584 1585 1586 1587 1588
	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;
1589 1590 1591

	if (dev->use_fast_reg) {
		state.sg = idb_sg;
1592
		sg_init_one(idb_sg, req->indirect_desc, idb_len);
1593
		idb_sg->dma_address = req->indirect_dma_addr; /* hack! */
1594 1595 1596
#ifdef CONFIG_NEED_SG_DMA_LENGTH
		idb_sg->dma_length = idb_sg->length;	      /* hack^2 */
#endif
1597
		ret = srp_map_finish_fr(&state, req, ch, 1, NULL);
1598 1599
		if (ret < 0)
			return ret;
1600
		WARN_ON_ONCE(ret < 1);
1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611
	} 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;
	}
1612 1613 1614

	*idb_rkey = idb_desc.key;

1615
	return 0;
1616 1617
}

1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641
static void srp_check_mapping(struct srp_map_state *state,
			      struct srp_rdma_ch *ch, struct srp_request *req,
			      struct scatterlist *scat, int count)
{
	struct srp_device *dev = ch->target->srp_host->srp_dev;
	struct srp_fr_desc **pfr;
	u64 desc_len = 0, mr_len = 0;
	int i;

	for (i = 0; i < state->ndesc; i++)
		desc_len += be32_to_cpu(req->indirect_desc[i].len);
	if (dev->use_fast_reg)
		for (i = 0, pfr = req->fr_list; i < state->nmdesc; i++, pfr++)
			mr_len += (*pfr)->mr->length;
	else if (dev->use_fmr)
		for (i = 0; i < state->nmdesc; i++)
			mr_len += be32_to_cpu(req->indirect_desc[i].len);
	if (desc_len != scsi_bufflen(req->scmnd) ||
	    mr_len > scsi_bufflen(req->scmnd))
		pr_err("Inconsistent: scsi len %d <> desc len %lld <> mr len %lld; ndesc %d; nmdesc = %d\n",
		       scsi_bufflen(req->scmnd), desc_len, mr_len,
		       state->ndesc, state->nmdesc);
}

1642 1643 1644 1645 1646 1647 1648 1649 1650
/**
 * 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.
 */
1651
static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_rdma_ch *ch,
1652 1653
			struct srp_request *req)
{
1654
	struct srp_target_port *target = ch->target;
1655
	struct scatterlist *scat;
1656
	struct srp_cmd *cmd = req->cmd->buf;
1657
	int len, nents, count, ret;
1658 1659
	struct srp_device *dev;
	struct ib_device *ibdev;
1660 1661
	struct srp_map_state state;
	struct srp_indirect_buf *indirect_hdr;
1662 1663
	u32 idb_len, table_len;
	__be32 idb_rkey;
1664
	u8 fmt;
1665

1666
	if (!scsi_sglist(scmnd) || scmnd->sc_data_direction == DMA_NONE)
1667 1668 1669 1670
		return sizeof (struct srp_cmd);

	if (scmnd->sc_data_direction != DMA_FROM_DEVICE &&
	    scmnd->sc_data_direction != DMA_TO_DEVICE) {
1671 1672 1673
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled data direction %d\n",
			     scmnd->sc_data_direction);
1674 1675 1676
		return -EINVAL;
	}

1677 1678
	nents = scsi_sg_count(scmnd);
	scat  = scsi_sglist(scmnd);
1679

1680
	dev = target->srp_host->srp_dev;
1681 1682 1683
	ibdev = dev->dev;

	count = ib_dma_map_sg(ibdev, scat, nents, scmnd->sc_data_direction);
1684 1685
	if (unlikely(count == 0))
		return -EIO;
1686 1687 1688

	fmt = SRP_DATA_DESC_DIRECT;
	len = sizeof (struct srp_cmd) +	sizeof (struct srp_direct_buf);
1689

B
Bart Van Assche 已提交
1690
	if (count == 1 && target->global_rkey) {
1691 1692 1693 1694 1695 1696
		/*
		 * 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.
		 */
1697
		struct srp_direct_buf *buf = (void *) cmd->add_data;
1698

1699
		buf->va  = cpu_to_be64(ib_sg_dma_address(ibdev, scat));
B
Bart Van Assche 已提交
1700
		buf->key = cpu_to_be32(target->global_rkey);
1701
		buf->len = cpu_to_be32(ib_sg_dma_len(ibdev, scat));
1702

1703
		req->nmdesc = 0;
1704 1705 1706
		goto map_complete;
	}

1707 1708 1709
	/*
	 * We have more than one scatter/gather entry, so build our indirect
	 * descriptor table, trying to merge as many entries as we can.
1710 1711 1712
	 */
	indirect_hdr = (void *) cmd->add_data;

1713 1714 1715
	ib_dma_sync_single_for_cpu(ibdev, req->indirect_dma_addr,
				   target->indirect_size, DMA_TO_DEVICE);

1716
	memset(&state, 0, sizeof(state));
B
Bart Van Assche 已提交
1717
	state.desc = req->indirect_desc;
S
Sagi Grimberg 已提交
1718
	if (dev->use_fast_reg)
1719
		ret = srp_map_sg_fr(&state, ch, req, scat, count);
S
Sagi Grimberg 已提交
1720
	else if (dev->use_fmr)
1721
		ret = srp_map_sg_fmr(&state, ch, req, scat, count);
S
Sagi Grimberg 已提交
1722
	else
1723 1724 1725 1726
		ret = srp_map_sg_dma(&state, ch, req, scat, count);
	req->nmdesc = state.nmdesc;
	if (ret < 0)
		goto unmap;
1727

1728 1729 1730
	{
		DEFINE_DYNAMIC_DEBUG_METADATA(ddm,
			"Memory mapping consistency check");
1731
		if (DYNAMIC_DEBUG_BRANCH(ddm))
1732 1733
			srp_check_mapping(&state, ch, req, scat, count);
	}
1734

1735 1736 1737 1738 1739
	/* 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.
1740 1741
	 */
	if (state.ndesc == 1) {
1742 1743
		/*
		 * Memory registration collapsed the sg-list into one entry,
1744 1745 1746
		 * so use a direct descriptor.
		 */
		struct srp_direct_buf *buf = (void *) cmd->add_data;
1747

1748
		*buf = req->indirect_desc[0];
1749
		goto map_complete;
1750 1751
	}

1752 1753 1754 1755
	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");
1756 1757
		ret = -EIO;
		goto unmap;
1758 1759 1760
	}

	count = min(state.ndesc, target->cmd_sg_cnt);
1761
	table_len = state.ndesc * sizeof (struct srp_direct_buf);
1762
	idb_len = sizeof(struct srp_indirect_buf) + table_len;
1763 1764 1765

	fmt = SRP_DATA_DESC_INDIRECT;
	len = sizeof(struct srp_cmd) + sizeof (struct srp_indirect_buf);
1766
	len += count * sizeof (struct srp_direct_buf);
1767

1768 1769
	memcpy(indirect_hdr->desc_list, req->indirect_desc,
	       count * sizeof (struct srp_direct_buf));
1770

B
Bart Van Assche 已提交
1771
	if (!target->global_rkey) {
1772 1773 1774
		ret = srp_map_idb(ch, req, state.gen.next, state.gen.end,
				  idb_len, &idb_rkey);
		if (ret < 0)
1775
			goto unmap;
1776 1777
		req->nmdesc++;
	} else {
B
Bart Van Assche 已提交
1778
		idb_rkey = cpu_to_be32(target->global_rkey);
1779 1780
	}

1781
	indirect_hdr->table_desc.va = cpu_to_be64(req->indirect_dma_addr);
1782
	indirect_hdr->table_desc.key = idb_rkey;
1783 1784 1785 1786
	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)
1787
		cmd->data_out_desc_cnt = count;
1788
	else
1789 1790 1791 1792
		cmd->data_in_desc_cnt = count;

	ib_dma_sync_single_for_device(ibdev, req->indirect_dma_addr, table_len,
				      DMA_TO_DEVICE);
1793 1794

map_complete:
1795 1796 1797 1798 1799 1800
	if (scmnd->sc_data_direction == DMA_TO_DEVICE)
		cmd->buf_fmt = fmt << 4;
	else
		cmd->buf_fmt = fmt;

	return len;
1801 1802 1803

unmap:
	srp_unmap_data(scmnd, ch, req);
1804 1805
	if (ret == -ENOMEM && req->nmdesc >= target->mr_pool_size)
		ret = -E2BIG;
1806
	return ret;
1807 1808
}

1809 1810 1811
/*
 * Return an IU and possible credit to the free pool
 */
1812
static void srp_put_tx_iu(struct srp_rdma_ch *ch, struct srp_iu *iu,
1813 1814 1815 1816
			  enum srp_iu_type iu_type)
{
	unsigned long flags;

1817 1818
	spin_lock_irqsave(&ch->lock, flags);
	list_add(&iu->list, &ch->free_tx);
1819
	if (iu_type != SRP_IU_RSP)
1820 1821
		++ch->req_lim;
	spin_unlock_irqrestore(&ch->lock, flags);
1822 1823
}

1824
/*
1825
 * Must be called with ch->lock held to protect req_lim and free_tx.
1826
 * If IU is not sent, it must be returned using srp_put_tx_iu().
1827 1828 1829 1830 1831 1832 1833 1834 1835 1836
 *
 * 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.
 */
1837
static struct srp_iu *__srp_get_tx_iu(struct srp_rdma_ch *ch,
1838 1839
				      enum srp_iu_type iu_type)
{
1840
	struct srp_target_port *target = ch->target;
1841 1842 1843
	s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE;
	struct srp_iu *iu;

