2 * Copyright (c) 2005 Cisco Systems. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/err.h>
39 #include <linux/string.h>
40 #include <linux/parser.h>
41 #include <linux/random.h>
42 #include <linux/jiffies.h>
44 #include <linux/atomic.h>
46 #include <scsi/scsi.h>
47 #include <scsi/scsi_device.h>
48 #include <scsi/scsi_dbg.h>
49 #include <scsi/scsi_tcq.h>
51 #include <scsi/scsi_transport_srp.h>
55 #define DRV_NAME "ib_srp"
56 #define PFX DRV_NAME ": "
57 #define DRV_VERSION "1.0"
58 #define DRV_RELDATE "July 1, 2013"
60 MODULE_AUTHOR("Roland Dreier");
61 MODULE_DESCRIPTION("InfiniBand SCSI RDMA Protocol initiator "
62 "v" DRV_VERSION " (" DRV_RELDATE ")");
63 MODULE_LICENSE("Dual BSD/GPL");
65 static unsigned int srp_sg_tablesize;
66 static unsigned int cmd_sg_entries;
67 static unsigned int indirect_sg_entries;
68 static bool allow_ext_sg;
69 static bool prefer_fr;
70 static bool register_always;
71 static int topspin_workarounds = 1;
73 module_param(srp_sg_tablesize, uint, 0444);
74 MODULE_PARM_DESC(srp_sg_tablesize, "Deprecated name for cmd_sg_entries");
76 module_param(cmd_sg_entries, uint, 0444);
77 MODULE_PARM_DESC(cmd_sg_entries,
78 "Default number of gather/scatter entries in the SRP command (default is 12, max 255)");
80 module_param(indirect_sg_entries, uint, 0444);
81 MODULE_PARM_DESC(indirect_sg_entries,
82 "Default max number of gather/scatter entries (default is 12, max is " __stringify(SCSI_MAX_SG_CHAIN_SEGMENTS) ")");
84 module_param(allow_ext_sg, bool, 0444);
85 MODULE_PARM_DESC(allow_ext_sg,
86 "Default behavior when there are more than cmd_sg_entries S/G entries after mapping; fails the request when false (default false)");
88 module_param(topspin_workarounds, int, 0444);
89 MODULE_PARM_DESC(topspin_workarounds,
90 "Enable workarounds for Topspin/Cisco SRP target bugs if != 0");
92 module_param(prefer_fr, bool, 0444);
93 MODULE_PARM_DESC(prefer_fr,
94 "Whether to use fast registration if both FMR and fast registration are supported");
96 module_param(register_always, bool, 0444);
97 MODULE_PARM_DESC(register_always,
98 "Use memory registration even for contiguous memory regions");
100 static struct kernel_param_ops srp_tmo_ops;
102 static int srp_reconnect_delay = 10;
103 module_param_cb(reconnect_delay, &srp_tmo_ops, &srp_reconnect_delay,
105 MODULE_PARM_DESC(reconnect_delay, "Time between successive reconnect attempts");
107 static int srp_fast_io_fail_tmo = 15;
108 module_param_cb(fast_io_fail_tmo, &srp_tmo_ops, &srp_fast_io_fail_tmo,
110 MODULE_PARM_DESC(fast_io_fail_tmo,
111 "Number of seconds between the observation of a transport"
112 " layer error and failing all I/O. \"off\" means that this"
113 " functionality is disabled.");
115 static int srp_dev_loss_tmo = 600;
116 module_param_cb(dev_loss_tmo, &srp_tmo_ops, &srp_dev_loss_tmo,
118 MODULE_PARM_DESC(dev_loss_tmo,
119 "Maximum number of seconds that the SRP transport should"
120 " insulate transport layer errors. After this time has been"
121 " exceeded the SCSI host is removed. Should be"
122 " between 1 and " __stringify(SCSI_DEVICE_BLOCK_MAX_TIMEOUT)
123 " if fast_io_fail_tmo has not been set. \"off\" means that"
124 " this functionality is disabled.");
126 static void srp_add_one(struct ib_device *device);
127 static void srp_remove_one(struct ib_device *device);
128 static void srp_recv_completion(struct ib_cq *cq, void *target_ptr);
129 static void srp_send_completion(struct ib_cq *cq, void *target_ptr);
130 static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event);
132 static struct scsi_transport_template *ib_srp_transport_template;
134 static struct ib_client srp_client = {
137 .remove = srp_remove_one
140 static struct ib_sa_client srp_sa_client;
142 static int srp_tmo_get(char *buffer, const struct kernel_param *kp)
144 int tmo = *(int *)kp->arg;
147 return sprintf(buffer, "%d", tmo);
149 return sprintf(buffer, "off");
152 static int srp_tmo_set(const char *val, const struct kernel_param *kp)
156 if (strncmp(val, "off", 3) != 0) {
157 res = kstrtoint(val, 0, &tmo);
163 if (kp->arg == &srp_reconnect_delay)
164 res = srp_tmo_valid(tmo, srp_fast_io_fail_tmo,
166 else if (kp->arg == &srp_fast_io_fail_tmo)
167 res = srp_tmo_valid(srp_reconnect_delay, tmo, srp_dev_loss_tmo);
169 res = srp_tmo_valid(srp_reconnect_delay, srp_fast_io_fail_tmo,
173 *(int *)kp->arg = tmo;
179 static struct kernel_param_ops srp_tmo_ops = {
184 static inline struct srp_target_port *host_to_target(struct Scsi_Host *host)
186 return (struct srp_target_port *) host->hostdata;
189 static const char *srp_target_info(struct Scsi_Host *host)
191 return host_to_target(host)->target_name;
194 static int srp_target_is_topspin(struct srp_target_port *target)
196 static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad };
197 static const u8 cisco_oui[3] = { 0x00, 0x1b, 0x0d };
199 return topspin_workarounds &&
200 (!memcmp(&target->ioc_guid, topspin_oui, sizeof topspin_oui) ||
201 !memcmp(&target->ioc_guid, cisco_oui, sizeof cisco_oui));
204 static struct srp_iu *srp_alloc_iu(struct srp_host *host, size_t size,
206 enum dma_data_direction direction)
210 iu = kmalloc(sizeof *iu, gfp_mask);
214 iu->buf = kzalloc(size, gfp_mask);
218 iu->dma = ib_dma_map_single(host->srp_dev->dev, iu->buf, size,
220 if (ib_dma_mapping_error(host->srp_dev->dev, iu->dma))
224 iu->direction = direction;
236 static void srp_free_iu(struct srp_host *host, struct srp_iu *iu)
241 ib_dma_unmap_single(host->srp_dev->dev, iu->dma, iu->size,
247 static void srp_qp_event(struct ib_event *event, void *context)
249 pr_debug("QP event %d\n", event->event);
252 static int srp_init_qp(struct srp_target_port *target,
255 struct ib_qp_attr *attr;
258 attr = kmalloc(sizeof *attr, GFP_KERNEL);
262 ret = ib_find_pkey(target->srp_host->srp_dev->dev,
263 target->srp_host->port,
264 be16_to_cpu(target->path.pkey),
269 attr->qp_state = IB_QPS_INIT;
270 attr->qp_access_flags = (IB_ACCESS_REMOTE_READ |
271 IB_ACCESS_REMOTE_WRITE);
272 attr->port_num = target->srp_host->port;
274 ret = ib_modify_qp(qp, attr,
285 static int srp_new_cm_id(struct srp_target_port *target)
287 struct ib_cm_id *new_cm_id;
289 new_cm_id = ib_create_cm_id(target->srp_host->srp_dev->dev,
290 srp_cm_handler, target);
291 if (IS_ERR(new_cm_id))
292 return PTR_ERR(new_cm_id);
295 ib_destroy_cm_id(target->cm_id);
296 target->cm_id = new_cm_id;
301 static struct ib_fmr_pool *srp_alloc_fmr_pool(struct srp_target_port *target)
303 struct srp_device *dev = target->srp_host->srp_dev;
304 struct ib_fmr_pool_param fmr_param;
306 memset(&fmr_param, 0, sizeof(fmr_param));
307 fmr_param.pool_size = target->scsi_host->can_queue;
308 fmr_param.dirty_watermark = fmr_param.pool_size / 4;
310 fmr_param.max_pages_per_fmr = dev->max_pages_per_mr;
311 fmr_param.page_shift = ilog2(dev->mr_page_size);
312 fmr_param.access = (IB_ACCESS_LOCAL_WRITE |
313 IB_ACCESS_REMOTE_WRITE |
314 IB_ACCESS_REMOTE_READ);
316 return ib_create_fmr_pool(dev->pd, &fmr_param);
320 * srp_destroy_fr_pool() - free the resources owned by a pool
321 * @pool: Fast registration pool to be destroyed.
323 static void srp_destroy_fr_pool(struct srp_fr_pool *pool)
326 struct srp_fr_desc *d;
331 for (i = 0, d = &pool->desc[0]; i < pool->size; i++, d++) {
333 ib_free_fast_reg_page_list(d->frpl);
341 * srp_create_fr_pool() - allocate and initialize a pool for fast registration
342 * @device: IB device to allocate fast registration descriptors for.
343 * @pd: Protection domain associated with the FR descriptors.
344 * @pool_size: Number of descriptors to allocate.
345 * @max_page_list_len: Maximum fast registration work request page list length.
347 static struct srp_fr_pool *srp_create_fr_pool(struct ib_device *device,
348 struct ib_pd *pd, int pool_size,
349 int max_page_list_len)
351 struct srp_fr_pool *pool;
352 struct srp_fr_desc *d;
354 struct ib_fast_reg_page_list *frpl;
355 int i, ret = -EINVAL;
360 pool = kzalloc(sizeof(struct srp_fr_pool) +
361 pool_size * sizeof(struct srp_fr_desc), GFP_KERNEL);
364 pool->size = pool_size;
365 pool->max_page_list_len = max_page_list_len;
366 spin_lock_init(&pool->lock);
367 INIT_LIST_HEAD(&pool->free_list);
369 for (i = 0, d = &pool->desc[0]; i < pool->size; i++, d++) {
370 mr = ib_alloc_fast_reg_mr(pd, max_page_list_len);
376 frpl = ib_alloc_fast_reg_page_list(device, max_page_list_len);
382 list_add_tail(&d->entry, &pool->free_list);
389 srp_destroy_fr_pool(pool);
397 * srp_fr_pool_get() - obtain a descriptor suitable for fast registration
398 * @pool: Pool to obtain descriptor from.
400 static struct srp_fr_desc *srp_fr_pool_get(struct srp_fr_pool *pool)
402 struct srp_fr_desc *d = NULL;
405 spin_lock_irqsave(&pool->lock, flags);
406 if (!list_empty(&pool->free_list)) {
407 d = list_first_entry(&pool->free_list, typeof(*d), entry);
410 spin_unlock_irqrestore(&pool->lock, flags);
416 * srp_fr_pool_put() - put an FR descriptor back in the free list
417 * @pool: Pool the descriptor was allocated from.
418 * @desc: Pointer to an array of fast registration descriptor pointers.
419 * @n: Number of descriptors to put back.
421 * Note: The caller must already have queued an invalidation request for
422 * desc->mr->rkey before calling this function.
