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) PFX 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>
50 #include <scsi/scsi_transport_srp.h>
54 #define DRV_NAME "ib_srp"
55 #define PFX DRV_NAME ": "
56 #define DRV_VERSION "0.2"
57 #define DRV_RELDATE "November 1, 2005"
59 MODULE_AUTHOR("Roland Dreier");
60 MODULE_DESCRIPTION("InfiniBand SCSI RDMA Protocol initiator "
61 "v" DRV_VERSION
" (" DRV_RELDATE
")");
62 MODULE_LICENSE("Dual BSD/GPL");
64 static unsigned int srp_sg_tablesize
;
65 static unsigned int cmd_sg_entries
;
66 static unsigned int indirect_sg_entries
;
67 static bool allow_ext_sg
;
68 static int topspin_workarounds
= 1;
70 module_param(srp_sg_tablesize
, uint
, 0444);
71 MODULE_PARM_DESC(srp_sg_tablesize
, "Deprecated name for cmd_sg_entries");
73 module_param(cmd_sg_entries
, uint
, 0444);
74 MODULE_PARM_DESC(cmd_sg_entries
,
75 "Default number of gather/scatter entries in the SRP command (default is 12, max 255)");
77 module_param(indirect_sg_entries
, uint
, 0444);
78 MODULE_PARM_DESC(indirect_sg_entries
,
79 "Default max number of gather/scatter entries (default is 12, max is " __stringify(SCSI_MAX_SG_CHAIN_SEGMENTS
) ")");
81 module_param(allow_ext_sg
, bool, 0444);
82 MODULE_PARM_DESC(allow_ext_sg
,
83 "Default behavior when there are more than cmd_sg_entries S/G entries after mapping; fails the request when false (default false)");
85 module_param(topspin_workarounds
, int, 0444);
86 MODULE_PARM_DESC(topspin_workarounds
,
87 "Enable workarounds for Topspin/Cisco SRP target bugs if != 0");
89 static void srp_add_one(struct ib_device
*device
);
90 static void srp_remove_one(struct ib_device
*device
);
91 static void srp_recv_completion(struct ib_cq
*cq
, void *target_ptr
);
92 static void srp_send_completion(struct ib_cq
*cq
, void *target_ptr
);
93 static int srp_cm_handler(struct ib_cm_id
*cm_id
, struct ib_cm_event
*event
);
95 static struct scsi_transport_template
*ib_srp_transport_template
;
96 static struct workqueue_struct
*srp_remove_wq
;
98 static struct ib_client srp_client
= {
101 .remove
= srp_remove_one
104 static struct ib_sa_client srp_sa_client
;
106 static inline struct srp_target_port
*host_to_target(struct Scsi_Host
*host
)
108 return (struct srp_target_port
*) host
->hostdata
;
111 static const char *srp_target_info(struct Scsi_Host
*host
)
113 return host_to_target(host
)->target_name
;
116 static int srp_target_is_topspin(struct srp_target_port
*target
)
118 static const u8 topspin_oui
[3] = { 0x00, 0x05, 0xad };
119 static const u8 cisco_oui
[3] = { 0x00, 0x1b, 0x0d };
121 return topspin_workarounds
&&
122 (!memcmp(&target
->ioc_guid
, topspin_oui
, sizeof topspin_oui
) ||
123 !memcmp(&target
->ioc_guid
, cisco_oui
, sizeof cisco_oui
));
126 static struct srp_iu
*srp_alloc_iu(struct srp_host
*host
, size_t size
,
128 enum dma_data_direction direction
)
132 iu
= kmalloc(sizeof *iu
, gfp_mask
);
136 iu
->buf
= kzalloc(size
, gfp_mask
);
140 iu
->dma
= ib_dma_map_single(host
->srp_dev
->dev
, iu
->buf
, size
,
142 if (ib_dma_mapping_error(host
->srp_dev
->dev
, iu
->dma
))
146 iu
->direction
= direction
;
158 static void srp_free_iu(struct srp_host
*host
, struct srp_iu
*iu
)
163 ib_dma_unmap_single(host
->srp_dev
->dev
, iu
->dma
, iu
->size
,
169 static void srp_qp_event(struct ib_event
*event
, void *context
)
171 pr_debug("QP event %d\n", event
->event
);
174 static int srp_init_qp(struct srp_target_port
*target
,
177 struct ib_qp_attr
*attr
;
180 attr
= kmalloc(sizeof *attr
, GFP_KERNEL
);
184 ret
= ib_find_pkey(target
->srp_host
->srp_dev
->dev
,
185 target
->srp_host
->port
,
186 be16_to_cpu(target
->path
.pkey
),
191 attr
->qp_state
= IB_QPS_INIT
;
192 attr
->qp_access_flags
= (IB_ACCESS_REMOTE_READ
|
193 IB_ACCESS_REMOTE_WRITE
);
194 attr
->port_num
= target
->srp_host
->port
;
196 ret
= ib_modify_qp(qp
, attr
,
207 static int srp_new_cm_id(struct srp_target_port
*target
)
209 struct ib_cm_id
*new_cm_id
;
211 new_cm_id
= ib_create_cm_id(target
->srp_host
->srp_dev
->dev
,
212 srp_cm_handler
, target
);
213 if (IS_ERR(new_cm_id
))
214 return PTR_ERR(new_cm_id
);
217 ib_destroy_cm_id(target
->cm_id
);
218 target
->cm_id
= new_cm_id
;
223 static int srp_create_target_ib(struct srp_target_port
*target
)
225 struct ib_qp_init_attr
*init_attr
;
226 struct ib_cq
*recv_cq
, *send_cq
;
230 init_attr
= kzalloc(sizeof *init_attr
, GFP_KERNEL
);
234 recv_cq
= ib_create_cq(target
->srp_host
->srp_dev
->dev
,
235 srp_recv_completion
, NULL
, target
, SRP_RQ_SIZE
, 0);
236 if (IS_ERR(recv_cq
)) {
237 ret
= PTR_ERR(recv_cq
);
241 send_cq
= ib_create_cq(target
->srp_host
->srp_dev
->dev
,
242 srp_send_completion
, NULL
, target
, SRP_SQ_SIZE
, 0);
243 if (IS_ERR(send_cq
)) {
244 ret
= PTR_ERR(send_cq
);
248 ib_req_notify_cq(recv_cq
, IB_CQ_NEXT_COMP
);
250 init_attr
->event_handler
= srp_qp_event
;
251 init_attr
->cap
.max_send_wr
= SRP_SQ_SIZE
;
252 init_attr
->cap
.max_recv_wr
= SRP_RQ_SIZE
;
253 init_attr
->cap
.max_recv_sge
= 1;
254 init_attr
->cap
.max_send_sge
= 1;
255 init_attr
->sq_sig_type
= IB_SIGNAL_ALL_WR
;
256 init_attr
->qp_type
= IB_QPT_RC
;
257 init_attr
->send_cq
= send_cq
;
258 init_attr
->recv_cq
= recv_cq
;
260 qp
= ib_create_qp(target
->srp_host
->srp_dev
->pd
, init_attr
);
266 ret
= srp_init_qp(target
, qp
);
271 ib_destroy_qp(target
->qp
);
273 ib_destroy_cq(target
->recv_cq
);
275 ib_destroy_cq(target
->send_cq
);
278 target
->recv_cq
= recv_cq
;
279 target
->send_cq
= send_cq
;
288 ib_destroy_cq(send_cq
);
291 ib_destroy_cq(recv_cq
);
298 static void srp_free_target_ib(struct srp_target_port
*target
)
302 ib_destroy_qp(target
->qp
);
303 ib_destroy_cq(target
->send_cq
);
304 ib_destroy_cq(target
->recv_cq
);
307 target
->send_cq
= target
->recv_cq
= NULL
;
309 for (i
= 0; i
< SRP_RQ_SIZE
; ++i
)
310 srp_free_iu(target
->srp_host
, target
->rx_ring
[i
]);
311 for (i
= 0; i
< SRP_SQ_SIZE
; ++i
)
312 srp_free_iu(target
->srp_host
, target
->tx_ring
[i
]);
315 static void srp_path_rec_completion(int status
,
316 struct ib_sa_path_rec
*pathrec
,
319 struct srp_target_port
*target
= target_ptr
;
321 target
->status
= status
;
323 shost_printk(KERN_ERR
, target
->scsi_host
,
324 PFX
"Got failed path rec status %d\n", status
);
326 target
->path
= *pathrec
;
327 complete(&target
->done
);
330 static int srp_lookup_path(struct srp_target_port
*target
)
332 target
->path
.numb_path
= 1;
334 init_completion(&target
->done
);
336 target
->path_query_id
= ib_sa_path_rec_get(&srp_sa_client
,
337 target
->srp_host
->srp_dev
->dev
,
338 target
->srp_host
->port
,
340 IB_SA_PATH_REC_SERVICE_ID
|
341 IB_SA_PATH_REC_DGID
|
342 IB_SA_PATH_REC_SGID
|
343 IB_SA_PATH_REC_NUMB_PATH
|
345 SRP_PATH_REC_TIMEOUT_MS
,
347 srp_path_rec_completion
,
348 target
, &target
->path_query
);
349 if (target
->path_query_id
< 0)
350 return target
->path_query_id
;
352 wait_for_completion(&target
->done
);
354 if (target
->status
< 0)
355 shost_printk(KERN_WARNING
, target
->scsi_host
,
356 PFX
"Path record query failed\n");
358 return target
->status
;
361 static int srp_send_req(struct srp_target_port
*target
)
364 struct ib_cm_req_param param
;
365 struct srp_login_req priv
;
369 req
= kzalloc(sizeof *req
, GFP_KERNEL
);
373 req
->param
.primary_path
= &target
->path
;
374 req
->param
.alternate_path
= NULL
;
375 req
->param
.service_id
= target
->service_id
;
376 req
->param
.qp_num
= target
->qp
->qp_num
;
377 req
->param
.qp_type
= target
->qp
->qp_type
;
378 req
->param
.private_data
= &req
->priv
;
379 req
->param
.private_data_len
= sizeof req
->priv
;
380 req
->param
.flow_control
= 1;
382 get_random_bytes(&req
->param
.starting_psn
, 4);
383 req
->param
.starting_psn
&= 0xffffff;
386 * Pick some arbitrary defaults here; we could make these
387 * module parameters if anyone cared about setting them.