1844 1845
	lockdep_assert_held(&ch->lock);

C
Christoph Hellwig 已提交
1846
	ib_process_cq_direct(ch->send_cq, -1);
1847

1848
	if (list_empty(&ch->free_tx))
1849 1850 1851
		return NULL;

	/* Initiator responses to target requests do not consume credits */
1852
	if (iu_type != SRP_IU_RSP) {
1853
		if (ch->req_lim <= rsv) {
1854 1855 1856 1857
			++target->zero_req_lim;
			return NULL;
		}

1858
		--ch->req_lim;
1859 1860
	}

1861
	iu = list_first_entry(&ch->free_tx, struct srp_iu, list);
1862
	list_del(&iu->list);
1863 1864 1865
	return iu;
}

1866 1867 1868 1869 1870
/*
 * Note: if this function is called from inside ib_drain_sq() then it will
 * be called without ch->lock being held. If ib_drain_sq() dequeues a WQE
 * with status IB_WC_SUCCESS then that's a bug.
 */
C
Christoph Hellwig 已提交
1871 1872 1873 1874 1875 1876 1877 1878 1879 1880
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;
	}

1881 1882
	lockdep_assert_held(&ch->lock);

C
Christoph Hellwig 已提交
1883 1884 1885
	list_add(&iu->list, &ch->free_tx);
}

1886
static int srp_post_send(struct srp_rdma_ch *ch, struct srp_iu *iu, int len)
1887
{
1888
	struct srp_target_port *target = ch->target;
1889 1890 1891 1892 1893
	struct ib_sge list;
	struct ib_send_wr wr, *bad_wr;

	list.addr   = iu->dma;
	list.length = len;
1894
	list.lkey   = target->lkey;
1895

C
Christoph Hellwig 已提交
1896 1897
	iu->cqe.done = srp_send_done;

1898
	wr.next       = NULL;
C
Christoph Hellwig 已提交
1899
	wr.wr_cqe     = &iu->cqe;
1900 1901 1902 1903 1904
	wr.sg_list    = &list;
	wr.num_sge    = 1;
	wr.opcode     = IB_WR_SEND;
	wr.send_flags = IB_SEND_SIGNALED;

1905
	return ib_post_send(ch->qp, &wr, &bad_wr);
1906 1907
}

1908
static int srp_post_recv(struct srp_rdma_ch *ch, struct srp_iu *iu)
1909
{
1910
	struct srp_target_port *target = ch->target;
1911
	struct ib_recv_wr wr, *bad_wr;
1912
	struct ib_sge list;
1913 1914 1915

	list.addr   = iu->dma;
	list.length = iu->size;
1916
	list.lkey   = target->lkey;
1917

C
Christoph Hellwig 已提交
1918 1919
	iu->cqe.done = srp_recv_done;

1920
	wr.next     = NULL;
C
Christoph Hellwig 已提交
1921
	wr.wr_cqe   = &iu->cqe;
1922 1923 1924
	wr.sg_list  = &list;
	wr.num_sge  = 1;

1925
	return ib_post_recv(ch->qp, &wr, &bad_wr);
1926 1927
}

1928
static void srp_process_rsp(struct srp_rdma_ch *ch, struct srp_rsp *rsp)
1929
{
1930
	struct srp_target_port *target = ch->target;
1931 1932 1933 1934 1935
	struct srp_request *req;
	struct scsi_cmnd *scmnd;
	unsigned long flags;

	if (unlikely(rsp->tag & SRP_TAG_TSK_MGMT)) {
1936 1937
		spin_lock_irqsave(&ch->lock, flags);
		ch->req_lim += be32_to_cpu(rsp->req_lim_delta);
1938 1939 1940 1941 1942 1943 1944 1945 1946 1947
		if (rsp->tag == ch->tsk_mgmt_tag) {
			ch->tsk_mgmt_status = -1;
			if (be32_to_cpu(rsp->resp_data_len) >= 4)
				ch->tsk_mgmt_status = rsp->data[3];
			complete(&ch->tsk_mgmt_done);
		} else {
			shost_printk(KERN_ERR, target->scsi_host,
				     "Received tsk mgmt response too late for tag %#llx\n",
				     rsp->tag);
		}
1948
		spin_unlock_irqrestore(&ch->lock, flags);
1949
	} else {
B
Bart Van Assche 已提交
1950
		scmnd = scsi_host_find_tag(target->scsi_host, rsp->tag);
1951
		if (scmnd && scmnd->host_scribble) {
B
Bart Van Assche 已提交
1952 1953
			req = (void *)scmnd->host_scribble;
			scmnd = srp_claim_req(ch, req, NULL, scmnd);
1954 1955
		} else {
			scmnd = NULL;
B
Bart Van Assche 已提交
1956
		}
B
Bart Van Assche 已提交
1957
		if (!scmnd) {
1958
			shost_printk(KERN_ERR, target->scsi_host,
B
Bart Van Assche 已提交
1959 1960
				     "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
1961

1962 1963 1964
			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
1965 1966 1967

			return;
		}
1968 1969 1970 1971 1972 1973 1974 1975 1976
		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
1977
		if (unlikely(rsp->flags & SRP_RSP_FLAG_DIUNDER))
1978
			scsi_set_resid(scmnd, be32_to_cpu(rsp->data_in_res_cnt));
B
Bart Van Assche 已提交
1979 1980 1981 1982 1983 1984
		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));
1985

1986
		srp_free_req(ch, req, scmnd,
B
Bart Van Assche 已提交
1987 1988
			     be32_to_cpu(rsp->req_lim_delta));

1989 1990
		scmnd->host_scribble = NULL;
		scmnd->scsi_done(scmnd);
1991 1992 1993
	}
}

1994
static int srp_response_common(struct srp_rdma_ch *ch, s32 req_delta,
1995 1996
			       void *rsp, int len)
{
1997
	struct srp_target_port *target = ch->target;
1998
	struct ib_device *dev = target->srp_host->srp_dev->dev;
1999 2000
	unsigned long flags;
	struct srp_iu *iu;
2001
	int err;
2002

2003 2004 2005 2006
	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);
2007

2008 2009 2010
	if (!iu) {
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "no IU available to send response\n");
2011
		return 1;
2012 2013 2014 2015 2016 2017
	}

	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);

2018
	err = srp_post_send(ch, iu, len);
2019
	if (err) {
2020 2021
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "unable to post response: %d\n", err);
2022
		srp_put_tx_iu(ch, iu, SRP_IU_RSP);
2023
	}
2024 2025 2026 2027

	return err;
}

2028
static void srp_process_cred_req(struct srp_rdma_ch *ch,
2029 2030 2031 2032 2033 2034 2035 2036
				 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);

2037 2038
	if (srp_response_common(ch, delta, &rsp, sizeof(rsp)))
		shost_printk(KERN_ERR, ch->target->scsi_host, PFX
2039 2040 2041
			     "problems processing SRP_CRED_REQ\n");
}

2042
static void srp_process_aer_req(struct srp_rdma_ch *ch,
2043 2044
				struct srp_aer_req *req)
{
2045
	struct srp_target_port *target = ch->target;
2046 2047 2048 2049 2050 2051 2052
	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
2053
		     "ignoring AER for LUN %llu\n", scsilun_to_int(&req->lun));
2054

2055
	if (srp_response_common(ch, delta, &rsp, sizeof(rsp)))
2056 2057 2058 2059
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "problems processing SRP_AER_REQ\n");
}

C
Christoph Hellwig 已提交
2060
static void srp_recv_done(struct ib_cq *cq, struct ib_wc *wc)
2061
{
C
Christoph Hellwig 已提交
2062 2063
	struct srp_iu *iu = container_of(wc->wr_cqe, struct srp_iu, cqe);
	struct srp_rdma_ch *ch = cq->cq_context;
2064
	struct srp_target_port *target = ch->target;
2065
	struct ib_device *dev = target->srp_host->srp_dev->dev;
2066
	int res;
2067 2068
	u8 opcode;

C
Christoph Hellwig 已提交
2069 2070 2071 2072 2073
	if (unlikely(wc->status != IB_WC_SUCCESS)) {
		srp_handle_qp_err(cq, wc, "RECV");
		return;
	}

2074
	ib_dma_sync_single_for_cpu(dev, iu->dma, ch->max_ti_iu_len,
2075
				   DMA_FROM_DEVICE);
2076 2077 2078 2079

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

	if (0) {
2080 2081
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "recv completion, opcode 0x%02x\n", opcode);
B
Bart Van Assche 已提交
2082 2083
		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 8, 1,
			       iu->buf, wc->byte_len, true);
2084 2085 2086 2087
	}

	switch (opcode) {
	case SRP_RSP:
2088
		srp_process_rsp(ch, iu->buf);
2089 2090
		break;

2091
	case SRP_CRED_REQ:
2092
		srp_process_cred_req(ch, iu->buf);
2093 2094 2095
		break;

	case SRP_AER_REQ:
2096
		srp_process_aer_req(ch, iu->buf);
2097 2098
		break;

2099 2100
	case SRP_T_LOGOUT:
		/* XXX Handle target logout */
2101 2102
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Got target logout request\n");
2103 2104 2105
		break;

	default:
2106 2107
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled SRP opcode 0x%02x\n", opcode);
2108 2109 2110
		break;
	}