424 static void srp_fr_pool_put(struct srp_fr_pool *pool, struct srp_fr_desc **desc,
430 spin_lock_irqsave(&pool->lock, flags);
431 for (i = 0; i < n; i++)
432 list_add(&desc[i]->entry, &pool->free_list);
433 spin_unlock_irqrestore(&pool->lock, flags);
436 static struct srp_fr_pool *srp_alloc_fr_pool(struct srp_target_port *target)
438 struct srp_device *dev = target->srp_host->srp_dev;
440 return srp_create_fr_pool(dev->dev, dev->pd,
441 target->scsi_host->can_queue,
442 dev->max_pages_per_mr);
445 static int srp_create_target_ib(struct srp_target_port *target)
447 struct srp_device *dev = target->srp_host->srp_dev;
448 struct ib_qp_init_attr *init_attr;
449 struct ib_cq *recv_cq, *send_cq;
451 struct ib_fmr_pool *fmr_pool = NULL;
452 struct srp_fr_pool *fr_pool = NULL;
453 const int m = 1 + dev->use_fast_reg;
456 init_attr = kzalloc(sizeof *init_attr, GFP_KERNEL);
460 recv_cq = ib_create_cq(dev->dev, srp_recv_completion, NULL, target,
461 target->queue_size, target->comp_vector);
462 if (IS_ERR(recv_cq)) {
463 ret = PTR_ERR(recv_cq);
467 send_cq = ib_create_cq(dev->dev, srp_send_completion, NULL, target,
468 m * target->queue_size, target->comp_vector);
469 if (IS_ERR(send_cq)) {
470 ret = PTR_ERR(send_cq);
474 ib_req_notify_cq(recv_cq, IB_CQ_NEXT_COMP);
476 init_attr->event_handler = srp_qp_event;
477 init_attr->cap.max_send_wr = m * target->queue_size;
478 init_attr->cap.max_recv_wr = target->queue_size;
479 init_attr->cap.max_recv_sge = 1;
480 init_attr->cap.max_send_sge = 1;
481 init_attr->sq_sig_type = IB_SIGNAL_REQ_WR;
482 init_attr->qp_type = IB_QPT_RC;
483 init_attr->send_cq = send_cq;
484 init_attr->recv_cq = recv_cq;
486 qp = ib_create_qp(dev->pd, init_attr);
492 ret = srp_init_qp(target, qp);
496 if (dev->use_fast_reg && dev->has_fr) {
497 fr_pool = srp_alloc_fr_pool(target);
498 if (IS_ERR(fr_pool)) {
499 ret = PTR_ERR(fr_pool);
500 shost_printk(KERN_WARNING, target->scsi_host, PFX
501 "FR pool allocation failed (%d)\n", ret);
505 srp_destroy_fr_pool(target->fr_pool);
506 target->fr_pool = fr_pool;
507 } else if (!dev->use_fast_reg && dev->has_fmr) {
508 fmr_pool = srp_alloc_fmr_pool(target);
509 if (IS_ERR(fmr_pool)) {
510 ret = PTR_ERR(fmr_pool);
511 shost_printk(KERN_WARNING, target->scsi_host, PFX
512 "FMR pool allocation failed (%d)\n", ret);
515 if (target->fmr_pool)
516 ib_destroy_fmr_pool(target->fmr_pool);
517 target->fmr_pool = fmr_pool;
521 ib_destroy_qp(target->qp);
523 ib_destroy_cq(target->recv_cq);
525 ib_destroy_cq(target->send_cq);
528 target->recv_cq = recv_cq;
529 target->send_cq = send_cq;
538 ib_destroy_cq(send_cq);
541 ib_destroy_cq(recv_cq);
549 * Note: this function may be called without srp_alloc_iu_bufs() having been
550 * invoked. Hence the target->[rt]x_ring checks.
552 static void srp_free_target_ib(struct srp_target_port *target)
554 struct srp_device *dev = target->srp_host->srp_dev;
557 if (dev->use_fast_reg) {
559 srp_destroy_fr_pool(target->fr_pool);
561 if (target->fmr_pool)
562 ib_destroy_fmr_pool(target->fmr_pool);
564 ib_destroy_qp(target->qp);
565 ib_destroy_cq(target->send_cq);
566 ib_destroy_cq(target->recv_cq);
569 target->send_cq = target->recv_cq = NULL;
571 if (target->rx_ring) {
572 for (i = 0; i < target->queue_size; ++i)
573 srp_free_iu(target->srp_host, target->rx_ring[i]);
574 kfree(target->rx_ring);
575 target->rx_ring = NULL;
577 if (target->tx_ring) {
578 for (i = 0; i < target->queue_size; ++i)
579 srp_free_iu(target->srp_host, target->tx_ring[i]);
580 kfree(target->tx_ring);
581 target->tx_ring = NULL;
585 static void srp_path_rec_completion(int status,
586 struct ib_sa_path_rec *pathrec,
589 struct srp_target_port *target = target_ptr;
591 target->status = status;
593 shost_printk(KERN_ERR, target->scsi_host,
594 PFX "Got failed path rec status %d\n", status);
596 target->path = *pathrec;
597 complete(&target->done);
600 static int srp_lookup_path(struct srp_target_port *target)
604 target->path.numb_path = 1;
606 init_completion(&target->done);
608 target->path_query_id = ib_sa_path_rec_get(&srp_sa_client,
609 target->srp_host->srp_dev->dev,
610 target->srp_host->port,
612 IB_SA_PATH_REC_SERVICE_ID |
613 IB_SA_PATH_REC_DGID |
614 IB_SA_PATH_REC_SGID |
615 IB_SA_PATH_REC_NUMB_PATH |
617 SRP_PATH_REC_TIMEOUT_MS,
619 srp_path_rec_completion,
620 target, &target->path_query);
621 if (target->path_query_id < 0)
622 return target->path_query_id;
624 ret = wait_for_completion_interruptible(&target->done);
628 if (target->status < 0)
629 shost_printk(KERN_WARNING, target->scsi_host,
630 PFX "Path record query failed\n");
632 return target->status;
635 static int srp_send_req(struct srp_target_port *target)
638 struct ib_cm_req_param param;
639 struct srp_login_req priv;
643 req = kzalloc(sizeof *req, GFP_KERNEL);
647 req->param.primary_path = &target->path;
648 req->param.alternate_path = NULL;
649 req->param.service_id = target->service_id;
650 req->param.qp_num = target->qp->qp_num;
651 req->param.qp_type = target->qp->qp_type;
652 req->param.private_data = &req->priv;
653 req->param.private_data_len = sizeof req->priv;
654 req->param.flow_control = 1;
656 get_random_bytes(&req->param.starting_psn, 4);
657 req->param.starting_psn &= 0xffffff;
660 * Pick some arbitrary defaults here; we could make these
661 * module parameters if anyone cared about setting them.
663 req->param.responder_resources = 4;
664 req->param.remote_cm_response_timeout = 20;
665 req->param.local_cm_response_timeout = 20;
666 req->param.retry_count = target->tl_retry_count;
667 req->param.rnr_retry_count = 7;
668 req->param.max_cm_retries = 15;
670 req->priv.opcode = SRP_LOGIN_REQ;
672 req->priv.req_it_iu_len = cpu_to_be32(target->max_iu_len);
673 req->priv.req_buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
674 SRP_BUF_FORMAT_INDIRECT);
676 * In the published SRP specification (draft rev. 16a), the
677 * port identifier format is 8 bytes of ID extension followed
678 * by 8 bytes of GUID. Older drafts put the two halves in the
679 * opposite order, so that the GUID comes first.
681 * Targets conforming to these obsolete drafts can be
682 * recognized by the I/O Class they report.
684 if (target->io_class == SRP_REV10_IB_IO_CLASS) {
685 memcpy(req->priv.initiator_port_id,
686 &target->path.sgid.global.interface_id, 8);
687 memcpy(req->priv.initiator_port_id + 8,
688 &target->initiator_ext, 8);
689 memcpy(req->priv.target_port_id, &target->ioc_guid, 8);
690 memcpy(req->priv.target_port_id + 8, &target->id_ext, 8);
692 memcpy(req->priv.initiator_port_id,
693 &target->initiator_ext, 8);
694 memcpy(req->priv.initiator_port_id + 8,
695 &target->path.sgid.global.interface_id, 8);
696 memcpy(req->priv.target_port_id, &target->id_ext, 8);
697 memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8);
701 * Topspin/Cisco SRP targets will reject our login unless we
702 * zero out the first 8 bytes of our initiator port ID and set
703 * the second 8 bytes to the local node GUID.
705 if (srp_target_is_topspin(target)) {
706 shost_printk(KERN_DEBUG, target->scsi_host,
707 PFX "Topspin/Cisco initiator port ID workaround "
708 "activated for target GUID %016llx\n",
709 (unsigned long long) be64_to_cpu(target->ioc_guid));
710 memset(req->priv.initiator_port_id, 0, 8);
711 memcpy(req->priv.initiator_port_id + 8,
712 &target->srp_host->srp_dev->dev->node_guid, 8);
715 status = ib_send_cm_req(target->cm_id, &req->param);
722 static bool srp_queue_remove_work(struct srp_target_port *target)
724 bool changed = false;
726 spin_lock_irq(&target->lock);
727 if (target->state != SRP_TARGET_REMOVED) {
728 target->state = SRP_TARGET_REMOVED;
731 spin_unlock_irq(&target->lock);
734 queue_work(system_long_wq, &target->remove_work);
739 static bool srp_change_conn_state(struct srp_target_port *target,
742 bool changed = false;
744 spin_lock_irq(&target->lock);
745 if (target->connected != connected) {
746 target->connected = connected;
749 spin_unlock_irq(&target->lock);
754 static void srp_disconnect_target(struct srp_target_port *target)
756 if (srp_change_conn_state(target, false)) {
757 /* XXX should send SRP_I_LOGOUT request */
759 if (ib_send_cm_dreq(target->cm_id, NULL, 0)) {
760 shost_printk(KERN_DEBUG, target->scsi_host,
761 PFX "Sending CM DREQ failed\n");
766 static void srp_free_req_data(struct srp_target_port *target)
768 struct srp_device *dev = target->srp_host->srp_dev;
769 struct ib_device *ibdev = dev->dev;
770 struct srp_request *req;
773 if (!target->req_ring)
776 for (i = 0; i < target->req_ring_size; ++i) {
777 req = &target->req_ring[i];
778 if (dev->use_fast_reg)
781 kfree(req->fmr_list);
782 kfree(req->map_page);
783 if (req->indirect_dma_addr) {
784 ib_dma_unmap_single(ibdev, req->indirect_dma_addr,
785 target->indirect_size,
788 kfree(req->indirect_desc);
791 kfree(target->req_ring);
792 target->req_ring = NULL;
795 static int srp_alloc_req_data(struct srp_target_port *target)
797 struct srp_device *srp_dev = target->srp_host->srp_dev;
798 struct ib_device *ibdev = srp_dev->dev;
799 struct srp_request *req;
802 int i, ret = -ENOMEM;
804 INIT_LIST_HEAD(&target->free_reqs);
806 target->req_ring = kzalloc(target->req_ring_size *
807 sizeof(*target->req_ring), GFP_KERNEL);
808 if (!target->req_ring)
811 for (i = 0; i < target->req_ring_size; ++i) {
812 req = &target->req_ring[i];
813 mr_list = kmalloc(target->cmd_sg_cnt * sizeof(void *),
817 if (srp_dev->use_fast_reg)
818 req->fr_list = mr_list;
820 req->fmr_list = mr_list;
821 req->map_page = kmalloc(srp_dev->max_pages_per_mr *
822 sizeof(void *), GFP_KERNEL);
825 req->indirect_desc = kmalloc(target->indirect_size, GFP_KERNEL);
826 if (!req->indirect_desc)
829 dma_addr = ib_dma_map_single(ibdev, req->indirect_desc,
830 target->indirect_size,
832 if (ib_dma_mapping_error(ibdev, dma_addr))
835 req->indirect_dma_addr = dma_addr;
837 list_add_tail(&req->list, &target->free_reqs);
846 * srp_del_scsi_host_attr() - Remove attributes defined in the host template.
847 * @shost: SCSI host whose attributes to remove from sysfs.
849 * Note: Any attributes defined in the host template and that did not exist
850 * before invocation of this function will be ignored.
852 static void srp_del_scsi_host_attr(struct Scsi_Host *shost)
854 struct device_attribute **attr;
856 for (attr = shost->hostt->shost_attrs; attr && *attr; ++attr)
857 device_remove_file(&shost->shost_dev, *attr);
860 static void srp_remove_target(struct srp_target_port *target)
862 WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);
864 srp_del_scsi_host_attr(target->scsi_host);
865 srp_rport_get(target->rport);
866 srp_remove_host(target->scsi_host);
867 scsi_remove_host(target->scsi_host);
868 srp_stop_rport_timers(target->rport);
869 srp_disconnect_target(target);
870 ib_destroy_cm_id(target->cm_id);
871 srp_free_target_ib(target);
872 cancel_work_sync(&target->tl_err_work);
873 srp_rport_put(target->rport);
874 srp_free_req_data(target);
876 spin_lock(&target->srp_host->target_lock);
877 list_del(&target->list);
878 spin_unlock(&target->srp_host->target_lock);
880 scsi_host_put(target->scsi_host);
883 static void srp_remove_work(struct work_struct *work)
885 struct srp_target_port *target =
886 container_of(work, struct srp_target_port, remove_work);
888 WARN_ON_ONCE(target->state != SRP_TARGET_REMOVED);
890 srp_remove_target(target);
893 static void srp_rport_delete(struct srp_rport *rport)
895 struct srp_target_port *target = rport->lld_data;
897 srp_queue_remove_work(target);
900 static int srp_connect_target(struct srp_target_port *target)
905 WARN_ON_ONCE(target->connected);
907 target->qp_in_error = false;
909 ret = srp_lookup_path(target);
914 init_completion(&target->done);
915 ret = srp_send_req(target);
918 ret = wait_for_completion_interruptible(&target->done);
923 * The CM event handling code will set status to
924 * SRP_PORT_REDIRECT if we get a port redirect REJ
925 * back, or SRP_DLID_REDIRECT if we get a lid/qp
928 switch (target->status) {
930 srp_change_conn_state(target, true);
933 case SRP_PORT_REDIRECT:
934 ret = srp_lookup_path(target);
939 case SRP_DLID_REDIRECT:
943 /* Our current CM id was stale, and is now in timewait.