389 req
->param
.responder_resources
= 4;
390 req
->param
.remote_cm_response_timeout
= 20;
391 req
->param
.local_cm_response_timeout
= 20;
392 req
->param
.retry_count
= 7;
393 req
->param
.rnr_retry_count
= 7;
394 req
->param
.max_cm_retries
= 15;
396 req
->priv
.opcode
= SRP_LOGIN_REQ
;
398 req
->priv
.req_it_iu_len
= cpu_to_be32(target
->max_iu_len
);
399 req
->priv
.req_buf_fmt
= cpu_to_be16(SRP_BUF_FORMAT_DIRECT
|
400 SRP_BUF_FORMAT_INDIRECT
);
402 * In the published SRP specification (draft rev. 16a), the
403 * port identifier format is 8 bytes of ID extension followed
404 * by 8 bytes of GUID. Older drafts put the two halves in the
405 * opposite order, so that the GUID comes first.
407 * Targets conforming to these obsolete drafts can be
408 * recognized by the I/O Class they report.
410 if (target
->io_class
== SRP_REV10_IB_IO_CLASS
) {
411 memcpy(req
->priv
.initiator_port_id
,
412 &target
->path
.sgid
.global
.interface_id
, 8);
413 memcpy(req
->priv
.initiator_port_id
+ 8,
414 &target
->initiator_ext
, 8);
415 memcpy(req
->priv
.target_port_id
, &target
->ioc_guid
, 8);
416 memcpy(req
->priv
.target_port_id
+ 8, &target
->id_ext
, 8);
418 memcpy(req
->priv
.initiator_port_id
,
419 &target
->initiator_ext
, 8);
420 memcpy(req
->priv
.initiator_port_id
+ 8,
421 &target
->path
.sgid
.global
.interface_id
, 8);
422 memcpy(req
->priv
.target_port_id
, &target
->id_ext
, 8);
423 memcpy(req
->priv
.target_port_id
+ 8, &target
->ioc_guid
, 8);
427 * Topspin/Cisco SRP targets will reject our login unless we
428 * zero out the first 8 bytes of our initiator port ID and set
429 * the second 8 bytes to the local node GUID.
431 if (srp_target_is_topspin(target
)) {
432 shost_printk(KERN_DEBUG
, target
->scsi_host
,
433 PFX
"Topspin/Cisco initiator port ID workaround "
434 "activated for target GUID %016llx\n",
435 (unsigned long long) be64_to_cpu(target
->ioc_guid
));
436 memset(req
->priv
.initiator_port_id
, 0, 8);
437 memcpy(req
->priv
.initiator_port_id
+ 8,
438 &target
->srp_host
->srp_dev
->dev
->node_guid
, 8);
441 status
= ib_send_cm_req(target
->cm_id
, &req
->param
);
448 static bool srp_queue_remove_work(struct srp_target_port
*target
)
450 bool changed
= false;
452 spin_lock_irq(&target
->lock
);
453 if (target
->state
!= SRP_TARGET_REMOVED
) {
454 target
->state
= SRP_TARGET_REMOVED
;
457 spin_unlock_irq(&target
->lock
);
460 queue_work(srp_remove_wq
, &target
->remove_work
);
465 static bool srp_change_conn_state(struct srp_target_port
*target
,
468 bool changed
= false;
470 spin_lock_irq(&target
->lock
);
471 if (target
->connected
!= connected
) {
472 target
->connected
= connected
;
475 spin_unlock_irq(&target
->lock
);
480 static void srp_disconnect_target(struct srp_target_port
*target
)
482 if (srp_change_conn_state(target
, false)) {
483 /* XXX should send SRP_I_LOGOUT request */
485 if (ib_send_cm_dreq(target
->cm_id
, NULL
, 0)) {
486 shost_printk(KERN_DEBUG
, target
->scsi_host
,
487 PFX
"Sending CM DREQ failed\n");
492 static void srp_free_req_data(struct srp_target_port
*target
)
494 struct ib_device
*ibdev
= target
->srp_host
->srp_dev
->dev
;
495 struct srp_request
*req
;
498 for (i
= 0, req
= target
->req_ring
; i
< SRP_CMD_SQ_SIZE
; ++i
, ++req
) {
499 kfree(req
->fmr_list
);
500 kfree(req
->map_page
);
501 if (req
->indirect_dma_addr
) {
502 ib_dma_unmap_single(ibdev
, req
->indirect_dma_addr
,
503 target
->indirect_size
,
506 kfree(req
->indirect_desc
);
511 * srp_del_scsi_host_attr() - Remove attributes defined in the host template.
512 * @shost: SCSI host whose attributes to remove from sysfs.
514 * Note: Any attributes defined in the host template and that did not exist
515 * before invocation of this function will be ignored.
517 static void srp_del_scsi_host_attr(struct Scsi_Host
*shost
)
519 struct device_attribute
**attr
;
521 for (attr
= shost
->hostt
->shost_attrs
; attr
&& *attr
; ++attr
)
522 device_remove_file(&shost
->shost_dev
, *attr
);
525 static void srp_remove_target(struct srp_target_port
*target
)
527 WARN_ON_ONCE(target
->state
!= SRP_TARGET_REMOVED
);
529 srp_del_scsi_host_attr(target
->scsi_host
);
530 srp_remove_host(target
->scsi_host
);
531 scsi_remove_host(target
->scsi_host
);
532 srp_disconnect_target(target
);
533 ib_destroy_cm_id(target
->cm_id
);
534 srp_free_target_ib(target
);
535 srp_free_req_data(target
);
536 scsi_host_put(target
->scsi_host
);
539 static void srp_remove_work(struct work_struct
*work
)
541 struct srp_target_port
*target
=
542 container_of(work
, struct srp_target_port
, remove_work
);
544 WARN_ON_ONCE(target
->state
!= SRP_TARGET_REMOVED
);
546 spin_lock(&target
->srp_host
->target_lock
);
547 list_del(&target
->list
);
548 spin_unlock(&target
->srp_host
->target_lock
);
550 srp_remove_target(target
);
553 static void srp_rport_delete(struct srp_rport
*rport
)
555 struct srp_target_port
*target
= rport
->lld_data
;
557 srp_queue_remove_work(target
);
560 static int srp_connect_target(struct srp_target_port
*target
)
565 WARN_ON_ONCE(target
->connected
);
567 target
->qp_in_error
= false;
569 ret
= srp_lookup_path(target
);
574 init_completion(&target
->done
);
575 ret
= srp_send_req(target
);
578 wait_for_completion(&target
->done
);
581 * The CM event handling code will set status to
582 * SRP_PORT_REDIRECT if we get a port redirect REJ
583 * back, or SRP_DLID_REDIRECT if we get a lid/qp
586 switch (target
->status
) {
588 srp_change_conn_state(target
, true);
591 case SRP_PORT_REDIRECT
:
592 ret
= srp_lookup_path(target
);
597 case SRP_DLID_REDIRECT
:
601 /* Our current CM id was stale, and is now in timewait.
602 * Try to reconnect with a new one.
604 if (!retries
-- || srp_new_cm_id(target
)) {
605 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
606 "giving up on stale connection\n");
607 target
->status
= -ECONNRESET
;
608 return target
->status
;
611 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
612 "retrying stale connection\n");
616 return target
->status
;
621 static void srp_unmap_data(struct scsi_cmnd
*scmnd
,
622 struct srp_target_port
*target
,
623 struct srp_request
*req
)
625 struct ib_device
*ibdev
= target
->srp_host
->srp_dev
->dev
;
626 struct ib_pool_fmr
**pfmr
;
628 if (!scsi_sglist(scmnd
) ||
629 (scmnd
->sc_data_direction
!= DMA_TO_DEVICE
&&
630 scmnd
->sc_data_direction
!= DMA_FROM_DEVICE
))
633 pfmr
= req
->fmr_list
;
635 ib_fmr_pool_unmap(*pfmr
++);
637 ib_dma_unmap_sg(ibdev
, scsi_sglist(scmnd
), scsi_sg_count(scmnd
),
638 scmnd
->sc_data_direction
);
642 * srp_claim_req - Take ownership of the scmnd associated with a request.
643 * @target: SRP target port.
645 * @scmnd: If NULL, take ownership of @req->scmnd. If not NULL, only take
646 * ownership of @req->scmnd if it equals @scmnd.
649 * Either NULL or a pointer to the SCSI command the caller became owner of.
651 static struct scsi_cmnd
*srp_claim_req(struct srp_target_port
*target
,
652 struct srp_request
*req
,
653 struct scsi_cmnd
*scmnd
)
657 spin_lock_irqsave(&target
->lock
, flags
);
661 } else if (req
->scmnd
== scmnd
) {
666 spin_unlock_irqrestore(&target
->lock
, flags
);
672 * srp_free_req() - Unmap data and add request to the free request list.
674 static void srp_free_req(struct srp_target_port
*target
,
675 struct srp_request
*req
, struct scsi_cmnd
*scmnd
,
680 srp_unmap_data(scmnd
, target
, req
);
682 spin_lock_irqsave(&target
->lock
, flags
);
683 target
->req_lim
+= req_lim_delta
;
684 list_add_tail(&req
->list
, &target
->free_reqs
);
685 spin_unlock_irqrestore(&target
->lock
, flags
);
688 static void srp_reset_req(struct srp_target_port
*target
, struct srp_request
*req
)
690 struct scsi_cmnd
*scmnd
= srp_claim_req(target
, req
, NULL
);
693 srp_free_req(target
, req
, scmnd
, 0);
694 scmnd
->result
= DID_RESET
<< 16;
695 scmnd
->scsi_done(scmnd
);
699 static int srp_reconnect_target(struct srp_target_port
*target
)
701 struct Scsi_Host
*shost
= target
->scsi_host
;
704 scsi_target_block(&shost
->shost_gendev
);
706 srp_disconnect_target(target
);
708 * Now get a new local CM ID so that we avoid confusing the target in
709 * case things are really fouled up. Doing so also ensures that all CM
710 * callbacks will have finished before a new QP is allocated.