2111
	ib_dma_sync_single_for_device(dev, iu->dma, ch->max_ti_iu_len,
2112
				      DMA_FROM_DEVICE);
2113

2114
	res = srp_post_recv(ch, iu);
2115 2116 2117
	if (res != 0)
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "Recv failed with error code %d\n", res);
2118 2119
}

2120 2121
/**
 * srp_tl_err_work() - handle a transport layer error
2122
 * @work: Work structure embedded in an SRP target port.
2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135
 *
 * 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
2136 2137
static void srp_handle_qp_err(struct ib_cq *cq, struct ib_wc *wc,
		const char *opname)
2138
{
C
Christoph Hellwig 已提交
2139
	struct srp_rdma_ch *ch = cq->cq_context;
2140 2141
	struct srp_target_port *target = ch->target;

2142
	if (ch->connected && !target->qp_in_error) {
C
Christoph Hellwig 已提交
2143 2144 2145 2146
		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);
2147
		queue_work(system_long_wq, &target->tl_err_work);
2148
	}
2149 2150 2151
	target->qp_in_error = true;
}

2152
static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
2153
{
2154
	struct srp_target_port *target = host_to_target(shost);
2155
	struct srp_rport *rport = target->rport;
2156
	struct srp_rdma_ch *ch;
2157 2158 2159
	struct srp_request *req;
	struct srp_iu *iu;
	struct srp_cmd *cmd;
2160
	struct ib_device *dev;
2161
	unsigned long flags;
B
Bart Van Assche 已提交
2162 2163
	u32 tag;
	u16 idx;
2164
	int len, ret;
2165 2166 2167 2168 2169 2170 2171 2172 2173 2174
	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);
2175

2176 2177 2178
	scmnd->result = srp_chkready(target->rport);
	if (unlikely(scmnd->result))
		goto err;
2179

B
Bart Van Assche 已提交
2180 2181
	WARN_ON_ONCE(scmnd->request->tag < 0);
	tag = blk_mq_unique_tag(scmnd->request);
B
Bart Van Assche 已提交
2182
	ch = &target->ch[blk_mq_unique_tag_to_hwq(tag)];
B
Bart Van Assche 已提交
2183 2184 2185 2186
	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);
2187 2188 2189 2190

	spin_lock_irqsave(&ch->lock, flags);
	iu = __srp_get_tx_iu(ch, SRP_IU_CMD);
	spin_unlock_irqrestore(&ch->lock, flags);
2191

B
Bart Van Assche 已提交
2192 2193 2194 2195
	if (!iu)
		goto err;

	req = &ch->req_ring[idx];
2196
	dev = target->srp_host->srp_dev->dev;
2197
	ib_dma_sync_single_for_cpu(dev, iu->dma, target->max_iu_len,
2198
				   DMA_TO_DEVICE);
2199

2200
	scmnd->host_scribble = (void *) req;
2201 2202 2203 2204 2205

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

	cmd->opcode = SRP_CMD;
B
Bart Van Assche 已提交
2206
	int_to_scsilun(scmnd->device->lun, &cmd->lun);
B
Bart Van Assche 已提交
2207
	cmd->tag    = tag;
2208 2209 2210 2211 2212
	memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);

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

2213
	len = srp_map_data(scmnd, ch, req);
2214
	if (len < 0) {
2215
		shost_printk(KERN_ERR, target->scsi_host,
2216 2217 2218 2219
			     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
2220
		 * max_pages_per_mr sg-list elements, tell the SCSI mid-layer
2221 2222 2223 2224
		 * to reduce queue depth temporarily.
		 */
		scmnd->result = len == -ENOMEM ?
			DID_OK << 16 | QUEUE_FULL << 1 : DID_ERROR << 16;
2225
		goto err_iu;
2226 2227
	}

2228
	ib_dma_sync_single_for_device(dev, iu->dma, target->max_iu_len,
2229
				      DMA_TO_DEVICE);
2230

2231
	if (srp_post_send(ch, iu, len)) {
2232
		shost_printk(KERN_ERR, target->scsi_host, PFX "Send failed\n");
2233 2234 2235
		goto err_unmap;
	}

2236 2237
	ret = 0;

2238 2239 2240 2241
unlock_rport:
	if (in_scsi_eh)
		mutex_unlock(&rport->mutex);

2242
	return ret;
2243 2244

err_unmap:
2245
	srp_unmap_data(scmnd, ch, req);
2246

2247
err_iu:
2248
	srp_put_tx_iu(ch, iu, SRP_IU_CMD);
2249

2250 2251 2252 2253 2254 2255
	/*
	 * Avoid that the loops that iterate over the request ring can
	 * encounter a dangling SCSI command pointer.
	 */
	req->scmnd = NULL;

2256 2257 2258 2259 2260 2261 2262
err:
	if (scmnd->result) {
		scmnd->scsi_done(scmnd);
		ret = 0;
	} else {
		ret = SCSI_MLQUEUE_HOST_BUSY;
	}
2263

2264
	goto unlock_rport;
2265 2266
}

2267 2268
/*
 * Note: the resources allocated in this function are freed in
2269
 * srp_free_ch_ib().
2270
 */
2271
static int srp_alloc_iu_bufs(struct srp_rdma_ch *ch)
2272
{
2273
	struct srp_target_port *target = ch->target;
2274 2275
	int i;

2276 2277 2278
	ch->rx_ring = kcalloc(target->queue_size, sizeof(*ch->rx_ring),
			      GFP_KERNEL);
	if (!ch->rx_ring)
2279
		goto err_no_ring;
2280 2281 2282
	ch->tx_ring = kcalloc(target->queue_size, sizeof(*ch->tx_ring),
			      GFP_KERNEL);
	if (!ch->tx_ring)
2283 2284 2285
		goto err_no_ring;

	for (i = 0; i < target->queue_size; ++i) {
2286 2287 2288 2289
		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])
2290 2291 2292
			goto err;
	}

2293
	for (i = 0; i < target->queue_size; ++i) {
2294 2295 2296 2297
		ch->tx_ring[i] = srp_alloc_iu(target->srp_host,
					      target->max_iu_len,
					      GFP_KERNEL, DMA_TO_DEVICE);
		if (!ch->tx_ring[i])
2298
			goto err;
2299

2300
		list_add(&ch->tx_ring[i]->list, &ch->free_tx);
2301 2302 2303 2304 2305
	}

	return 0;

err:
2306
	for (i = 0; i < target->queue_size; ++i) {
2307 2308
		srp_free_iu(target->srp_host, ch->rx_ring[i]);
		srp_free_iu(target->srp_host, ch->tx_ring[i]);
2309 2310
	}

2311 2312

err_no_ring:
2313 2314 2315 2316
	kfree(ch->tx_ring);
	ch->tx_ring = NULL;
	kfree(ch->rx_ring);
	ch->rx_ring = NULL;
2317

2318 2319 2320
	return -ENOMEM;
}

2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347
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;
}

2348
static void srp_cm_rep_handler(struct ib_cm_id *cm_id,
2349
			       const struct srp_login_rsp *lrsp,
2350
			       struct srp_rdma_ch *ch)
2351
{
2352
	struct srp_target_port *target = ch->target;
2353 2354 2355 2356 2357 2358
	struct ib_qp_attr *qp_attr = NULL;
	int attr_mask = 0;
	int ret;
	int i;

	if (lrsp->opcode == SRP_LOGIN_RSP) {
2359 2360
		ch->max_ti_iu_len = be32_to_cpu(lrsp->max_ti_iu_len);
		ch->req_lim       = be32_to_cpu(lrsp->req_lim_delta);
2361 2362 2363 2364 2365 2366

		/*
		 * Reserve credits for task management so we don't
		 * bounce requests back to the SCSI mid-layer.
		 */
		target->scsi_host->can_queue
2367
			= min(ch->req_lim - SRP_TSK_MGMT_SQ_SIZE,
2368
			      target->scsi_host->can_queue);
2369 2370 2371
		target->scsi_host->cmd_per_lun
			= min_t(int, target->scsi_host->can_queue,
				target->scsi_host->cmd_per_lun);
2372 2373 2374 2375 2376 2377 2378
	} else {
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled RSP opcode %#x\n", lrsp->opcode);
		ret = -ECONNRESET;
		goto error;
	}

2379 2380
	if (!ch->rx_ring) {
		ret = srp_alloc_iu_bufs(ch);
2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394
		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;

2395
	ret = ib_modify_qp(ch->qp, qp_attr, attr_mask);
2396 2397 2398
	if (ret)
		goto error_free;

2399
	for (i = 0; i < target->queue_size; i++) {
2400 2401 2402
		struct srp_iu *iu = ch->rx_ring[i];

		ret = srp_post_recv(ch, iu);
2403 2404 2405 2406 2407 2408 2409 2410 2411
		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;

2412 2413
	target->rq_tmo_jiffies = srp_compute_rq_tmo(qp_attr, attr_mask);