944 * Try to reconnect with a new one.
946 if (!retries-- || srp_new_cm_id(target)) {
947 shost_printk(KERN_ERR, target->scsi_host, PFX
948 "giving up on stale connection\n");
949 target->status = -ECONNRESET;
950 return target->status;
953 shost_printk(KERN_ERR, target->scsi_host, PFX
954 "retrying stale connection\n");
958 return target->status;
963 static int srp_inv_rkey(struct srp_target_port *target, u32 rkey)
965 struct ib_send_wr *bad_wr;
966 struct ib_send_wr wr = {
967 .opcode = IB_WR_LOCAL_INV,
968 .wr_id = LOCAL_INV_WR_ID_MASK,
972 .ex.invalidate_rkey = rkey,
975 return ib_post_send(target->qp, &wr, &bad_wr);
978 static void srp_unmap_data(struct scsi_cmnd *scmnd,
979 struct srp_target_port *target,
980 struct srp_request *req)
982 struct srp_device *dev = target->srp_host->srp_dev;
983 struct ib_device *ibdev = dev->dev;
986 if (!scsi_sglist(scmnd) ||
987 (scmnd->sc_data_direction != DMA_TO_DEVICE &&
988 scmnd->sc_data_direction != DMA_FROM_DEVICE))
991 if (dev->use_fast_reg) {
992 struct srp_fr_desc **pfr;
994 for (i = req->nmdesc, pfr = req->fr_list; i > 0; i--, pfr++) {
995 res = srp_inv_rkey(target, (*pfr)->mr->rkey);
997 shost_printk(KERN_ERR, target->scsi_host, PFX
998 "Queueing INV WR for rkey %#x failed (%d)\n",
999 (*pfr)->mr->rkey, res);
1000 queue_work(system_long_wq,
1001 &target->tl_err_work);
1005 srp_fr_pool_put(target->fr_pool, req->fr_list,
1008 struct ib_pool_fmr **pfmr;
1010 for (i = req->nmdesc, pfmr = req->fmr_list; i > 0; i--, pfmr++)
1011 ib_fmr_pool_unmap(*pfmr);
1014 ib_dma_unmap_sg(ibdev, scsi_sglist(scmnd), scsi_sg_count(scmnd),
1015 scmnd->sc_data_direction);
1019 * srp_claim_req - Take ownership of the scmnd associated with a request.
1020 * @target: SRP target port.
1021 * @req: SRP request.
1022 * @sdev: If not NULL, only take ownership for this SCSI device.
1023 * @scmnd: If NULL, take ownership of @req->scmnd. If not NULL, only take
1024 * ownership of @req->scmnd if it equals @scmnd.
1027 * Either NULL or a pointer to the SCSI command the caller became owner of.
1029 static struct scsi_cmnd *srp_claim_req(struct srp_target_port *target,
1030 struct srp_request *req,
1031 struct scsi_device *sdev,
1032 struct scsi_cmnd *scmnd)
1034 unsigned long flags;
1036 spin_lock_irqsave(&target->lock, flags);
1038 (!sdev || req->scmnd->device == sdev) &&
1039 (!scmnd || req->scmnd == scmnd)) {
1045 spin_unlock_irqrestore(&target->lock, flags);
1051 * srp_free_req() - Unmap data and add request to the free request list.
1052 * @target: SRP target port.
1053 * @req: Request to be freed.
1054 * @scmnd: SCSI command associated with @req.
1055 * @req_lim_delta: Amount to be added to @target->req_lim.
1057 static void srp_free_req(struct srp_target_port *target,
1058 struct srp_request *req, struct scsi_cmnd *scmnd,
1061 unsigned long flags;
1063 srp_unmap_data(scmnd, target, req);
1065 spin_lock_irqsave(&target->lock, flags);
1066 target->req_lim += req_lim_delta;
1067 list_add_tail(&req->list, &target->free_reqs);
1068 spin_unlock_irqrestore(&target->lock, flags);
1071 static void srp_finish_req(struct srp_target_port *target,
1072 struct srp_request *req, struct scsi_device *sdev,
1075 struct scsi_cmnd *scmnd = srp_claim_req(target, req, sdev, NULL);
1078 srp_free_req(target, req, scmnd, 0);
1079 scmnd->result = result;
1080 scmnd->scsi_done(scmnd);
1084 static void srp_terminate_io(struct srp_rport *rport)
1086 struct srp_target_port *target = rport->lld_data;
1087 struct Scsi_Host *shost = target->scsi_host;
1088 struct scsi_device *sdev;
1092 * Invoking srp_terminate_io() while srp_queuecommand() is running
1093 * is not safe. Hence the warning statement below.
1095 shost_for_each_device(sdev, shost)
1096 WARN_ON_ONCE(sdev->request_queue->request_fn_active);
1098 for (i = 0; i < target->req_ring_size; ++i) {
1099 struct srp_request *req = &target->req_ring[i];
1100 srp_finish_req(target, req, NULL, DID_TRANSPORT_FAILFAST << 16);
1105 * It is up to the caller to ensure that srp_rport_reconnect() calls are
1106 * serialized and that no concurrent srp_queuecommand(), srp_abort(),
1107 * srp_reset_device() or srp_reset_host() calls will occur while this function
1108 * is in progress. One way to realize that is not to call this function
1109 * directly but to call srp_reconnect_rport() instead since that last function
1110 * serializes calls of this function via rport->mutex and also blocks
1111 * srp_queuecommand() calls before invoking this function.
1113 static int srp_rport_reconnect(struct srp_rport *rport)
1115 struct srp_target_port *target = rport->lld_data;
1118 srp_disconnect_target(target);
1120 * Now get a new local CM ID so that we avoid confusing the target in
1121 * case things are really fouled up. Doing so also ensures that all CM
1122 * callbacks will have finished before a new QP is allocated.
1124 ret = srp_new_cm_id(target);
1126 for (i = 0; i < target->req_ring_size; ++i) {
1127 struct srp_request *req = &target->req_ring[i];
1128 srp_finish_req(target, req, NULL, DID_RESET << 16);
1132 * Whether or not creating a new CM ID succeeded, create a new
1133 * QP. This guarantees that all callback functions for the old QP have
1134 * finished before any send requests are posted on the new QP.
1136 ret += srp_create_target_ib(target);
1138 INIT_LIST_HEAD(&target->free_tx);
1139 for (i = 0; i < target->queue_size; ++i)
1140 list_add(&target->tx_ring[i]->list, &target->free_tx);
1143 ret = srp_connect_target(target);
1146 shost_printk(KERN_INFO, target->scsi_host,
1147 PFX "reconnect succeeded\n");
1152 static void srp_map_desc(struct srp_map_state *state, dma_addr_t dma_addr,
1153 unsigned int dma_len, u32 rkey)
1155 struct srp_direct_buf *desc = state->desc;
1157 desc->va = cpu_to_be64(dma_addr);
1158 desc->key = cpu_to_be32(rkey);
1159 desc->len = cpu_to_be32(dma_len);
1161 state->total_len += dma_len;
1166 static int srp_map_finish_fmr(struct srp_map_state *state,
1167 struct srp_target_port *target)
1169 struct ib_pool_fmr *fmr;
1172 fmr = ib_fmr_pool_map_phys(target->fmr_pool, state->pages,
1173 state->npages, io_addr);
1175 return PTR_ERR(fmr);
1177 *state->next_fmr++ = fmr;
1180 srp_map_desc(state, 0, state->dma_len, fmr->fmr->rkey);
1185 static int srp_map_finish_fr(struct srp_map_state *state,
1186 struct srp_target_port *target)
1188 struct srp_device *dev = target->srp_host->srp_dev;
1189 struct ib_send_wr *bad_wr;
1190 struct ib_send_wr wr;
1191 struct srp_fr_desc *desc;
1194 desc = srp_fr_pool_get(target->fr_pool);
1198 rkey = ib_inc_rkey(desc->mr->rkey);
1199 ib_update_fast_reg_key(desc->mr, rkey);
1201 memcpy(desc->frpl->page_list, state->pages,
1202 sizeof(state->pages[0]) * state->npages);
1204 memset(&wr, 0, sizeof(wr));
1205 wr.opcode = IB_WR_FAST_REG_MR;
1206 wr.wr_id = FAST_REG_WR_ID_MASK;
1207 wr.wr.fast_reg.iova_start = state->base_dma_addr;
1208 wr.wr.fast_reg.page_list = desc->frpl;
1209 wr.wr.fast_reg.page_list_len = state->npages;
1210 wr.wr.fast_reg.page_shift = ilog2(dev->mr_page_size);
1211 wr.wr.fast_reg.length = state->dma_len;
1212 wr.wr.fast_reg.access_flags = (IB_ACCESS_LOCAL_WRITE |
1213 IB_ACCESS_REMOTE_READ |
1214 IB_ACCESS_REMOTE_WRITE);
1215 wr.wr.fast_reg.rkey = desc->mr->lkey;
1217 *state->next_fr++ = desc;
1220 srp_map_desc(state, state->base_dma_addr, state->dma_len,
1223 return ib_post_send(target->qp, &wr, &bad_wr);
1226 static int srp_finish_mapping(struct srp_map_state *state,
1227 struct srp_target_port *target)
1231 if (state->npages == 0)
1234 if (state->npages == 1 && !register_always)
1235 srp_map_desc(state, state->base_dma_addr, state->dma_len,
1238 ret = target->srp_host->srp_dev->use_fast_reg ?
1239 srp_map_finish_fr(state, target) :
1240 srp_map_finish_fmr(state, target);
1250 static void srp_map_update_start(struct srp_map_state *state,
1251 struct scatterlist *sg, int sg_index,
1252 dma_addr_t dma_addr)
1254 state->unmapped_sg = sg;
1255 state->unmapped_index = sg_index;
1256 state->unmapped_addr = dma_addr;
1259 static int srp_map_sg_entry(struct srp_map_state *state,
1260 struct srp_target_port *target,
1261 struct scatterlist *sg, int sg_index,
1264 struct srp_device *dev = target->srp_host->srp_dev;
1265 struct ib_device *ibdev = dev->dev;
1266 dma_addr_t dma_addr = ib_sg_dma_address(ibdev, sg);
1267 unsigned int dma_len = ib_sg_dma_len(ibdev, sg);
1276 * Once we're in direct map mode for a request, we don't
1277 * go back to FMR or FR mode, so no need to update anything
1278 * other than the descriptor.
1280 srp_map_desc(state, dma_addr, dma_len, target->rkey);
1285 * Since not all RDMA HW drivers support non-zero page offsets for
1286 * FMR, if we start at an offset into a page, don't merge into the
1287 * current FMR mapping. Finish it out, and use the kernel's MR for
1290 if ((!dev->use_fast_reg && dma_addr & ~dev->mr_page_mask) ||
1291 dma_len > dev->mr_max_size) {
1292 ret = srp_finish_mapping(state, target);
1296 srp_map_desc(state, dma_addr, dma_len, target->rkey);
1297 srp_map_update_start(state, NULL, 0, 0);
1302 * If this is the first sg that will be mapped via FMR or via FR, save
1303 * our position. We need to know the first unmapped entry, its index,
1304 * and the first unmapped address within that entry to be able to
1305 * restart mapping after an error.