712 ret
= srp_new_cm_id(target
);
714 * Whether or not creating a new CM ID succeeded, create a new
715 * QP. This guarantees that all completion callback function
716 * invocations have finished before request resetting starts.
719 ret
= srp_create_target_ib(target
);
721 srp_create_target_ib(target
);
723 for (i
= 0; i
< SRP_CMD_SQ_SIZE
; ++i
) {
724 struct srp_request
*req
= &target
->req_ring
[i
];
726 srp_reset_req(target
, req
);
729 INIT_LIST_HEAD(&target
->free_tx
);
730 for (i
= 0; i
< SRP_SQ_SIZE
; ++i
)
731 list_add(&target
->tx_ring
[i
]->list
, &target
->free_tx
);
734 ret
= srp_connect_target(target
);
736 scsi_target_unblock(&shost
->shost_gendev
, ret
== 0 ? SDEV_RUNNING
:
737 SDEV_TRANSPORT_OFFLINE
);
738 target
->transport_offline
= !!ret
;
743 shost_printk(KERN_INFO
, target
->scsi_host
, PFX
"reconnect succeeded\n");
748 shost_printk(KERN_ERR
, target
->scsi_host
,
749 PFX
"reconnect failed (%d), removing target port.\n", ret
);
752 * We couldn't reconnect, so kill our target port off.
753 * However, we have to defer the real removal because we
754 * are in the context of the SCSI error handler now, which
755 * will deadlock if we call scsi_remove_host().
757 srp_queue_remove_work(target
);
762 static void srp_map_desc(struct srp_map_state
*state
, dma_addr_t dma_addr
,
763 unsigned int dma_len
, u32 rkey
)
765 struct srp_direct_buf
*desc
= state
->desc
;
767 desc
->va
= cpu_to_be64(dma_addr
);
768 desc
->key
= cpu_to_be32(rkey
);
769 desc
->len
= cpu_to_be32(dma_len
);
771 state
->total_len
+= dma_len
;
776 static int srp_map_finish_fmr(struct srp_map_state
*state
,
777 struct srp_target_port
*target
)
779 struct srp_device
*dev
= target
->srp_host
->srp_dev
;
780 struct ib_pool_fmr
*fmr
;
786 if (state
->npages
== 1) {
787 srp_map_desc(state
, state
->base_dma_addr
, state
->fmr_len
,
789 state
->npages
= state
->fmr_len
= 0;
793 fmr
= ib_fmr_pool_map_phys(dev
->fmr_pool
, state
->pages
,
794 state
->npages
, io_addr
);
798 *state
->next_fmr
++ = fmr
;
801 srp_map_desc(state
, 0, state
->fmr_len
, fmr
->fmr
->rkey
);
802 state
->npages
= state
->fmr_len
= 0;
806 static void srp_map_update_start(struct srp_map_state
*state
,
807 struct scatterlist
*sg
, int sg_index
,
810 state
->unmapped_sg
= sg
;
811 state
->unmapped_index
= sg_index
;
812 state
->unmapped_addr
= dma_addr
;
815 static int srp_map_sg_entry(struct srp_map_state
*state
,
816 struct srp_target_port
*target
,
817 struct scatterlist
*sg
, int sg_index
,
820 struct srp_device
*dev
= target
->srp_host
->srp_dev
;
821 struct ib_device
*ibdev
= dev
->dev
;
822 dma_addr_t dma_addr
= ib_sg_dma_address(ibdev
, sg
);
823 unsigned int dma_len
= ib_sg_dma_len(ibdev
, sg
);
830 if (use_fmr
== SRP_MAP_NO_FMR
) {
831 /* Once we're in direct map mode for a request, we don't
832 * go back to FMR mode, so no need to update anything
833 * other than the descriptor.
835 srp_map_desc(state
, dma_addr
, dma_len
, target
->rkey
);
839 /* If we start at an offset into the FMR page, don't merge into
840 * the current FMR. Finish it out, and use the kernel's MR for this
841 * sg entry. This is to avoid potential bugs on some SRP targets
842 * that were never quite defined, but went away when the initiator
843 * avoided using FMR on such page fragments.
845 if (dma_addr
& ~dev
->fmr_page_mask
|| dma_len
> dev
->fmr_max_size
) {
846 ret
= srp_map_finish_fmr(state
, target
);
850 srp_map_desc(state
, dma_addr
, dma_len
, target
->rkey
);
851 srp_map_update_start(state
, NULL
, 0, 0);
855 /* If this is the first sg to go into the FMR, save our position.
856 * We need to know the first unmapped entry, its index, and the
857 * first unmapped address within that entry to be able to restart
858 * mapping after an error.
860 if (!state
->unmapped_sg
)
861 srp_map_update_start(state
, sg
, sg_index
, dma_addr
);
864 if (state
->npages
== SRP_FMR_SIZE
) {
865 ret
= srp_map_finish_fmr(state
, target
);
869 srp_map_update_start(state
, sg
, sg_index
, dma_addr
);
872 len
= min_t(unsigned int, dma_len
, dev
->fmr_page_size
);
875 state
->base_dma_addr
= dma_addr
;
876 state
->pages
[state
->npages
++] = dma_addr
;
877 state
->fmr_len
+= len
;
882 /* If the last entry of the FMR wasn't a full page, then we need to
883 * close it out and start a new one -- we can only merge at page
887 if (len
!= dev
->fmr_page_size
) {
888 ret
= srp_map_finish_fmr(state
, target
);
890 srp_map_update_start(state
, NULL
, 0, 0);
895 static int srp_map_data(struct scsi_cmnd
*scmnd
, struct srp_target_port
*target
,
896 struct srp_request
*req
)
898 struct scatterlist
*scat
, *sg
;
899 struct srp_cmd
*cmd
= req
->cmd
->buf
;
900 int i
, len
, nents
, count
, use_fmr
;
901 struct srp_device
*dev
;
902 struct ib_device
*ibdev
;
903 struct srp_map_state state
;
904 struct srp_indirect_buf
*indirect_hdr
;
908 if (!scsi_sglist(scmnd
) || scmnd
->sc_data_direction
== DMA_NONE
)
909 return sizeof (struct srp_cmd
);
911 if (scmnd
->sc_data_direction
!= DMA_FROM_DEVICE
&&
912 scmnd
->sc_data_direction
!= DMA_TO_DEVICE
) {
913 shost_printk(KERN_WARNING
, target
->scsi_host
,
914 PFX
"Unhandled data direction %d\n",
915 scmnd
->sc_data_direction
);
919 nents
= scsi_sg_count(scmnd
);
920 scat
= scsi_sglist(scmnd
);
922 dev
= target
->srp_host
->srp_dev
;
925 count
= ib_dma_map_sg(ibdev
, scat
, nents
, scmnd
->sc_data_direction
);
926 if (unlikely(count
== 0))
929 fmt
= SRP_DATA_DESC_DIRECT
;
930 len
= sizeof (struct srp_cmd
) + sizeof (struct srp_direct_buf
);
934 * The midlayer only generated a single gather/scatter
935 * entry, or DMA mapping coalesced everything to a
936 * single entry. So a direct descriptor along with
937 * the DMA MR suffices.
939 struct srp_direct_buf
*buf
= (void *) cmd
->add_data
;
941 buf
->va
= cpu_to_be64(ib_sg_dma_address(ibdev
, scat
));
942 buf
->key
= cpu_to_be32(target
->rkey
);
943 buf
->len
= cpu_to_be32(ib_sg_dma_len(ibdev
, scat
));
949 /* We have more than one scatter/gather entry, so build our indirect
950 * descriptor table, trying to merge as many entries with FMR as we
953 indirect_hdr
= (void *) cmd
->add_data
;
955 ib_dma_sync_single_for_cpu(ibdev
, req
->indirect_dma_addr
,
956 target
->indirect_size
, DMA_TO_DEVICE
);
958 memset(&state
, 0, sizeof(state
));
959 state
.desc
= req
->indirect_desc
;
960 state
.pages
= req
->map_page
;
961 state
.next_fmr
= req
->fmr_list
;
963 use_fmr
= dev
->fmr_pool
? SRP_MAP_ALLOW_FMR
: SRP_MAP_NO_FMR
;
965 for_each_sg(scat
, sg
, count
, i
) {
966 if (srp_map_sg_entry(&state
, target
, sg
, i
, use_fmr
)) {
967 /* FMR mapping failed, so backtrack to the first
968 * unmapped entry and continue on without using FMR.
971 unsigned int dma_len
;
974 sg
= state
.unmapped_sg
;
975 i
= state
.unmapped_index
;
977 dma_addr
= ib_sg_dma_address(ibdev
, sg
);
978 dma_len
= ib_sg_dma_len(ibdev
, sg
);
979 dma_len
-= (state
.unmapped_addr
- dma_addr
);
980 dma_addr
= state
.unmapped_addr
;
981 use_fmr
= SRP_MAP_NO_FMR
;
982 srp_map_desc(&state
, dma_addr
, dma_len
, target
->rkey
);
986 if (use_fmr
== SRP_MAP_ALLOW_FMR
&& srp_map_finish_fmr(&state
, target
))
989 /* We've mapped the request, now pull as much of the indirect
990 * descriptor table as we can into the command buffer. If this
991 * target is not using an external indirect table, we are
992 * guaranteed to fit into the command, as the SCSI layer won't
993 * give us more S/G entries than we allow.
995 req
->nfmr
= state
.nfmr
;
996 if (state
.ndesc
== 1) {
997 /* FMR mapping was able to collapse this to one entry,
998 * so use a direct descriptor.