2414
	ret = ib_modify_qp(ch->qp, qp_attr, attr_mask);
2415 2416 2417 2418 2419 2420 2421 2422 2423
	if (ret)
		goto error_free;

	ret = ib_send_cm_rtu(cm_id, NULL, 0);

error_free:
	kfree(qp_attr);

error:
2424
	ch->status = ret;
2425 2426
}

2427 2428
static void srp_cm_rej_handler(struct ib_cm_id *cm_id,
			       struct ib_cm_event *event,
2429
			       struct srp_rdma_ch *ch)
2430
{
2431
	struct srp_target_port *target = ch->target;
2432
	struct Scsi_Host *shost = target->scsi_host;
2433 2434 2435 2436 2437 2438
	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;
2439
		sa_path_set_dlid(&ch->path, ntohs(cpi->redirect_lid));
2440
		ch->path.pkey = cpi->redirect_pkey;
2441
		cm_id->remote_cm_qpn = be32_to_cpu(cpi->redirect_qp) & 0x00ffffff;
2442
		memcpy(ch->path.dgid.raw, cpi->redirect_gid, 16);
2443

2444
		ch->status = sa_path_get_dlid(&ch->path) ?
2445 2446 2447 2448
			SRP_DLID_REDIRECT : SRP_PORT_REDIRECT;
		break;

	case IB_CM_REJ_PORT_REDIRECT:
2449
		if (srp_target_is_topspin(target)) {
2450 2451 2452 2453 2454
			/*
			 * Topspin/Cisco SRP gateways incorrectly send
			 * reject reason code 25 when they mean 24
			 * (port redirect).
			 */
2455
			memcpy(ch->path.dgid.raw,
2456 2457
			       event->param.rej_rcvd.ari, 16);

2458 2459
			shost_printk(KERN_DEBUG, shost,
				     PFX "Topspin/Cisco redirect to target port GID %016llx%016llx\n",
2460 2461
				     be64_to_cpu(ch->path.dgid.global.subnet_prefix),
				     be64_to_cpu(ch->path.dgid.global.interface_id));
2462

2463
			ch->status = SRP_PORT_REDIRECT;
2464
		} else {
2465 2466
			shost_printk(KERN_WARNING, shost,
				     "  REJ reason: IB_CM_REJ_PORT_REDIRECT\n");
2467
			ch->status = -ECONNRESET;
2468 2469 2470 2471
		}
		break;

	case IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID:
2472 2473
		shost_printk(KERN_WARNING, shost,
			    "  REJ reason: IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID\n");
2474
		ch->status = -ECONNRESET;
2475 2476 2477 2478 2479 2480 2481 2482 2483
		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)
2484 2485
				shost_printk(KERN_WARNING, shost,
					     PFX "SRP_LOGIN_REJ: requested max_it_iu_len too large\n");
2486
			else
B
Bart Van Assche 已提交
2487 2488
				shost_printk(KERN_WARNING, shost, PFX
					     "SRP LOGIN from %pI6 to %pI6 REJECTED, reason 0x%08x\n",
2489 2490
					     target->sgid.raw,
					     target->orig_dgid.raw, reason);
2491
		} else
2492 2493 2494
			shost_printk(KERN_WARNING, shost,
				     "  REJ reason: IB_CM_REJ_CONSUMER_DEFINED,"
				     " opcode 0x%02x\n", opcode);
2495
		ch->status = -ECONNRESET;
2496 2497
		break;

D
David Dillow 已提交
2498 2499
	case IB_CM_REJ_STALE_CONN:
		shost_printk(KERN_WARNING, shost, "  REJ reason: stale connection\n");
2500
		ch->status = SRP_STALE_CONN;
D
David Dillow 已提交
2501 2502
		break;

2503
	default:
2504 2505
		shost_printk(KERN_WARNING, shost, "  REJ reason 0x%x\n",
			     event->param.rej_rcvd.reason);
2506
		ch->status = -ECONNRESET;
2507 2508 2509 2510 2511
	}
}

static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
{
2512 2513
	struct srp_rdma_ch *ch = cm_id->context;
	struct srp_target_port *target = ch->target;
2514 2515 2516 2517
	int comp = 0;

	switch (event->event) {
	case IB_CM_REQ_ERROR:
2518 2519
		shost_printk(KERN_DEBUG, target->scsi_host,
			     PFX "Sending CM REQ failed\n");
2520
		comp = 1;
2521
		ch->status = -ECONNRESET;
2522 2523 2524 2525
		break;

	case IB_CM_REP_RECEIVED:
		comp = 1;
2526
		srp_cm_rep_handler(cm_id, event->private_data, ch);
2527 2528 2529
		break;

	case IB_CM_REJ_RECEIVED:
2530
		shost_printk(KERN_DEBUG, target->scsi_host, PFX "REJ received\n");
2531 2532
		comp = 1;

2533
		srp_cm_rej_handler(cm_id, event, ch);
2534 2535
		break;

2536
	case IB_CM_DREQ_RECEIVED:
2537 2538
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "DREQ received - connection closed\n");
2539
		ch->connected = false;
2540
		if (ib_send_cm_drep(cm_id, NULL, 0))
2541 2542
			shost_printk(KERN_ERR, target->scsi_host,
				     PFX "Sending CM DREP failed\n");
2543
		queue_work(system_long_wq, &target->tl_err_work);
2544 2545 2546
		break;

	case IB_CM_TIMEWAIT_EXIT:
2547 2548
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "connection closed\n");
2549
		comp = 1;
2550

2551
		ch->status = 0;
2552 2553
		break;

2554 2555 2556 2557 2558
	case IB_CM_MRA_RECEIVED:
	case IB_CM_DREQ_ERROR:
	case IB_CM_DREP_RECEIVED:
		break;

2559
	default:
2560 2561
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled CM event %d\n", event->event);
2562 2563 2564 2565
		break;
	}

	if (comp)
2566
		complete(&ch->done);
2567 2568 2569 2570

	return 0;
}

2571 2572 2573 2574 2575 2576 2577 2578
/**
 * srp_change_queue_depth - setting device queue depth
 * @sdev: scsi device struct
 * @qdepth: requested queue depth
 *
 * Returns queue depth.
 */
static int
2579
srp_change_queue_depth(struct scsi_device *sdev, int qdepth)
2580
{
2581
	if (!sdev->tagged_supported)
2582
		qdepth = 1;
2583
	return scsi_change_queue_depth(sdev, qdepth);
2584 2585
}

B
Bart Van Assche 已提交
2586
static int srp_send_tsk_mgmt(struct srp_rdma_ch *ch, u64 req_tag, u64 lun,
2587
			     u8 func, u8 *status)
2588
{
2589
	struct srp_target_port *target = ch->target;
2590
	struct srp_rport *rport = target->rport;
2591
	struct ib_device *dev = target->srp_host->srp_dev->dev;
2592 2593
	struct srp_iu *iu;
	struct srp_tsk_mgmt *tsk_mgmt;
2594
	int res;
2595

2596
	if (!ch->connected || target->qp_in_error)
2597 2598
		return -1;

2599
	/*
2600
	 * Lock the rport mutex to avoid that srp_create_ch_ib() is
2601 2602 2603
	 * invoked while a task management function is being sent.
	 */
	mutex_lock(&rport->mutex);
2604 2605 2606
	spin_lock_irq(&ch->lock);
	iu = __srp_get_tx_iu(ch, SRP_IU_TSK_MGMT);
	spin_unlock_irq(&ch->lock);
2607

2608 2609 2610
	if (!iu) {
		mutex_unlock(&rport->mutex);

2611
		return -1;
2612
	}
2613

2614 2615
	ib_dma_sync_single_for_cpu(dev, iu->dma, sizeof *tsk_mgmt,
				   DMA_TO_DEVICE);
2616 2617 2618 2619
	tsk_mgmt = iu->buf;
	memset(tsk_mgmt, 0, sizeof *tsk_mgmt);

	tsk_mgmt->opcode 	= SRP_TSK_MGMT;
B
Bart Van Assche 已提交
2620
	int_to_scsilun(lun, &tsk_mgmt->lun);
2621
	tsk_mgmt->tsk_mgmt_func = func;
2622
	tsk_mgmt->task_tag	= req_tag;
2623

2624 2625 2626 2627 2628 2629 2630
	spin_lock_irq(&ch->lock);
	ch->tsk_mgmt_tag = (ch->tsk_mgmt_tag + 1) | SRP_TAG_TSK_MGMT;
	tsk_mgmt->tag = ch->tsk_mgmt_tag;
	spin_unlock_irq(&ch->lock);

	init_completion(&ch->tsk_mgmt_done);

2631 2632
	ib_dma_sync_single_for_device(dev, iu->dma, sizeof *tsk_mgmt,
				      DMA_TO_DEVICE);
2633 2634
	if (srp_post_send(ch, iu, sizeof(*tsk_mgmt))) {
		srp_put_tx_iu(ch, iu, SRP_IU_TSK_MGMT);
2635 2636
		mutex_unlock(&rport->mutex);

2637 2638
		return -1;
	}
2639 2640 2641 2642
	res = wait_for_completion_timeout(&ch->tsk_mgmt_done,
					msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS));
	if (res > 0 && status)
		*status = ch->tsk_mgmt_status;
2643
	mutex_unlock(&rport->mutex);
2644

2645
	WARN_ON_ONCE(res < 0);
2646

2647
	return res > 0 ? 0 : -1;
2648 2649
}

2650 2651
static int srp_abort(struct scsi_cmnd *scmnd)
{
2652
	struct srp_target_port *target = host_to_target(scmnd->device->host);
2653
	struct srp_request *req = (struct srp_request *) scmnd->host_scribble;
B
Bart Van Assche 已提交
2654
	u32 tag;
B
Bart Van Assche 已提交
2655
	u16 ch_idx;
2656
	struct srp_rdma_ch *ch;
2657
	int ret;
2658