1307 if (!state->unmapped_sg)
1308 srp_map_update_start(state, sg, sg_index, dma_addr);
1311 unsigned offset = dma_addr & ~dev->mr_page_mask;
1312 if (state->npages == dev->max_pages_per_mr || offset != 0) {
1313 ret = srp_finish_mapping(state, target);
1317 srp_map_update_start(state, sg, sg_index, dma_addr);
1320 len = min_t(unsigned int, dma_len, dev->mr_page_size - offset);
1323 state->base_dma_addr = dma_addr;
1324 state->pages[state->npages++] = dma_addr & dev->mr_page_mask;
1325 state->dma_len += len;
1331 * If the last entry of the MR wasn't a full page, then we need to
1332 * close it out and start a new one -- we can only merge at page
1336 if (len != dev->mr_page_size) {
1337 ret = srp_finish_mapping(state, target);
1339 srp_map_update_start(state, NULL, 0, 0);
1344 static int srp_map_sg(struct srp_map_state *state,
1345 struct srp_target_port *target, struct srp_request *req,
1346 struct scatterlist *scat, int count)
1348 struct srp_device *dev = target->srp_host->srp_dev;
1349 struct ib_device *ibdev = dev->dev;
1350 struct scatterlist *sg;
1354 state->desc = req->indirect_desc;
1355 state->pages = req->map_page;
1356 if (dev->use_fast_reg) {
1357 state->next_fr = req->fr_list;
1358 use_mr = !!target->fr_pool;
1360 state->next_fmr = req->fmr_list;
1361 use_mr = !!target->fmr_pool;
1364 for_each_sg(scat, sg, count, i) {
1365 if (srp_map_sg_entry(state, target, sg, i, use_mr)) {
1367 * Memory registration failed, so backtrack to the
1368 * first unmapped entry and continue on without using
1369 * memory registration.
1371 dma_addr_t dma_addr;
1372 unsigned int dma_len;
1375 sg = state->unmapped_sg;
1376 i = state->unmapped_index;
1378 dma_addr = ib_sg_dma_address(ibdev, sg);
1379 dma_len = ib_sg_dma_len(ibdev, sg);
1380 dma_len -= (state->unmapped_addr - dma_addr);
1381 dma_addr = state->unmapped_addr;
1383 srp_map_desc(state, dma_addr, dma_len, target->rkey);
1387 if (use_mr && srp_finish_mapping(state, target))
1390 req->nmdesc = state->nmdesc;
1395 static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target,
1396 struct srp_request *req)
1398 struct scatterlist *scat;
1399 struct srp_cmd *cmd = req->cmd->buf;
1400 int len, nents, count;
1401 struct srp_device *dev;
1402 struct ib_device *ibdev;
1403 struct srp_map_state state;
1404 struct srp_indirect_buf *indirect_hdr;
1408 if (!scsi_sglist(scmnd) || scmnd->sc_data_direction == DMA_NONE)
1409 return sizeof (struct srp_cmd);
1411 if (scmnd->sc_data_direction != DMA_FROM_DEVICE &&
1412 scmnd->sc_data_direction != DMA_TO_DEVICE) {
1413 shost_printk(KERN_WARNING, target->scsi_host,
1414 PFX "Unhandled data direction %d\n",
1415 scmnd->sc_data_direction);
1419 nents = scsi_sg_count(scmnd);
1420 scat = scsi_sglist(scmnd);
1422 dev = target->srp_host->srp_dev;
1425 count = ib_dma_map_sg(ibdev, scat, nents, scmnd->sc_data_direction);
1426 if (unlikely(count == 0))
1429 fmt = SRP_DATA_DESC_DIRECT;
1430 len = sizeof (struct srp_cmd) + sizeof (struct srp_direct_buf);
1432 if (count == 1 && !register_always) {
1434 * The midlayer only generated a single gather/scatter
1435 * entry, or DMA mapping coalesced everything to a
1436 * single entry. So a direct descriptor along with
1437 * the DMA MR suffices.
1439 struct srp_direct_buf *buf = (void *) cmd->add_data;
1441 buf->va = cpu_to_be64(ib_sg_dma_address(ibdev, scat));
1442 buf->key = cpu_to_be32(target->rkey);
1443 buf->len = cpu_to_be32(ib_sg_dma_len(ibdev, scat));
1450 * We have more than one scatter/gather entry, so build our indirect
1451 * descriptor table, trying to merge as many entries as we can.
1453 indirect_hdr = (void *) cmd->add_data;
1455 ib_dma_sync_single_for_cpu(ibdev, req->indirect_dma_addr,
1456 target->indirect_size, DMA_TO_DEVICE);
1458 memset(&state, 0, sizeof(state));
1459 srp_map_sg(&state, target, req, scat, count);
1461 /* We've mapped the request, now pull as much of the indirect
1462 * descriptor table as we can into the command buffer. If this
1463 * target is not using an external indirect table, we are
1464 * guaranteed to fit into the command, as the SCSI layer won't
1465 * give us more S/G entries than we allow.
1467 if (state.ndesc == 1) {
1469 * Memory registration collapsed the sg-list into one entry,
1470 * so use a direct descriptor.
1472 struct srp_direct_buf *buf = (void *) cmd->add_data;
1474 *buf = req->indirect_desc[0];
1478 if (unlikely(target->cmd_sg_cnt < state.ndesc &&
1479 !target->allow_ext_sg)) {
1480 shost_printk(KERN_ERR, target->scsi_host,
1481 "Could not fit S/G list into SRP_CMD\n");
1485 count = min(state.ndesc, target->cmd_sg_cnt);
1486 table_len = state.ndesc * sizeof (struct srp_direct_buf);
1488 fmt = SRP_DATA_DESC_INDIRECT;
1489 len = sizeof(struct srp_cmd) + sizeof (struct srp_indirect_buf);
1490 len += count * sizeof (struct srp_direct_buf);
1492 memcpy(indirect_hdr->desc_list, req->indirect_desc,
1493 count * sizeof (struct srp_direct_buf));
1495 indirect_hdr->table_desc.va = cpu_to_be64(req->indirect_dma_addr);
1496 indirect_hdr->table_desc.key = cpu_to_be32(target->rkey);
1497 indirect_hdr->table_desc.len = cpu_to_be32(table_len);
1498 indirect_hdr->len = cpu_to_be32(state.total_len);
1500 if (scmnd->sc_data_direction == DMA_TO_DEVICE)
1501 cmd->data_out_desc_cnt = count;
1503 cmd->data_in_desc_cnt = count;
1505 ib_dma_sync_single_for_device(ibdev, req->indirect_dma_addr, table_len,
1509 if (scmnd->sc_data_direction == DMA_TO_DEVICE)
1510 cmd->buf_fmt = fmt << 4;
1518 * Return an IU and possible credit to the free pool
1520 static void srp_put_tx_iu(struct srp_target_port *target, struct srp_iu *iu,
1521 enum srp_iu_type iu_type)
1523 unsigned long flags;
1525 spin_lock_irqsave(&target->lock, flags);
1526 list_add(&iu->list, &target->free_tx);
1527 if (iu_type != SRP_IU_RSP)
1529 spin_unlock_irqrestore(&target->lock, flags);
1533 * Must be called with target->lock held to protect req_lim and free_tx.
1534 * If IU is not sent, it must be returned using srp_put_tx_iu().
1537 * An upper limit for the number of allocated information units for each
1539 * - SRP_IU_CMD: SRP_CMD_SQ_SIZE, since the SCSI mid-layer never queues
1540 * more than Scsi_Host.can_queue requests.
1541 * - SRP_IU_TSK_MGMT: SRP_TSK_MGMT_SQ_SIZE.
1542 * - SRP_IU_RSP: 1, since a conforming SRP target never sends more than
1543 * one unanswered SRP request to an initiator.
1545 static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target,
1546 enum srp_iu_type iu_type)
1548 s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE;
1551 srp_send_completion(target->send_cq, target);
1553 if (list_empty(&target->free_tx))
1556 /* Initiator responses to target requests do not consume credits */
1557 if (iu_type != SRP_IU_RSP) {
1558 if (target->req_lim <= rsv) {
1559 ++target->zero_req_lim;
1566 iu = list_first_entry(&target->free_tx, struct srp_iu, list);
1567 list_del(&iu->list);
1571 static int srp_post_send(struct srp_target_port *target,
1572 struct srp_iu *iu, int len)
1575 struct ib_send_wr wr, *bad_wr;
1577 list.addr = iu->dma;
1579 list.lkey = target->lkey;
1582 wr.wr_id = (uintptr_t) iu;
1585 wr.opcode = IB_WR_SEND;
1586 wr.send_flags = IB_SEND_SIGNALED;
1588 return ib_post_send(target->qp, &wr, &bad_wr);
1591 static int srp_post_recv(struct srp_target_port *target, struct srp_iu *iu)
1593 struct ib_recv_wr wr, *bad_wr;
1596 list.addr = iu->dma;
1597 list.length = iu->size;
1598 list.lkey = target->lkey;
1601 wr.wr_id = (uintptr_t) iu;
1605 return ib_post_recv(target->qp, &wr, &bad_wr);
1608 static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
1610 struct srp_request *req;
1611 struct scsi_cmnd *scmnd;
1612 unsigned long flags;
1614 if (unlikely(rsp->tag & SRP_TAG_TSK_MGMT)) {
1615 spin_lock_irqsave(&target->lock, flags);
1616 target->req_lim += be32_to_cpu(rsp->req_lim_delta);
1617 spin_unlock_irqrestore(&target->lock, flags);
1619 target->tsk_mgmt_status = -1;
1620 if (be32_to_cpu(rsp->resp_data_len) >= 4)
1621 target->tsk_mgmt_status = rsp->data[3];
1622 complete(&target->tsk_mgmt_done);
1624 req = &target->req_ring[rsp->tag];
1625 scmnd = srp_claim_req(target, req, NULL, NULL);
1627 shost_printk(KERN_ERR, target->scsi_host,
1628 "Null scmnd for RSP w/tag %016llx\n",
1629 (unsigned long long) rsp->tag);
1631 spin_lock_irqsave(&target->lock, flags);
1632 target->req_lim += be32_to_cpu(rsp->req_lim_delta);
1633 spin_unlock_irqrestore(&target->lock, flags);
1637 scmnd->result = rsp->status;
1639 if (rsp->flags & SRP_RSP_FLAG_SNSVALID) {
1640 memcpy(scmnd->sense_buffer, rsp->data +
1641 be32_to_cpu(rsp->resp_data_len),
1642 min_t(int, be32_to_cpu(rsp->sense_data_len),
1643 SCSI_SENSE_BUFFERSIZE));
1646 if (rsp->flags & (SRP_RSP_FLAG_DOOVER | SRP_RSP_FLAG_DOUNDER))
1647 scsi_set_resid(scmnd, be32_to_cpu(rsp->data_out_res_cnt));
1648 else if (rsp->flags & (SRP_RSP_FLAG_DIOVER | SRP_RSP_FLAG_DIUNDER))
1649 scsi_set_resid(scmnd, be32_to_cpu(rsp->data_in_res_cnt));
1651 srp_free_req(target, req, scmnd,
1652 be32_to_cpu(rsp->req_lim_delta));
1654 scmnd->host_scribble = NULL;
1655 scmnd->scsi_done(scmnd);
1659 static int srp_response_common(struct srp_target_port *target, s32 req_delta,
1662 struct ib_device *dev = target->srp_host->srp_dev->dev;
1663 unsigned long flags;
1667 spin_lock_irqsave(&target->lock, flags);
1668 target->req_lim += req_delta;
1669 iu = __srp_get_tx_iu(target, SRP_IU_RSP);
1670 spin_unlock_irqrestore(&target->lock, flags);
1673 shost_printk(KERN_ERR, target->scsi_host, PFX
1674 "no IU available to send response\n");
1678 ib_dma_sync_single_for_cpu(dev, iu->dma, len, DMA_TO_DEVICE);
1679 memcpy(iu->buf, rsp, len);
1680 ib_dma_sync_single_for_device(dev, iu->dma, len, DMA_TO_DEVICE);
1682 err = srp_post_send(target, iu, len);
1684 shost_printk(KERN_ERR, target->scsi_host, PFX
1685 "unable to post response: %d\n", err);
1686 srp_put_tx_iu(target, iu, SRP_IU_RSP);
1692 static void srp_process_cred_req(struct srp_target_port *target,
1693 struct srp_cred_req *req)
1695 struct srp_cred_rsp rsp = {
1696 .opcode = SRP_CRED_RSP,
1699 s32 delta = be32_to_cpu(req->req_lim_delta);
1701 if (srp_response_common(target, delta, &rsp, sizeof rsp))
1702 shost_printk(KERN_ERR, target->scsi_host, PFX
1703 "problems processing SRP_CRED_REQ\n");
1706 static void srp_process_aer_req(struct srp_target_port *target,
1707 struct srp_aer_req *req)
1709 struct srp_aer_rsp rsp = {
1710 .opcode = SRP_AER_RSP,
1713 s32 delta = be32_to_cpu(req->req_lim_delta);
1715 shost_printk(KERN_ERR, target->scsi_host, PFX
1716 "ignoring AER for LUN %llu\n", be64_to_cpu(req->lun));
1718 if (srp_response_common(target, delta, &rsp, sizeof rsp))
1719 shost_printk(KERN_ERR, target->scsi_host, PFX
1720 "problems processing SRP_AER_REQ\n");
1723 static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc)
1725 struct ib_device *dev = target->srp_host->srp_dev->dev;
1726 struct srp_iu *iu = (struct srp_iu *) (uintptr_t) wc->wr_id;
1730 ib_dma_sync_single_for_cpu(dev, iu->dma, target->max_ti_iu_len,
1733 opcode = *(u8 *) iu->buf;
1736 shost_printk(KERN_ERR, target->scsi_host,
1737 PFX "recv completion, opcode 0x%02x\n", opcode);
1738 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 8, 1,
1739 iu->buf, wc->byte_len, true);
1744 srp_process_rsp(target, iu->buf);
1748 srp_process_cred_req(target, iu->buf);
1752 srp_process_aer_req(target, iu->buf);
1756 /* XXX Handle target logout */
1757 shost_printk(KERN_WARNING, target->scsi_host,
1758 PFX "Got target logout request\n");
1762 shost_printk(KERN_WARNING, target->scsi_host,
1763 PFX "Unhandled SRP opcode 0x%02x\n", opcode);
1767 ib_dma_sync_single_for_device(dev, iu->dma, target->max_ti_iu_len,
1770 res = srp_post_recv(target, iu);
1772 shost_printk(KERN_ERR, target->scsi_host,
1773 PFX "Recv failed with error code %d\n", res);
1777 * srp_tl_err_work() - handle a transport layer error
1778 * @work: Work structure embedded in an SRP target port.