1000 struct srp_direct_buf
*buf
= (void *) cmd
->add_data
;
1002 *buf
= req
->indirect_desc
[0];
1006 if (unlikely(target
->cmd_sg_cnt
< state
.ndesc
&&
1007 !target
->allow_ext_sg
)) {
1008 shost_printk(KERN_ERR
, target
->scsi_host
,
1009 "Could not fit S/G list into SRP_CMD\n");
1013 count
= min(state
.ndesc
, target
->cmd_sg_cnt
);
1014 table_len
= state
.ndesc
* sizeof (struct srp_direct_buf
);
1016 fmt
= SRP_DATA_DESC_INDIRECT
;
1017 len
= sizeof(struct srp_cmd
) + sizeof (struct srp_indirect_buf
);
1018 len
+= count
* sizeof (struct srp_direct_buf
);
1020 memcpy(indirect_hdr
->desc_list
, req
->indirect_desc
,
1021 count
* sizeof (struct srp_direct_buf
));
1023 indirect_hdr
->table_desc
.va
= cpu_to_be64(req
->indirect_dma_addr
);
1024 indirect_hdr
->table_desc
.key
= cpu_to_be32(target
->rkey
);
1025 indirect_hdr
->table_desc
.len
= cpu_to_be32(table_len
);
1026 indirect_hdr
->len
= cpu_to_be32(state
.total_len
);
1028 if (scmnd
->sc_data_direction
== DMA_TO_DEVICE
)
1029 cmd
->data_out_desc_cnt
= count
;
1031 cmd
->data_in_desc_cnt
= count
;
1033 ib_dma_sync_single_for_device(ibdev
, req
->indirect_dma_addr
, table_len
,
1037 if (scmnd
->sc_data_direction
== DMA_TO_DEVICE
)
1038 cmd
->buf_fmt
= fmt
<< 4;
1046 * Return an IU and possible credit to the free pool
1048 static void srp_put_tx_iu(struct srp_target_port
*target
, struct srp_iu
*iu
,
1049 enum srp_iu_type iu_type
)
1051 unsigned long flags
;
1053 spin_lock_irqsave(&target
->lock
, flags
);
1054 list_add(&iu
->list
, &target
->free_tx
);
1055 if (iu_type
!= SRP_IU_RSP
)
1057 spin_unlock_irqrestore(&target
->lock
, flags
);
1061 * Must be called with target->lock held to protect req_lim and free_tx.
1062 * If IU is not sent, it must be returned using srp_put_tx_iu().
1065 * An upper limit for the number of allocated information units for each
1067 * - SRP_IU_CMD: SRP_CMD_SQ_SIZE, since the SCSI mid-layer never queues
1068 * more than Scsi_Host.can_queue requests.
1069 * - SRP_IU_TSK_MGMT: SRP_TSK_MGMT_SQ_SIZE.
1070 * - SRP_IU_RSP: 1, since a conforming SRP target never sends more than
1071 * one unanswered SRP request to an initiator.
1073 static struct srp_iu
*__srp_get_tx_iu(struct srp_target_port
*target
,
1074 enum srp_iu_type iu_type
)
1076 s32 rsv
= (iu_type
== SRP_IU_TSK_MGMT
) ? 0 : SRP_TSK_MGMT_SQ_SIZE
;
1079 srp_send_completion(target
->send_cq
, target
);
1081 if (list_empty(&target
->free_tx
))
1084 /* Initiator responses to target requests do not consume credits */
1085 if (iu_type
!= SRP_IU_RSP
) {
1086 if (target
->req_lim
<= rsv
) {
1087 ++target
->zero_req_lim
;
1094 iu
= list_first_entry(&target
->free_tx
, struct srp_iu
, list
);
1095 list_del(&iu
->list
);
1099 static int srp_post_send(struct srp_target_port
*target
,
1100 struct srp_iu
*iu
, int len
)
1103 struct ib_send_wr wr
, *bad_wr
;
1105 list
.addr
= iu
->dma
;
1107 list
.lkey
= target
->lkey
;
1110 wr
.wr_id
= (uintptr_t) iu
;
1113 wr
.opcode
= IB_WR_SEND
;
1114 wr
.send_flags
= IB_SEND_SIGNALED
;
1116 return ib_post_send(target
->qp
, &wr
, &bad_wr
);
1119 static int srp_post_recv(struct srp_target_port
*target
, struct srp_iu
*iu
)
1121 struct ib_recv_wr wr
, *bad_wr
;
1124 list
.addr
= iu
->dma
;
1125 list
.length
= iu
->size
;
1126 list
.lkey
= target
->lkey
;
1129 wr
.wr_id
= (uintptr_t) iu
;
1133 return ib_post_recv(target
->qp
, &wr
, &bad_wr
);
1136 static void srp_process_rsp(struct srp_target_port
*target
, struct srp_rsp
*rsp
)
1138 struct srp_request
*req
;
1139 struct scsi_cmnd
*scmnd
;
1140 unsigned long flags
;
1142 if (unlikely(rsp
->tag
& SRP_TAG_TSK_MGMT
)) {
1143 spin_lock_irqsave(&target
->lock
, flags
);
1144 target
->req_lim
+= be32_to_cpu(rsp
->req_lim_delta
);
1145 spin_unlock_irqrestore(&target
->lock
, flags
);
1147 target
->tsk_mgmt_status
= -1;
1148 if (be32_to_cpu(rsp
->resp_data_len
) >= 4)
1149 target
->tsk_mgmt_status
= rsp
->data
[3];
1150 complete(&target
->tsk_mgmt_done
);
1152 req
= &target
->req_ring
[rsp
->tag
];
1153 scmnd
= srp_claim_req(target
, req
, NULL
);
1155 shost_printk(KERN_ERR
, target
->scsi_host
,
1156 "Null scmnd for RSP w/tag %016llx\n",
1157 (unsigned long long) rsp
->tag
);
1159 spin_lock_irqsave(&target
->lock
, flags
);
1160 target
->req_lim
+= be32_to_cpu(rsp
->req_lim_delta
);
1161 spin_unlock_irqrestore(&target
->lock
, flags
);
1165 scmnd
->result
= rsp
->status
;
1167 if (rsp
->flags
& SRP_RSP_FLAG_SNSVALID
) {
1168 memcpy(scmnd
->sense_buffer
, rsp
->data
+
1169 be32_to_cpu(rsp
->resp_data_len
),
1170 min_t(int, be32_to_cpu(rsp
->sense_data_len
),
1171 SCSI_SENSE_BUFFERSIZE
));
1174 if (rsp
->flags
& (SRP_RSP_FLAG_DOOVER
| SRP_RSP_FLAG_DOUNDER
))
1175 scsi_set_resid(scmnd
, be32_to_cpu(rsp
->data_out_res_cnt
));
1176 else if (rsp
->flags
& (SRP_RSP_FLAG_DIOVER
| SRP_RSP_FLAG_DIUNDER
))
1177 scsi_set_resid(scmnd
, be32_to_cpu(rsp
->data_in_res_cnt
));
1179 srp_free_req(target
, req
, scmnd
,
1180 be32_to_cpu(rsp
->req_lim_delta
));
1182 scmnd
->host_scribble
= NULL
;
1183 scmnd
->scsi_done(scmnd
);
1187 static int srp_response_common(struct srp_target_port
*target
, s32 req_delta
,
1190 struct ib_device
*dev
= target
->srp_host
->srp_dev
->dev
;
1191 unsigned long flags
;
1195 spin_lock_irqsave(&target
->lock
, flags
);
1196 target
->req_lim
+= req_delta
;
1197 iu
= __srp_get_tx_iu(target
, SRP_IU_RSP
);
1198 spin_unlock_irqrestore(&target
->lock
, flags
);
1201 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
1202 "no IU available to send response\n");
1206 ib_dma_sync_single_for_cpu(dev
, iu
->dma
, len
, DMA_TO_DEVICE
);
1207 memcpy(iu
->buf
, rsp
, len
);
1208 ib_dma_sync_single_for_device(dev
, iu
->dma
, len
, DMA_TO_DEVICE
);
1210 err
= srp_post_send(target
, iu
, len
);
1212 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
1213 "unable to post response: %d\n", err
);
1214 srp_put_tx_iu(target
, iu
, SRP_IU_RSP
);
1220 static void srp_process_cred_req(struct srp_target_port
*target
,
1221 struct srp_cred_req
*req
)
1223 struct srp_cred_rsp rsp
= {
1224 .opcode
= SRP_CRED_RSP
,
1227 s32 delta
= be32_to_cpu(req
->req_lim_delta
);
1229 if (srp_response_common(target
, delta
, &rsp
, sizeof rsp
))
1230 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
1231 "problems processing SRP_CRED_REQ\n");
1234 static void srp_process_aer_req(struct srp_target_port
*target
,
1235 struct srp_aer_req
*req
)
1237 struct srp_aer_rsp rsp
= {
1238 .opcode
= SRP_AER_RSP
,
1241 s32 delta
= be32_to_cpu(req
->req_lim_delta
);
1243 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
1244 "ignoring AER for LUN %llu\n", be64_to_cpu(req
->lun
));
1246 if (srp_response_common(target
, delta
, &rsp
, sizeof rsp
))
1247 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
1248 "problems processing SRP_AER_REQ\n");
1251 static void srp_handle_recv(struct srp_target_port
*target
, struct ib_wc
*wc
)
1253 struct ib_device
*dev
= target
->srp_host
->srp_dev
->dev
;
1254 struct srp_iu
*iu
= (struct srp_iu
*) (uintptr_t) wc
->wr_id
;
1258 ib_dma_sync_single_for_cpu(dev
, iu
->dma
, target
->max_ti_iu_len
,
1261 opcode
= *(u8
*) iu
->buf
;
1264 shost_printk(KERN_ERR
, target
->scsi_host
,
1265 PFX
"recv completion, opcode 0x%02x\n", opcode
);
1266 print_hex_dump(KERN_ERR
, "", DUMP_PREFIX_OFFSET
, 8, 1,
1267 iu
->buf
, wc
->byte_len