2659
	shost_printk(KERN_ERR, target->scsi_host, "SRP abort called\n");
2660

B
Bart Van Assche 已提交
2661
	if (!req)
2662
		return SUCCESS;
B
Bart Van Assche 已提交
2663
	tag = blk_mq_unique_tag(scmnd->request);
B
Bart Van Assche 已提交
2664 2665 2666 2667 2668 2669 2670 2671
	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
2672
	if (srp_send_tsk_mgmt(ch, tag, scmnd->device->lun,
2673
			      SRP_TSK_ABORT_TASK, NULL) == 0)
2674
		ret = SUCCESS;
2675
	else if (target->rport->state == SRP_RPORT_LOST)
2676
		ret = FAST_IO_FAIL;
2677 2678
	else
		ret = FAILED;
2679
	srp_free_req(ch, req, scmnd, 0);
B
Bart Van Assche 已提交
2680
	scmnd->result = DID_ABORT << 16;
2681
	scmnd->scsi_done(scmnd);
2682

2683
	return ret;
2684 2685 2686 2687
}

static int srp_reset_device(struct scsi_cmnd *scmnd)
{
2688
	struct srp_target_port *target = host_to_target(scmnd->device->host);
B
Bart Van Assche 已提交
2689
	struct srp_rdma_ch *ch;
2690
	int i;
2691
	u8 status;
2692

2693
	shost_printk(KERN_ERR, target->scsi_host, "SRP reset_device called\n");
2694

B
Bart Van Assche 已提交
2695
	ch = &target->ch[0];
2696
	if (srp_send_tsk_mgmt(ch, SRP_TAG_NO_REQ, scmnd->device->lun,
2697
			      SRP_TSK_LUN_RESET, &status))
2698
		return FAILED;
2699
	if (status)
2700 2701
		return FAILED;

B
Bart Van Assche 已提交
2702 2703 2704 2705
	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];
2706

B
Bart Van Assche 已提交
2707 2708
			srp_finish_req(ch, req, scmnd->device, DID_RESET << 16);
		}
2709
	}
2710 2711

	return SUCCESS;
2712 2713 2714 2715 2716 2717
}

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

2718
	shost_printk(KERN_ERR, target->scsi_host, PFX "SRP reset_host called\n");
2719

2720
	return srp_reconnect_rport(target->rport) == 0 ? SUCCESS : FAILED;
2721 2722
}

2723 2724 2725 2726 2727 2728
static int srp_slave_alloc(struct scsi_device *sdev)
{
	struct Scsi_Host *shost = sdev->host;
	struct srp_target_port *target = host_to_target(shost);
	struct srp_device *srp_dev = target->srp_host->srp_dev;

2729
	if (true)
2730 2731 2732 2733 2734 2735
		blk_queue_virt_boundary(sdev->request_queue,
					~srp_dev->mr_page_mask);

	return 0;
}

2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750
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;
}

2751 2752
static ssize_t show_id_ext(struct device *dev, struct device_attribute *attr,
			   char *buf)
2753
{
2754
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2755

2756
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->id_ext));
2757 2758
}

2759 2760
static ssize_t show_ioc_guid(struct device *dev, struct device_attribute *attr,
			     char *buf)
2761
{
2762
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2763

2764
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->ioc_guid));
2765 2766
}

2767 2768
static ssize_t show_service_id(struct device *dev,
			       struct device_attribute *attr, char *buf)
2769
{
2770
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2771

2772
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->service_id));
2773 2774
}

2775 2776
static ssize_t show_pkey(struct device *dev, struct device_attribute *attr,
			 char *buf)
2777
{
2778
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2779

2780
	return sprintf(buf, "0x%04x\n", be16_to_cpu(target->pkey));
2781 2782
}

B
Bart Van Assche 已提交
2783 2784 2785 2786 2787
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));

2788
	return sprintf(buf, "%pI6\n", target->sgid.raw);
B
Bart Van Assche 已提交
2789 2790
}

2791 2792
static ssize_t show_dgid(struct device *dev, struct device_attribute *attr,
			 char *buf)
2793
{
2794
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
B
Bart Van Assche 已提交
2795
	struct srp_rdma_ch *ch = &target->ch[0];
2796

2797
	return sprintf(buf, "%pI6\n", ch->path.dgid.raw);
2798 2799
}

2800 2801
static ssize_t show_orig_dgid(struct device *dev,
			      struct device_attribute *attr, char *buf)
2802
{
2803
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2804

2805
	return sprintf(buf, "%pI6\n", target->orig_dgid.raw);
2806 2807
}

2808 2809 2810 2811
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
2812 2813
	struct srp_rdma_ch *ch;
	int i, req_lim = INT_MAX;
2814

B
Bart Van Assche 已提交
2815 2816 2817 2818 2819
	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);
2820 2821
}

2822 2823
static ssize_t show_zero_req_lim(struct device *dev,
				 struct device_attribute *attr, char *buf)
2824
{
2825
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2826 2827 2828 2829

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

2830 2831
static ssize_t show_local_ib_port(struct device *dev,
				  struct device_attribute *attr, char *buf)
2832
{
2833
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2834 2835 2836 2837

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

2838 2839
static ssize_t show_local_ib_device(struct device *dev,
				    struct device_attribute *attr, char *buf)
2840
{
2841
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2842

2843
	return sprintf(buf, "%s\n", target->srp_host->srp_dev->dev->name);
2844 2845
}

B
Bart Van Assche 已提交
2846 2847 2848 2849 2850 2851 2852 2853
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);
}

2854 2855 2856 2857 2858 2859 2860 2861
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);
}

2862 2863 2864 2865 2866 2867 2868 2869
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);
}

2870 2871 2872 2873 2874 2875 2876 2877
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);
}

2878 2879 2880 2881 2882 2883 2884 2885
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");
}

2886 2887 2888 2889
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
2890
static DEVICE_ATTR(sgid,	    S_IRUGO, show_sgid,		   NULL);
2891 2892
static DEVICE_ATTR(dgid,	    S_IRUGO, show_dgid,		   NULL);
static DEVICE_ATTR(orig_dgid,	    S_IRUGO, show_orig_dgid,	   NULL);
2893
static DEVICE_ATTR(req_lim,         S_IRUGO, show_req_lim,         NULL);
2894 2895 2896
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
2897
static DEVICE_ATTR(ch_count,        S_IRUGO, show_ch_count,        NULL);
2898
static DEVICE_ATTR(comp_vector,     S_IRUGO, show_comp_vector,     NULL);
2899
static DEVICE_ATTR(tl_retry_count,  S_IRUGO, show_tl_retry_count,  NULL);
2900
static DEVICE_ATTR(cmd_sg_entries,  S_IRUGO, show_cmd_sg_entries,  NULL);
2901
static DEVICE_ATTR(allow_ext_sg,    S_IRUGO, show_allow_ext_sg,    NULL);
2902 2903 2904 2905 2906 2907

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
2908
	&dev_attr_sgid,
2909 2910
	&dev_attr_dgid,
	&dev_attr_orig_dgid,
2911
	&dev_attr_req_lim,
2912 2913 2914
	&dev_attr_zero_req_lim,
	&dev_attr_local_ib_port,
	&dev_attr_local_ib_device,
B
Bart Van Assche 已提交
2915
	&dev_attr_ch_count,
2916
	&dev_attr_comp_vector,
2917
	&dev_attr_tl_retry_count,
2918
	&dev_attr_cmd_sg_entries,
2919
	&dev_attr_allow_ext_sg,
2920 2921 2922
	NULL
};

2923 2924
static struct scsi_host_template srp_template = {
	.module				= THIS_MODULE,
R
Roland Dreier 已提交
2925 2926
	.name				= "InfiniBand SRP initiator",
	.proc_name			= DRV_NAME,
2927
	.slave_alloc			= srp_slave_alloc,
2928
	.slave_configure		= srp_slave_configure,
2929 2930
	.info				= srp_target_info,
	.queuecommand			= srp_queuecommand,
2931
	.change_queue_depth             = srp_change_queue_depth,
2932
	.eh_timed_out			= srp_timed_out,
2933 2934 2935
	.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
2936
	.skip_settle_delay		= true,
2937
	.sg_tablesize			= SRP_DEF_SG_TABLESIZE,
2938
	.can_queue			= SRP_DEFAULT_CMD_SQ_SIZE,
2939
	.this_id			= -1,
2940
	.cmd_per_lun			= SRP_DEFAULT_CMD_SQ_SIZE,
2941
	.use_clustering			= ENABLE_CLUSTERING,
B
Bart Van Assche 已提交
2942
	.shost_attrs			= srp_host_attrs,
2943
	.track_queue_depth		= 1,
2944 2945
};

2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956
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;
}

2957 2958 2959 2960 2961 2962 2963
/*
 * 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.
 */
2964 2965
static int srp_add_target(struct srp_host *host, struct srp_target_port *target)
{
2966 2967 2968
	struct srp_rport_identifiers ids;
	struct srp_rport *rport;

2969
	target->state = SRP_TARGET_SCANNING;
2970
	sprintf(target->target_name, "SRP.T10:%016llX",
2971
		be64_to_cpu(target->id_ext));
2972

