ib_srp.c 97.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
/*
 * Copyright (c) 2005 Cisco Systems.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

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

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

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

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

#include "ib_srp.h"

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

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

67 68 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);

311 312 313 314 315 316 317
	if (ch->cm_id)
		ib_destroy_cm_id(ch->cm_id);
	ch->cm_id = new_cm_id;
	ch->path.sgid = target->sgid;
	ch->path.dgid = target->orig_dgid;
	ch->path.pkey = target->pkey;
	ch->path.service_id = target->service_id;
D
David Dillow 已提交
318 319 320 321

	return 0;
}

322 323 324 325 326 327
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));
328
	fmr_param.pool_size	    = target->mr_pool_size;
329 330
	fmr_param.dirty_watermark   = fmr_param.pool_size / 4;
	fmr_param.cache		    = 1;
331 332
	fmr_param.max_pages_per_fmr = dev->max_pages_per_mr;
	fmr_param.page_shift	    = ilog2(dev->mr_page_size);
333 334 335 336 337 338 339
	fmr_param.access	    = (IB_ACCESS_LOCAL_WRITE |
				       IB_ACCESS_REMOTE_WRITE |
				       IB_ACCESS_REMOTE_READ);

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

340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387
/**
 * srp_destroy_fr_pool() - free the resources owned by a pool
 * @pool: Fast registration pool to be destroyed.
 */
static void srp_destroy_fr_pool(struct srp_fr_pool *pool)
{
	int i;
	struct srp_fr_desc *d;

	if (!pool)
		return;

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

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

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

	for (i = 0, d = &pool->desc[0]; i < pool->size; i++, d++) {
S
Sagi Grimberg 已提交
388 389
		mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG,
				 max_page_list_len);
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452
		if (IS_ERR(mr)) {
			ret = PTR_ERR(mr);
			goto destroy_pool;
		}
		d->mr = mr;
		list_add_tail(&d->entry, &pool->free_list);
	}

out:
	return pool;

destroy_pool:
	srp_destroy_fr_pool(pool);

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

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

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

	return d;
}

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

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

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

453
	return srp_create_fr_pool(dev->dev, dev->pd, target->mr_pool_size,
454 455 456
				  dev->max_pages_per_mr);
}

457 458
/**
 * srp_destroy_qp() - destroy an RDMA queue pair
459
 * @qp: RDMA queue pair.
460
 *
S
Steve Wise 已提交
461 462
 * Drain the qp before destroying it.  This avoids that the receive
 * completion handler can access the queue pair while it is
463 464
 * being destroyed.
 */
465
static void srp_destroy_qp(struct ib_qp *qp)
466
{
467 468
	ib_drain_rq(qp);
	ib_destroy_qp(qp);
469 470
}

471
static int srp_create_ch_ib(struct srp_rdma_ch *ch)
472
{
473
	struct srp_target_port *target = ch->target;
474
	struct srp_device *dev = target->srp_host->srp_dev;
475
	struct ib_qp_init_attr *init_attr;
476 477
	struct ib_cq *recv_cq, *send_cq;
	struct ib_qp *qp;
478
	struct ib_fmr_pool *fmr_pool = NULL;
479
	struct srp_fr_pool *fr_pool = NULL;
480
	const int m = 1 + dev->use_fast_reg * target->mr_per_cmd * 2;
481 482 483 484 485 486
	int ret;

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

S
Steve Wise 已提交
487
	/* queue_size + 1 for ib_drain_rq() */
C
Christoph Hellwig 已提交
488 489
	recv_cq = ib_alloc_cq(dev->dev, ch, target->queue_size + 1,
				ch->comp_vector, IB_POLL_SOFTIRQ);
490 491
	if (IS_ERR(recv_cq)) {
		ret = PTR_ERR(recv_cq);
492
		goto err;
493 494
	}

C
Christoph Hellwig 已提交
495 496
	send_cq = ib_alloc_cq(dev->dev, ch, m * target->queue_size,
				ch->comp_vector, IB_POLL_DIRECT);
497 498
	if (IS_ERR(send_cq)) {
		ret = PTR_ERR(send_cq);
499
		goto err_recv_cq;
500 501
	}

502
	init_attr->event_handler       = srp_qp_event;
503
	init_attr->cap.max_send_wr     = m * target->queue_size;
504
	init_attr->cap.max_recv_wr     = target->queue_size + 1;
505 506
	init_attr->cap.max_recv_sge    = 1;
	init_attr->cap.max_send_sge    = 1;
507
	init_attr->sq_sig_type         = IB_SIGNAL_REQ_WR;
508
	init_attr->qp_type             = IB_QPT_RC;
509 510
	init_attr->send_cq             = send_cq;
	init_attr->recv_cq             = recv_cq;
511

512
	qp = ib_create_qp(dev->pd, init_attr);
513 514
	if (IS_ERR(qp)) {
		ret = PTR_ERR(qp);
515
		goto err_send_cq;
516 517
	}

518
	ret = srp_init_qp(target, qp);
519 520
	if (ret)
		goto err_qp;
521

522
	if (dev->use_fast_reg) {
523 524 525 526 527 528 529
		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;
		}
530
	} else if (dev->use_fmr) {
531 532 533 534 535 536 537 538 539
		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;
		}
	}

540
	if (ch->qp)
541
		srp_destroy_qp(ch->qp);
542
	if (ch->recv_cq)
C
Christoph Hellwig 已提交
543
		ib_free_cq(ch->recv_cq);
544
	if (ch->send_cq)
C
Christoph Hellwig 已提交
545
		ib_free_cq(ch->send_cq);
546

547 548 549
	ch->qp = qp;
	ch->recv_cq = recv_cq;
	ch->send_cq = send_cq;
550

551 552 553 554 555 556 557 558 559 560
	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;
	}

561 562 563 564
	kfree(init_attr);
	return 0;

err_qp:
565
	srp_destroy_qp(qp);
566 567

err_send_cq:
C
Christoph Hellwig 已提交
568
	ib_free_cq(send_cq);
569 570

err_recv_cq:
C
Christoph Hellwig 已提交
571
	ib_free_cq(recv_cq);
572 573

err:
574 575 576 577
	kfree(init_attr);
	return ret;
}

578 579
/*
 * Note: this function may be called without srp_alloc_iu_bufs() having been
580
 * invoked. Hence the ch->[rt]x_ring checks.
581
 */
582 583
static void srp_free_ch_ib(struct srp_target_port *target,
			   struct srp_rdma_ch *ch)
584
{
585
	struct srp_device *dev = target->srp_host->srp_dev;
586 587
	int i;

B
Bart Van Assche 已提交
588 589 590
	if (!ch->target)
		return;

591 592 593
	if (ch->cm_id) {
		ib_destroy_cm_id(ch->cm_id);
		ch->cm_id = NULL;
594 595
	}

B
Bart Van Assche 已提交
596 597 598 599
	/* If srp_new_cm_id() succeeded but srp_create_ch_ib() not, return. */
	if (!ch->qp)
		return;

600
	if (dev->use_fast_reg) {
601 602
		if (ch->fr_pool)
			srp_destroy_fr_pool(ch->fr_pool);
603
	} else if (dev->use_fmr) {
604 605
		if (ch->fmr_pool)
			ib_destroy_fmr_pool(ch->fmr_pool);
606
	}
C
Christoph Hellwig 已提交
607

608
	srp_destroy_qp(ch->qp);
C
Christoph Hellwig 已提交
609 610
	ib_free_cq(ch->send_cq);
	ib_free_cq(ch->recv_cq);
611

B
Bart Van Assche 已提交
612 613 614 615 616 617 618 619
	/*
	 * 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;

620 621
	ch->qp = NULL;
	ch->send_cq = ch->recv_cq = NULL;
622

623
	if (ch->rx_ring) {
624
		for (i = 0; i < target->queue_size; ++i)
625 626 627
			srp_free_iu(target->srp_host, ch->rx_ring[i]);
		kfree(ch->rx_ring);
		ch->rx_ring = NULL;
628
	}
629
	if (ch->tx_ring) {
630
		for (i = 0; i < target->queue_size; ++i)
631 632 633
			srp_free_iu(target->srp_host, ch->tx_ring[i]);
		kfree(ch->tx_ring);
		ch->tx_ring = NULL;
634
	}
635 636 637 638
}

static void srp_path_rec_completion(int status,
				    struct ib_sa_path_rec *pathrec,
639
				    void *ch_ptr)
640
{
641 642
	struct srp_rdma_ch *ch = ch_ptr;
	struct srp_target_port *target = ch->target;
643

644
	ch->status = status;
645
	if (status)
646 647
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "Got failed path rec status %d\n", status);
648
	else
649 650
		ch->path = *pathrec;
	complete(&ch->done);
651 652
}

653
static int srp_lookup_path(struct srp_rdma_ch *ch)
654
{
655
	struct srp_target_port *target = ch->target;
656 657
	int ret;

658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678
	ch->path.numb_path = 1;

	init_completion(&ch->done);

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

	ret = wait_for_completion_interruptible(&ch->done);
679 680
	if (ret < 0)
		return ret;
681

682
	if (ch->status < 0)
683 684
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Path record query failed\n");
685

686
	return ch->status;
687 688
}

B
Bart Van Assche 已提交
689
static int srp_send_req(struct srp_rdma_ch *ch, bool multich)
690
{
691
	struct srp_target_port *target = ch->target;
692 693 694 695 696 697 698 699 700 701
	struct {
		struct ib_cm_req_param param;
		struct srp_login_req   priv;
	} *req = NULL;
	int status;

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

702
	req->param.primary_path		      = &ch->path;
703 704
	req->param.alternate_path 	      = NULL;
	req->param.service_id 		      = target->service_id;
705 706
	req->param.qp_num		      = ch->qp->qp_num;
	req->param.qp_type		      = ch->qp->qp_type;
707 708 709 710 711 712 713 714 715 716 717 718 719 720
	req->param.private_data 	      = &req->priv;
	req->param.private_data_len 	      = sizeof req->priv;
	req->param.flow_control 	      = 1;

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

	/*
	 * Pick some arbitrary defaults here; we could make these
	 * module parameters if anyone cared about setting them.
	 */
	req->param.responder_resources	      = 4;
	req->param.remote_cm_response_timeout = 20;
	req->param.local_cm_response_timeout  = 20;
721
	req->param.retry_count                = target->tl_retry_count;
722 723 724 725 726
	req->param.rnr_retry_count 	      = 7;
	req->param.max_cm_retries 	      = 15;

	req->priv.opcode     	= SRP_LOGIN_REQ;
	req->priv.tag        	= 0;
727
	req->priv.req_it_iu_len = cpu_to_be32(target->max_iu_len);
728 729
	req->priv.req_buf_fmt 	= cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
					      SRP_BUF_FORMAT_INDIRECT);
B
Bart Van Assche 已提交
730 731
	req->priv.req_flags	= (multich ? SRP_MULTICHAN_MULTI :
				   SRP_MULTICHAN_SINGLE);
732
	/*
R
Roland Dreier 已提交
733
	 * In the published SRP specification (draft rev. 16a), the
734 735 736 737 738 739 740 741 742
	 * 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,
743
		       &target->sgid.global.interface_id, 8);
744
		memcpy(req->priv.initiator_port_id + 8,
745
		       &target->initiator_ext, 8);
746 747 748 749
		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,
750 751
		       &target->initiator_ext, 8);
		memcpy(req->priv.initiator_port_id + 8,
752
		       &target->sgid.global.interface_id, 8);
753 754 755 756
		memcpy(req->priv.target_port_id,     &target->id_ext, 8);
		memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8);
	}

757 758
	/*
	 * Topspin/Cisco SRP targets will reject our login unless we
759 760
	 * zero out the first 8 bytes of our initiator port ID and set
	 * the second 8 bytes to the local node GUID.
761
	 */
762
	if (srp_target_is_topspin(target)) {
763 764 765
		shost_printk(KERN_DEBUG, target->scsi_host,
			     PFX "Topspin/Cisco initiator port ID workaround "
			     "activated for target GUID %016llx\n",
766
			     be64_to_cpu(target->ioc_guid));
767
		memset(req->priv.initiator_port_id, 0, 8);
768
		memcpy(req->priv.initiator_port_id + 8,
769
		       &target->srp_host->srp_dev->dev->node_guid, 8);
770 771
	}

772
	status = ib_send_cm_req(ch->cm_id, &req->param);
773 774 775 776 777 778

	kfree(req);

	return status;
}

779 780 781 782 783 784 785 786 787 788 789 790
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)
791
		queue_work(srp_remove_wq, &target->remove_work);
792 793 794 795

	return changed;
}

796 797
static void srp_disconnect_target(struct srp_target_port *target)
{
B
Bart Van Assche 已提交
798 799
	struct srp_rdma_ch *ch;
	int i;
800

801
	/* XXX should send SRP_I_LOGOUT request */
802

803 804 805 806 807 808
	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");
809
		}
810
	}
811 812
}