1780 * Note: This function may get invoked before the rport has been created,
1781 * hence the target->rport test.
1783 static void srp_tl_err_work(struct work_struct *work)
1785 struct srp_target_port *target;
1787 target = container_of(work, struct srp_target_port, tl_err_work);
1789 srp_start_tl_fail_timers(target->rport);
1792 static void srp_handle_qp_err(u64 wr_id, enum ib_wc_status wc_status,
1793 bool send_err, struct srp_target_port *target)
1795 if (target->connected && !target->qp_in_error) {
1796 if (wr_id & LOCAL_INV_WR_ID_MASK) {
1797 shost_printk(KERN_ERR, target->scsi_host, PFX
1798 "LOCAL_INV failed with status %d\n",
1800 } else if (wr_id & FAST_REG_WR_ID_MASK) {
1801 shost_printk(KERN_ERR, target->scsi_host, PFX
1802 "FAST_REG_MR failed status %d\n",
1805 shost_printk(KERN_ERR, target->scsi_host,
1806 PFX "failed %s status %d for iu %p\n",
1807 send_err ? "send" : "receive",
1808 wc_status, (void *)(uintptr_t)wr_id);
1810 queue_work(system_long_wq, &target->tl_err_work);
1812 target->qp_in_error = true;
1815 static void srp_recv_completion(struct ib_cq *cq, void *target_ptr)
1817 struct srp_target_port *target = target_ptr;
1820 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
1821 while (ib_poll_cq(cq, 1, &wc) > 0) {
1822 if (likely(wc.status == IB_WC_SUCCESS)) {
1823 srp_handle_recv(target, &wc);
1825 srp_handle_qp_err(wc.wr_id, wc.status, false, target);
1830 static void srp_send_completion(struct ib_cq *cq, void *target_ptr)
1832 struct srp_target_port *target = target_ptr;
1836 while (ib_poll_cq(cq, 1, &wc) > 0) {
1837 if (likely(wc.status == IB_WC_SUCCESS)) {
1838 iu = (struct srp_iu *) (uintptr_t) wc.wr_id;
1839 list_add(&iu->list, &target->free_tx);
1841 srp_handle_qp_err(wc.wr_id, wc.status, true, target);
1846 static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
1848 struct srp_target_port *target = host_to_target(shost);
1849 struct srp_rport *rport = target->rport;
1850 struct srp_request *req;
1852 struct srp_cmd *cmd;
1853 struct ib_device *dev;
1854 unsigned long flags;
1856 const bool in_scsi_eh = !in_interrupt() && current == shost->ehandler;
1859 * The SCSI EH thread is the only context from which srp_queuecommand()
1860 * can get invoked for blocked devices (SDEV_BLOCK /
1861 * SDEV_CREATED_BLOCK). Avoid racing with srp_reconnect_rport() by
1862 * locking the rport mutex if invoked from inside the SCSI EH.
1865 mutex_lock(&rport->mutex);
1867 scmnd->result = srp_chkready(target->rport);
1868 if (unlikely(scmnd->result))
1871 spin_lock_irqsave(&target->lock, flags);
1872 iu = __srp_get_tx_iu(target, SRP_IU_CMD);
1876 req = list_first_entry(&target->free_reqs, struct srp_request, list);
1877 list_del(&req->list);
1878 spin_unlock_irqrestore(&target->lock, flags);
1880 dev = target->srp_host->srp_dev->dev;
1881 ib_dma_sync_single_for_cpu(dev, iu->dma, target->max_iu_len,
1884 scmnd->host_scribble = (void *) req;
1887 memset(cmd, 0, sizeof *cmd);
1889 cmd->opcode = SRP_CMD;
1890 cmd->lun = cpu_to_be64((u64) scmnd->device->lun << 48);
1891 cmd->tag = req->index;
1892 memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);
1897 len = srp_map_data(scmnd, target, req);
1899 shost_printk(KERN_ERR, target->scsi_host,
1900 PFX "Failed to map data (%d)\n", len);
1902 * If we ran out of memory descriptors (-ENOMEM) because an
1903 * application is queuing many requests with more than
1904 * max_pages_per_mr sg-list elements, tell the SCSI mid-layer
1905 * to reduce queue depth temporarily.
1907 scmnd->result = len == -ENOMEM ?
1908 DID_OK << 16 | QUEUE_FULL << 1 : DID_ERROR << 16;
1912 ib_dma_sync_single_for_device(dev, iu->dma, target->max_iu_len,
1915 if (srp_post_send(target, iu, len)) {
1916 shost_printk(KERN_ERR, target->scsi_host, PFX "Send failed\n");
1924 mutex_unlock(&rport->mutex);
1929 srp_unmap_data(scmnd, target, req);
1932 srp_put_tx_iu(target, iu, SRP_IU_CMD);
1935 * Avoid that the loops that iterate over the request ring can
1936 * encounter a dangling SCSI command pointer.
1940 spin_lock_irqsave(&target->lock, flags);
1941 list_add(&req->list, &target->free_reqs);
1944 spin_unlock_irqrestore(&target->lock, flags);
1947 if (scmnd->result) {
1948 scmnd->scsi_done(scmnd);
1951 ret = SCSI_MLQUEUE_HOST_BUSY;
1958 * Note: the resources allocated in this function are freed in
1959 * srp_free_target_ib().
1961 static int srp_alloc_iu_bufs(struct srp_target_port *target)
1965 target->rx_ring = kzalloc(target->queue_size * sizeof(*target->rx_ring),
1967 if (!target->rx_ring)
1969 target->tx_ring = kzalloc(target->queue_size * sizeof(*target->tx_ring),
1971 if (!target->tx_ring)
1974 for (i = 0; i < target->queue_size; ++i) {
1975 target->rx_ring[i] = srp_alloc_iu(target->srp_host,
1976 target->max_ti_iu_len,
1977 GFP_KERNEL, DMA_FROM_DEVICE);
1978 if (!target->rx_ring[i])
1982 for (i = 0; i < target->queue_size; ++i) {
1983 target->tx_ring[i] = srp_alloc_iu(target->srp_host,
1985 GFP_KERNEL, DMA_TO_DEVICE);
1986 if (!target->tx_ring[i])
1989 list_add(&target->tx_ring[i]->list, &target->free_tx);
1995 for (i = 0; i < target->queue_size; ++i) {
1996 srp_free_iu(target->srp_host, target->rx_ring[i]);
1997 srp_free_iu(target->srp_host, target->tx_ring[i]);
2002 kfree(target->tx_ring);
2003 target->tx_ring = NULL;
2004 kfree(target->rx_ring);
2005 target->rx_ring = NULL;
2010 static uint32_t srp_compute_rq_tmo(struct ib_qp_attr *qp_attr, int attr_mask)
2012 uint64_t T_tr_ns, max_compl_time_ms;
2013 uint32_t rq_tmo_jiffies;
2016 * According to section 11.2.4.2 in the IBTA spec (Modify Queue Pair,
2017 * table 91), both the QP timeout and the retry count have to be set
2018 * for RC QP's during the RTR to RTS transition.
2020 WARN_ON_ONCE((attr_mask & (IB_QP_TIMEOUT | IB_QP_RETRY_CNT)) !=
2021 (IB_QP_TIMEOUT | IB_QP_RETRY_CNT));
2024 * Set target->rq_tmo_jiffies to one second more than the largest time
2025 * it can take before an error completion is generated. See also
2026 * C9-140..142 in the IBTA spec for more information about how to
2027 * convert the QP Local ACK Timeout value to nanoseconds.
2029 T_tr_ns = 4096 * (1ULL << qp_attr->timeout);
2030 max_compl_time_ms = qp_attr->retry_cnt * 4 * T_tr_ns;
2031 do_div(max_compl_time_ms, NSEC_PER_MSEC);
2032 rq_tmo_jiffies = msecs_to_jiffies(max_compl_time_ms + 1000);
2034 return rq_tmo_jiffies;
2037 static void srp_cm_rep_handler(struct ib_cm_id *cm_id,
2038 struct srp_login_rsp *lrsp,
2039 struct srp_target_port *target)
2041 struct ib_qp_attr *qp_attr = NULL;
2046 if (lrsp->opcode == SRP_LOGIN_RSP) {
2047 target->max_ti_iu_len = be32_to_cpu(lrsp->max_ti_iu_len);
2048 target->req_lim = be32_to_cpu(lrsp->req_lim_delta);
2051 * Reserve credits for task management so we don't
2052 * bounce requests back to the SCSI mid-layer.
2054 target->scsi_host->can_queue
2055 = min(target->req_lim - SRP_TSK_MGMT_SQ_SIZE,
2056 target->scsi_host->can_queue);
2057 target->scsi_host->cmd_per_lun
2058 = min_t(int, target->scsi_host->can_queue,
2059 target->scsi_host->cmd_per_lun);
2061 shost_printk(KERN_WARNING, target->scsi_host,
2062 PFX "Unhandled RSP opcode %#x\n", lrsp->opcode);
2067 if (!target->rx_ring) {
2068 ret = srp_alloc_iu_bufs(target);
2074 qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL);
2078 qp_attr->qp_state = IB_QPS_RTR;
2079 ret = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
2083 ret = ib_modify_qp(target->qp, qp_attr, attr_mask);
2087 for (i = 0; i < target->queue_size; i++) {
2088 struct srp_iu *iu = target->rx_ring[i];
2089 ret = srp_post_recv(target, iu);
2094 qp_attr->qp_state = IB_QPS_RTS;
2095 ret = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
2099 target->rq_tmo_jiffies = srp_compute_rq_tmo(qp_attr, attr_mask);
2101 ret = ib_modify_qp(target->qp, qp_attr, attr_mask);
2105 ret = ib_send_cm_rtu(cm_id, NULL, 0);
2111 target->status = ret;
2114 static void srp_cm_rej_handler(struct ib_cm_id *cm_id,
2115 struct ib_cm_event *event,
2116 struct srp_target_port *target)
2118 struct Scsi_Host *shost = target->scsi_host;
2119 struct ib_class_port_info *cpi;
2122 switch (event->param.rej_rcvd.reason) {
2123 case IB_CM_REJ_PORT_CM_REDIRECT:
2124 cpi = event->param.rej_rcvd.ari;
2125 target->path.dlid = cpi->redirect_lid;
2126 target->path.pkey = cpi->redirect_pkey;
2127 cm_id->remote_cm_qpn = be32_to_cpu(cpi->redirect_qp) & 0x00ffffff;
2128 memcpy(target->path.dgid.raw, cpi->redirect_gid, 16);
2130 target->status = target->path.dlid ?