2973
	if (scsi_add_host(target->scsi_host, host->srp_dev->dev->dev.parent))
2974 2975
		return -ENODEV;

2976 2977
	memcpy(ids.port_id, &target->id_ext, 8);
	memcpy(ids.port_id + 8, &target->ioc_guid, 8);
2978
	ids.roles = SRP_RPORT_ROLE_TARGET;
2979 2980 2981 2982 2983 2984
	rport = srp_rport_add(target->scsi_host, &ids);
	if (IS_ERR(rport)) {
		scsi_remove_host(target->scsi_host);
		return PTR_ERR(rport);
	}

2985
	rport->lld_data = target;
2986
	target->rport = rport;
2987

2988
	spin_lock(&host->target_lock);
2989
	list_add_tail(&target->list, &host->target_list);
2990
	spin_unlock(&host->target_lock);
2991 2992

	scsi_scan_target(&target->scsi_host->shost_gendev,
2993
			 0, target->scsi_id, SCAN_WILD_CARD, SCSI_SCAN_INITIAL);
2994

2995 2996
	if (srp_connected_ch(target) < target->ch_count ||
	    target->qp_in_error) {
2997 2998 2999 3000 3001 3002
		shost_printk(KERN_INFO, target->scsi_host,
			     PFX "SCSI scan failed - removing SCSI host\n");
		srp_queue_remove_work(target);
		goto out;
	}

3003
	pr_debug("%s: SCSI scan succeeded - detected %d LUNs\n",
3004 3005 3006 3007 3008 3009 3010 3011 3012
		 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:
3013 3014 3015
	return 0;
}

3016
static void srp_release_dev(struct device *dev)
3017 3018
{
	struct srp_host *host =
3019
		container_of(dev, struct srp_host, dev);
3020 3021 3022 3023 3024 3025

	complete(&host->released);
}

static struct class srp_class = {
	.name    = "infiniband_srp",
3026
	.dev_release = srp_release_dev
3027 3028
};

3029 3030
/**
 * srp_conn_unique() - check whether the connection to a target is unique
3031 3032
 * @host:   SRP host.
 * @target: SRP target port.
3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060
 */
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;
}

3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076
/*
 * 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,
3077
	SRP_OPT_MAX_CMD_PER_LUN	= 1 << 6,
3078
	SRP_OPT_IO_CLASS	= 1 << 7,
3079
	SRP_OPT_INITIATOR_EXT	= 1 << 8,
3080
	SRP_OPT_CMD_SG_ENTRIES	= 1 << 9,
3081 3082
	SRP_OPT_ALLOW_EXT_SG	= 1 << 10,
	SRP_OPT_SG_TABLESIZE	= 1 << 11,
3083
	SRP_OPT_COMP_VECTOR	= 1 << 12,
3084
	SRP_OPT_TL_RETRY_COUNT	= 1 << 13,
3085
	SRP_OPT_QUEUE_SIZE	= 1 << 14,
3086 3087 3088 3089 3090 3091 3092
	SRP_OPT_ALL		= (SRP_OPT_ID_EXT	|
				   SRP_OPT_IOC_GUID	|
				   SRP_OPT_DGID		|
				   SRP_OPT_PKEY		|
				   SRP_OPT_SERVICE_ID),
};

3093
static const match_table_t srp_opt_tokens = {
3094 3095 3096 3097 3098 3099 3100
	{ 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" 	},
3101
	{ SRP_OPT_IO_CLASS,		"io_class=%x"		},
3102
	{ SRP_OPT_INITIATOR_EXT,	"initiator_ext=%s"	},
3103
	{ SRP_OPT_CMD_SG_ENTRIES,	"cmd_sg_entries=%u"	},
3104 3105
	{ SRP_OPT_ALLOW_EXT_SG,		"allow_ext_sg=%u"	},
	{ SRP_OPT_SG_TABLESIZE,		"sg_tablesize=%u"	},
3106
	{ SRP_OPT_COMP_VECTOR,		"comp_vector=%u"	},
3107
	{ SRP_OPT_TL_RETRY_COUNT,	"tl_retry_count=%u"	},
3108
	{ SRP_OPT_QUEUE_SIZE,		"queue_size=%d"		},
3109
	{ SRP_OPT_ERR,			NULL 			}
3110 3111 3112 3113 3114 3115 3116
};

static int srp_parse_options(const char *buf, struct srp_target_port *target)
{
	char *options, *sep_opt;
	char *p;
	substring_t args[MAX_OPT_ARGS];
3117
	unsigned long long ull;
3118 3119 3120 3121 3122 3123 3124 3125 3126 3127
	int opt_mask = 0;
	int token;
	int ret = -EINVAL;
	int i;

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

	sep_opt = options;
3128
	while ((p = strsep(&sep_opt, ",\n")) != NULL) {
3129 3130 3131 3132 3133 3134 3135 3136 3137
		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);
3138 3139 3140 3141
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3142 3143 3144 3145 3146 3147 3148
			ret = kstrtoull(p, 16, &ull);
			if (ret) {
				pr_warn("invalid id_ext parameter '%s'\n", p);
				kfree(p);
				goto out;
			}
			target->id_ext = cpu_to_be64(ull);
3149 3150 3151 3152 3153
			kfree(p);
			break;

		case SRP_OPT_IOC_GUID:
			p = match_strdup(args);
3154 3155 3156 3157
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3158 3159 3160 3161 3162 3163 3164
			ret = kstrtoull(p, 16, &ull);
			if (ret) {
				pr_warn("invalid ioc_guid parameter '%s'\n", p);
				kfree(p);
				goto out;
			}
			target->ioc_guid = cpu_to_be64(ull);
3165 3166 3167 3168 3169
			kfree(p);
			break;

		case SRP_OPT_DGID:
			p = match_strdup(args);
3170 3171 3172 3173
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3174
			if (strlen(p) != 32) {
3175
				pr_warn("bad dest GID parameter '%s'\n", p);
3176
				kfree(p);
3177 3178 3179
				goto out;
			}

3180
			ret = hex2bin(target->orig_dgid.raw, p, 16);
3181
			kfree(p);
3182 3183
			if (ret < 0)
				goto out;
3184 3185 3186 3187
			break;

		case SRP_OPT_PKEY:
			if (match_hex(args, &token)) {
3188
				pr_warn("bad P_Key parameter '%s'\n", p);
3189 3190
				goto out;
			}
3191
			target->pkey = cpu_to_be16(token);
3192 3193 3194 3195
			break;

		case SRP_OPT_SERVICE_ID:
			p = match_strdup(args);
3196 3197 3198 3199
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3200 3201 3202 3203 3204 3205 3206
			ret = kstrtoull(p, 16, &ull);
			if (ret) {
				pr_warn("bad service_id parameter '%s'\n", p);
				kfree(p);
				goto out;
			}
			target->service_id = cpu_to_be64(ull);
3207 3208 3209 3210 3211
			kfree(p);
			break;

		case SRP_OPT_MAX_SECT:
			if (match_int(args, &token)) {
3212
				pr_warn("bad max sect parameter '%s'\n", p);
3213 3214 3215 3216 3217
				goto out;
			}
			target->scsi_host->max_sectors = token;
			break;

3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229
		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;

3230
		case SRP_OPT_MAX_CMD_PER_LUN:
3231
			if (match_int(args, &token) || token < 1) {
3232 3233
				pr_warn("bad max cmd_per_lun parameter '%s'\n",
					p);
3234 3235
				goto out;
			}
3236
			target->scsi_host->cmd_per_lun = token;
3237 3238
			break;

3239 3240
		case SRP_OPT_IO_CLASS:
			if (match_hex(args, &token)) {
3241
				pr_warn("bad IO class parameter '%s'\n", p);
3242 3243 3244 3245
				goto out;
			}
			if (token != SRP_REV10_IB_IO_CLASS &&
			    token != SRP_REV16A_IB_IO_CLASS) {
3246 3247 3248
				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);
3249 3250 3251 3252 3253
				goto out;
			}
			target->io_class = token;
			break;

3254 3255
		case SRP_OPT_INITIATOR_EXT:
			p = match_strdup(args);
3256 3257 3258 3259
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3260 3261 3262 3263 3264 3265 3266
			ret = kstrtoull(p, 16, &ull);
			if (ret) {
				pr_warn("bad initiator_ext value '%s'\n", p);
				kfree(p);
				goto out;
			}
			target->initiator_ext = cpu_to_be64(ull);
3267 3268 3269
			kfree(p);
			break;

3270 3271
		case SRP_OPT_CMD_SG_ENTRIES:
			if (match_int(args, &token) || token < 1 || token > 255) {
3272 3273
				pr_warn("bad max cmd_sg_entries parameter '%s'\n",
					p);
3274 3275 3276 3277 3278
				goto out;
			}
			target->cmd_sg_cnt = token;
			break;

3279 3280
		case SRP_OPT_ALLOW_EXT_SG:
			if (match_int(args, &token)) {
3281
				pr_warn("bad allow_ext_sg parameter '%s'\n", p);
3282 3283 3284 3285 3286 3287 3288
				goto out;
			}
			target->allow_ext_sg = !!token;
			break;

		case SRP_OPT_SG_TABLESIZE:
			if (match_int(args, &token) || token < 1 ||
3289
					token > SG_MAX_SEGMENTS) {
3290 3291
				pr_warn("bad max sg_tablesize parameter '%s'\n",
					p);
3292 3293 3294 3295 3296
				goto out;
			}
			target->sg_tablesize = token;
			break;