813 814
static void srp_free_req_data(struct srp_target_port *target,
			      struct srp_rdma_ch *ch)
815
{
816 817
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = dev->dev;
818 819 820
	struct srp_request *req;
	int i;

821
	if (!ch->req_ring)
822 823 824
		return;

	for (i = 0; i < target->req_ring_size; ++i) {
825
		req = &ch->req_ring[i];
826
		if (dev->use_fast_reg) {
827
			kfree(req->fr_list);
828
		} else {
829
			kfree(req->fmr_list);
830 831
			kfree(req->map_page);
		}
832 833 834 835 836 837
		if (req->indirect_dma_addr) {
			ib_dma_unmap_single(ibdev, req->indirect_dma_addr,
					    target->indirect_size,
					    DMA_TO_DEVICE);
		}
		kfree(req->indirect_desc);
838
	}
839

840 841
	kfree(ch->req_ring);
	ch->req_ring = NULL;
842 843
}

844
static int srp_alloc_req_data(struct srp_rdma_ch *ch)
845
{
846
	struct srp_target_port *target = ch->target;
847 848 849
	struct srp_device *srp_dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = srp_dev->dev;
	struct srp_request *req;
850
	void *mr_list;
851 852 853
	dma_addr_t dma_addr;
	int i, ret = -ENOMEM;

854 855 856
	ch->req_ring = kcalloc(target->req_ring_size, sizeof(*ch->req_ring),
			       GFP_KERNEL);
	if (!ch->req_ring)
857 858 859
		goto out;

	for (i = 0; i < target->req_ring_size; ++i) {
860
		req = &ch->req_ring[i];
861
		mr_list = kmalloc(target->mr_per_cmd * sizeof(void *),
862 863 864
				  GFP_KERNEL);
		if (!mr_list)
			goto out;
865
		if (srp_dev->use_fast_reg) {
866
			req->fr_list = mr_list;
867
		} else {
868
			req->fmr_list = mr_list;
869 870 871 872 873
			req->map_page = kmalloc(srp_dev->max_pages_per_mr *
						sizeof(void *), GFP_KERNEL);
			if (!req->map_page)
				goto out;
		}
874
		req->indirect_desc = kmalloc(target->indirect_size, GFP_KERNEL);
875
		if (!req->indirect_desc)
876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
			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;
}

892 893 894 895 896 897 898 899 900 901 902 903 904 905 906
/**
 * 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);
}

907 908
static void srp_remove_target(struct srp_target_port *target)
{
B
Bart Van Assche 已提交
909 910
	struct srp_rdma_ch *ch;
	int i;
911

912 913
	WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);

914
	srp_del_scsi_host_attr(target->scsi_host);
915
	srp_rport_get(target->rport);
916 917
	srp_remove_host(target->scsi_host);
	scsi_remove_host(target->scsi_host);
918
	srp_stop_rport_timers(target->rport);
919
	srp_disconnect_target(target);
B
Bart Van Assche 已提交
920 921 922 923
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_ch_ib(target, ch);
	}
924
	cancel_work_sync(&target->tl_err_work);
925
	srp_rport_put(target->rport);
B
Bart Van Assche 已提交
926 927 928 929 930 931
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_req_data(target, ch);
	}
	kfree(target->ch);
	target->ch = NULL;
932 933 934 935 936

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

937 938 939
	scsi_host_put(target->scsi_host);
}

D
David Howells 已提交
940
static void srp_remove_work(struct work_struct *work)
941
{
D
David Howells 已提交
942
	struct srp_target_port *target =
943
		container_of(work, struct srp_target_port, remove_work);
944

945
	WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);
946

947
	srp_remove_target(target);
948 949
}

950 951 952 953 954 955 956
static void srp_rport_delete(struct srp_rport *rport)
{
	struct srp_target_port *target = rport->lld_data;

	srp_queue_remove_work(target);
}

957 958 959 960 961 962 963 964 965 966 967 968 969 970
/**
 * 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
971
static int srp_connect_ch(struct srp_rdma_ch *ch, bool multich)
972
{
973
	struct srp_target_port *target = ch->target;
974 975
	int ret;

976
	WARN_ON_ONCE(!multich && srp_connected_ch(target) > 0);
977

978
	ret = srp_lookup_path(ch);
979
	if (ret)
B
Bart Van Assche 已提交
980
		goto out;
981 982

	while (1) {
983
		init_completion(&ch->done);
B
Bart Van Assche 已提交
984
		ret = srp_send_req(ch, multich);
985
		if (ret)
B
Bart Van Assche 已提交
986
			goto out;
987
		ret = wait_for_completion_interruptible(&ch->done);
988
		if (ret < 0)
B
Bart Van Assche 已提交
989
			goto out;
990 991 992 993 994 995 996

		/*
		 * 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
997 998
		ret = ch->status;
		switch (ret) {
999
		case 0:
1000
			ch->connected = true;
B
Bart Van Assche 已提交
1001
			goto out;
1002 1003

		case SRP_PORT_REDIRECT:
1004
			ret = srp_lookup_path(ch);
1005
			if (ret)
B
Bart Van Assche 已提交
1006
				goto out;
1007 1008 1009 1010 1011
			break;

		case SRP_DLID_REDIRECT:
			break;

D
David Dillow 已提交
1012 1013
		case SRP_STALE_CONN:
			shost_printk(KERN_ERR, target->scsi_host, PFX
1014
				     "giving up on stale connection\n");
B
Bart Van Assche 已提交
1015 1016
			ret = -ECONNRESET;
			goto out;
D
David Dillow 已提交
1017

1018
		default:
B
Bart Van Assche 已提交
1019
			goto out;
1020 1021
		}
	}
B
Bart Van Assche 已提交
1022 1023 1024

out:
	return ret <= 0 ? ret : -ENODEV;
1025 1026
}

C
Christoph Hellwig 已提交
1027 1028 1029 1030 1031 1032 1033
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)
1034 1035 1036 1037 1038 1039 1040 1041 1042 1043
{
	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
1044 1045
	wr.wr_cqe = &req->reg_cqe;
	req->reg_cqe.done = srp_inv_rkey_err_done;
1046
	return ib_post_send(ch->qp, &wr, &bad_wr);
1047 1048
}

1049
static void srp_unmap_data(struct scsi_cmnd *scmnd,
1050
			   struct srp_rdma_ch *ch,
1051 1052
			   struct srp_request *req)
{
1053
	struct srp_target_port *target = ch->target;
1054 1055 1056
	struct srp_device *dev = target->srp_host->srp_dev;
	struct ib_device *ibdev = dev->dev;
	int i, res;
1057

1058
	if (!scsi_sglist(scmnd) ||
1059 1060 1061 1062
	    (scmnd->sc_data_direction != DMA_TO_DEVICE &&
	     scmnd->sc_data_direction != DMA_FROM_DEVICE))
		return;

1063 1064 1065 1066
	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
1067
			res = srp_inv_rkey(req, ch, (*pfr)->mr->rkey);
1068 1069 1070 1071 1072 1073 1074 1075 1076
			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)
1077
			srp_fr_pool_put(ch->fr_pool, req->fr_list,
1078
					req->nmdesc);
1079
	} else if (dev->use_fmr) {
1080 1081 1082 1083 1084
		struct ib_pool_fmr **pfmr;

		for (i = req->nmdesc, pfmr = req->fmr_list; i > 0; i--, pfmr++)
			ib_fmr_pool_unmap(*pfmr);
	}
1085

1086 1087
	ib_dma_unmap_sg(ibdev, scsi_sglist(scmnd), scsi_sg_count(scmnd),
			scmnd->sc_data_direction);
1088 1089
}

B
Bart Van Assche 已提交
1090 1091
/**
 * srp_claim_req - Take ownership of the scmnd associated with a request.
1092
 * @ch: SRP RDMA channel.
B
Bart Van Assche 已提交
1093
 * @req: SRP request.
1094
 * @sdev: If not NULL, only take ownership for this SCSI device.
B
Bart Van Assche 已提交
1095 1096 1097 1098 1099 1100
 * @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.
 */
1101
static struct scsi_cmnd *srp_claim_req(struct srp_rdma_ch *ch,
B
Bart Van Assche 已提交
1102
				       struct srp_request *req,
1103
				       struct scsi_device *sdev,
B
Bart Van Assche 已提交
1104 1105 1106 1107
				       struct scsi_cmnd *scmnd)
{
	unsigned long flags;

1108
	spin_lock_irqsave(&ch->lock, flags);
1109 1110 1111
	if (req->scmnd &&
	    (!sdev || req->scmnd->device == sdev) &&
	    (!scmnd || req->scmnd == scmnd)) {
B
Bart Van Assche 已提交
1112 1113 1114 1115 1116
		scmnd = req->scmnd;
		req->scmnd = NULL;
	} else {
		scmnd = NULL;
	}
1117
	spin_unlock_irqrestore(&ch->lock, flags);
B
Bart Van Assche 已提交
1118 1119 1120 1121 1122

	return scmnd;
}

/**
B
Bart Van Assche 已提交
1123
 * srp_free_req() - Unmap data and adjust ch->req_lim.
1124
 * @ch:     SRP RDMA channel.
1125 1126 1127
 * @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
1128
 */
1129 1130
static void srp_free_req(struct srp_rdma_ch *ch, struct srp_request *req,
			 struct scsi_cmnd *scmnd, s32 req_lim_delta)
1131
{
1132 1133
	unsigned long flags;

1134
	srp_unmap_data(scmnd, ch, req);
B
Bart Van Assche 已提交
1135

1136 1137 1138
	spin_lock_irqsave(&ch->lock, flags);
	ch->req_lim += req_lim_delta;
	spin_unlock_irqrestore(&ch->lock, flags);
1139 1140
}

1141 1142
static void srp_finish_req(struct srp_rdma_ch *ch, struct srp_request *req,
			   struct scsi_device *sdev, int result)
1143
{
1144
	struct scsi_cmnd *scmnd = srp_claim_req(ch, req, sdev, NULL);
B
Bart Van Assche 已提交
1145 1146

	if (scmnd) {
1147
		srp_free_req(ch, req, scmnd, 0);
1148
		scmnd->result = result;
B
Bart Van Assche 已提交
1149 1150
		scmnd->scsi_done(scmnd);
	}
1151 1152
}

1153
static void srp_terminate_io(struct srp_rport *rport)
1154
{
1155
	struct srp_target_port *target = rport->lld_data;
B
Bart Van Assche 已提交
1156
	struct srp_rdma_ch *ch;
1157 1158
	struct Scsi_Host *shost = target->scsi_host;
	struct scsi_device *sdev;
B
Bart Van Assche 已提交
1159
	int i, j;
1160

1161 1162 1163 1164 1165 1166 1167
	/*
	 * 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
1168 1169
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
1170

B
Bart Van Assche 已提交
1171 1172 1173 1174 1175 1176
		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);
		}
1177 1178
	}
}
1179

1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
/*
 * 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
1192 1193 1194
	struct srp_rdma_ch *ch;
	int i, j, ret = 0;
	bool multich = false;
1195

1196
	srp_disconnect_target(target);
1197 1198 1199 1200

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

1201
	/*
1202 1203 1204
	 * 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.
1205
	 */
B
Bart Van Assche 已提交
1206 1207 1208
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		ret += srp_new_cm_id(ch);
1209
	}
B
Bart Van Assche 已提交
1210 1211 1212 1213
	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];
1214

B
Bart Van Assche 已提交
1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225
			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);
1226

B
Bart Van Assche 已提交
1227 1228 1229 1230
		INIT_LIST_HEAD(&ch->free_tx);
		for (j = 0; j < target->queue_size; ++j)
			list_add(&ch->tx_ring[j]->list, &ch->free_tx);
	}
1231 1232 1233

	target->qp_in_error = false;

B
Bart Van Assche 已提交
1234 1235
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
1236
		if (ret)
B
Bart Van Assche 已提交
1237 1238 1239 1240
			break;
		ret = srp_connect_ch(ch, multich);
		multich = true;
	}
1241

1242 1243 1244
	if (ret == 0)
		shost_printk(KERN_INFO, target->scsi_host,
			     PFX "reconnect succeeded\n");
1245 1246 1247 1248

	return ret;
}

1249 1250
static void srp_map_desc(struct srp_map_state *state, dma_addr_t dma_addr,
			 unsigned int dma_len, u32 rkey)
1251
{
1252
	struct srp_direct_buf *desc = state->desc;
1253

1254 1255
	WARN_ON_ONCE(!dma_len);

1256 1257 1258
	desc->va = cpu_to_be64(dma_addr);
	desc->key = cpu_to_be32(rkey);
	desc->len = cpu_to_be32(dma_len);
1259

1260 1261 1262 1263
	state->total_len += dma_len;
	state->desc++;
	state->ndesc++;
}
1264