, true);
1272 srp_process_rsp(target
, iu
->buf
);
1276 srp_process_cred_req(target
, iu
->buf
);
1280 srp_process_aer_req(target
, iu
->buf
);
1284 /* XXX Handle target logout */
1285 shost_printk(KERN_WARNING
, target
->scsi_host
,
1286 PFX
"Got target logout request\n");
1290 shost_printk(KERN_WARNING
, target
->scsi_host
,
1291 PFX
"Unhandled SRP opcode 0x%02x\n", opcode
);
1295 ib_dma_sync_single_for_device(dev
, iu
->dma
, target
->max_ti_iu_len
,
1298 res
= srp_post_recv(target
, iu
);
1300 shost_printk(KERN_ERR
, target
->scsi_host
,
1301 PFX
"Recv failed with error code %d\n", res
);
1304 static void srp_handle_qp_err(enum ib_wc_status wc_status
, bool send_err
,
1305 struct srp_target_port
*target
)
1307 if (target
->connected
&& !target
->qp_in_error
) {
1308 shost_printk(KERN_ERR
, target
->scsi_host
,
1309 PFX
"failed %s status %d\n",
1310 send_err
? "send" : "receive",
1313 target
->qp_in_error
= true;
1316 static void srp_recv_completion(struct ib_cq
*cq
, void *target_ptr
)
1318 struct srp_target_port
*target
= target_ptr
;
1321 ib_req_notify_cq(cq
, IB_CQ_NEXT_COMP
);
1322 while (ib_poll_cq(cq
, 1, &wc
) > 0) {
1323 if (likely(wc
.status
== IB_WC_SUCCESS
)) {
1324 srp_handle_recv(target
, &wc
);
1326 srp_handle_qp_err(wc
.status
, false, target
);
1331 static void srp_send_completion(struct ib_cq
*cq
, void *target_ptr
)
1333 struct srp_target_port
*target
= target_ptr
;
1337 while (ib_poll_cq(cq
, 1, &wc
) > 0) {
1338 if (likely(wc
.status
== IB_WC_SUCCESS
)) {
1339 iu
= (struct srp_iu
*) (uintptr_t) wc
.wr_id
;
1340 list_add(&iu
->list
, &target
->free_tx
);
1342 srp_handle_qp_err(wc
.status
, true, target
);
1347 static int srp_queuecommand(struct Scsi_Host
*shost
, struct scsi_cmnd
*scmnd
)
1349 struct srp_target_port
*target
= host_to_target(shost
);
1350 struct srp_request
*req
;
1352 struct srp_cmd
*cmd
;
1353 struct ib_device
*dev
;
1354 unsigned long flags
;
1357 if (unlikely(target
->transport_offline
)) {
1358 scmnd
->result
= DID_NO_CONNECT
<< 16;
1359 scmnd
->scsi_done(scmnd
);
1363 spin_lock_irqsave(&target
->lock
, flags
);
1364 iu
= __srp_get_tx_iu(target
, SRP_IU_CMD
);
1368 req
= list_first_entry(&target
->free_reqs
, struct srp_request
, list
);
1369 list_del(&req
->list
);
1370 spin_unlock_irqrestore(&target
->lock
, flags
);
1372 dev
= target
->srp_host
->srp_dev
->dev
;
1373 ib_dma_sync_single_for_cpu(dev
, iu
->dma
, target
->max_iu_len
,
1377 scmnd
->host_scribble
= (void *) req
;
1380 memset(cmd
, 0, sizeof *cmd
);
1382 cmd
->opcode
= SRP_CMD
;
1383 cmd
->lun
= cpu_to_be64((u64
) scmnd
->device
->lun
<< 48);
1384 cmd
->tag
= req
->index
;
1385 memcpy(cmd
->cdb
, scmnd
->cmnd
, scmnd
->cmd_len
);
1390 len
= srp_map_data(scmnd
, target
, req
);
1392 shost_printk(KERN_ERR
, target
->scsi_host
,
1393 PFX
"Failed to map data\n");
1397 ib_dma_sync_single_for_device(dev
, iu
->dma
, target
->max_iu_len
,
1400 if (srp_post_send(target
, iu
, len
)) {
1401 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
"Send failed\n");
1408 srp_unmap_data(scmnd
, target
, req
);
1411 srp_put_tx_iu(target
, iu
, SRP_IU_CMD
);
1414 * Avoid that the loops that iterate over the request ring can
1415 * encounter a dangling SCSI command pointer.
1419 spin_lock_irqsave(&target
->lock
, flags
);
1420 list_add(&req
->list
, &target
->free_reqs
);
1423 spin_unlock_irqrestore(&target
->lock
, flags
);
1425 return SCSI_MLQUEUE_HOST_BUSY
;
1428 static int srp_alloc_iu_bufs(struct srp_target_port
*target
)
1432 for (i
= 0; i
< SRP_RQ_SIZE
; ++i
) {
1433 target
->rx_ring
[i
] = srp_alloc_iu(target
->srp_host
,
1434 target
->max_ti_iu_len
,
1435 GFP_KERNEL
, DMA_FROM_DEVICE
);
1436 if (!target
->rx_ring
[i
])
1440 for (i
= 0; i
< SRP_SQ_SIZE
; ++i
) {
1441 target
->tx_ring
[i
] = srp_alloc_iu(target
->srp_host
,
1443 GFP_KERNEL
, DMA_TO_DEVICE
);
1444 if (!target
->tx_ring
[i
])
1447 list_add(&target
->tx_ring
[i
]->list
, &target
->free_tx
);
1453 for (i
= 0; i
< SRP_RQ_SIZE
; ++i
) {
1454 srp_free_iu(target
->srp_host
, target
->rx_ring
[i
]);
1455 target
->rx_ring
[i
] = NULL
;
1458 for (i
= 0; i
< SRP_SQ_SIZE
; ++i
) {
1459 srp_free_iu(target
->srp_host
, target
->tx_ring
[i
]);
1460 target
->tx_ring
[i
] = NULL
;
1466 static uint32_t srp_compute_rq_tmo(struct ib_qp_attr
*qp_attr
, int attr_mask
)
1468 uint64_t T_tr_ns
, max_compl_time_ms
;
1469 uint32_t rq_tmo_jiffies
;
1472 * According to section 11.2.4.2 in the IBTA spec (Modify Queue Pair,
1473 * table 91), both the QP timeout and the retry count have to be set
1474 * for RC QP's during the RTR to RTS transition.
1476 WARN_ON_ONCE((attr_mask
& (IB_QP_TIMEOUT
| IB_QP_RETRY_CNT
)) !=
1477 (IB_QP_TIMEOUT
| IB_QP_RETRY_CNT
));
1480 * Set target->rq_tmo_jiffies to one second more than the largest time
1481 * it can take before an error completion is generated. See also
1482 * C9-140..142 in the IBTA spec for more information about how to
1483 * convert the QP Local ACK Timeout value to nanoseconds.
1485 T_tr_ns
= 4096 * (1ULL << qp_attr
->timeout
);
1486 max_compl_time_ms
= qp_attr
->retry_cnt
* 4 * T_tr_ns
;
1487 do_div(max_compl_time_ms
, NSEC_PER_MSEC
);
1488 rq_tmo_jiffies
= msecs_to_jiffies(max_compl_time_ms
+ 1000);
1490 return rq_tmo_jiffies
;
1493 static void srp_cm_rep_handler(struct ib_cm_id
*cm_id
,
1494 struct srp_login_rsp
*lrsp
,
1495 struct srp_target_port
*target
)
1497 struct ib_qp_attr
*qp_attr
= NULL
;
1502 if (lrsp
->opcode
== SRP_LOGIN_RSP
) {
1503 target
->max_ti_iu_len
= be32_to_cpu(lrsp
->max_ti_iu_len
);
1504 target
->req_lim
= be32_to_cpu(lrsp
->req_lim_delta
);
1507 * Reserve credits for task management so we don't
1508 * bounce requests back to the SCSI mid-layer.
1510 target
->scsi_host
->can_queue
1511 = min(target
->req_lim
- SRP_TSK_MGMT_SQ_SIZE
,
1512 target
->scsi_host
->can_queue
);
1514 shost_printk(KERN_WARNING
, target
->scsi_host
,
1515 PFX
"Unhandled RSP opcode %#x\n", lrsp
->opcode
);
1520 if (!target
->rx_ring
[0]) {
1521 ret
= srp_alloc_iu_bufs(target
);
1527 qp_attr
= kmalloc(sizeof *qp_attr
, GFP_KERNEL
);
1531 qp_attr
->qp_state
= IB_QPS_RTR
;
1532 ret
= ib_cm_init_qp_attr(cm_id
, qp_attr
, &attr_mask
);
1536 ret
= ib_modify_qp(target
->qp
, qp_attr
, attr_mask
);
1540 for (i
= 0; i
< SRP_RQ_SIZE
; i
++) {
1541 struct srp_iu
*iu
= target
->rx_ring
[i
];
1542 ret
= srp_post_recv(target
, iu
);
1547 qp_attr
->qp_state
= IB_QPS_RTS
;
1548 ret
= ib_cm_init_qp_attr(cm_id
, qp_attr
, &attr_mask
);
1552 target
->rq_tmo_jiffies
= srp_compute_rq_tmo(qp_attr
, attr_mask
);
1554 ret
= ib_modify_qp(target
->qp
, qp_attr
, attr_mask
);
1558 ret
= ib_send_cm_rtu(cm_id
, NULL
, 0);
1564 target
->status
= ret
;
1567 static void srp_cm_rej_handler(struct ib_cm_id
*cm_id
,
1568 struct ib_cm_event
*event
,
1569 struct srp_target_port
*target
)
1571 struct Scsi_Host
*shost
= target
->scsi_host
;
1572 struct ib_class_port_info
*cpi
;
1575 switch (event
->param
.rej_rcvd
.reason
) {
1576 case IB_CM_REJ_PORT_CM_REDIRECT
:
1577 cpi
= event
->param
.rej_rcvd
.ari
;
1578 target
->path
.dlid
= cpi
->redirect_lid
;
1579 target
->path
.pkey
= cpi
->redirect_pkey
;
1580 cm_id
->remote_cm_qpn
= be32_to_cpu(cpi
->redirect_qp
) & 0x00ffffff;
1581 memcpy(target
->path
.dgid
.raw
, cpi
->redirect_gid
, 16);
1583 target
->status
= target
->path
.dlid
?