3297 3298 3299 3300 3301 3302 3303 3304
		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;

3305 3306 3307 3308 3309 3310 3311 3312 3313
		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;

3314
		default:
3315 3316
			pr_warn("unknown parameter or missing value '%s' in target creation request\n",
				p);
3317 3318 3319 3320 3321 3322 3323 3324 3325 3326
			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))
3327 3328
				pr_warn("target creation request is missing parameter '%s'\n",
					srp_opt_tokens[i].pattern);
3329

3330 3331 3332 3333 3334 3335
	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);

3336 3337 3338 3339 3340
out:
	kfree(options);
	return ret;
}

3341 3342
static ssize_t srp_create_target(struct device *dev,
				 struct device_attribute *attr,
3343 3344 3345
				 const char *buf, size_t count)
{
	struct srp_host *host =
3346
		container_of(dev, struct srp_host, dev);
3347 3348
	struct Scsi_Host *target_host;
	struct srp_target_port *target;
3349
	struct srp_rdma_ch *ch;
3350 3351
	struct srp_device *srp_dev = host->srp_dev;
	struct ib_device *ibdev = srp_dev->dev;
B
Bart Van Assche 已提交
3352
	int ret, node_idx, node, cpu, i;
3353
	unsigned int max_sectors_per_mr, mr_per_cmd = 0;
B
Bart Van Assche 已提交
3354
	bool multich = false;
3355 3356 3357 3358 3359 3360

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

3361
	target_host->transportt  = ib_srp_transport_template;
3362 3363
	target_host->max_channel = 0;
	target_host->max_id      = 1;
B
Bart Van Assche 已提交
3364
	target_host->max_lun     = -1LL;
A
Arne Redlich 已提交
3365
	target_host->max_cmd_len = sizeof ((struct srp_cmd *) (void *) 0L)->cdb;
R
Roland Dreier 已提交
3366

3367 3368
	target = host_to_target(target_host);

3369 3370 3371
	target->io_class	= SRP_REV16A_IB_IO_CLASS;
	target->scsi_host	= target_host;
	target->srp_host	= host;
J
Jason Gunthorpe 已提交
3372
	target->lkey		= host->srp_dev->pd->local_dma_lkey;
B
Bart Van Assche 已提交
3373
	target->global_rkey	= host->srp_dev->global_rkey;
3374
	target->cmd_sg_cnt	= cmd_sg_entries;
3375 3376
	target->sg_tablesize	= indirect_sg_entries ? : cmd_sg_entries;
	target->allow_ext_sg	= allow_ext_sg;
3377
	target->tl_retry_count	= 7;
3378
	target->queue_size	= SRP_DEFAULT_QUEUE_SIZE;
3379

3380 3381 3382 3383 3384 3385
	/*
	 * Avoid that the SCSI host can be removed by srp_remove_target()
	 * before this function returns.
	 */
	scsi_host_get(target->scsi_host);

3386 3387 3388
	ret = mutex_lock_interruptible(&host->add_target_mutex);
	if (ret < 0)
		goto put;
3389

3390 3391
	ret = srp_parse_options(buf, target);
	if (ret)
3392
		goto out;
3393

3394 3395
	target->req_ring_size = target->queue_size - SRP_TSK_MGMT_SQ_SIZE;

3396 3397 3398 3399 3400 3401 3402
	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;
3403
		goto out;
3404 3405
	}

3406
	if (!srp_dev->has_fmr && !srp_dev->has_fr && !target->allow_ext_sg &&
3407
	    target->cmd_sg_cnt < target->sg_tablesize) {
3408
		pr_warn("No MR pool and no external indirect descriptors, limiting sg_tablesize to cmd_sg_cnt\n");
3409 3410 3411
		target->sg_tablesize = target->cmd_sg_cnt;
	}

3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435
	if (srp_dev->use_fast_reg || srp_dev->use_fmr) {
		/*
		 * FR and FMR can only map one HCA page per entry. If the
		 * start address is not aligned on a HCA page boundary two
		 * entries will be used for the head and the tail although
		 * these two entries combined contain at most one HCA page of
		 * data. Hence the "+ 1" in the calculation below.
		 *
		 * The indirect data buffer descriptor is contiguous so the
		 * memory for that buffer will only be registered if
		 * register_always is true. Hence add one to mr_per_cmd if
		 * register_always has been set.
		 */
		max_sectors_per_mr = srp_dev->max_pages_per_mr <<
				  (ilog2(srp_dev->mr_page_size) - 9);
		mr_per_cmd = register_always +
			(target->scsi_host->max_sectors + 1 +
			 max_sectors_per_mr - 1) / max_sectors_per_mr;
		pr_debug("max_sectors = %u; max_pages_per_mr = %u; mr_page_size = %u; max_sectors_per_mr = %u; mr_per_cmd = %u\n",
			 target->scsi_host->max_sectors,
			 srp_dev->max_pages_per_mr, srp_dev->mr_page_size,
			 max_sectors_per_mr, mr_per_cmd);
	}

3436
	target_host->sg_tablesize = target->sg_tablesize;
3437 3438
	target->mr_pool_size = target->scsi_host->can_queue * mr_per_cmd;
	target->mr_per_cmd = mr_per_cmd;
3439 3440
	target->indirect_size = target->sg_tablesize *
				sizeof (struct srp_direct_buf);
3441 3442 3443 3444
	target->max_iu_len = sizeof (struct srp_cmd) +
			     sizeof (struct srp_indirect_buf) +
			     target->cmd_sg_cnt * sizeof (struct srp_direct_buf);

3445
	INIT_WORK(&target->tl_err_work, srp_tl_err_work);
3446
	INIT_WORK(&target->remove_work, srp_remove_work);
3447
	spin_lock_init(&target->lock);
3448
	ret = ib_query_gid(ibdev, host->port, 0, &target->sgid, NULL);
3449
	if (ret)
3450
		goto out;
3451

B
Bart Van Assche 已提交
3452 3453 3454 3455 3456 3457 3458 3459 3460
	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)
3461
		goto out;
3462

B
Bart Van Assche 已提交
3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490
	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;
3491

B
Bart Van Assche 已提交
3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502
			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,
3503
					     PFX "Connection %d/%d to %pI6 failed\n",
B
Bart Van Assche 已提交
3504
					     ch_start + cpu_idx,
3505 3506
					     target->ch_count,
					     ch->target->orig_dgid.raw);
B
Bart Van Assche 已提交
3507
				if (node_idx == 0 && cpu_idx == 0) {
B
Bart Van Assche 已提交
3508
					goto free_ch;
B
Bart Van Assche 已提交
3509 3510 3511 3512
				} else {
					srp_free_ch_ib(target, ch);
					srp_free_req_data(target, ch);
					target->ch_count = ch - target->ch;
3513
					goto connected;
B
Bart Van Assche 已提交
3514 3515 3516 3517 3518 3519 3520
				}
			}

			multich = true;
			cpu_idx++;
		}
		node_idx++;
3521 3522
	}

3523
connected:
B
Bart Van Assche 已提交
3524 3525
	target->scsi_host->nr_hw_queues = target->ch_count;

3526 3527 3528 3529
	ret = srp_add_target(host, target);
	if (ret)
		goto err_disconnect;

3530 3531 3532 3533 3534
	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),
3535
			     be16_to_cpu(target->pkey),
3536
			     be64_to_cpu(target->service_id),
3537
			     target->sgid.raw, target->orig_dgid.raw);
3538
	}
B
Bart Van Assche 已提交
3539

3540 3541 3542 3543
	ret = count;

out:
	mutex_unlock(&host->add_target_mutex);
3544

3545
put:
3546
	scsi_host_put(target->scsi_host);
3547 3548
	if (ret < 0)
		scsi_host_put(target->scsi_host);
3549

3550
	return ret;
3551 3552 3553 3554

err_disconnect:
	srp_disconnect_target(target);

B
Bart Van Assche 已提交
3555
free_ch:
B
Bart Van Assche 已提交
3556 3557 3558 3559 3560
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_ch_ib(target, ch);
		srp_free_req_data(target, ch);
	}
3561

B
Bart Van Assche 已提交
3562
	kfree(target->ch);
3563
	goto out;
3564 3565
}

3566
static DEVICE_ATTR(add_target, S_IWUSR, NULL, srp_create_target);
3567

3568 3569
static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
			  char *buf)
3570
{
3571
	struct srp_host *host = container_of(dev, struct srp_host, dev);
3572

3573
	return sprintf(buf, "%s\n", host->srp_dev->dev->name);
3574 3575
}

3576
static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
3577

3578 3579
static ssize_t show_port(struct device *dev, struct device_attribute *attr,
			 char *buf)
3580
{
3581
	struct srp_host *host = container_of(dev, struct srp_host, dev);
3582 3583 3584 3585

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

3586
static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
3587

3588
static struct srp_host *srp_add_port(struct srp_device *device, u8 port)
3589 3590 3591 3592 3593 3594 3595 3596
{
	struct srp_host *host;

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

	INIT_LIST_HEAD(&host->target_list);
3597
	spin_lock_init(&host->target_lock);
3598
	init_completion(&host->released);
3599
	mutex_init(&host->add_target_mutex);
3600
	host->srp_dev = device;
3601 3602
	host->port = port;