1265
static int srp_map_finish_fmr(struct srp_map_state *state,
1266
			      struct srp_rdma_ch *ch)
1267
{
1268 1269
	struct srp_target_port *target = ch->target;
	struct srp_device *dev = target->srp_host->srp_dev;
1270
	struct ib_pd *pd = target->pd;
1271 1272
	struct ib_pool_fmr *fmr;
	u64 io_addr = 0;
1273

1274 1275 1276
	if (state->fmr.next >= state->fmr.end)
		return -ENOMEM;

S
Sagi Grimberg 已提交
1277 1278 1279 1280 1281
	WARN_ON_ONCE(!dev->use_fmr);

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

1282
	if (state->npages == 1 && (pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)) {
S
Sagi Grimberg 已提交
1283
		srp_map_desc(state, state->base_dma_addr, state->dma_len,
1284
			     pd->unsafe_global_rkey);
S
Sagi Grimberg 已提交
1285 1286 1287
		goto reset_state;
	}

1288
	fmr = ib_fmr_pool_map_phys(ch->fmr_pool, state->pages,
1289 1290 1291
				   state->npages, io_addr);
	if (IS_ERR(fmr))
		return PTR_ERR(fmr);
1292

1293
	*state->fmr.next++ = fmr;
1294
	state->nmdesc++;
1295

1296 1297
	srp_map_desc(state, state->base_dma_addr & ~dev->mr_page_mask,
		     state->dma_len, fmr->fmr->rkey);
1298

S
Sagi Grimberg 已提交
1299 1300 1301 1302
reset_state:
	state->npages = 0;
	state->dma_len = 0;

1303 1304 1305
	return 0;
}

C
Christoph Hellwig 已提交
1306 1307 1308 1309 1310
static void srp_reg_mr_err_done(struct ib_cq *cq, struct ib_wc *wc)
{
	srp_handle_qp_err(cq, wc, "FAST REG");
}

1311 1312 1313 1314 1315 1316
/*
 * 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.
 */
1317
static int srp_map_finish_fr(struct srp_map_state *state,
C
Christoph Hellwig 已提交
1318
			     struct srp_request *req,
1319 1320
			     struct srp_rdma_ch *ch, int sg_nents,
			     unsigned int *sg_offset_p)
1321
{
1322
	struct srp_target_port *target = ch->target;
1323
	struct srp_device *dev = target->srp_host->srp_dev;
1324
	struct ib_pd *pd = target->pd;
1325
	struct ib_send_wr *bad_wr;
1326
	struct ib_reg_wr wr;
1327 1328
	struct srp_fr_desc *desc;
	u32 rkey;
1329
	int n, err;
1330

1331 1332 1333
	if (state->fr.next >= state->fr.end)
		return -ENOMEM;

S
Sagi Grimberg 已提交
1334 1335
	WARN_ON_ONCE(!dev->use_fast_reg);

1336
	if (sg_nents == 1 && (pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)) {
1337 1338 1339 1340
		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,
1341
			     pd->unsafe_global_rkey);
1342 1343
		if (sg_offset_p)
			*sg_offset_p = 0;
1344
		return 1;
S
Sagi Grimberg 已提交
1345 1346
	}

1347
	desc = srp_fr_pool_get(ch->fr_pool);
1348 1349 1350 1351 1352 1353
	if (!desc)
		return -ENOMEM;

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

1354 1355
	n = ib_map_mr_sg(desc->mr, state->sg, sg_nents, sg_offset_p,
			 dev->mr_page_size);
1356 1357
	if (unlikely(n < 0)) {
		srp_fr_pool_put(ch->fr_pool, &desc, 1);
1358
		pr_debug("%s: ib_map_mr_sg(%d, %d) returned %d.\n",
1359
			 dev_name(&req->scmnd->device->sdev_gendev), sg_nents,
1360
			 sg_offset_p ? *sg_offset_p : -1, n);
1361
		return n;
1362
	}
1363

1364
	WARN_ON_ONCE(desc->mr->length == 0);
1365

C
Christoph Hellwig 已提交
1366 1367
	req->reg_cqe.done = srp_reg_mr_err_done;

1368 1369
	wr.wr.next = NULL;
	wr.wr.opcode = IB_WR_REG_MR;
C
Christoph Hellwig 已提交
1370
	wr.wr.wr_cqe = &req->reg_cqe;
1371 1372 1373 1374 1375 1376 1377
	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);
1378

1379
	*state->fr.next++ = desc;
1380 1381
	state->nmdesc++;

1382 1383
	srp_map_desc(state, desc->mr->iova,
		     desc->mr->length, desc->mr->rkey);
1384

S
Sagi Grimberg 已提交
1385
	err = ib_post_send(ch->qp, &wr.wr, &bad_wr);
1386 1387
	if (unlikely(err)) {
		WARN_ON_ONCE(err == -ENOMEM);
S
Sagi Grimberg 已提交
1388
		return err;
1389
	}
S
Sagi Grimberg 已提交
1390

1391
	return n;
1392 1393
}

1394
static int srp_map_sg_entry(struct srp_map_state *state,
1395
			    struct srp_rdma_ch *ch,
1396
			    struct scatterlist *sg)
1397
{
1398
	struct srp_target_port *target = ch->target;
1399 1400 1401 1402
	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);
1403
	unsigned int len = 0;
1404 1405
	int ret;

1406
	WARN_ON_ONCE(!dma_len);
1407

1408
	while (dma_len) {
1409
		unsigned offset = dma_addr & ~dev->mr_page_mask;
1410 1411 1412

		if (state->npages == dev->max_pages_per_mr ||
		    (state->npages > 0 && offset != 0)) {
1413
			ret = srp_map_finish_fmr(state, ch);
1414 1415 1416 1417
			if (ret)
				return ret;
		}

1418
		len = min_t(unsigned int, dma_len, dev->mr_page_size - offset);
1419

1420 1421
		if (!state->npages)
			state->base_dma_addr = dma_addr;
1422
		state->pages[state->npages++] = dma_addr & dev->mr_page_mask;
1423
		state->dma_len += len;
1424 1425 1426 1427
		dma_addr += len;
		dma_len -= len;
	}

1428
	/*
1429
	 * If the end of the MR is not on a page boundary then we need to
1430
	 * close it out and start a new one -- we can only merge at page
1431
	 * boundaries.
1432 1433
	 */
	ret = 0;
1434
	if ((dma_addr & ~dev->mr_page_mask) != 0)
1435
		ret = srp_map_finish_fmr(state, ch);
1436 1437 1438
	return ret;
}

S
Sagi Grimberg 已提交
1439 1440 1441
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)
1442 1443
{
	struct scatterlist *sg;
1444
	int i, ret;
1445

S
Sagi Grimberg 已提交
1446 1447
	state->pages = req->map_page;
	state->fmr.next = req->fmr_list;
1448
	state->fmr.end = req->fmr_list + ch->target->mr_per_cmd;
S
Sagi Grimberg 已提交
1449 1450

	for_each_sg(scat, sg, count, i) {
1451
		ret = srp_map_sg_entry(state, ch, sg);
S
Sagi Grimberg 已提交
1452 1453
		if (ret)
			return ret;
1454
	}
1455

1456
	ret = srp_map_finish_fmr(state, ch);
S
Sagi Grimberg 已提交
1457 1458 1459 1460 1461 1462 1463 1464 1465 1466
	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)
{
1467 1468
	unsigned int sg_offset = 0;

1469
	state->fr.next = req->fr_list;
1470
	state->fr.end = req->fr_list + ch->target->mr_per_cmd;
1471
	state->sg = scat;
S
Sagi Grimberg 已提交
1472

1473 1474 1475
	if (count == 0)
		return 0;

B
Bart Van Assche 已提交
1476
	while (count) {
1477
		int i, n;
S
Sagi Grimberg 已提交
1478

1479
		n = srp_map_finish_fr(state, req, ch, count, &sg_offset);
1480 1481 1482
		if (unlikely(n < 0))
			return n;

B
Bart Van Assche 已提交
1483
		count -= n;
1484 1485 1486
		for (i = 0; i < n; i++)
			state->sg = sg_next(state->sg);
	}
S
Sagi Grimberg 已提交
1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502

	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),
1503
			     target->pd->unsafe_global_rkey);
1504
	}
1505

S
Sagi Grimberg 已提交
1506
	return 0;
1507 1508
}

1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524
/*
 * 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];
1525
	struct scatterlist idb_sg[1];
1526 1527 1528 1529 1530 1531 1532 1533 1534
	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;
1535 1536 1537

	if (dev->use_fast_reg) {
		state.sg = idb_sg;
1538
		sg_init_one(idb_sg, req->indirect_desc, idb_len);
1539
		idb_sg->dma_address = req->indirect_dma_addr; /* hack! */
1540 1541 1542
#ifdef CONFIG_NEED_SG_DMA_LENGTH
		idb_sg->dma_length = idb_sg->length;	      /* hack^2 */
#endif
1543
		ret = srp_map_finish_fr(&state, req, ch, 1, NULL);
1544 1545
		if (ret < 0)
			return ret;
1546
		WARN_ON_ONCE(ret < 1);
1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
	} 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;
	}
1558 1559 1560

	*idb_rkey = idb_desc.key;

1561
	return 0;
1562 1563
}

1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587
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);
}

1588 1589 1590 1591 1592 1593 1594 1595 1596
/**
 * 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.
 */
1597
static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_rdma_ch *ch,
1598 1599
			struct srp_request *req)
{
1600
	struct srp_target_port *target = ch->target;
1601
	struct ib_pd *pd = target->pd;
1602
	struct scatterlist *scat;
1603
	struct srp_cmd *cmd = req->cmd->buf;
1604
	int len, nents, count, ret;
1605 1606
	struct srp_device *dev;
	struct ib_device *ibdev;
1607 1608
	struct srp_map_state state;
	struct srp_indirect_buf *indirect_hdr;
1609 1610
	u32 idb_len, table_len;
	__be32 idb_rkey;
1611
	u8 fmt;
1612

1613
	if (!scsi_sglist(scmnd) || scmnd->sc_data_direction == DMA_NONE)
1614 1615 1616 1617
		return sizeof (struct srp_cmd);

	if (scmnd->sc_data_direction != DMA_FROM_DEVICE &&
	    scmnd->sc_data_direction != DMA_TO_DEVICE) {
1618 1619 1620
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled data direction %d\n",
			     scmnd->sc_data_direction);
1621 1622 1623
		return -EINVAL;
	}

1624 1625
	nents = scsi_sg_count(scmnd);
	scat  = scsi_sglist(scmnd);
1626

1627
	dev = target->srp_host->srp_dev;
1628 1629 1630
	ibdev = dev->dev;

	count = ib_dma_map_sg(ibdev, scat, nents, scmnd->sc_data_direction);
1631 1632
	if (unlikely(count == 0))
		return -EIO;
1633 1634 1635

	fmt = SRP_DATA_DESC_DIRECT;
	len = sizeof (struct srp_cmd) +	sizeof (struct srp_direct_buf);
1636

1637
	if (count == 1 && (pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)) {
1638 1639 1640 1641 1642 1643
		/*
		 * 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.
		 */
1644
		struct srp_direct_buf *buf = (void *) cmd->add_data;
1645

1646
		buf->va  = cpu_to_be64(ib_sg_dma_address(ibdev, scat));
1647
		buf->key = cpu_to_be32(pd->unsafe_global_rkey);
1648
		buf->len = cpu_to_be32(ib_sg_dma_len(ibdev, scat));
1649

1650
		req->nmdesc = 0;
1651 1652 1653
		goto map_complete;
	}

1654 1655 1656
	/*
	 * We have more than one scatter/gather entry, so build our indirect
	 * descriptor table, trying to merge as many entries as we can.
1657 1658 1659
	 */
	indirect_hdr = (void *) cmd->add_data;

1660 1661 1662
	ib_dma_sync_single_for_cpu(ibdev, req->indirect_dma_addr,
				   target->indirect_size, DMA_TO_DEVICE);

1663
	memset(&state, 0, sizeof(state));
B
Bart Van Assche 已提交
1664
	state.desc = req->indirect_desc;
S
Sagi Grimberg 已提交
1665
	if (dev->use_fast_reg)
1666
		ret = srp_map_sg_fr(&state, ch, req, scat, count);
S
Sagi Grimberg 已提交
1667
	else if (dev->use_fmr)
1668
		ret = srp_map_sg_fmr(&state, ch, req, scat, count);
S
Sagi Grimberg 已提交
1669
	else
1670 1671 1672 1673
		ret = srp_map_sg_dma(&state, ch, req, scat, count);
	req->nmdesc = state.nmdesc;
	if (ret < 0)
		goto unmap;
1674

1675 1676 1677
	{
		DEFINE_DYNAMIC_DEBUG_METADATA(ddm,
			"Memory mapping consistency check");
1678
		if (DYNAMIC_DEBUG_BRANCH(ddm))
1679 1680
			srp_check_mapping(&state, ch, req, scat, count);
	}
1681