1584 SRP_DLID_REDIRECT
: SRP_PORT_REDIRECT
;
1587 case IB_CM_REJ_PORT_REDIRECT
:
1588 if (srp_target_is_topspin(target
)) {
1590 * Topspin/Cisco SRP gateways incorrectly send
1591 * reject reason code 25 when they mean 24
1594 memcpy(target
->path
.dgid
.raw
,
1595 event
->param
.rej_rcvd
.ari
, 16);
1597 shost_printk(KERN_DEBUG
, shost
,
1598 PFX
"Topspin/Cisco redirect to target port GID %016llx%016llx\n",
1599 (unsigned long long) be64_to_cpu(target
->path
.dgid
.global
.subnet_prefix
),
1600 (unsigned long long) be64_to_cpu(target
->path
.dgid
.global
.interface_id
));
1602 target
->status
= SRP_PORT_REDIRECT
;
1604 shost_printk(KERN_WARNING
, shost
,
1605 " REJ reason: IB_CM_REJ_PORT_REDIRECT\n");
1606 target
->status
= -ECONNRESET
;
1610 case IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID
:
1611 shost_printk(KERN_WARNING
, shost
,
1612 " REJ reason: IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID\n");
1613 target
->status
= -ECONNRESET
;
1616 case IB_CM_REJ_CONSUMER_DEFINED
:
1617 opcode
= *(u8
*) event
->private_data
;
1618 if (opcode
== SRP_LOGIN_REJ
) {
1619 struct srp_login_rej
*rej
= event
->private_data
;
1620 u32 reason
= be32_to_cpu(rej
->reason
);
1622 if (reason
== SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE
)
1623 shost_printk(KERN_WARNING
, shost
,
1624 PFX
"SRP_LOGIN_REJ: requested max_it_iu_len too large\n");
1626 shost_printk(KERN_WARNING
, shost
,
1627 PFX
"SRP LOGIN REJECTED, reason 0x%08x\n", reason
);
1629 shost_printk(KERN_WARNING
, shost
,
1630 " REJ reason: IB_CM_REJ_CONSUMER_DEFINED,"
1631 " opcode 0x%02x\n", opcode
);
1632 target
->status
= -ECONNRESET
;
1635 case IB_CM_REJ_STALE_CONN
:
1636 shost_printk(KERN_WARNING
, shost
, " REJ reason: stale connection\n");
1637 target
->status
= SRP_STALE_CONN
;
1641 shost_printk(KERN_WARNING
, shost
, " REJ reason 0x%x\n",
1642 event
->param
.rej_rcvd
.reason
);
1643 target
->status
= -ECONNRESET
;
1647 static int srp_cm_handler(struct ib_cm_id
*cm_id
, struct ib_cm_event
*event
)
1649 struct srp_target_port
*target
= cm_id
->context
;
1652 switch (event
->event
) {
1653 case IB_CM_REQ_ERROR
:
1654 shost_printk(KERN_DEBUG
, target
->scsi_host
,
1655 PFX
"Sending CM REQ failed\n");
1657 target
->status
= -ECONNRESET
;
1660 case IB_CM_REP_RECEIVED
:
1662 srp_cm_rep_handler(cm_id
, event
->private_data
, target
);
1665 case IB_CM_REJ_RECEIVED
:
1666 shost_printk(KERN_DEBUG
, target
->scsi_host
, PFX
"REJ received\n");
1669 srp_cm_rej_handler(cm_id
, event
, target
);
1672 case IB_CM_DREQ_RECEIVED
:
1673 shost_printk(KERN_WARNING
, target
->scsi_host
,
1674 PFX
"DREQ received - connection closed\n");
1675 srp_change_conn_state(target
, false);
1676 if (ib_send_cm_drep(cm_id
, NULL
, 0))
1677 shost_printk(KERN_ERR
, target
->scsi_host
,
1678 PFX
"Sending CM DREP failed\n");
1681 case IB_CM_TIMEWAIT_EXIT
:
1682 shost_printk(KERN_ERR
, target
->scsi_host
,
1683 PFX
"connection closed\n");
1688 case IB_CM_MRA_RECEIVED
:
1689 case IB_CM_DREQ_ERROR
:
1690 case IB_CM_DREP_RECEIVED
:
1694 shost_printk(KERN_WARNING
, target
->scsi_host
,
1695 PFX
"Unhandled CM event %d\n", event
->event
);
1700 complete(&target
->done
);
1705 static int srp_send_tsk_mgmt(struct srp_target_port
*target
,
1706 u64 req_tag
, unsigned int lun
, u8 func
)
1708 struct ib_device
*dev
= target
->srp_host
->srp_dev
->dev
;
1710 struct srp_tsk_mgmt
*tsk_mgmt
;
1712 if (!target
->connected
|| target
->qp_in_error
)
1715 init_completion(&target
->tsk_mgmt_done
);
1717 spin_lock_irq(&target
->lock
);
1718 iu
= __srp_get_tx_iu(target
, SRP_IU_TSK_MGMT
);
1719 spin_unlock_irq(&target
->lock
);
1724 ib_dma_sync_single_for_cpu(dev
, iu
->dma
, sizeof *tsk_mgmt
,
1727 memset(tsk_mgmt
, 0, sizeof *tsk_mgmt
);
1729 tsk_mgmt
->opcode
= SRP_TSK_MGMT
;
1730 tsk_mgmt
->lun
= cpu_to_be64((u64
) lun
<< 48);
1731 tsk_mgmt
->tag
= req_tag
| SRP_TAG_TSK_MGMT
;
1732 tsk_mgmt
->tsk_mgmt_func
= func
;
1733 tsk_mgmt
->task_tag
= req_tag
;
1735 ib_dma_sync_single_for_device(dev
, iu
->dma
, sizeof *tsk_mgmt
,
1737 if (srp_post_send(target
, iu
, sizeof *tsk_mgmt
)) {
1738 srp_put_tx_iu(target
, iu
, SRP_IU_TSK_MGMT
);
1742 if (!wait_for_completion_timeout(&target
->tsk_mgmt_done
,
1743 msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS
)))
1749 static int srp_abort(struct scsi_cmnd
*scmnd
)
1751 struct srp_target_port
*target
= host_to_target(scmnd
->device
->host
);
1752 struct srp_request
*req
= (struct srp_request
*) scmnd
->host_scribble
;
1754 shost_printk(KERN_ERR
, target
->scsi_host
, "SRP abort called\n");
1756 if (!req
|| !srp_claim_req(target
, req
, scmnd
))
1758 srp_send_tsk_mgmt(target
, req
->index
, scmnd
->device
->lun
,
1759 SRP_TSK_ABORT_TASK
);
1760 srp_free_req(target
, req
, scmnd
, 0);
1761 scmnd
->result
= DID_ABORT
<< 16;
1762 scmnd
->scsi_done(scmnd
);
1767 static int srp_reset_device(struct scsi_cmnd
*scmnd
)
1769 struct srp_target_port
*target
= host_to_target(scmnd
->device
->host
);
1772 shost_printk(KERN_ERR
, target
->scsi_host
, "SRP reset_device called\n");
1774 if (srp_send_tsk_mgmt(target
, SRP_TAG_NO_REQ
, scmnd
->device
->lun
,
1777 if (target
->tsk_mgmt_status
)
1780 for (i
= 0; i
< SRP_CMD_SQ_SIZE
; ++i
) {
1781 struct srp_request
*req
= &target
->req_ring
[i
];
1782 if (req
->scmnd
&& req
->scmnd
->device
== scmnd
->device
)
1783 srp_reset_req(target
, req
);
1789 static int srp_reset_host(struct scsi_cmnd
*scmnd
)
1791 struct srp_target_port
*target
= host_to_target(scmnd
->device
->host
);
1794 shost_printk(KERN_ERR
, target
->scsi_host
, PFX
"SRP reset_host called\n");
1796 if (!srp_reconnect_target(target
))
1802 static int srp_slave_configure(struct scsi_device
*sdev
)
1804 struct Scsi_Host
*shost
= sdev
->host
;
1805 struct srp_target_port
*target
= host_to_target(shost
);
1806 struct request_queue
*q
= sdev
->request_queue
;
1807 unsigned long timeout
;
1809 if (sdev
->type
== TYPE_DISK
) {
1810 timeout
= max_t(unsigned, 30 * HZ
, target
->rq_tmo_jiffies
);
1811 blk_queue_rq_timeout(q
, timeout
);
1817 static ssize_t
show_id_ext(struct device
*dev
, struct device_attribute
*attr
,
1820 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1822 return sprintf(buf
, "0x%016llx\n",
1823 (unsigned long long) be64_to_cpu(target
->id_ext
));
1826 static ssize_t
show_ioc_guid(struct device
*dev
, struct device_attribute
*attr
,
1829 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1831 return sprintf(buf
, "0x%016llx\n",
1832 (unsigned long long) be64_to_cpu(target
->ioc_guid
));
1835 static ssize_t
show_service_id(struct device
*dev
,
1836 struct device_attribute
*attr
, char *buf
)
1838 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1840 return sprintf(buf
, "0x%016llx\n",
1841 (unsigned long long) be64_to_cpu(target
->service_id
));
1844 static ssize_t
show_pkey(struct device
*dev
, struct device_attribute
*attr
,
1847 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1849 return sprintf(buf
, "0x%04x\n", be16_to_cpu(target
->path
.pkey
));
1852 static ssize_t
show_dgid(struct device
*dev
, struct device_attribute
*attr
,