2131 SRP_DLID_REDIRECT : SRP_PORT_REDIRECT;
2134 case IB_CM_REJ_PORT_REDIRECT:
2135 if (srp_target_is_topspin(target)) {
2137 * Topspin/Cisco SRP gateways incorrectly send
2138 * reject reason code 25 when they mean 24
2141 memcpy(target->path.dgid.raw,
2142 event->param.rej_rcvd.ari, 16);
2144 shost_printk(KERN_DEBUG, shost,
2145 PFX "Topspin/Cisco redirect to target port GID %016llx%016llx\n",
2146 (unsigned long long) be64_to_cpu(target->path.dgid.global.subnet_prefix),
2147 (unsigned long long) be64_to_cpu(target->path.dgid.global.interface_id));
2149 target->status = SRP_PORT_REDIRECT;
2151 shost_printk(KERN_WARNING, shost,
2152 " REJ reason: IB_CM_REJ_PORT_REDIRECT\n");
2153 target->status = -ECONNRESET;
2157 case IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID:
2158 shost_printk(KERN_WARNING, shost,
2159 " REJ reason: IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID\n");
2160 target->status = -ECONNRESET;
2163 case IB_CM_REJ_CONSUMER_DEFINED:
2164 opcode = *(u8 *) event->private_data;
2165 if (opcode == SRP_LOGIN_REJ) {
2166 struct srp_login_rej *rej = event->private_data;
2167 u32 reason = be32_to_cpu(rej->reason);
2169 if (reason == SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE)
2170 shost_printk(KERN_WARNING, shost,
2171 PFX "SRP_LOGIN_REJ: requested max_it_iu_len too large\n");
2173 shost_printk(KERN_WARNING, shost, PFX
2174 "SRP LOGIN from %pI6 to %pI6 REJECTED, reason 0x%08x\n",
2175 target->path.sgid.raw,
2176 target->orig_dgid, reason);
2178 shost_printk(KERN_WARNING, shost,
2179 " REJ reason: IB_CM_REJ_CONSUMER_DEFINED,"
2180 " opcode 0x%02x\n", opcode);
2181 target->status = -ECONNRESET;
2184 case IB_CM_REJ_STALE_CONN:
2185 shost_printk(KERN_WARNING, shost, " REJ reason: stale connection\n");
2186 target->status = SRP_STALE_CONN;
2190 shost_printk(KERN_WARNING, shost, " REJ reason 0x%x\n",
2191 event->param.rej_rcvd.reason);
2192 target->status = -ECONNRESET;
2196 static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
2198 struct srp_target_port *target = cm_id->context;
2201 switch (event->event) {
2202 case IB_CM_REQ_ERROR:
2203 shost_printk(KERN_DEBUG, target->scsi_host,
2204 PFX "Sending CM REQ failed\n");
2206 target->status = -ECONNRESET;
2209 case IB_CM_REP_RECEIVED:
2211 srp_cm_rep_handler(cm_id, event->private_data, target);
2214 case IB_CM_REJ_RECEIVED:
2215 shost_printk(KERN_DEBUG, target->scsi_host, PFX "REJ received\n");
2218 srp_cm_rej_handler(cm_id, event, target);
2221 case IB_CM_DREQ_RECEIVED:
2222 shost_printk(KERN_WARNING, target->scsi_host,
2223 PFX "DREQ received - connection closed\n");
2224 srp_change_conn_state(target, false);
2225 if (ib_send_cm_drep(cm_id, NULL, 0))
2226 shost_printk(KERN_ERR, target->scsi_host,
2227 PFX "Sending CM DREP failed\n");
2228 queue_work(system_long_wq, &target->tl_err_work);
2231 case IB_CM_TIMEWAIT_EXIT:
2232 shost_printk(KERN_ERR, target->scsi_host,
2233 PFX "connection closed\n");
2239 case IB_CM_MRA_RECEIVED:
2240 case IB_CM_DREQ_ERROR:
2241 case IB_CM_DREP_RECEIVED:
2245 shost_printk(KERN_WARNING, target->scsi_host,
2246 PFX "Unhandled CM event %d\n", event->event);
2251 complete(&target->done);
2257 * srp_change_queue_type - changing device queue tag type
2258 * @sdev: scsi device struct
2259 * @tag_type: requested tag type
2261 * Returns queue tag type.
2264 srp_change_queue_type(struct scsi_device *sdev, int tag_type)
2266 if (sdev->tagged_supported) {
2267 scsi_set_tag_type(sdev, tag_type);
2269 scsi_activate_tcq(sdev, sdev->queue_depth);
2271 scsi_deactivate_tcq(sdev, sdev->queue_depth);
2279 * srp_change_queue_depth - setting device queue depth
2280 * @sdev: scsi device struct
2281 * @qdepth: requested queue depth
2282 * @reason: SCSI_QDEPTH_DEFAULT/SCSI_QDEPTH_QFULL/SCSI_QDEPTH_RAMP_UP
2283 * (see include/scsi/scsi_host.h for definition)
2285 * Returns queue depth.
2288 srp_change_queue_depth(struct scsi_device *sdev, int qdepth, int reason)
2290 struct Scsi_Host *shost = sdev->host;
2292 if (reason == SCSI_QDEPTH_DEFAULT || reason == SCSI_QDEPTH_RAMP_UP) {
2293 max_depth = shost->can_queue;
2294 if (!sdev->tagged_supported)
2296 if (qdepth > max_depth)
2298 scsi_adjust_queue_depth(sdev, scsi_get_tag_type(sdev), qdepth);
2299 } else if (reason == SCSI_QDEPTH_QFULL)
2300 scsi_track_queue_full(sdev, qdepth);
2304 return sdev->queue_depth;
2307 static int srp_send_tsk_mgmt(struct srp_target_port *target,
2308 u64 req_tag, unsigned int lun, u8 func)
2310 struct srp_rport *rport = target->rport;
2311 struct ib_device *dev = target->srp_host->srp_dev->dev;
2313 struct srp_tsk_mgmt *tsk_mgmt;
2315 if (!target->connected || target->qp_in_error)
2318 init_completion(&target->tsk_mgmt_done);
2321 * Lock the rport mutex to avoid that srp_create_target_ib() is
2322 * invoked while a task management function is being sent.
2324 mutex_lock(&rport->mutex);
2325 spin_lock_irq(&target->lock);
2326 iu = __srp_get_tx_iu(target, SRP_IU_TSK_MGMT);
2327 spin_unlock_irq(&target->lock);
2330 mutex_unlock(&rport->mutex);
2335 ib_dma_sync_single_for_cpu(dev, iu->dma, sizeof *tsk_mgmt,
2338 memset(tsk_mgmt, 0, sizeof *tsk_mgmt);
2340 tsk_mgmt->opcode = SRP_TSK_MGMT;
2341 tsk_mgmt->lun = cpu_to_be64((u64) lun << 48);
2342 tsk_mgmt->tag = req_tag | SRP_TAG_TSK_MGMT;
2343 tsk_mgmt->tsk_mgmt_func = func;
2344 tsk_mgmt->task_tag = req_tag;
2346 ib_dma_sync_single_for_device(dev, iu->dma, sizeof *tsk_mgmt,
2348 if (srp_post_send(target, iu, sizeof *tsk_mgmt)) {
2349 srp_put_tx_iu(target, iu, SRP_IU_TSK_MGMT);
2350 mutex_unlock(&rport->mutex);
2354 mutex_unlock(&rport->mutex);
2356 if (!wait_for_completion_timeout(&target->tsk_mgmt_done,
2357 msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS)))
2363 static int srp_abort(struct scsi_cmnd *scmnd)
2365 struct srp_target_port *target = host_to_target(scmnd->device->host);
2366 struct srp_request *req = (struct srp_request *) scmnd->host_scribble;
2369 shost_printk(KERN_ERR, target->scsi_host, "SRP abort called\n");
2371 if (!req || !srp_claim_req(target, req, NULL, scmnd))
2373 if (srp_send_tsk_mgmt(target, req->index, scmnd->device->lun,
2374 SRP_TSK_ABORT_TASK) == 0)
2376 else if (target->rport->state == SRP_RPORT_LOST)
2380 srp_free_req(target, req, scmnd, 0);
2381 scmnd->result = DID_ABORT << 16;
2382 scmnd->scsi_done(scmnd);
2387 static int srp_reset_device(struct scsi_cmnd *scmnd)
2389 struct srp_target_port *target = host_to_target(scmnd->device->host);
2392 shost_printk(KERN_ERR, target->scsi_host, "SRP reset_device called\n");
2394 if (srp_send_tsk_mgmt(target, SRP_TAG_NO_REQ, scmnd->device->lun,
2397 if (target->tsk_mgmt_status)
2400 for (i = 0; i < target->req_ring_size; ++i) {
2401 struct srp_request *req = &target->req_ring[i];
2402 srp_finish_req(target, req, scmnd->device, DID_RESET << 16);
2408 static int srp_reset_host(struct scsi_cmnd *scmnd)
2410 struct srp_target_port *target = host_to_target(scmnd->device->host);
2412 shost_printk(KERN_ERR, target->scsi_host, PFX "SRP reset_host called\n");
2414 return srp_reconnect_rport(target->rport) == 0 ? SUCCESS : FAILED;
2417 static int srp_slave_configure(struct scsi_device *sdev)
2419 struct Scsi_Host *shost = sdev->host;
2420 struct srp_target_port *target = host_to_target(shost);
2421 struct request_queue *q = sdev->request_queue;
2422 unsigned long timeout;
2424 if (sdev->type == TYPE_DISK) {
2425 timeout = max_t(unsigned, 30 * HZ, target->rq_tmo_jiffies);
2426 blk_queue_rq_timeout(q, timeout);
2432 static ssize_t show_id_ext(struct device *dev, struct device_attribute *attr,
2435 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2437 return sprintf(buf, "0x%016llx\n",
2438 (unsigned long long) be64_to_cpu(target->id_ext));
2441 static ssize_t show_ioc_guid(struct device *dev, struct device_attribute *attr,
2444 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2446 return sprintf(buf, "0x%016llx\n",
2447 (unsigned long long) be64_to_cpu(target->ioc_guid));
2450 static ssize_t show_service_id(struct device *dev,
2451 struct device_attribute *attr, char *buf)
2453 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2455 return sprintf(buf, "0x%016llx\n",
2456 (unsigned long long) be64_to_cpu(target->service_id));
2459 static ssize_t show_pkey(struct device *dev, struct device_attribute *attr,
2462 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2464 return sprintf(buf, "0x%04x\n", be16_to_cpu(target->path.pkey));
2467 static ssize_t show_sgid(struct device *dev, struct device_attribute *attr,
2470 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2472 return sprintf(buf, "%pI6\n", target->path.sgid.raw);
2475 static ssize_t show_dgid(struct device *dev, struct device_attribute *attr,
2478 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2480 return sprintf(buf, "%pI6\n", target->path.dgid.raw);
2483 static ssize_t show_orig_dgid(struct device *dev,
2484 struct device_attribute *attr, char *buf)
2486 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2488 return sprintf(buf, "%pI6\n", target->orig_dgid);
2491 static ssize_t show_req_lim(struct device *dev,
2492 struct device_attribute *attr, char *buf)
2494 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2496 return sprintf(buf, "%d\n", target->req_lim);
2499 static ssize_t show_zero_req_lim(struct device *dev,
2500 struct device_attribute *attr, char *buf)
2502 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2504 return sprintf(buf, "%d\n", target->zero_req_lim);
2507 static ssize_t show_local_ib_port(struct device *dev,
2508 struct device_attribute *attr, char *buf)
2510 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2512 return sprintf(buf, "%d\n", target->srp_host->port);
2515 static ssize_t show_local_ib_device(struct device *dev,
2516 struct device_attribute *attr, char *buf)
2518 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2520 return sprintf(buf, "%s\n", target->srp_host->srp_dev->dev->name);
2523 static ssize_t show_comp_vector(struct device *dev,
2524 struct device_attribute *attr, char *buf)