1682 1683 1684 1685 1686
	/* 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.
1687 1688
	 */
	if (state.ndesc == 1) {
1689 1690
		/*
		 * Memory registration collapsed the sg-list into one entry,
1691 1692 1693
		 * so use a direct descriptor.
		 */
		struct srp_direct_buf *buf = (void *) cmd->add_data;
1694

1695
		*buf = req->indirect_desc[0];
1696
		goto map_complete;
1697 1698
	}

1699 1700 1701 1702
	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");
1703 1704
		ret = -EIO;
		goto unmap;
1705 1706 1707
	}

	count = min(state.ndesc, target->cmd_sg_cnt);
1708
	table_len = state.ndesc * sizeof (struct srp_direct_buf);
1709
	idb_len = sizeof(struct srp_indirect_buf) + table_len;
1710 1711 1712

	fmt = SRP_DATA_DESC_INDIRECT;
	len = sizeof(struct srp_cmd) + sizeof (struct srp_indirect_buf);
1713
	len += count * sizeof (struct srp_direct_buf);
1714

1715 1716
	memcpy(indirect_hdr->desc_list, req->indirect_desc,
	       count * sizeof (struct srp_direct_buf));
1717

1718
	if (!(pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)) {
1719 1720 1721
		ret = srp_map_idb(ch, req, state.gen.next, state.gen.end,
				  idb_len, &idb_rkey);
		if (ret < 0)
1722
			goto unmap;
1723 1724
		req->nmdesc++;
	} else {
1725
		idb_rkey = cpu_to_be32(pd->unsafe_global_rkey);
1726 1727
	}

1728
	indirect_hdr->table_desc.va = cpu_to_be64(req->indirect_dma_addr);
1729
	indirect_hdr->table_desc.key = idb_rkey;
1730 1731 1732 1733
	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)
1734
		cmd->data_out_desc_cnt = count;
1735
	else
1736 1737 1738 1739
		cmd->data_in_desc_cnt = count;

	ib_dma_sync_single_for_device(ibdev, req->indirect_dma_addr, table_len,
				      DMA_TO_DEVICE);
1740 1741

map_complete:
1742 1743 1744 1745 1746 1747
	if (scmnd->sc_data_direction == DMA_TO_DEVICE)
		cmd->buf_fmt = fmt << 4;
	else
		cmd->buf_fmt = fmt;

	return len;
1748 1749 1750

unmap:
	srp_unmap_data(scmnd, ch, req);
1751 1752
	if (ret == -ENOMEM && req->nmdesc >= target->mr_pool_size)
		ret = -E2BIG;
1753
	return ret;
1754 1755
}

1756 1757 1758
/*
 * Return an IU and possible credit to the free pool
 */
1759
static void srp_put_tx_iu(struct srp_rdma_ch *ch, struct srp_iu *iu,
1760 1761 1762 1763
			  enum srp_iu_type iu_type)
{
	unsigned long flags;

1764 1765
	spin_lock_irqsave(&ch->lock, flags);
	list_add(&iu->list, &ch->free_tx);
1766
	if (iu_type != SRP_IU_RSP)
1767 1768
		++ch->req_lim;
	spin_unlock_irqrestore(&ch->lock, flags);
1769 1770
}

1771
/*
1772
 * Must be called with ch->lock held to protect req_lim and free_tx.
1773
 * If IU is not sent, it must be returned using srp_put_tx_iu().
1774 1775 1776 1777 1778 1779 1780 1781 1782 1783
 *
 * 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.
 */
1784
static struct srp_iu *__srp_get_tx_iu(struct srp_rdma_ch *ch,
1785 1786
				      enum srp_iu_type iu_type)
{
1787
	struct srp_target_port *target = ch->target;
1788 1789 1790
	s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE;
	struct srp_iu *iu;

C
Christoph Hellwig 已提交
1791
	ib_process_cq_direct(ch->send_cq, -1);
1792

1793
	if (list_empty(&ch->free_tx))
1794 1795 1796
		return NULL;

	/* Initiator responses to target requests do not consume credits */
1797
	if (iu_type != SRP_IU_RSP) {
1798
		if (ch->req_lim <= rsv) {
1799 1800 1801 1802
			++target->zero_req_lim;
			return NULL;
		}

1803
		--ch->req_lim;
1804 1805
	}

1806
	iu = list_first_entry(&ch->free_tx, struct srp_iu, list);
1807
	list_del(&iu->list);
1808 1809 1810
	return iu;
}

C
Christoph Hellwig 已提交
1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823
static void srp_send_done(struct ib_cq *cq, struct ib_wc *wc)
{
	struct srp_iu *iu = container_of(wc->wr_cqe, struct srp_iu, cqe);
	struct srp_rdma_ch *ch = cq->cq_context;

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

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

1824
static int srp_post_send(struct srp_rdma_ch *ch, struct srp_iu *iu, int len)
1825
{
1826
	struct srp_target_port *target = ch->target;
1827 1828 1829 1830 1831
	struct ib_sge list;
	struct ib_send_wr wr, *bad_wr;

	list.addr   = iu->dma;
	list.length = len;
1832
	list.lkey   = target->lkey;
1833

C
Christoph Hellwig 已提交
1834 1835
	iu->cqe.done = srp_send_done;

1836
	wr.next       = NULL;
C
Christoph Hellwig 已提交
1837
	wr.wr_cqe     = &iu->cqe;
1838 1839 1840 1841 1842
	wr.sg_list    = &list;
	wr.num_sge    = 1;
	wr.opcode     = IB_WR_SEND;
	wr.send_flags = IB_SEND_SIGNALED;

1843
	return ib_post_send(ch->qp, &wr, &bad_wr);
1844 1845
}

1846
static int srp_post_recv(struct srp_rdma_ch *ch, struct srp_iu *iu)
1847
{
1848
	struct srp_target_port *target = ch->target;
1849
	struct ib_recv_wr wr, *bad_wr;
1850
	struct ib_sge list;
1851 1852 1853

	list.addr   = iu->dma;
	list.length = iu->size;
1854
	list.lkey   = target->lkey;
1855

C
Christoph Hellwig 已提交
1856 1857
	iu->cqe.done = srp_recv_done;

1858
	wr.next     = NULL;
C
Christoph Hellwig 已提交
1859
	wr.wr_cqe   = &iu->cqe;
1860 1861 1862
	wr.sg_list  = &list;
	wr.num_sge  = 1;

1863
	return ib_post_recv(ch->qp, &wr, &bad_wr);
1864 1865
}

1866
static void srp_process_rsp(struct srp_rdma_ch *ch, struct srp_rsp *rsp)
1867
{
1868
	struct srp_target_port *target = ch->target;
1869 1870 1871 1872 1873
	struct srp_request *req;
	struct scsi_cmnd *scmnd;
	unsigned long flags;

	if (unlikely(rsp->tag & SRP_TAG_TSK_MGMT)) {
1874 1875 1876
		spin_lock_irqsave(&ch->lock, flags);
		ch->req_lim += be32_to_cpu(rsp->req_lim_delta);
		spin_unlock_irqrestore(&ch->lock, flags);
1877

1878
		ch->tsk_mgmt_status = -1;
1879
		if (be32_to_cpu(rsp->resp_data_len) >= 4)
1880 1881
			ch->tsk_mgmt_status = rsp->data[3];
		complete(&ch->tsk_mgmt_done);
1882
	} else {
B
Bart Van Assche 已提交
1883 1884 1885 1886 1887
		scmnd = scsi_host_find_tag(target->scsi_host, rsp->tag);
		if (scmnd) {
			req = (void *)scmnd->host_scribble;
			scmnd = srp_claim_req(ch, req, NULL, scmnd);
		}
B
Bart Van Assche 已提交
1888
		if (!scmnd) {
1889
			shost_printk(KERN_ERR, target->scsi_host,
B
Bart Van Assche 已提交
1890 1891
				     "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
1892

1893 1894 1895
			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
1896 1897 1898

			return;
		}
1899 1900 1901 1902 1903 1904 1905 1906 1907
		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
1908
		if (unlikely(rsp->flags & SRP_RSP_FLAG_DIUNDER))
1909
			scsi_set_resid(scmnd, be32_to_cpu(rsp->data_in_res_cnt));
B
Bart Van Assche 已提交
1910 1911 1912 1913 1914 1915
		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));
1916

1917
		srp_free_req(ch, req, scmnd,
B
Bart Van Assche 已提交
1918 1919
			     be32_to_cpu(rsp->req_lim_delta));

1920 1921
		scmnd->host_scribble = NULL;
		scmnd->scsi_done(scmnd);
1922 1923 1924
	}
}

1925
static int srp_response_common(struct srp_rdma_ch *ch, s32 req_delta,
1926 1927
			       void *rsp, int len)
{
1928
	struct srp_target_port *target = ch->target;
1929
	struct ib_device *dev = target->srp_host->srp_dev->dev;
1930 1931
	unsigned long flags;
	struct srp_iu *iu;
1932
	int err;
1933

1934 1935 1936 1937
	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);
1938

1939 1940 1941
	if (!iu) {
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "no IU available to send response\n");
1942
		return 1;
1943 1944 1945 1946 1947 1948
	}

	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);

1949
	err = srp_post_send(ch, iu, len);
1950
	if (err) {
1951 1952
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "unable to post response: %d\n", err);
1953
		srp_put_tx_iu(ch, iu, SRP_IU_RSP);
1954
	}
1955 1956 1957 1958

	return err;
}

1959
static void srp_process_cred_req(struct srp_rdma_ch *ch,
1960 1961 1962 1963 1964 1965 1966 1967
				 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);

1968 1969
	if (srp_response_common(ch, delta, &rsp, sizeof(rsp)))
		shost_printk(KERN_ERR, ch->target->scsi_host, PFX
1970 1971 1972
			     "problems processing SRP_CRED_REQ\n");
}

1973
static void srp_process_aer_req(struct srp_rdma_ch *ch,
1974 1975
				struct srp_aer_req *req)
{
1976
	struct srp_target_port *target = ch->target;
1977 1978 1979 1980 1981 1982 1983
	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
1984
		     "ignoring AER for LUN %llu\n", scsilun_to_int(&req->lun));
1985

1986
	if (srp_response_common(ch, delta, &rsp, sizeof(rsp)))
1987 1988 1989 1990
		shost_printk(KERN_ERR, target->scsi_host, PFX
			     "problems processing SRP_AER_REQ\n");
}

C
Christoph Hellwig 已提交
1991
static void srp_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1992
{
C
Christoph Hellwig 已提交
1993 1994
	struct srp_iu *iu = container_of(wc->wr_cqe, struct srp_iu, cqe);
	struct srp_rdma_ch *ch = cq->cq_context;
1995
	struct srp_target_port *target = ch->target;
1996
	struct ib_device *dev = target->srp_host->srp_dev->dev;
1997
	int res;
1998 1999
	u8 opcode;

C
Christoph Hellwig 已提交
2000 2001 2002 2003 2004
	if (unlikely(wc->status != IB_WC_SUCCESS)) {
		srp_handle_qp_err(cq, wc, "RECV");
		return;
	}

2005
	ib_dma_sync_single_for_cpu(dev, iu->dma, ch->max_ti_iu_len,
2006
				   DMA_FROM_DEVICE);
2007 2008 2009 2010

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

	if (0) {
2011 2012
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "recv completion, opcode 0x%02x\n", opcode);
B
Bart Van Assche 已提交
2013 2014
		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 8, 1,
			       iu->buf, wc->byte_len, true);
2015 2016 2017 2018
	}

	switch (opcode) {
	case SRP_RSP:
2019
		srp_process_rsp(ch, iu->buf);
2020 2021
		break;

2022
	case SRP_CRED_REQ:
2023
		srp_process_cred_req(ch, iu->buf);
2024 2025 2026
		break;

	case SRP_AER_REQ:
2027
		srp_process_aer_req(ch, iu->buf);
2028 2029
		break;

2030 2031
	case SRP_T_LOGOUT:
		/* XXX Handle target logout */
2032 2033
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Got target logout request\n");
2034 2035 2036
		break;

	default:
2037 2038
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled SRP opcode 0x%02x\n", opcode);
2039 2040 2041
		break;
	}

2042
	ib_dma_sync_single_for_device(dev, iu->dma, ch->max_ti_iu_len,
2043
				      DMA_FROM_DEVICE);
2044

2045
	res = srp_post_recv(ch, iu);
2046 2047 2048
	if (res != 0)
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "Recv failed with error code %d\n", res);
2049 2050
}

2051 2052
/**
 * srp_tl_err_work() - handle a transport layer error
2053
 * @work: Work structure embedded in an SRP target port.
2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066
 *
 * 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
2067 2068
static void srp_handle_qp_err(struct ib_cq *cq, struct ib_wc *wc,
		const char *opname)
2069
{
C
Christoph Hellwig 已提交
2070
	struct srp_rdma_ch *ch = cq->cq_context;
2071 2072
	struct srp_target_port *target = ch->target;