3603
	host->dev.class = &srp_class;
3604
	host->dev.parent = device->dev->dev.parent;
3605
	dev_set_name(&host->dev, "srp-%s-%d", device->dev->name, port);
3606

3607
	if (device_register(&host->dev))
3608
		goto free_host;
3609
	if (device_create_file(&host->dev, &dev_attr_add_target))
3610
		goto err_class;
3611
	if (device_create_file(&host->dev, &dev_attr_ibdev))
3612
		goto err_class;
3613
	if (device_create_file(&host->dev, &dev_attr_port))
3614 3615 3616 3617 3618
		goto err_class;

	return host;

err_class:
3619
	device_unregister(&host->dev);
3620

3621
free_host:
3622 3623 3624 3625 3626 3627 3628
	kfree(host);

	return NULL;
}

static void srp_add_one(struct ib_device *device)
{
3629
	struct srp_device *srp_dev;
3630
	struct ib_device_attr *attr = &device->attrs;
3631
	struct srp_host *host;
3632
	int mr_page_shift, p;
3633
	u64 max_pages_per_mr;
3634
	unsigned int flags = 0;
3635

3636
	srp_dev = kzalloc(sizeof(*srp_dev), GFP_KERNEL);
3637
	if (!srp_dev)
3638
		return;
3639 3640 3641

	/*
	 * Use the smallest page size supported by the HCA, down to a
3642 3643
	 * minimum of 4096 bytes. We're unlikely to build large sglists
	 * out of smaller entries.
3644
	 */
3645
	mr_page_shift		= max(12, ffs(attr->page_size_cap) - 1);
3646 3647
	srp_dev->mr_page_size	= 1 << mr_page_shift;
	srp_dev->mr_page_mask	= ~((u64) srp_dev->mr_page_size - 1);
3648
	max_pages_per_mr	= attr->max_mr_size;
3649
	do_div(max_pages_per_mr, srp_dev->mr_page_size);
3650
	pr_debug("%s: %llu / %u = %llu <> %u\n", __func__,
3651
		 attr->max_mr_size, srp_dev->mr_page_size,
3652
		 max_pages_per_mr, SRP_MAX_PAGES_PER_MR);
3653 3654
	srp_dev->max_pages_per_mr = min_t(u64, SRP_MAX_PAGES_PER_MR,
					  max_pages_per_mr);
3655 3656 3657

	srp_dev->has_fmr = (device->alloc_fmr && device->dealloc_fmr &&
			    device->map_phys_fmr && device->unmap_fmr);
3658
	srp_dev->has_fr = (attr->device_cap_flags &
3659
			   IB_DEVICE_MEM_MGT_EXTENSIONS);
3660
	if (!never_register && !srp_dev->has_fmr && !srp_dev->has_fr) {
3661
		dev_warn(&device->dev, "neither FMR nor FR is supported\n");
3662
	} else if (!never_register &&
3663
		   attr->max_mr_size >= 2 * srp_dev->mr_page_size) {
3664 3665 3666 3667
		srp_dev->use_fast_reg = (srp_dev->has_fr &&
					 (!srp_dev->has_fmr || prefer_fr));
		srp_dev->use_fmr = !srp_dev->use_fast_reg && srp_dev->has_fmr;
	}
3668

3669 3670 3671 3672
	if (never_register || !register_always ||
	    (!srp_dev->has_fmr && !srp_dev->has_fr))
		flags |= IB_PD_UNSAFE_GLOBAL_RKEY;

3673 3674 3675
	if (srp_dev->use_fast_reg) {
		srp_dev->max_pages_per_mr =
			min_t(u32, srp_dev->max_pages_per_mr,
3676
			      attr->max_fast_reg_page_list_len);
3677
	}
3678 3679
	srp_dev->mr_max_size	= srp_dev->mr_page_size *
				   srp_dev->max_pages_per_mr;
3680
	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",
3681 3682
		 device->name, mr_page_shift, attr->max_mr_size,
		 attr->max_fast_reg_page_list_len,
3683
		 srp_dev->max_pages_per_mr, srp_dev->mr_max_size);
3684 3685 3686 3687

	INIT_LIST_HEAD(&srp_dev->dev_list);

	srp_dev->dev = device;
3688
	srp_dev->pd  = ib_alloc_pd(device, flags);
3689 3690 3691
	if (IS_ERR(srp_dev->pd))
		goto free_dev;

B
Bart Van Assche 已提交
3692 3693 3694 3695
	if (flags & IB_PD_UNSAFE_GLOBAL_RKEY) {
		srp_dev->global_rkey = srp_dev->pd->unsafe_global_rkey;
		WARN_ON_ONCE(srp_dev->global_rkey == 0);
	}
3696

3697
	for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) {
3698
		host = srp_add_port(srp_dev, p);
3699
		if (host)
3700
			list_add_tail(&host->list, &srp_dev->dev_list);
3701 3702
	}

3703
	ib_set_client_data(device, &srp_client, srp_dev);
3704
	return;
3705 3706 3707

free_dev:
	kfree(srp_dev);
3708 3709
}

3710
static void srp_remove_one(struct ib_device *device, void *client_data)
3711
{
3712
	struct srp_device *srp_dev;
3713
	struct srp_host *host, *tmp_host;
3714
	struct srp_target_port *target;
3715

3716
	srp_dev = client_data;
3717 3718
	if (!srp_dev)
		return;
3719

3720
	list_for_each_entry_safe(host, tmp_host, &srp_dev->dev_list, list) {
3721
		device_unregister(&host->dev);
3722 3723 3724 3725 3726 3727 3728
		/*
		 * Wait for the sysfs entry to go away, so that no new
		 * target ports can be created.
		 */
		wait_for_completion(&host->released);

		/*
3729
		 * Remove all target ports.
3730
		 */
3731
		spin_lock(&host->target_lock);
3732 3733
		list_for_each_entry(target, &host->target_list, list)
			srp_queue_remove_work(target);
3734
		spin_unlock(&host->target_lock);
3735 3736

		/*
3737
		 * Wait for tl_err and target port removal tasks.
3738
		 */
3739
		flush_workqueue(system_long_wq);
3740
		flush_workqueue(srp_remove_wq);
3741 3742 3743 3744

		kfree(host);
	}

3745 3746 3747
	ib_dealloc_pd(srp_dev->pd);

	kfree(srp_dev);
3748 3749
}

3750
static struct srp_function_template ib_srp_transport_functions = {
3751 3752
	.has_rport_state	 = true,
	.reset_timer_if_blocked	 = true,
3753
	.reconnect_delay	 = &srp_reconnect_delay,
3754 3755 3756
	.fast_io_fail_tmo	 = &srp_fast_io_fail_tmo,
	.dev_loss_tmo		 = &srp_dev_loss_tmo,
	.reconnect		 = srp_rport_reconnect,
3757
	.rport_delete		 = srp_rport_delete,
3758
	.terminate_rport_io	 = srp_terminate_io,
3759 3760
};

3761 3762 3763 3764
static int __init srp_init_module(void)
{
	int ret;

3765
	if (srp_sg_tablesize) {
3766
		pr_warn("srp_sg_tablesize is deprecated, please use cmd_sg_entries\n");
3767 3768 3769 3770 3771 3772 3773 3774
		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) {
3775
		pr_warn("Clamping cmd_sg_entries to 255\n");
3776
		cmd_sg_entries = 255;
3777 3778
	}

3779 3780 3781
	if (!indirect_sg_entries)
		indirect_sg_entries = cmd_sg_entries;
	else if (indirect_sg_entries < cmd_sg_entries) {
3782 3783
		pr_warn("Bumping up indirect_sg_entries to match cmd_sg_entries (%u)\n",
			cmd_sg_entries);
3784 3785 3786
		indirect_sg_entries = cmd_sg_entries;
	}

3787 3788 3789 3790 3791 3792
	if (indirect_sg_entries > SG_MAX_SEGMENTS) {
		pr_warn("Clamping indirect_sg_entries to %u\n",
			SG_MAX_SEGMENTS);
		indirect_sg_entries = SG_MAX_SEGMENTS;
	}

3793
	srp_remove_wq = create_workqueue("srp_remove");
3794 3795
	if (!srp_remove_wq) {
		ret = -ENOMEM;
3796 3797 3798 3799
		goto out;
	}

	ret = -ENOMEM;
3800 3801 3802
	ib_srp_transport_template =
		srp_attach_transport(&ib_srp_transport_functions);
	if (!ib_srp_transport_template)
3803
		goto destroy_wq;
3804

3805 3806
	ret = class_register(&srp_class);
	if (ret) {
3807
		pr_err("couldn't register class infiniband_srp\n");
3808
		goto release_tr;
3809 3810
	}

3811 3812
	ib_sa_register_client(&srp_sa_client);

3813 3814
	ret = ib_register_client(&srp_client);
	if (ret) {
3815
		pr_err("couldn't register IB client\n");
3816
		goto unreg_sa;
3817 3818
	}

3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831
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;
3832 3833 3834 3835 3836
}

static void __exit srp_cleanup_module(void)
{
	ib_unregister_client(&srp_client);
3837
	ib_sa_unregister_client(&srp_sa_client);
3838
	class_unregister(&srp_class);
3839
	srp_release_transport(ib_srp_transport_template);
3840
	destroy_workqueue(srp_remove_wq);
3841 3842 3843 3844
}

module_init(srp_init_module);
module_exit(srp_cleanup_module);