2526 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2528 return sprintf(buf, "%d\n", target->comp_vector);
2531 static ssize_t show_tl_retry_count(struct device *dev,
2532 struct device_attribute *attr, char *buf)
2534 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2536 return sprintf(buf, "%d\n", target->tl_retry_count);
2539 static ssize_t show_cmd_sg_entries(struct device *dev,
2540 struct device_attribute *attr, char *buf)
2542 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2544 return sprintf(buf, "%u\n", target->cmd_sg_cnt);
2547 static ssize_t show_allow_ext_sg(struct device *dev,
2548 struct device_attribute *attr, char *buf)
2550 struct srp_target_port *target = host_to_target(class_to_shost(dev));
2552 return sprintf(buf, "%s\n", target->allow_ext_sg ? "true" : "false");
2555 static DEVICE_ATTR(id_ext, S_IRUGO, show_id_ext, NULL);
2556 static DEVICE_ATTR(ioc_guid, S_IRUGO, show_ioc_guid, NULL);
2557 static DEVICE_ATTR(service_id, S_IRUGO, show_service_id, NULL);
2558 static DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL);
2559 static DEVICE_ATTR(sgid, S_IRUGO, show_sgid, NULL);
2560 static DEVICE_ATTR(dgid, S_IRUGO, show_dgid, NULL);
2561 static DEVICE_ATTR(orig_dgid, S_IRUGO, show_orig_dgid, NULL);
2562 static DEVICE_ATTR(req_lim, S_IRUGO, show_req_lim, NULL);
2563 static DEVICE_ATTR(zero_req_lim, S_IRUGO, show_zero_req_lim, NULL);
2564 static DEVICE_ATTR(local_ib_port, S_IRUGO, show_local_ib_port, NULL);
2565 static DEVICE_ATTR(local_ib_device, S_IRUGO, show_local_ib_device, NULL);
2566 static DEVICE_ATTR(comp_vector, S_IRUGO, show_comp_vector, NULL);
2567 static DEVICE_ATTR(tl_retry_count, S_IRUGO, show_tl_retry_count, NULL);
2568 static DEVICE_ATTR(cmd_sg_entries, S_IRUGO, show_cmd_sg_entries, NULL);
2569 static DEVICE_ATTR(allow_ext_sg, S_IRUGO, show_allow_ext_sg, NULL);
2571 static struct device_attribute *srp_host_attrs[] = {
2574 &dev_attr_service_id,
2578 &dev_attr_orig_dgid,
2580 &dev_attr_zero_req_lim,
2581 &dev_attr_local_ib_port,
2582 &dev_attr_local_ib_device,
2583 &dev_attr_comp_vector,
2584 &dev_attr_tl_retry_count,
2585 &dev_attr_cmd_sg_entries,
2586 &dev_attr_allow_ext_sg,
2590 static struct scsi_host_template srp_template = {
2591 .module = THIS_MODULE,
2592 .name = "InfiniBand SRP initiator",
2593 .proc_name = DRV_NAME,
2594 .slave_configure = srp_slave_configure,
2595 .info = srp_target_info,
2596 .queuecommand = srp_queuecommand,
2597 .change_queue_depth = srp_change_queue_depth,
2598 .change_queue_type = srp_change_queue_type,
2599 .eh_abort_handler = srp_abort,
2600 .eh_device_reset_handler = srp_reset_device,
2601 .eh_host_reset_handler = srp_reset_host,
2602 .skip_settle_delay = true,
2603 .sg_tablesize = SRP_DEF_SG_TABLESIZE,
2604 .can_queue = SRP_DEFAULT_CMD_SQ_SIZE,
2606 .cmd_per_lun = SRP_DEFAULT_CMD_SQ_SIZE,
2607 .use_clustering = ENABLE_CLUSTERING,
2608 .shost_attrs = srp_host_attrs
2611 static int srp_add_target(struct srp_host *host, struct srp_target_port *target)
2613 struct srp_rport_identifiers ids;
2614 struct srp_rport *rport;
2616 sprintf(target->target_name, "SRP.T10:%016llX",
2617 (unsigned long long) be64_to_cpu(target->id_ext));
2619 if (scsi_add_host(target->scsi_host, host->srp_dev->dev->dma_device))
2622 memcpy(ids.port_id, &target->id_ext, 8);
2623 memcpy(ids.port_id + 8, &target->ioc_guid, 8);
2624 ids.roles = SRP_RPORT_ROLE_TARGET;
2625 rport = srp_rport_add(target->scsi_host, &ids);
2626 if (IS_ERR(rport)) {
2627 scsi_remove_host(target->scsi_host);
2628 return PTR_ERR(rport);
2631 rport->lld_data = target;
2632 target->rport = rport;
2634 spin_lock(&host->target_lock);
2635 list_add_tail(&target->list, &host->target_list);
2636 spin_unlock(&host->target_lock);
2638 target->state = SRP_TARGET_LIVE;
2640 scsi_scan_target(&target->scsi_host->shost_gendev,
2641 0, target->scsi_id, SCAN_WILD_CARD, 0);
2646 static void srp_release_dev(struct device *dev)
2648 struct srp_host *host =
2649 container_of(dev, struct srp_host, dev);
2651 complete(&host->released);
2654 static struct class srp_class = {
2655 .name = "infiniband_srp",
2656 .dev_release = srp_release_dev
2660 * srp_conn_unique() - check whether the connection to a target is unique
2662 * @target: SRP target port.
2664 static bool srp_conn_unique(struct srp_host *host,
2665 struct srp_target_port *target)
2667 struct srp_target_port *t;
2670 if (target->state == SRP_TARGET_REMOVED)
2675 spin_lock(&host->target_lock);
2676 list_for_each_entry(t, &host->target_list, list) {
2678 target->id_ext == t->id_ext &&
2679 target->ioc_guid == t->ioc_guid &&
2680 target->initiator_ext == t->initiator_ext) {
2685 spin_unlock(&host->target_lock);
2692 * Target ports are added by writing
2694 * id_ext=<SRP ID ext>,ioc_guid=<SRP IOC GUID>,dgid=<dest GID>,
2695 * pkey=<P_Key>,service_id=<service ID>
2697 * to the add_target sysfs attribute.
2701 SRP_OPT_ID_EXT = 1 << 0,
2702 SRP_OPT_IOC_GUID = 1 << 1,
2703 SRP_OPT_DGID = 1 << 2,
2704 SRP_OPT_PKEY = 1 << 3,
2705 SRP_OPT_SERVICE_ID = 1 << 4,
2706 SRP_OPT_MAX_SECT = 1 << 5,
2707 SRP_OPT_MAX_CMD_PER_LUN = 1 << 6,
2708 SRP_OPT_IO_CLASS = 1 << 7,
2709 SRP_OPT_INITIATOR_EXT = 1 << 8,
2710 SRP_OPT_CMD_SG_ENTRIES = 1 << 9,
2711 SRP_OPT_ALLOW_EXT_SG = 1 << 10,
2712 SRP_OPT_SG_TABLESIZE = 1 << 11,
2713 SRP_OPT_COMP_VECTOR = 1 << 12,
2714 SRP_OPT_TL_RETRY_COUNT = 1 << 13,
2715 SRP_OPT_QUEUE_SIZE = 1 << 14,
2716 SRP_OPT_ALL = (SRP_OPT_ID_EXT |
2720 SRP_OPT_SERVICE_ID),
2723 static const match_table_t srp_opt_tokens = {
2724 { SRP_OPT_ID_EXT, "id_ext=%s" },
2725 { SRP_OPT_IOC_GUID, "ioc_guid=%s" },
2726 { SRP_OPT_DGID, "dgid=%s" },
2727 { SRP_OPT_PKEY, "pkey=%x" },
2728 { SRP_OPT_SERVICE_ID, "service_id=%s" },
2729 { SRP_OPT_MAX_SECT, "max_sect=%d" },
2730 { SRP_OPT_MAX_CMD_PER_LUN, "max_cmd_per_lun=%d" },
2731 { SRP_OPT_IO_CLASS, "io_class=%x" },
2732 { SRP_OPT_INITIATOR_EXT, "initiator_ext=%s" },
2733 { SRP_OPT_CMD_SG_ENTRIES, "cmd_sg_entries=%u" },
2734 { SRP_OPT_ALLOW_EXT_SG, "allow_ext_sg=%u" },
2735 { SRP_OPT_SG_TABLESIZE, "sg_tablesize=%u" },
2736 { SRP_OPT_COMP_VECTOR, "comp_vector=%u" },
2737 { SRP_OPT_TL_RETRY_COUNT, "tl_retry_count=%u" },
2738 { SRP_OPT_QUEUE_SIZE, "queue_size=%d" },
2739 { SRP_OPT_ERR, NULL }
2742 static int srp_parse_options(const char *buf, struct srp_target_port *target)
2744 char *options, *sep_opt;
2747 substring_t args[MAX_OPT_ARGS];
2753 options = kstrdup(buf, GFP_KERNEL);
2758 while ((p = strsep(&sep_opt, ",")) != NULL) {
2762 token = match_token(p, srp_opt_tokens, args);
2766 case SRP_OPT_ID_EXT:
2767 p = match_strdup(args);
2772 target->id_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
2776 case SRP_OPT_IOC_GUID:
2777 p = match_strdup(args);
2782 target->ioc_guid = cpu_to_be64(simple_strtoull(p, NULL, 16));
2787 p = match_strdup(args);
2792 if (strlen(p) != 32) {
2793 pr_warn("bad dest GID parameter '%s'\n", p);
2798 for (i = 0; i < 16; ++i) {
2799 strlcpy(dgid, p + i * 2, 3);
2800 target->path.dgid.raw[i] = simple_strtoul(dgid, NULL, 16);
2803 memcpy(target->orig_dgid, target->path.dgid.raw, 16);
2807 if (match_hex(args, &token)) {
2808 pr_warn("bad P_Key parameter '%s'\n", p);
2811 target->path.pkey = cpu_to_be16(token);
2814 case SRP_OPT_SERVICE_ID:
2815 p = match_strdup(args);
2820 target->service_id = cpu_to_be64(simple_strtoull(p, NULL, 16));
2821 target->path.service_id = target->service_id;
2825 case SRP_OPT_MAX_SECT:
2826 if (match_int(args, &token)) {
2827 pr_warn("bad max sect parameter '%s'\n", p);
2830 target->scsi_host->max_sectors = token;
2833 case SRP_OPT_QUEUE_SIZE:
2834 if (match_int(args, &token) || token < 1) {
2835 pr_warn("bad queue_size parameter '%s'\n", p);
2838 target->scsi_host->can_queue = token;
2839 target->queue_size = token + SRP_RSP_SQ_SIZE +
2840 SRP_TSK_MGMT_SQ_SIZE;
2841 if (!(opt_mask & SRP_OPT_MAX_CMD_PER_LUN))
2842 target->scsi_host->cmd_per_lun = token;
2845 case SRP_OPT_MAX_CMD_PER_LUN:
2846 if (match_int(args, &token) || token < 1) {
2847 pr_warn("bad max cmd_per_lun parameter '%s'\n",
2851 target->scsi_host->cmd_per_lun = token;
2854 case SRP_OPT_IO_CLASS:
2855 if (match_hex(args, &token)) {
2856 pr_warn("bad IO class parameter '%s'\n", p);
2859 if (token != SRP_REV10_IB_IO_CLASS &&
2860 token != SRP_REV16A_IB_IO_CLASS) {
2861 pr_warn("unknown IO class parameter value %x specified (use %x or %x).\n",
2862 token, SRP_REV10_IB_IO_CLASS,
2863 SRP_REV16A_IB_IO_CLASS);
2866 target->io_class = token;
2869 case SRP_OPT_INITIATOR_EXT:
2870 p = match_strdup(args);
2875 target->initiator_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
2879 case SRP_OPT_CMD_SG_ENTRIES:
2880 if (match_int(args, &token) || token < 1 || token > 255) {
2881 pr_warn("bad max cmd_sg_entries parameter '%s'\n",
2885 target->cmd_sg_cnt = token;
2888 case SRP_OPT_ALLOW_EXT_SG:
2889 if (match_int(args, &token)) {
2890 pr_warn("bad allow_ext_sg parameter '%s'\n", p);
2893 target->allow_ext_sg = !!token;
2896 case SRP_OPT_SG_TABLESIZE:
2897 if (match_int(args, &token) || token < 1 ||
2898 token > SCSI_MAX_SG_CHAIN_SEGMENTS) {
2899 pr_warn("bad max sg_tablesize parameter '%s'\n",
2903 target->sg_tablesize = token;
2906 case SRP_OPT_COMP_VECTOR:
2907 if (match_int(args, &token) || token < 0) {
2908 pr_warn("bad comp_vector parameter '%s'\n", p);
2911 target->comp_vector = token;
2914 case SRP_OPT_TL_RETRY_COUNT:
2915 if (match_int(args, &token) || token < 2 || token > 7) {
2916 pr_warn("bad tl_retry_count parameter '%s' (must be a number between 2 and 7)\n",
2920 target->tl_retry_count = token;
2924 pr_warn("unknown parameter or missing value '%s' in target creation request\n",
2930 if ((opt_mask & SRP_OPT_ALL) == SRP_OPT_ALL)
2933 for (i = 0; i < ARRAY_SIZE(srp_opt_tokens); ++i)
2934 if ((srp_opt_tokens[i].token & SRP_OPT_ALL) &&
2935 !(srp_opt_tokens[i].token & opt_mask))
2936 pr_warn("target creation request is missing parameter '%s'\n",
2937 srp_opt_tokens[i].pattern);
2939 if (target->scsi_host->cmd_per_lun > target->scsi_host->can_queue
2940 && (opt_mask & SRP_OPT_MAX_CMD_PER_LUN))
2941 pr_warn("cmd_per_lun = %d > queue_size = %d\n",