2073
	if (ch->connected && !target->qp_in_error) {
C
Christoph Hellwig 已提交
2074 2075 2076 2077
		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);
2078
		queue_work(system_long_wq, &target->tl_err_work);
2079
	}
2080 2081 2082
	target->qp_in_error = true;
}

2083
static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
2084
{
2085
	struct srp_target_port *target = host_to_target(shost);
2086
	struct srp_rport *rport = target->rport;
2087
	struct srp_rdma_ch *ch;
2088 2089 2090
	struct srp_request *req;
	struct srp_iu *iu;
	struct srp_cmd *cmd;
2091
	struct ib_device *dev;
2092
	unsigned long flags;
B
Bart Van Assche 已提交
2093 2094
	u32 tag;
	u16 idx;
2095
	int len, ret;
2096 2097 2098 2099 2100 2101 2102 2103 2104 2105
	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);
2106

2107 2108 2109
	scmnd->result = srp_chkready(target->rport);
	if (unlikely(scmnd->result))
		goto err;
2110

B
Bart Van Assche 已提交
2111 2112
	WARN_ON_ONCE(scmnd->request->tag < 0);
	tag = blk_mq_unique_tag(scmnd->request);
B
Bart Van Assche 已提交
2113
	ch = &target->ch[blk_mq_unique_tag_to_hwq(tag)];
B
Bart Van Assche 已提交
2114 2115 2116 2117
	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);
2118 2119 2120 2121

	spin_lock_irqsave(&ch->lock, flags);
	iu = __srp_get_tx_iu(ch, SRP_IU_CMD);
	spin_unlock_irqrestore(&ch->lock, flags);
2122

B
Bart Van Assche 已提交
2123 2124 2125 2126
	if (!iu)
		goto err;

	req = &ch->req_ring[idx];
2127
	dev = target->srp_host->srp_dev->dev;
2128
	ib_dma_sync_single_for_cpu(dev, iu->dma, target->max_iu_len,
2129
				   DMA_TO_DEVICE);
2130

2131
	scmnd->host_scribble = (void *) req;
2132 2133 2134 2135 2136

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

	cmd->opcode = SRP_CMD;
B
Bart Van Assche 已提交
2137
	int_to_scsilun(scmnd->device->lun, &cmd->lun);
B
Bart Van Assche 已提交
2138
	cmd->tag    = tag;
2139 2140 2141 2142 2143
	memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);

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

2144
	len = srp_map_data(scmnd, ch, req);
2145
	if (len < 0) {
2146
		shost_printk(KERN_ERR, target->scsi_host,
2147 2148 2149 2150
			     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
2151
		 * max_pages_per_mr sg-list elements, tell the SCSI mid-layer
2152 2153 2154 2155
		 * to reduce queue depth temporarily.
		 */
		scmnd->result = len == -ENOMEM ?
			DID_OK << 16 | QUEUE_FULL << 1 : DID_ERROR << 16;
2156
		goto err_iu;
2157 2158
	}

2159
	ib_dma_sync_single_for_device(dev, iu->dma, target->max_iu_len,
2160
				      DMA_TO_DEVICE);
2161

2162
	if (srp_post_send(ch, iu, len)) {
2163
		shost_printk(KERN_ERR, target->scsi_host, PFX "Send failed\n");
2164 2165 2166
		goto err_unmap;
	}

2167 2168
	ret = 0;

2169 2170 2171 2172
unlock_rport:
	if (in_scsi_eh)
		mutex_unlock(&rport->mutex);

2173
	return ret;
2174 2175

err_unmap:
2176
	srp_unmap_data(scmnd, ch, req);
2177

2178
err_iu:
2179
	srp_put_tx_iu(ch, iu, SRP_IU_CMD);
2180

2181 2182 2183 2184 2185 2186
	/*
	 * Avoid that the loops that iterate over the request ring can
	 * encounter a dangling SCSI command pointer.
	 */
	req->scmnd = NULL;

2187 2188 2189 2190 2191 2192 2193
err:
	if (scmnd->result) {
		scmnd->scsi_done(scmnd);
		ret = 0;
	} else {
		ret = SCSI_MLQUEUE_HOST_BUSY;
	}
2194

2195
	goto unlock_rport;
2196 2197
}

2198 2199
/*
 * Note: the resources allocated in this function are freed in
2200
 * srp_free_ch_ib().
2201
 */
2202
static int srp_alloc_iu_bufs(struct srp_rdma_ch *ch)
2203
{
2204
	struct srp_target_port *target = ch->target;
2205 2206
	int i;

2207 2208 2209
	ch->rx_ring = kcalloc(target->queue_size, sizeof(*ch->rx_ring),
			      GFP_KERNEL);
	if (!ch->rx_ring)
2210
		goto err_no_ring;
2211 2212 2213
	ch->tx_ring = kcalloc(target->queue_size, sizeof(*ch->tx_ring),
			      GFP_KERNEL);
	if (!ch->tx_ring)
2214 2215 2216
		goto err_no_ring;

	for (i = 0; i < target->queue_size; ++i) {
2217 2218 2219 2220
		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])
2221 2222 2223
			goto err;
	}

2224
	for (i = 0; i < target->queue_size; ++i) {
2225 2226 2227 2228
		ch->tx_ring[i] = srp_alloc_iu(target->srp_host,
					      target->max_iu_len,
					      GFP_KERNEL, DMA_TO_DEVICE);
		if (!ch->tx_ring[i])
2229
			goto err;
2230

2231
		list_add(&ch->tx_ring[i]->list, &ch->free_tx);
2232 2233 2234 2235 2236
	}

	return 0;

err:
2237
	for (i = 0; i < target->queue_size; ++i) {
2238 2239
		srp_free_iu(target->srp_host, ch->rx_ring[i]);
		srp_free_iu(target->srp_host, ch->tx_ring[i]);
2240 2241
	}

2242 2243

err_no_ring:
2244 2245 2246 2247
	kfree(ch->tx_ring);
	ch->tx_ring = NULL;
	kfree(ch->rx_ring);
	ch->rx_ring = NULL;
2248

2249 2250 2251
	return -ENOMEM;
}

2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278
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;
}

2279
static void srp_cm_rep_handler(struct ib_cm_id *cm_id,
2280
			       const struct srp_login_rsp *lrsp,
2281
			       struct srp_rdma_ch *ch)
2282
{
2283
	struct srp_target_port *target = ch->target;
2284 2285 2286 2287 2288 2289
	struct ib_qp_attr *qp_attr = NULL;
	int attr_mask = 0;
	int ret;
	int i;

	if (lrsp->opcode == SRP_LOGIN_RSP) {
2290 2291
		ch->max_ti_iu_len = be32_to_cpu(lrsp->max_ti_iu_len);
		ch->req_lim       = be32_to_cpu(lrsp->req_lim_delta);
2292 2293 2294 2295 2296 2297

		/*
		 * Reserve credits for task management so we don't
		 * bounce requests back to the SCSI mid-layer.
		 */
		target->scsi_host->can_queue
2298
			= min(ch->req_lim - SRP_TSK_MGMT_SQ_SIZE,
2299
			      target->scsi_host->can_queue);
2300 2301 2302
		target->scsi_host->cmd_per_lun
			= min_t(int, target->scsi_host->can_queue,
				target->scsi_host->cmd_per_lun);
2303 2304 2305 2306 2307 2308 2309
	} else {
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled RSP opcode %#x\n", lrsp->opcode);
		ret = -ECONNRESET;
		goto error;
	}

2310 2311
	if (!ch->rx_ring) {
		ret = srp_alloc_iu_bufs(ch);
2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325
		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;

2326
	ret = ib_modify_qp(ch->qp, qp_attr, attr_mask);
2327 2328 2329
	if (ret)
		goto error_free;

2330
	for (i = 0; i < target->queue_size; i++) {
2331 2332 2333
		struct srp_iu *iu = ch->rx_ring[i];

		ret = srp_post_recv(ch, iu);
2334 2335 2336 2337 2338 2339 2340 2341 2342
		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;

2343 2344
	target->rq_tmo_jiffies = srp_compute_rq_tmo(qp_attr, attr_mask);

2345
	ret = ib_modify_qp(ch->qp, qp_attr, attr_mask);
2346 2347 2348 2349 2350 2351 2352 2353 2354
	if (ret)
		goto error_free;

	ret = ib_send_cm_rtu(cm_id, NULL, 0);

error_free:
	kfree(qp_attr);

error:
2355
	ch->status = ret;
2356 2357
}

2358 2359
static void srp_cm_rej_handler(struct ib_cm_id *cm_id,
			       struct ib_cm_event *event,
2360
			       struct srp_rdma_ch *ch)
2361
{
2362
	struct srp_target_port *target = ch->target;
2363
	struct Scsi_Host *shost = target->scsi_host;
2364 2365 2366 2367 2368 2369
	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;
2370 2371
		ch->path.dlid = cpi->redirect_lid;
		ch->path.pkey = cpi->redirect_pkey;
2372
		cm_id->remote_cm_qpn = be32_to_cpu(cpi->redirect_qp) & 0x00ffffff;
2373
		memcpy(ch->path.dgid.raw, cpi->redirect_gid, 16);
2374

2375
		ch->status = ch->path.dlid ?
2376 2377 2378 2379
			SRP_DLID_REDIRECT : SRP_PORT_REDIRECT;
		break;

	case IB_CM_REJ_PORT_REDIRECT:
2380
		if (srp_target_is_topspin(target)) {
2381 2382 2383 2384 2385
			/*
			 * Topspin/Cisco SRP gateways incorrectly send
			 * reject reason code 25 when they mean 24
			 * (port redirect).
			 */
2386
			memcpy(ch->path.dgid.raw,
2387 2388
			       event->param.rej_rcvd.ari, 16);

2389 2390
			shost_printk(KERN_DEBUG, shost,
				     PFX "Topspin/Cisco redirect to target port GID %016llx%016llx\n",
2391 2392
				     be64_to_cpu(ch->path.dgid.global.subnet_prefix),
				     be64_to_cpu(ch->path.dgid.global.interface_id));
2393

2394
			ch->status = SRP_PORT_REDIRECT;
2395
		} else {
2396 2397
			shost_printk(KERN_WARNING, shost,
				     "  REJ reason: IB_CM_REJ_PORT_REDIRECT\n");
2398
			ch->status = -ECONNRESET;
2399 2400 2401 2402
		}
		break;

	case IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID:
2403 2404
		shost_printk(KERN_WARNING, shost,
			    "  REJ reason: IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID\n");
2405
		ch->status = -ECONNRESET;
2406 2407 2408 2409 2410 2411 2412 2413 2414
		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)
2415 2416
				shost_printk(KERN_WARNING, shost,
					     PFX "SRP_LOGIN_REJ: requested max_it_iu_len too large\n");
2417
			else
B
Bart Van Assche 已提交
2418 2419
				shost_printk(KERN_WARNING, shost, PFX
					     "SRP LOGIN from %pI6 to %pI6 REJECTED, reason 0x%08x\n",
2420 2421
					     target->sgid.raw,
					     target->orig_dgid.raw, reason);
2422
		} else
2423 2424 2425
			shost_printk(KERN_WARNING, shost,
				     "  REJ reason: IB_CM_REJ_CONSUMER_DEFINED,"
				     " opcode 0x%02x\n", opcode);
2426
		ch->status = -ECONNRESET;
2427 2428
		break;

D
David Dillow 已提交
2429 2430
	case IB_CM_REJ_STALE_CONN:
		shost_printk(KERN_WARNING, shost, "  REJ reason: stale connection\n");
2431
		ch->status = SRP_STALE_CONN;
D
David Dillow 已提交
2432 2433
		break;

2434
	default:
2435 2436
		shost_printk(KERN_WARNING, shost, "  REJ reason 0x%x\n",
			     event->param.rej_rcvd.reason);
2437
		ch->status = -ECONNRESET;
2438 2439 2440 2441 2442
	}
}

static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
{
2443 2444
	struct srp_rdma_ch *ch = cm_id->context;
	struct srp_target_port *target = ch->target;
2445 2446 2447 2448
	int comp = 0;

	switch (event->event) {
	case IB_CM_REQ_ERROR:
2449 2450
		shost_printk(KERN_DEBUG, target->scsi_host,
			     PFX "Sending CM REQ failed\n");
2451
		comp = 1;
2452
		ch->status = -ECONNRESET;
2453 2454 2455 2456
		break;

	case IB_CM_REP_RECEIVED:
		comp = 1;
2457
		srp_cm_rep_handler(cm_id, event->private_data, ch);
2458 2459 2460
		break;

	case IB_CM_REJ_RECEIVED:
2461
		shost_printk(KERN_DEBUG, target->scsi_host, PFX "REJ received\n");
2462 2463
		comp = 1;