1855 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1857 return sprintf(buf
, "%pI6\n", target
->path
.dgid
.raw
);
1860 static ssize_t
show_orig_dgid(struct device
*dev
,
1861 struct device_attribute
*attr
, char *buf
)
1863 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1865 return sprintf(buf
, "%pI6\n", target
->orig_dgid
);
1868 static ssize_t
show_req_lim(struct device
*dev
,
1869 struct device_attribute
*attr
, char *buf
)
1871 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1873 return sprintf(buf
, "%d\n", target
->req_lim
);
1876 static ssize_t
show_zero_req_lim(struct device
*dev
,
1877 struct device_attribute
*attr
, char *buf
)
1879 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1881 return sprintf(buf
, "%d\n", target
->zero_req_lim
);
1884 static ssize_t
show_local_ib_port(struct device
*dev
,
1885 struct device_attribute
*attr
, char *buf
)
1887 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1889 return sprintf(buf
, "%d\n", target
->srp_host
->port
);
1892 static ssize_t
show_local_ib_device(struct device
*dev
,
1893 struct device_attribute
*attr
, char *buf
)
1895 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1897 return sprintf(buf
, "%s\n", target
->srp_host
->srp_dev
->dev
->name
);
1900 static ssize_t
show_cmd_sg_entries(struct device
*dev
,
1901 struct device_attribute
*attr
, char *buf
)
1903 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1905 return sprintf(buf
, "%u\n", target
->cmd_sg_cnt
);
1908 static ssize_t
show_allow_ext_sg(struct device
*dev
,
1909 struct device_attribute
*attr
, char *buf
)
1911 struct srp_target_port
*target
= host_to_target(class_to_shost(dev
));
1913 return sprintf(buf
, "%s\n", target
->allow_ext_sg
? "true" : "false");
1916 static DEVICE_ATTR(id_ext
, S_IRUGO
, show_id_ext
, NULL
);
1917 static DEVICE_ATTR(ioc_guid
, S_IRUGO
, show_ioc_guid
, NULL
);
1918 static DEVICE_ATTR(service_id
, S_IRUGO
, show_service_id
, NULL
);
1919 static DEVICE_ATTR(pkey
, S_IRUGO
, show_pkey
, NULL
);
1920 static DEVICE_ATTR(dgid
, S_IRUGO
, show_dgid
, NULL
);
1921 static DEVICE_ATTR(orig_dgid
, S_IRUGO
, show_orig_dgid
, NULL
);
1922 static DEVICE_ATTR(req_lim
, S_IRUGO
, show_req_lim
, NULL
);
1923 static DEVICE_ATTR(zero_req_lim
, S_IRUGO
, show_zero_req_lim
, NULL
);
1924 static DEVICE_ATTR(local_ib_port
, S_IRUGO
, show_local_ib_port
, NULL
);
1925 static DEVICE_ATTR(local_ib_device
, S_IRUGO
, show_local_ib_device
, NULL
);
1926 static DEVICE_ATTR(cmd_sg_entries
, S_IRUGO
, show_cmd_sg_entries
, NULL
);
1927 static DEVICE_ATTR(allow_ext_sg
, S_IRUGO
, show_allow_ext_sg
, NULL
);
1929 static struct device_attribute
*srp_host_attrs
[] = {
1932 &dev_attr_service_id
,
1935 &dev_attr_orig_dgid
,
1937 &dev_attr_zero_req_lim
,
1938 &dev_attr_local_ib_port
,
1939 &dev_attr_local_ib_device
,
1940 &dev_attr_cmd_sg_entries
,
1941 &dev_attr_allow_ext_sg
,
1945 static struct scsi_host_template srp_template
= {
1946 .module
= THIS_MODULE
,
1947 .name
= "InfiniBand SRP initiator",
1948 .proc_name
= DRV_NAME
,
1949 .slave_configure
= srp_slave_configure
,
1950 .info
= srp_target_info
,
1951 .queuecommand
= srp_queuecommand
,
1952 .eh_abort_handler
= srp_abort
,
1953 .eh_device_reset_handler
= srp_reset_device
,
1954 .eh_host_reset_handler
= srp_reset_host
,
1955 .sg_tablesize
= SRP_DEF_SG_TABLESIZE
,
1956 .can_queue
= SRP_CMD_SQ_SIZE
,
1958 .cmd_per_lun
= SRP_CMD_SQ_SIZE
,
1959 .use_clustering
= ENABLE_CLUSTERING
,
1960 .shost_attrs
= srp_host_attrs
1963 static int srp_add_target(struct srp_host
*host
, struct srp_target_port
*target
)
1965 struct srp_rport_identifiers ids
;
1966 struct srp_rport
*rport
;
1968 sprintf(target
->target_name
, "SRP.T10:%016llX",
1969 (unsigned long long) be64_to_cpu(target
->id_ext
));
1971 if (scsi_add_host(target
->scsi_host
, host
->srp_dev
->dev
->dma_device
))
1974 memcpy(ids
.port_id
, &target
->id_ext
, 8);
1975 memcpy(ids
.port_id
+ 8, &target
->ioc_guid
, 8);
1976 ids
.roles
= SRP_RPORT_ROLE_TARGET
;
1977 rport
= srp_rport_add(target
->scsi_host
, &ids
);
1978 if (IS_ERR(rport
)) {
1979 scsi_remove_host(target
->scsi_host
);
1980 return PTR_ERR(rport
);
1983 rport
->lld_data
= target
;
1985 spin_lock(&host
->target_lock
);
1986 list_add_tail(&target
->list
, &host
->target_list
);
1987 spin_unlock(&host
->target_lock
);
1989 target
->state
= SRP_TARGET_LIVE
;
1991 scsi_scan_target(&target
->scsi_host
->shost_gendev
,
1992 0, target
->scsi_id
, SCAN_WILD_CARD
, 0);
1997 static void srp_release_dev(struct device
*dev
)
1999 struct srp_host
*host
=
2000 container_of(dev
, struct srp_host
, dev
);
2002 complete(&host
->released
);
2005 static struct class srp_class
= {
2006 .name
= "infiniband_srp",
2007 .dev_release
= srp_release_dev
2011 * Target ports are added by writing
2013 * id_ext=<SRP ID ext>,ioc_guid=<SRP IOC GUID>,dgid=<dest GID>,
2014 * pkey=<P_Key>,service_id=<service ID>
2016 * to the add_target sysfs attribute.
2020 SRP_OPT_ID_EXT
= 1 << 0,
2021 SRP_OPT_IOC_GUID
= 1 << 1,
2022 SRP_OPT_DGID
= 1 << 2,
2023 SRP_OPT_PKEY
= 1 << 3,
2024 SRP_OPT_SERVICE_ID
= 1 << 4,
2025 SRP_OPT_MAX_SECT
= 1 << 5,
2026 SRP_OPT_MAX_CMD_PER_LUN
= 1 << 6,
2027 SRP_OPT_IO_CLASS
= 1 << 7,
2028 SRP_OPT_INITIATOR_EXT
= 1 << 8,
2029 SRP_OPT_CMD_SG_ENTRIES
= 1 << 9,
2030 SRP_OPT_ALLOW_EXT_SG
= 1 << 10,
2031 SRP_OPT_SG_TABLESIZE
= 1 << 11,
2032 SRP_OPT_ALL
= (SRP_OPT_ID_EXT
|
2036 SRP_OPT_SERVICE_ID
),
2039 static const match_table_t srp_opt_tokens
= {
2040 { SRP_OPT_ID_EXT
, "id_ext=%s" },
2041 { SRP_OPT_IOC_GUID
, "ioc_guid=%s" },
2042 { SRP_OPT_DGID
, "dgid=%s" },
2043 { SRP_OPT_PKEY
, "pkey=%x" },
2044 { SRP_OPT_SERVICE_ID
, "service_id=%s" },
2045 { SRP_OPT_MAX_SECT
, "max_sect=%d" },
2046 { SRP_OPT_MAX_CMD_PER_LUN
, "max_cmd_per_lun=%d" },
2047 { SRP_OPT_IO_CLASS
, "io_class=%x" },
2048 { SRP_OPT_INITIATOR_EXT
, "initiator_ext=%s" },
2049 { SRP_OPT_CMD_SG_ENTRIES
, "cmd_sg_entries=%u" },
2050 { SRP_OPT_ALLOW_EXT_SG
, "allow_ext_sg=%u" },
2051 { SRP_OPT_SG_TABLESIZE
, "sg_tablesize=%u" },
2052 { SRP_OPT_ERR
, NULL
}
2055 static int srp_parse_options(const char *buf
, struct srp_target_port
*target
)
2057 char *options
, *sep_opt
;
2060 substring_t args
[MAX_OPT_ARGS
];
2066 options
= kstrdup(buf
, GFP_KERNEL
);
2071 while ((p
= strsep(&sep_opt
, ",")) != NULL
) {
2075 token
= match_token(p
, srp_opt_tokens
, args
);
2079 case SRP_OPT_ID_EXT
:
2080 p
= match_strdup(args
);
2085 target
->id_ext
= cpu_to_be64(simple_strtoull(p
, NULL
, 16));
2089 case SRP_OPT_IOC_GUID
:
2090 p
= match_strdup(args
);
2095 target
->ioc_guid
= cpu_to_be64(simple_strtoull(p
, NULL
, 16));
2100 p
= match_strdup(args
);
2105 if (strlen(p
) != 32) {
2106 pr_warn("bad dest GID parameter '%s'\n", p
);
2111 for (i
= 0; i
< 16; ++i
) {
2112 strlcpy(dgid
, p
+ i
* 2, 3);
2113 target
->path
.dgid
.raw
[i
] = simple_strtoul(dgid
, NULL
, 16);
2116 memcpy(target
->orig_dgid
, target
->path
.dgid
.raw
, 16);
2120 if (match_hex(args
, &token
)) {
2121 pr_warn("bad P_Key parameter '%s'\n", p
);
2124 target
->path
.pkey
= cpu_to_be16(token
);
2127 case SRP_OPT_SERVICE_ID
:
2128 p
= match_strdup(args
);
2133 target
->service_id