2942 target->scsi_host->cmd_per_lun,
2943 target->scsi_host->can_queue);
2950 static ssize_t srp_create_target(struct device *dev,
2951 struct device_attribute *attr,
2952 const char *buf, size_t count)
2954 struct srp_host *host =
2955 container_of(dev, struct srp_host, dev);
2956 struct Scsi_Host *target_host;
2957 struct srp_target_port *target;
2958 struct srp_device *srp_dev = host->srp_dev;
2959 struct ib_device *ibdev = srp_dev->dev;
2962 target_host = scsi_host_alloc(&srp_template,
2963 sizeof (struct srp_target_port));
2967 target_host->transportt = ib_srp_transport_template;
2968 target_host->max_channel = 0;
2969 target_host->max_id = 1;
2970 target_host->max_lun = SRP_MAX_LUN;
2971 target_host->max_cmd_len = sizeof ((struct srp_cmd *) (void *) 0L)->cdb;
2973 target = host_to_target(target_host);
2975 target->io_class = SRP_REV16A_IB_IO_CLASS;
2976 target->scsi_host = target_host;
2977 target->srp_host = host;
2978 target->lkey = host->srp_dev->mr->lkey;
2979 target->rkey = host->srp_dev->mr->rkey;
2980 target->cmd_sg_cnt = cmd_sg_entries;
2981 target->sg_tablesize = indirect_sg_entries ? : cmd_sg_entries;
2982 target->allow_ext_sg = allow_ext_sg;
2983 target->tl_retry_count = 7;
2984 target->queue_size = SRP_DEFAULT_QUEUE_SIZE;
2986 mutex_lock(&host->add_target_mutex);
2988 ret = srp_parse_options(buf, target);
2992 target->req_ring_size = target->queue_size - SRP_TSK_MGMT_SQ_SIZE;
2994 if (!srp_conn_unique(target->srp_host, target)) {
2995 shost_printk(KERN_INFO, target->scsi_host,
2996 PFX "Already connected to target port with id_ext=%016llx;ioc_guid=%016llx;initiator_ext=%016llx\n",
2997 be64_to_cpu(target->id_ext),
2998 be64_to_cpu(target->ioc_guid),
2999 be64_to_cpu(target->initiator_ext));
3004 if (!srp_dev->has_fmr && !srp_dev->has_fr && !target->allow_ext_sg &&
3005 target->cmd_sg_cnt < target->sg_tablesize) {
3006 pr_warn("No MR pool and no external indirect descriptors, limiting sg_tablesize to cmd_sg_cnt\n");
3007 target->sg_tablesize = target->cmd_sg_cnt;
3010 target_host->sg_tablesize = target->sg_tablesize;
3011 target->indirect_size = target->sg_tablesize *
3012 sizeof (struct srp_direct_buf);
3013 target->max_iu_len = sizeof (struct srp_cmd) +
3014 sizeof (struct srp_indirect_buf) +
3015 target->cmd_sg_cnt * sizeof (struct srp_direct_buf);
3017 INIT_WORK(&target->tl_err_work, srp_tl_err_work);
3018 INIT_WORK(&target->remove_work, srp_remove_work);
3019 spin_lock_init(&target->lock);
3020 INIT_LIST_HEAD(&target->free_tx);
3021 ret = srp_alloc_req_data(target);
3025 ret = ib_query_gid(ibdev, host->port, 0, &target->path.sgid);
3029 ret = srp_create_target_ib(target);
3033 ret = srp_new_cm_id(target);
3037 ret = srp_connect_target(target);
3039 shost_printk(KERN_ERR, target->scsi_host,
3040 PFX "Connection failed\n");
3044 ret = srp_add_target(host, target);
3046 goto err_disconnect;
3048 shost_printk(KERN_DEBUG, target->scsi_host, PFX
3049 "new target: id_ext %016llx ioc_guid %016llx pkey %04x service_id %016llx sgid %pI6 dgid %pI6\n",
3050 be64_to_cpu(target->id_ext),
3051 be64_to_cpu(target->ioc_guid),
3052 be16_to_cpu(target->path.pkey),
3053 be64_to_cpu(target->service_id),
3054 target->path.sgid.raw, target->path.dgid.raw);
3059 mutex_unlock(&host->add_target_mutex);
3063 srp_disconnect_target(target);
3066 ib_destroy_cm_id(target->cm_id);
3069 srp_free_target_ib(target);
3072 srp_free_req_data(target);
3075 scsi_host_put(target_host);
3079 static DEVICE_ATTR(add_target, S_IWUSR, NULL, srp_create_target);
3081 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
3084 struct srp_host *host = container_of(dev, struct srp_host, dev);
3086 return sprintf(buf, "%s\n", host->srp_dev->dev->name);
3089 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
3091 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
3094 struct srp_host *host = container_of(dev, struct srp_host, dev);
3096 return sprintf(buf, "%d\n", host->port);
3099 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
3101 static struct srp_host *srp_add_port(struct srp_device *device, u8 port)
3103 struct srp_host *host;
3105 host = kzalloc(sizeof *host, GFP_KERNEL);
3109 INIT_LIST_HEAD(&host->target_list);
3110 spin_lock_init(&host->target_lock);
3111 init_completion(&host->released);
3112 mutex_init(&host->add_target_mutex);
3113 host->srp_dev = device;
3116 host->dev.class = &srp_class;
3117 host->dev.parent = device->dev->dma_device;
3118 dev_set_name(&host->dev, "srp-%s-%d", device->dev->name, port);
3120 if (device_register(&host->dev))
3122 if (device_create_file(&host->dev, &dev_attr_add_target))
3124 if (device_create_file(&host->dev, &dev_attr_ibdev))
3126 if (device_create_file(&host->dev, &dev_attr_port))
3132 device_unregister(&host->dev);
3140 static void srp_add_one(struct ib_device *device)
3142 struct srp_device *srp_dev;
3143 struct ib_device_attr *dev_attr;
3144 struct srp_host *host;
3145 int mr_page_shift, s, e, p;
3146 u64 max_pages_per_mr;
3148 dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
3152 if (ib_query_device(device, dev_attr)) {
3153 pr_warn("Query device failed for %s\n", device->name);
3157 srp_dev = kmalloc(sizeof *srp_dev, GFP_KERNEL);
3161 srp_dev->has_fmr = (device->alloc_fmr && device->dealloc_fmr &&
3162 device->map_phys_fmr && device->unmap_fmr);
3163 srp_dev->has_fr = (dev_attr->device_cap_flags &
3164 IB_DEVICE_MEM_MGT_EXTENSIONS);
3165 if (!srp_dev->has_fmr && !srp_dev->has_fr)
3166 dev_warn(&device->dev, "neither FMR nor FR is supported\n");
3168 srp_dev->use_fast_reg = (srp_dev->has_fr &&
3169 (!srp_dev->has_fmr || prefer_fr));
3172 * Use the smallest page size supported by the HCA, down to a
3173 * minimum of 4096 bytes. We're unlikely to build large sglists
3174 * out of smaller entries.
3176 mr_page_shift = max(12, ffs(dev_attr->page_size_cap) - 1);
3177 srp_dev->mr_page_size = 1 << mr_page_shift;
3178 srp_dev->mr_page_mask = ~((u64) srp_dev->mr_page_size - 1);
3179 max_pages_per_mr = dev_attr->max_mr_size;
3180 do_div(max_pages_per_mr, srp_dev->mr_page_size);
3181 srp_dev->max_pages_per_mr = min_t(u64, SRP_MAX_PAGES_PER_MR,
3183 if (srp_dev->use_fast_reg) {
3184 srp_dev->max_pages_per_mr =
3185 min_t(u32, srp_dev->max_pages_per_mr,
3186 dev_attr->max_fast_reg_page_list_len);
3188 srp_dev->mr_max_size = srp_dev->mr_page_size *
3189 srp_dev->max_pages_per_mr;
3190 pr_debug("%s: mr_page_shift = %d, dev_attr->max_mr_size = %#llx, dev_attr->max_fast_reg_page_list_len = %u, max_pages_per_mr = %d, mr_max_size = %#x\n",
3191 device->name, mr_page_shift, dev_attr->max_mr_size,
3192 dev_attr->max_fast_reg_page_list_len,
3193 srp_dev->max_pages_per_mr, srp_dev->mr_max_size);
3195 INIT_LIST_HEAD(&srp_dev->dev_list);
3197 srp_dev->dev = device;
3198 srp_dev->pd = ib_alloc_pd(device);
3199 if (IS_ERR(srp_dev->pd))
3202 srp_dev->mr = ib_get_dma_mr(srp_dev->pd,
3203 IB_ACCESS_LOCAL_WRITE |
3204 IB_ACCESS_REMOTE_READ |
3205 IB_ACCESS_REMOTE_WRITE);
3206 if (IS_ERR(srp_dev->mr))
3209 if (device->node_type == RDMA_NODE_IB_SWITCH) {
3214 e = device->phys_port_cnt;
3217 for (p = s; p <= e; ++p) {
3218 host = srp_add_port(srp_dev, p);
3220 list_add_tail(&host->list, &srp_dev->dev_list);
3223 ib_set_client_data(device, &srp_client, srp_dev);
3228 ib_dealloc_pd(srp_dev->pd);
3237 static void srp_remove_one(struct ib_device *device)
3239 struct srp_device *srp_dev;
3240 struct srp_host *host, *tmp_host;
3241 struct srp_target_port *target;
3243 srp_dev = ib_get_client_data(device, &srp_client);
3247 list_for_each_entry_safe(host, tmp_host, &srp_dev->dev_list, list) {
3248 device_unregister(&host->dev);
3250 * Wait for the sysfs entry to go away, so that no new
3251 * target ports can be created.
3253 wait_for_completion(&host->released);
3256 * Remove all target ports.
3258 spin_lock(&host->target_lock);
3259 list_for_each_entry(target, &host->target_list, list)
3260 srp_queue_remove_work(target);
3261 spin_unlock(&host->target_lock);
3264 * Wait for target port removal tasks.
3266 flush_workqueue(system_long_wq);
3271 ib_dereg_mr(srp_dev->mr);
3272 ib_dealloc_pd(srp_dev->pd);
3277 static struct srp_function_template ib_srp_transport_functions = {
3278 .has_rport_state = true,
3279 .reset_timer_if_blocked = true,
3280 .reconnect_delay = &srp_reconnect_delay,
3281 .fast_io_fail_tmo = &srp_fast_io_fail_tmo,
3282 .dev_loss_tmo = &srp_dev_loss_tmo,
3283 .reconnect = srp_rport_reconnect,
3284 .rport_delete = srp_rport_delete,
3285 .terminate_rport_io = srp_terminate_io,
3288 static int __init srp_init_module(void)
3292 BUILD_BUG_ON(FIELD_SIZEOF(struct ib_wc, wr_id) < sizeof(void *));
3294 if (srp_sg_tablesize) {
3295 pr_warn("srp_sg_tablesize is deprecated, please use cmd_sg_entries\n");
3296 if (!cmd_sg_entries)
3297 cmd_sg_entries = srp_sg_tablesize;
3300 if (!cmd_sg_entries)
3301 cmd_sg_entries = SRP_DEF_SG_TABLESIZE;
3303 if (cmd_sg_entries > 255) {
3304 pr_warn("Clamping cmd_sg_entries to 255\n");
3305 cmd_sg_entries = 255;
3308 if (!indirect_sg_entries)
3309 indirect_sg_entries = cmd_sg_entries;
3310 else if (indirect_sg_entries < cmd_sg_entries) {
3311 pr_warn("Bumping up indirect_sg_entries to match cmd_sg_entries (%u)\n",
3313 indirect_sg_entries = cmd_sg_entries;
3316 ib_srp_transport_template =
3317 srp_attach_transport(&ib_srp_transport_functions);
3318 if (!ib_srp_transport_template)
3321 ret = class_register(&srp_class);
3323 pr_err("couldn't register class infiniband_srp\n");
3324 srp_release_transport(ib_srp_transport_template);
3328 ib_sa_register_client(&srp_sa_client);
3330 ret = ib_register_client(&srp_client);
3332 pr_err("couldn't register IB client\n");
3333 srp_release_transport(ib_srp_transport_template);
3334 ib_sa_unregister_client(&srp_sa_client);
3335 class_unregister(&srp_class);
3342 static void __exit srp_cleanup_module(void)
3344 ib_unregister_client(&srp_client);
3345 ib_sa_unregister_client(&srp_sa_client);
3346 class_unregister(&srp_class);
3347 srp_release_transport(ib_srp_transport_template);
3350 module_init(srp_init_module);
3351 module_exit(srp_cleanup_module);