2464
		srp_cm_rej_handler(cm_id, event, ch);
2465 2466
		break;

2467
	case IB_CM_DREQ_RECEIVED:
2468 2469
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "DREQ received - connection closed\n");
2470
		ch->connected = false;
2471
		if (ib_send_cm_drep(cm_id, NULL, 0))
2472 2473
			shost_printk(KERN_ERR, target->scsi_host,
				     PFX "Sending CM DREP failed\n");
2474
		queue_work(system_long_wq, &target->tl_err_work);
2475 2476 2477
		break;

	case IB_CM_TIMEWAIT_EXIT:
2478 2479
		shost_printk(KERN_ERR, target->scsi_host,
			     PFX "connection closed\n");
2480
		comp = 1;
2481

2482
		ch->status = 0;
2483 2484
		break;

2485 2486 2487 2488 2489
	case IB_CM_MRA_RECEIVED:
	case IB_CM_DREQ_ERROR:
	case IB_CM_DREP_RECEIVED:
		break;

2490
	default:
2491 2492
		shost_printk(KERN_WARNING, target->scsi_host,
			     PFX "Unhandled CM event %d\n", event->event);
2493 2494 2495 2496
		break;
	}

	if (comp)
2497
		complete(&ch->done);
2498 2499 2500 2501

	return 0;
}

2502 2503 2504 2505 2506 2507 2508 2509
/**
 * srp_change_queue_depth - setting device queue depth
 * @sdev: scsi device struct
 * @qdepth: requested queue depth
 *
 * Returns queue depth.
 */
static int
2510
srp_change_queue_depth(struct scsi_device *sdev, int qdepth)
2511
{
2512
	if (!sdev->tagged_supported)
2513
		qdepth = 1;
2514
	return scsi_change_queue_depth(sdev, qdepth);
2515 2516
}

B
Bart Van Assche 已提交
2517 2518
static int srp_send_tsk_mgmt(struct srp_rdma_ch *ch, u64 req_tag, u64 lun,
			     u8 func)
2519
{
2520
	struct srp_target_port *target = ch->target;
2521
	struct srp_rport *rport = target->rport;
2522
	struct ib_device *dev = target->srp_host->srp_dev->dev;
2523 2524 2525
	struct srp_iu *iu;
	struct srp_tsk_mgmt *tsk_mgmt;

2526
	if (!ch->connected || target->qp_in_error)
2527 2528
		return -1;

2529
	init_completion(&ch->tsk_mgmt_done);
2530

2531
	/*
2532
	 * Lock the rport mutex to avoid that srp_create_ch_ib() is
2533 2534 2535
	 * invoked while a task management function is being sent.
	 */
	mutex_lock(&rport->mutex);
2536 2537 2538
	spin_lock_irq(&ch->lock);
	iu = __srp_get_tx_iu(ch, SRP_IU_TSK_MGMT);
	spin_unlock_irq(&ch->lock);
2539

2540 2541 2542
	if (!iu) {
		mutex_unlock(&rport->mutex);

2543
		return -1;
2544
	}
2545

2546 2547
	ib_dma_sync_single_for_cpu(dev, iu->dma, sizeof *tsk_mgmt,
				   DMA_TO_DEVICE);
2548 2549 2550 2551
	tsk_mgmt = iu->buf;
	memset(tsk_mgmt, 0, sizeof *tsk_mgmt);

	tsk_mgmt->opcode 	= SRP_TSK_MGMT;
B
Bart Van Assche 已提交
2552
	int_to_scsilun(lun, &tsk_mgmt->lun);
2553
	tsk_mgmt->tag		= req_tag | SRP_TAG_TSK_MGMT;
2554
	tsk_mgmt->tsk_mgmt_func = func;
2555
	tsk_mgmt->task_tag	= req_tag;
2556

2557 2558
	ib_dma_sync_single_for_device(dev, iu->dma, sizeof *tsk_mgmt,
				      DMA_TO_DEVICE);
2559 2560
	if (srp_post_send(ch, iu, sizeof(*tsk_mgmt))) {
		srp_put_tx_iu(ch, iu, SRP_IU_TSK_MGMT);
2561 2562
		mutex_unlock(&rport->mutex);

2563 2564
		return -1;
	}
2565
	mutex_unlock(&rport->mutex);
2566

2567
	if (!wait_for_completion_timeout(&ch->tsk_mgmt_done,
2568
					 msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS)))
2569
		return -1;
2570

2571 2572 2573
	return 0;
}

2574 2575
static int srp_abort(struct scsi_cmnd *scmnd)
{
2576
	struct srp_target_port *target = host_to_target(scmnd->device->host);
2577
	struct srp_request *req = (struct srp_request *) scmnd->host_scribble;
B
Bart Van Assche 已提交
2578
	u32 tag;
B
Bart Van Assche 已提交
2579
	u16 ch_idx;
2580
	struct srp_rdma_ch *ch;
2581
	int ret;
2582

2583
	shost_printk(KERN_ERR, target->scsi_host, "SRP abort called\n");
2584

B
Bart Van Assche 已提交
2585
	if (!req)
2586
		return SUCCESS;
B
Bart Van Assche 已提交
2587
	tag = blk_mq_unique_tag(scmnd->request);
B
Bart Van Assche 已提交
2588 2589 2590 2591 2592 2593 2594 2595
	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
2596
	if (srp_send_tsk_mgmt(ch, tag, scmnd->device->lun,
2597
			      SRP_TSK_ABORT_TASK) == 0)
2598
		ret = SUCCESS;
2599
	else if (target->rport->state == SRP_RPORT_LOST)
2600
		ret = FAST_IO_FAIL;
2601 2602
	else
		ret = FAILED;
2603
	srp_free_req(ch, req, scmnd, 0);
B
Bart Van Assche 已提交
2604
	scmnd->result = DID_ABORT << 16;
2605
	scmnd->scsi_done(scmnd);
2606

2607
	return ret;
2608 2609 2610 2611
}

static int srp_reset_device(struct scsi_cmnd *scmnd)
{
2612
	struct srp_target_port *target = host_to_target(scmnd->device->host);
B
Bart Van Assche 已提交
2613
	struct srp_rdma_ch *ch;
2614
	int i;
2615

2616
	shost_printk(KERN_ERR, target->scsi_host, "SRP reset_device called\n");
2617

B
Bart Van Assche 已提交
2618
	ch = &target->ch[0];
2619
	if (srp_send_tsk_mgmt(ch, SRP_TAG_NO_REQ, scmnd->device->lun,
2620
			      SRP_TSK_LUN_RESET))
2621
		return FAILED;
2622
	if (ch->tsk_mgmt_status)
2623 2624
		return FAILED;

B
Bart Van Assche 已提交
2625 2626 2627 2628
	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];
2629

B
Bart Van Assche 已提交
2630 2631
			srp_finish_req(ch, req, scmnd->device, DID_RESET << 16);
		}
2632
	}
2633 2634

	return SUCCESS;
2635 2636 2637 2638 2639 2640
}

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

2641
	shost_printk(KERN_ERR, target->scsi_host, PFX "SRP reset_host called\n");
2642

2643
	return srp_reconnect_rport(target->rport) == 0 ? SUCCESS : FAILED;
2644 2645
}

2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659
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;
	struct ib_device *ibdev = srp_dev->dev;

	if (!(ibdev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG))
		blk_queue_virt_boundary(sdev->request_queue,
					~srp_dev->mr_page_mask);

	return 0;
}

2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674
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;
}

2675 2676
static ssize_t show_id_ext(struct device *dev, struct device_attribute *attr,
			   char *buf)
2677
{
2678
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2679

2680
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->id_ext));
2681 2682
}

2683 2684
static ssize_t show_ioc_guid(struct device *dev, struct device_attribute *attr,
			     char *buf)
2685
{
2686
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2687

2688
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->ioc_guid));
2689 2690
}

2691 2692
static ssize_t show_service_id(struct device *dev,
			       struct device_attribute *attr, char *buf)
2693
{
2694
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2695

2696
	return sprintf(buf, "0x%016llx\n", be64_to_cpu(target->service_id));
2697 2698
}

2699 2700
static ssize_t show_pkey(struct device *dev, struct device_attribute *attr,
			 char *buf)
2701
{
2702
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2703

2704
	return sprintf(buf, "0x%04x\n", be16_to_cpu(target->pkey));
2705 2706
}

B
Bart Van Assche 已提交
2707 2708 2709 2710 2711
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));

2712
	return sprintf(buf, "%pI6\n", target->sgid.raw);
B
Bart Van Assche 已提交
2713 2714
}

2715 2716
static ssize_t show_dgid(struct device *dev, struct device_attribute *attr,
			 char *buf)
2717
{
2718
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
B
Bart Van Assche 已提交
2719
	struct srp_rdma_ch *ch = &target->ch[0];
2720

2721
	return sprintf(buf, "%pI6\n", ch->path.dgid.raw);
2722 2723
}

2724 2725
static ssize_t show_orig_dgid(struct device *dev,
			      struct device_attribute *attr, char *buf)
2726
{
2727
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2728

2729
	return sprintf(buf, "%pI6\n", target->orig_dgid.raw);
2730 2731
}

2732 2733 2734 2735
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
2736 2737
	struct srp_rdma_ch *ch;
	int i, req_lim = INT_MAX;
2738

B
Bart Van Assche 已提交
2739 2740 2741 2742 2743
	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);
2744 2745
}

2746 2747
static ssize_t show_zero_req_lim(struct device *dev,
				 struct device_attribute *attr, char *buf)
2748
{
2749
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2750 2751 2752 2753

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

2754 2755
static ssize_t show_local_ib_port(struct device *dev,
				  struct device_attribute *attr, char *buf)
2756
{
2757
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2758 2759 2760 2761

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

2762 2763
static ssize_t show_local_ib_device(struct device *dev,
				    struct device_attribute *attr, char *buf)
2764
{
2765
	struct srp_target_port *target = host_to_target(class_to_shost(dev));
2766

2767
	return sprintf(buf, "%s\n", target->srp_host->srp_dev->dev->name);
2768 2769
}

B
Bart Van Assche 已提交
2770 2771 2772 2773 2774 2775 2776 2777
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);
}

2778 2779 2780 2781 2782 2783 2784 2785
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);
}

2786 2787 2788 2789 2790 2791 2792 2793
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);
}

2794 2795 2796 2797 2798 2799 2800 2801
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);
}

2802 2803 2804 2805 2806 2807 2808 2809
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");
}

2810 2811 2812 2813
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
2814
static DEVICE_ATTR(sgid,	    S_IRUGO, show_sgid,		   NULL);
2815 2816
static DEVICE_ATTR(dgid,	    S_IRUGO, show_dgid,		   NULL);
static DEVICE_ATTR(orig_dgid,	    S_IRUGO, show_orig_dgid,	   NULL);
2817
static DEVICE_ATTR(req_lim,         S_IRUGO, show_req_lim,         NULL);
2818 2819 2820
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
2821
static DEVICE_ATTR(ch_count,        S_IRUGO, show_ch_count,        NULL);
2822
static DEVICE_ATTR(comp_vector,     S_IRUGO, show_comp_vector,     NULL);
2823
static DEVICE_ATTR(tl_retry_count,  S_IRUGO, show_tl_retry_count,  NULL);
2824
static DEVICE_ATTR(cmd_sg_entries,  S_IRUGO, show_cmd_sg_entries,  NULL);
2825
static DEVICE_ATTR(allow_ext_sg,    S_IRUGO, show_allow_ext_sg,    NULL);
2826 2827 2828 2829 2830 2831

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
2832
	&dev_attr_sgid,
2833 2834
	&dev_attr_dgid,
	&dev_attr_orig_dgid,
2835
	&dev_attr_req_lim,
2836 2837 2838
	&dev_attr_zero_req_lim,
	&dev_attr_local_ib_port,
	&dev_attr_local_ib_device,
B
Bart Van Assche 已提交
2839
	&dev_attr_ch_count,
2840
	&dev_attr_comp_vector,
2841
	&dev_attr_tl_retry_count,
2842
	&dev_attr_cmd_sg_entries,
2843
	&dev_attr_allow_ext_sg,
2844 2845 2846
	NULL
};

2847 2848
static struct scsi_host_template srp_template = {
	.module				= THIS_MODULE,
R
Roland Dreier 已提交
2849 2850
	.name				= "InfiniBand SRP initiator",
	.proc_name			= DRV_NAME,
2851
	.slave_alloc			= srp_slave_alloc,
2852
	.slave_configure		= srp_slave_configure,
2853 2854
	.info				= srp_target_info,
	.queuecommand			= srp_queuecommand,
2855
	.change_queue_depth             = srp_change_queue_depth,
2856 2857 2858
	.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
2859
	.skip_settle_delay		= true,
2860
	.sg_tablesize			= SRP_DEF_SG_TABLESIZE,
2861
	.can_queue			= SRP_DEFAULT_CMD_SQ_SIZE,
2862
	.this_id			= -1,
2863
	.cmd_per_lun			= SRP_DEFAULT_CMD_SQ_SIZE,
2864
	.use_clustering			= ENABLE_CLUSTERING,
B
Bart Van Assche 已提交
2865
	.shost_attrs			= srp_host_attrs,
2866
	.track_queue_depth		= 1,
2867 2868
};

2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879
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;
}

2880 2881 2882 2883 2884 2885 2886
/*
 * 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.
 */
2887 2888
static int srp_add_target(struct srp_host *host, struct srp_target_port *target)
{
2889 2890 2891
	struct srp_rport_identifiers ids;
	struct srp_rport *rport;

2892
	target->state = SRP_TARGET_SCANNING;
2893
	sprintf(target->target_name, "SRP.T10:%016llX",
2894
		be64_to_cpu(target->id_ext));
2895

2896
	if (scsi_add_host(target->scsi_host, host->srp_dev->dev->dma_device))
2897 2898
		return -ENODEV;

2899 2900
	memcpy(ids.port_id, &target->id_ext, 8);
	memcpy(ids.port_id + 8, &target->ioc_guid, 8);
2901
	ids.roles = SRP_RPORT_ROLE_TARGET;
2902 2903 2904 2905 2906 2907
	rport = srp_rport_add(target->scsi_host, &ids);
	if (IS_ERR(rport)) {
		scsi_remove_host(target->scsi_host);
		return PTR_ERR(rport);
	}

2908
	rport->lld_data = target;
2909
	target->rport = rport;
2910

2911
	spin_lock(&host->target_lock);
2912
	list_add_tail(&target->list, &host->target_list);
2913
	spin_unlock(&host->target_lock);
2914 2915

	scsi_scan_target(&target->scsi_host->shost_gendev,
2916
			 0, target->scsi_id, SCAN_WILD_CARD, SCSI_SCAN_INITIAL);
2917

2918 2919
	if (srp_connected_ch(target) < target->ch_count ||
	    target->qp_in_error) {
2920 2921 2922 2923 2924 2925
		shost_printk(KERN_INFO, target->scsi_host,
			     PFX "SCSI scan failed - removing SCSI host\n");
		srp_queue_remove_work(target);
		goto out;
	}

2926
	pr_debug("%s: SCSI scan succeeded - detected %d LUNs\n",
2927 2928 2929 2930 2931 2932 2933 2934 2935
		 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:
2936 2937 2938
	return 0;
}

2939
static void srp_release_dev(struct device *dev)
2940 2941
{
	struct srp_host *host =
2942
		container_of(dev, struct srp_host, dev);
2943 2944 2945 2946 2947 2948

	complete(&host->released);
}

static struct class srp_class = {
	.name    = "infiniband_srp",
2949
	.dev_release = srp_release_dev
2950 2951
};

2952 2953
/**
 * srp_conn_unique() - check whether the connection to a target is unique
2954 2955
 * @host:   SRP host.
 * @target: SRP target port.
2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983
 */
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;
}

2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999
/*
 * 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,
3000
	SRP_OPT_MAX_CMD_PER_LUN	= 1 << 6,
3001
	SRP_OPT_IO_CLASS	= 1 << 7,
3002
	SRP_OPT_INITIATOR_EXT	= 1 << 8,
3003
	SRP_OPT_CMD_SG_ENTRIES	= 1 << 9,
3004 3005
	SRP_OPT_ALLOW_EXT_SG	= 1 << 10,
	SRP_OPT_SG_TABLESIZE	= 1 << 11,
3006
	SRP_OPT_COMP_VECTOR	= 1 << 12,
3007
	SRP_OPT_TL_RETRY_COUNT	= 1 << 13,
3008
	SRP_OPT_QUEUE_SIZE	= 1 << 14,
3009 3010 3011 3012 3013 3014 3015
	SRP_OPT_ALL		= (SRP_OPT_ID_EXT	|
				   SRP_OPT_IOC_GUID	|
				   SRP_OPT_DGID		|
				   SRP_OPT_PKEY		|
				   SRP_OPT_SERVICE_ID),
};

3016
static const match_table_t srp_opt_tokens = {
3017 3018 3019 3020 3021 3022 3023
	{ 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" 	},
3024
	{ SRP_OPT_IO_CLASS,		"io_class=%x"		},
3025
	{ SRP_OPT_INITIATOR_EXT,	"initiator_ext=%s"	},
3026
	{ SRP_OPT_CMD_SG_ENTRIES,	"cmd_sg_entries=%u"	},
3027 3028
	{ SRP_OPT_ALLOW_EXT_SG,		"allow_ext_sg=%u"	},
	{ SRP_OPT_SG_TABLESIZE,		"sg_tablesize=%u"	},
3029
	{ SRP_OPT_COMP_VECTOR,		"comp_vector=%u"	},
3030
	{ SRP_OPT_TL_RETRY_COUNT,	"tl_retry_count=%u"	},
3031
	{ SRP_OPT_QUEUE_SIZE,		"queue_size=%d"		},
3032
	{ SRP_OPT_ERR,			NULL 			}
3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050
};

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

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

	sep_opt = options;
3051
	while ((p = strsep(&sep_opt, ",\n")) != NULL) {
3052 3053 3054 3055 3056 3057 3058 3059 3060
		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);
3061 3062 3063 3064
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3065 3066 3067 3068 3069 3070
			target->id_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

		case SRP_OPT_IOC_GUID:
			p = match_strdup(args);
3071 3072 3073 3074
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3075 3076 3077 3078 3079 3080
			target->ioc_guid = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

		case SRP_OPT_DGID:
			p = match_strdup(args);
3081 3082 3083 3084
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3085
			if (strlen(p) != 32) {
3086
				pr_warn("bad dest GID parameter '%s'\n", p);
3087
				kfree(p);
3088 3089 3090 3091
				goto out;
			}

			for (i = 0; i < 16; ++i) {
3092 3093 3094 3095 3096 3097 3098
				strlcpy(dgid, p + i * 2, sizeof(dgid));
				if (sscanf(dgid, "%hhx",
					   &target->orig_dgid.raw[i]) < 1) {
					ret = -EINVAL;
					kfree(p);
					goto out;
				}
3099
			}
3100
			kfree(p);
3101 3102 3103 3104
			break;

		case SRP_OPT_PKEY:
			if (match_hex(args, &token)) {
3105
				pr_warn("bad P_Key parameter '%s'\n", p);
3106 3107
				goto out;
			}
3108
			target->pkey = cpu_to_be16(token);
3109 3110 3111 3112
			break;

		case SRP_OPT_SERVICE_ID:
			p = match_strdup(args);
3113 3114 3115 3116
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3117 3118 3119 3120 3121 3122
			target->service_id = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

		case SRP_OPT_MAX_SECT:
			if (match_int(args, &token)) {
3123
				pr_warn("bad max sect parameter '%s'\n", p);
3124 3125 3126 3127 3128
				goto out;
			}
			target->scsi_host->max_sectors = token;
			break;

3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140
		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;

3141
		case SRP_OPT_MAX_CMD_PER_LUN:
3142
			if (match_int(args, &token) || token < 1) {
3143 3144
				pr_warn("bad max cmd_per_lun parameter '%s'\n",
					p);
3145 3146
				goto out;
			}
3147
			target->scsi_host->cmd_per_lun = token;
3148 3149
			break;

3150 3151
		case SRP_OPT_IO_CLASS:
			if (match_hex(args, &token)) {
3152
				pr_warn("bad IO class parameter '%s'\n", p);
3153 3154 3155 3156
				goto out;
			}
			if (token != SRP_REV10_IB_IO_CLASS &&
			    token != SRP_REV16A_IB_IO_CLASS) {
3157 3158 3159
				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);
3160 3161 3162 3163 3164
				goto out;
			}
			target->io_class = token;
			break;

3165 3166
		case SRP_OPT_INITIATOR_EXT:
			p = match_strdup(args);
3167 3168 3169 3170
			if (!p) {
				ret = -ENOMEM;
				goto out;
			}
3171 3172 3173 3174
			target->initiator_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
			kfree(p);
			break;

3175 3176
		case SRP_OPT_CMD_SG_ENTRIES:
			if (match_int(args, &token) || token < 1 || token > 255) {
3177 3178
				pr_warn("bad max cmd_sg_entries parameter '%s'\n",
					p);
3179 3180 3181 3182 3183
				goto out;
			}
			target->cmd_sg_cnt = token;
			break;

3184 3185
		case SRP_OPT_ALLOW_EXT_SG:
			if (match_int(args, &token)) {
3186
				pr_warn("bad allow_ext_sg parameter '%s'\n", p);
3187 3188 3189 3190 3191 3192 3193
				goto out;
			}
			target->allow_ext_sg = !!token;
			break;

		case SRP_OPT_SG_TABLESIZE:
			if (match_int(args, &token) || token < 1 ||
3194
					token > SG_MAX_SEGMENTS) {
3195 3196
				pr_warn("bad max sg_tablesize parameter '%s'\n",
					p);
3197 3198 3199 3200 3201
				goto out;
			}
			target->sg_tablesize = token;
			break;

3202 3203 3204 3205 3206 3207 3208 3209
		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;

3210 3211 3212 3213 3214 3215 3216 3217 3218
		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;

3219
		default:
3220 3221
			pr_warn("unknown parameter or missing value '%s' in target creation request\n",
				p);
3222 3223 3224 3225 3226 3227 3228 3229 3230 3231
			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))
3232 3233
				pr_warn("target creation request is missing parameter '%s'\n",
					srp_opt_tokens[i].pattern);
3234

3235 3236 3237 3238 3239 3240
	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);

3241 3242 3243 3244 3245
out:
	kfree(options);
	return ret;
}

3246 3247
static ssize_t srp_create_target(struct device *dev,
				 struct device_attribute *attr,
3248 3249 3250
				 const char *buf, size_t count)
{
	struct srp_host *host =
3251
		container_of(dev, struct srp_host, dev);
3252 3253
	struct Scsi_Host *target_host;
	struct srp_target_port *target;
3254
	struct srp_rdma_ch *ch;
3255 3256
	struct srp_device *srp_dev = host->srp_dev;
	struct ib_device *ibdev = srp_dev->dev;
B
Bart Van Assche 已提交
3257
	int ret, node_idx, node, cpu, i;
3258
	unsigned int max_sectors_per_mr, mr_per_cmd = 0;
B
Bart Van Assche 已提交
3259
	bool multich = false;
3260 3261 3262 3263 3264 3265

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

3266
	target_host->transportt  = ib_srp_transport_template;
3267 3268
	target_host->max_channel = 0;
	target_host->max_id      = 1;
B
Bart Van Assche 已提交
3269
	target_host->max_lun     = -1LL;
A
Arne Redlich 已提交
3270
	target_host->max_cmd_len = sizeof ((struct srp_cmd *) (void *) 0L)->cdb;
R
Roland Dreier 已提交
3271

3272 3273
	target = host_to_target(target_host);

3274 3275 3276
	target->io_class	= SRP_REV16A_IB_IO_CLASS;
	target->scsi_host	= target_host;
	target->srp_host	= host;
3277
	target->pd		= host->srp_dev->pd;
J
Jason Gunthorpe 已提交
3278
	target->lkey		= host->srp_dev->pd->local_dma_lkey;
3279
	target->cmd_sg_cnt	= cmd_sg_entries;
3280 3281
	target->sg_tablesize	= indirect_sg_entries ? : cmd_sg_entries;
	target->allow_ext_sg	= allow_ext_sg;
3282
	target->tl_retry_count	= 7;
3283
	target->queue_size	= SRP_DEFAULT_QUEUE_SIZE;
3284

3285 3286 3287 3288 3289 3290
	/*
	 * Avoid that the SCSI host can be removed by srp_remove_target()
	 * before this function returns.
	 */
	scsi_host_get(target->scsi_host);

3291 3292
	mutex_lock(&host->add_target_mutex);

3293 3294
	ret = srp_parse_options(buf, target);
	if (ret)
3295
		goto out;
3296

3297 3298
	target->req_ring_size = target->queue_size - SRP_TSK_MGMT_SQ_SIZE;

3299 3300 3301 3302 3303 3304 3305
	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;
3306
		goto out;
3307 3308
	}

3309
	if (!srp_dev->has_fmr && !srp_dev->has_fr && !target->allow_ext_sg &&
3310
	    target->cmd_sg_cnt < target->sg_tablesize) {
3311
		pr_warn("No MR pool and no external indirect descriptors, limiting sg_tablesize to cmd_sg_cnt\n");
3312 3313 3314
		target->sg_tablesize = target->cmd_sg_cnt;
	}

3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338
	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);
	}

3339
	target_host->sg_tablesize = target->sg_tablesize;
3340 3341
	target->mr_pool_size = target->scsi_host->can_queue * mr_per_cmd;
	target->mr_per_cmd = mr_per_cmd;
3342 3343
	target->indirect_size = target->sg_tablesize *
				sizeof (struct srp_direct_buf);
3344 3345 3346 3347
	target->max_iu_len = sizeof (struct srp_cmd) +
			     sizeof (struct srp_indirect_buf) +
			     target->cmd_sg_cnt * sizeof (struct srp_direct_buf);

3348
	INIT_WORK(&target->tl_err_work, srp_tl_err_work);
3349
	INIT_WORK(&target->remove_work, srp_remove_work);
3350
	spin_lock_init(&target->lock);
3351
	ret = ib_query_gid(ibdev, host->port, 0, &target->sgid, NULL);
3352
	if (ret)
3353
		goto out;
3354

B
Bart Van Assche 已提交
3355 3356 3357 3358 3359 3360 3361 3362 3363
	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)
3364
		goto out;
3365

B
Bart Van Assche 已提交
3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393
	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;
3394

B
Bart Van Assche 已提交
3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414
			ret = srp_create_ch_ib(ch);
			if (ret)
				goto err_disconnect;

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

			ret = srp_connect_ch(ch, multich);
			if (ret) {
				shost_printk(KERN_ERR, target->scsi_host,
					     PFX "Connection %d/%d failed\n",
					     ch_start + cpu_idx,
					     target->ch_count);
				if (node_idx == 0 && cpu_idx == 0) {
					goto err_disconnect;
				} else {
					srp_free_ch_ib(target, ch);
					srp_free_req_data(target, ch);
					target->ch_count = ch - target->ch;
3415
					goto connected;
B
Bart Van Assche 已提交
3416 3417 3418 3419 3420 3421 3422
				}
			}

			multich = true;
			cpu_idx++;
		}
		node_idx++;
3423 3424
	}

3425
connected:
B
Bart Van Assche 已提交
3426 3427
	target->scsi_host->nr_hw_queues = target->ch_count;

3428 3429 3430 3431
	ret = srp_add_target(host, target);
	if (ret)
		goto err_disconnect;

3432 3433 3434 3435 3436
	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),
3437
			     be16_to_cpu(target->pkey),
3438
			     be64_to_cpu(target->service_id),
3439
			     target->sgid.raw, target->orig_dgid.raw);
3440
	}
B
Bart Van Assche 已提交
3441

3442 3443 3444 3445
	ret = count;

out:
	mutex_unlock(&host->add_target_mutex);
3446 3447

	scsi_host_put(target->scsi_host);
3448 3449
	if (ret < 0)
		scsi_host_put(target->scsi_host);
3450

3451
	return ret;
3452 3453 3454 3455

err_disconnect:
	srp_disconnect_target(target);

B
Bart Van Assche 已提交
3456 3457 3458 3459 3460
	for (i = 0; i < target->ch_count; i++) {
		ch = &target->ch[i];
		srp_free_ch_ib(target, ch);
		srp_free_req_data(target, ch);
	}
3461

B
Bart Van Assche 已提交
3462
	kfree(target->ch);
3463
	goto out;
3464 3465
}

3466
static DEVICE_ATTR(add_target, S_IWUSR, NULL, srp_create_target);
3467

3468 3469
static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
			  char *buf)
3470
{
3471
	struct srp_host *host = container_of(dev, struct srp_host, dev);
3472

3473
	return sprintf(buf, "%s\n", host->srp_dev->dev->name);
3474 3475
}

3476
static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
3477

3478 3479
static ssize_t show_port(struct device *dev, struct device_attribute *attr,
			 char *buf)
3480
{
3481
	struct srp_host *host = container_of(dev, struct srp_host, dev);
3482 3483 3484 3485

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

3486
static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
3487

3488
static struct srp_host *srp_add_port(struct srp_device *device, u8 port)
3489 3490 3491 3492 3493 3494 3495 3496
{
	struct srp_host *host;

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

	INIT_LIST_HEAD(&host->target_list);
3497
	spin_lock_init(&host->target_lock);
3498
	init_completion(&host->released);
3499
	mutex_init(&host->add_target_mutex);
3500
	host->srp_dev = device;
3501 3502
	host->port = port;

3503 3504
	host->dev.class = &srp_class;
	host->dev.parent = device->dev->dma_device;
3505
	dev_set_name(&host->dev, "srp-%s-%d", device->dev->name, port);
3506

3507
	if (device_register(&host->dev))
3508
		goto free_host;
3509
	if (device_create_file(&host->dev, &dev_attr_add_target))
3510
		goto err_class;
3511
	if (device_create_file(&host->dev, &dev_attr_ibdev))
3512
		goto err_class;
3513
	if (device_create_file(&host->dev, &dev_attr_port))
3514 3515 3516 3517 3518
		goto err_class;

	return host;

err_class:
3519
	device_unregister(&host->dev);
3520

3521
free_host:
3522 3523 3524 3525 3526 3527 3528
	kfree(host);

	return NULL;
}

static void srp_add_one(struct ib_device *device)
{
3529
	struct srp_device *srp_dev;
3530
	struct srp_host *host;
3531
	int mr_page_shift, p;
3532
	u64 max_pages_per_mr;
3533
	unsigned int flags = 0;
3534

3535
	srp_dev = kzalloc(sizeof(*srp_dev), GFP_KERNEL);
3536
	if (!srp_dev)
3537
		return;
3538 3539 3540

	/*
	 * Use the smallest page size supported by the HCA, down to a
3541 3542
	 * minimum of 4096 bytes. We're unlikely to build large sglists
	 * out of smaller entries.
3543
	 */
3544
	mr_page_shift		= max(12, ffs(device->attrs.page_size_cap) - 1);
3545 3546
	srp_dev->mr_page_size	= 1 << mr_page_shift;
	srp_dev->mr_page_mask	= ~((u64) srp_dev->mr_page_size - 1);
3547
	max_pages_per_mr	= device->attrs.max_mr_size;
3548
	do_div(max_pages_per_mr, srp_dev->mr_page_size);
3549 3550 3551
	pr_debug("%s: %llu / %u = %llu <> %u\n", __func__,
		 device->attrs.max_mr_size, srp_dev->mr_page_size,
		 max_pages_per_mr, SRP_MAX_PAGES_PER_MR);
3552 3553
	srp_dev->max_pages_per_mr = min_t(u64, SRP_MAX_PAGES_PER_MR,
					  max_pages_per_mr);
3554 3555 3556 3557 3558

	srp_dev->has_fmr = (device->alloc_fmr && device->dealloc_fmr &&
			    device->map_phys_fmr && device->unmap_fmr);
	srp_dev->has_fr = (device->attrs.device_cap_flags &
			   IB_DEVICE_MEM_MGT_EXTENSIONS);
3559
	if (!never_register && !srp_dev->has_fmr && !srp_dev->has_fr) {
3560
		dev_warn(&device->dev, "neither FMR nor FR is supported\n");
3561 3562
	} else if (!never_register &&
		   device->attrs.max_mr_size >= 2 * srp_dev->mr_page_size) {
3563 3564 3565 3566
		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;
	}
3567

3568 3569 3570 3571
	if (never_register || !register_always ||
	    (!srp_dev->has_fmr && !srp_dev->has_fr))
		flags |= IB_PD_UNSAFE_GLOBAL_RKEY;

3572 3573 3574
	if (srp_dev->use_fast_reg) {
		srp_dev->max_pages_per_mr =
			min_t(u32, srp_dev->max_pages_per_mr,
3575
			      device->attrs.max_fast_reg_page_list_len);
3576
	}
3577 3578
	srp_dev->mr_max_size	= srp_dev->mr_page_size *
				   srp_dev->max_pages_per_mr;
3579 3580 3581
	pr_debug("%s: mr_page_shift = %d, device->max_mr_size = %#llx, device->max_fast_reg_page_list_len = %u, max_pages_per_mr = %d, mr_max_size = %#x\n",
		 device->name, mr_page_shift, device->attrs.max_mr_size,
		 device->attrs.max_fast_reg_page_list_len,
3582
		 srp_dev->max_pages_per_mr, srp_dev->mr_max_size);
3583 3584 3585 3586

	INIT_LIST_HEAD(&srp_dev->dev_list);

	srp_dev->dev = device;
3587
	srp_dev->pd  = ib_alloc_pd(device, flags);
3588 3589 3590 3591
	if (IS_ERR(srp_dev->pd))
		goto free_dev;


3592
	for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) {
3593
		host = srp_add_port(srp_dev, p);
3594
		if (host)
3595
			list_add_tail(&host->list, &srp_dev->dev_list);
3596 3597
	}

3598
	ib_set_client_data(device, &srp_client, srp_dev);
3599
	return;
3600 3601 3602

free_dev:
	kfree(srp_dev);
3603 3604
}

3605
static void srp_remove_one(struct ib_device *device, void *client_data)
3606
{
3607
	struct srp_device *srp_dev;
3608
	struct srp_host *host, *tmp_host;
3609
	struct srp_target_port *target;
3610

3611
	srp_dev = client_data;
3612 3613
	if (!srp_dev)
		return;
3614

3615
	list_for_each_entry_safe(host, tmp_host, &srp_dev->dev_list, list) {
3616
		device_unregister(&host->dev);
3617 3618 3619 3620 3621 3622 3623
		/*
		 * Wait for the sysfs entry to go away, so that no new
		 * target ports can be created.
		 */
		wait_for_completion(&host->released);

		/*
3624
		 * Remove all target ports.
3625
		 */
3626
		spin_lock(&host->target_lock);
3627 3628
		list_for_each_entry(target, &host->target_list, list)
			srp_queue_remove_work(target);
3629
		spin_unlock(&host->target_lock);
3630 3631

		/*
3632
		 * Wait for tl_err and target port removal tasks.
3633
		 */
3634
		flush_workqueue(system_long_wq);
3635
		flush_workqueue(srp_remove_wq);
3636 3637 3638 3639

		kfree(host);
	}

3640 3641 3642
	ib_dealloc_pd(srp_dev->pd);

	kfree(srp_dev);
3643 3644
}

3645
static struct srp_function_template ib_srp_transport_functions = {
3646 3647
	.has_rport_state	 = true,
	.reset_timer_if_blocked	 = true,
3648
	.reconnect_delay	 = &srp_reconnect_delay,
3649 3650 3651
	.fast_io_fail_tmo	 = &srp_fast_io_fail_tmo,
	.dev_loss_tmo		 = &srp_dev_loss_tmo,
	.reconnect		 = srp_rport_reconnect,
3652
	.rport_delete		 = srp_rport_delete,
3653
	.terminate_rport_io	 = srp_terminate_io,
3654 3655
};

3656 3657 3658 3659
static int __init srp_init_module(void)
{
	int ret;

3660
	if (srp_sg_tablesize) {
3661
		pr_warn("srp_sg_tablesize is deprecated, please use cmd_sg_entries\n");
3662 3663 3664 3665 3666 3667 3668 3669
		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) {
3670
		pr_warn("Clamping cmd_sg_entries to 255\n");
3671
		cmd_sg_entries = 255;
3672 3673
	}

3674 3675 3676
	if (!indirect_sg_entries)
		indirect_sg_entries = cmd_sg_entries;
	else if (indirect_sg_entries < cmd_sg_entries) {
3677 3678
		pr_warn("Bumping up indirect_sg_entries to match cmd_sg_entries (%u)\n",
			cmd_sg_entries);
3679 3680 3681
		indirect_sg_entries = cmd_sg_entries;
	}

3682
	srp_remove_wq = create_workqueue("srp_remove");
3683 3684
	if (!srp_remove_wq) {
		ret = -ENOMEM;
3685 3686 3687 3688
		goto out;
	}

	ret = -ENOMEM;
3689 3690 3691
	ib_srp_transport_template =
		srp_attach_transport(&ib_srp_transport_functions);
	if (!ib_srp_transport_template)
3692
		goto destroy_wq;
3693

3694 3695
	ret = class_register(&srp_class);
	if (ret) {
3696
		pr_err("couldn't register class infiniband_srp\n");
3697
		goto release_tr;
3698 3699
	}

3700 3701
	ib_sa_register_client(&srp_sa_client);

3702 3703
	ret = ib_register_client(&srp_client);
	if (ret) {
3704
		pr_err("couldn't register IB client\n");
3705
		goto unreg_sa;
3706 3707
	}

3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720
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;
3721 3722 3723 3724 3725
}

static void __exit srp_cleanup_module(void)
{
	ib_unregister_client(&srp_client);
3726
	ib_sa_unregister_client(&srp_sa_client);
3727
	class_unregister(&srp_class);
3728
	srp_release_transport(ib_srp_transport_template);
3729
	destroy_workqueue(srp_remove_wq);
3730 3731 3732 3733
}

module_init(srp_init_module);
module_exit(srp_cleanup_module);