= cpu_to_be64(simple_strtoull(p
, NULL
, 16));
2134 target
->path
.service_id
= target
->service_id
;
2138 case SRP_OPT_MAX_SECT
:
2139 if (match_int(args
, &token
)) {
2140 pr_warn("bad max sect parameter '%s'\n", p
);
2143 target
->scsi_host
->max_sectors
= token
;
2146 case SRP_OPT_MAX_CMD_PER_LUN
:
2147 if (match_int(args
, &token
)) {
2148 pr_warn("bad max cmd_per_lun parameter '%s'\n",
2152 target
->scsi_host
->cmd_per_lun
= min(token
, SRP_CMD_SQ_SIZE
);
2155 case SRP_OPT_IO_CLASS
:
2156 if (match_hex(args
, &token
)) {
2157 pr_warn("bad IO class parameter '%s'\n", p
);
2160 if (token
!= SRP_REV10_IB_IO_CLASS
&&
2161 token
!= SRP_REV16A_IB_IO_CLASS
) {
2162 pr_warn("unknown IO class parameter value %x specified (use %x or %x).\n",
2163 token
, SRP_REV10_IB_IO_CLASS
,
2164 SRP_REV16A_IB_IO_CLASS
);
2167 target
->io_class
= token
;
2170 case SRP_OPT_INITIATOR_EXT
:
2171 p
= match_strdup(args
);
2176 target
->initiator_ext
= cpu_to_be64(simple_strtoull(p
, NULL
, 16));
2180 case SRP_OPT_CMD_SG_ENTRIES
:
2181 if (match_int(args
, &token
) || token
< 1 || token
> 255) {
2182 pr_warn("bad max cmd_sg_entries parameter '%s'\n",
2186 target
->cmd_sg_cnt
= token
;
2189 case SRP_OPT_ALLOW_EXT_SG
:
2190 if (match_int(args
, &token
)) {
2191 pr_warn("bad allow_ext_sg parameter '%s'\n", p
);
2194 target
->allow_ext_sg
= !!token
;
2197 case SRP_OPT_SG_TABLESIZE
:
2198 if (match_int(args
, &token
) || token
< 1 ||
2199 token
> SCSI_MAX_SG_CHAIN_SEGMENTS
) {
2200 pr_warn("bad max sg_tablesize parameter '%s'\n",
2204 target
->sg_tablesize
= token
;
2208 pr_warn("unknown parameter or missing value '%s' in target creation request\n",
2214 if ((opt_mask
& SRP_OPT_ALL
) == SRP_OPT_ALL
)
2217 for (i
= 0; i
< ARRAY_SIZE(srp_opt_tokens
); ++i
)
2218 if ((srp_opt_tokens
[i
].token
& SRP_OPT_ALL
) &&
2219 !(srp_opt_tokens
[i
].token
& opt_mask
))
2220 pr_warn("target creation request is missing parameter '%s'\n",
2221 srp_opt_tokens
[i
].pattern
);
2228 static ssize_t
srp_create_target(struct device
*dev
,
2229 struct device_attribute
*attr
,
2230 const char *buf
, size_t count
)
2232 struct srp_host
*host
=
2233 container_of(dev
, struct srp_host
, dev
);
2234 struct Scsi_Host
*target_host
;
2235 struct srp_target_port
*target
;
2236 struct ib_device
*ibdev
= host
->srp_dev
->dev
;
2237 dma_addr_t dma_addr
;
2240 target_host
= scsi_host_alloc(&srp_template
,
2241 sizeof (struct srp_target_port
));
2245 target_host
->transportt
= ib_srp_transport_template
;
2246 target_host
->max_channel
= 0;
2247 target_host
->max_id
= 1;
2248 target_host
->max_lun
= SRP_MAX_LUN
;
2249 target_host
->max_cmd_len
= sizeof ((struct srp_cmd
*) (void *) 0L)->cdb
;
2251 target
= host_to_target(target_host
);
2253 target
->io_class
= SRP_REV16A_IB_IO_CLASS
;
2254 target
->scsi_host
= target_host
;
2255 target
->srp_host
= host
;
2256 target
->lkey
= host
->srp_dev
->mr
->lkey
;
2257 target
->rkey
= host
->srp_dev
->mr
->rkey
;
2258 target
->cmd_sg_cnt
= cmd_sg_entries
;
2259 target
->sg_tablesize
= indirect_sg_entries
? : cmd_sg_entries
;
2260 target
->allow_ext_sg
= allow_ext_sg
;
2262 ret
= srp_parse_options(buf
, target
);
2266 if (!host
->srp_dev
->fmr_pool
&& !target
->allow_ext_sg
&&
2267 target
->cmd_sg_cnt
< target
->sg_tablesize
) {
2268 pr_warn("No FMR pool and no external indirect descriptors, limiting sg_tablesize to cmd_sg_cnt\n");
2269 target
->sg_tablesize
= target
->cmd_sg_cnt
;
2272 target_host
->sg_tablesize
= target
->sg_tablesize
;
2273 target
->indirect_size
= target
->sg_tablesize
*
2274 sizeof (struct srp_direct_buf
);
2275 target
->max_iu_len
= sizeof (struct srp_cmd
) +
2276 sizeof (struct srp_indirect_buf
) +
2277 target
->cmd_sg_cnt
* sizeof (struct srp_direct_buf
);
2279 INIT_WORK(&target
->remove_work
, srp_remove_work
);
2280 spin_lock_init(&target
->lock
);
2281 INIT_LIST_HEAD(&target
->free_tx
);
2282 INIT_LIST_HEAD(&target
->free_reqs
);
2283 for (i
= 0; i
< SRP_CMD_SQ_SIZE
; ++i
) {
2284 struct srp_request
*req
= &target
->req_ring
[i
];
2286 req
->fmr_list
= kmalloc(target
->cmd_sg_cnt
* sizeof (void *),
2288 req
->map_page
= kmalloc(SRP_FMR_SIZE
* sizeof (void *),
2290 req
->indirect_desc
= kmalloc(target
->indirect_size
, GFP_KERNEL
);
2291 if (!req
->fmr_list
|| !req
->map_page
|| !req
->indirect_desc
)
2294 dma_addr
= ib_dma_map_single(ibdev
, req
->indirect_desc
,
2295 target
->indirect_size
,
2297 if (ib_dma_mapping_error(ibdev
, dma_addr
))
2300 req
->indirect_dma_addr
= dma_addr
;
2302 list_add_tail(&req
->list
, &target
->free_reqs
);
2305 ib_query_gid(ibdev
, host
->port
, 0, &target
->path
.sgid
);
2307 shost_printk(KERN_DEBUG
, target
->scsi_host
, PFX
2308 "new target: id_ext %016llx ioc_guid %016llx pkey %04x "
2309 "service_id %016llx dgid %pI6\n",
2310 (unsigned long long) be64_to_cpu(target
->id_ext
),
2311 (unsigned long long) be64_to_cpu(target
->ioc_guid
),
2312 be16_to_cpu(target
->path
.pkey
),
2313 (unsigned long long) be64_to_cpu(target
->service_id
),
2314 target
->path
.dgid
.raw
);
2316 ret
= srp_create_target_ib(target
);
2320 ret
= srp_new_cm_id(target
);
2324 ret
= srp_connect_target(target
);
2326 shost_printk(KERN_ERR
, target
->scsi_host
,
2327 PFX
"Connection failed\n");
2331 ret
= srp_add_target(host
, target
);
2333 goto err_disconnect
;
2338 srp_disconnect_target(target
);
2341 ib_destroy_cm_id(target
->cm_id
);
2344 srp_free_target_ib(target
);
2347 srp_free_req_data(target
);
2350 scsi_host_put(target_host
);
2355 static DEVICE_ATTR(add_target
, S_IWUSR
, NULL
, srp_create_target
);
2357 static ssize_t
show_ibdev(struct device
*dev
, struct device_attribute
*attr
,
2360 struct srp_host
*host
= container_of(dev
, struct srp_host
, dev
);
2362 return sprintf(buf
, "%s\n", host
->srp_dev
->dev
->name
);
2365 static DEVICE_ATTR(ibdev
, S_IRUGO
, show_ibdev
, NULL
);
2367 static ssize_t
show_port(struct device
*dev
, struct device_attribute
*attr
,
2370 struct srp_host
*host
= container_of(dev
, struct srp_host
, dev
);
2372 return sprintf(buf
, "%d\n", host
->port
);
2375 static DEVICE_ATTR(port
, S_IRUGO
, show_port
, NULL
);
2377 static struct srp_host
*srp_add_port(struct srp_device
*device
, u8 port
)
2379 struct srp_host
*host
;
2381 host
= kzalloc(sizeof *host
, GFP_KERNEL
);
2385 INIT_LIST_HEAD(&host
->target_list
);
2386 spin_lock_init(&host
->target_lock
);
2387 init_completion(&host
->released
);
2388 host
->srp_dev
= device
;
2391 host
->dev
.class = &srp_class
;
2392 host
->dev
.parent
= device
->dev
->dma_device
;
2393 dev_set_name(&host
->dev
, "srp-%s-%d", device
->dev
->name
, port
);
2395 if (device_register(&host
->dev
))
2397 if (device_create_file(&host
->dev
, &dev_attr_add_target
))
2399 if (device_create_file(&host
->dev
, &dev_attr_ibdev
))
2401 if (device_create_file(&host
->dev
, &dev_attr_port
))
2407 device_unregister(&host
->dev
);
2415 static void srp_add_one(struct ib_device
*device
)
2417 struct srp_device
*srp_dev
;
2418 struct ib_device_attr
*dev_attr
;
2419 struct ib_fmr_pool_param fmr_param
;
2420 struct srp_host
*host
;
2421 int max_pages_per_fmr
, fmr_page_shift
, s
, e
, p
;
2423 dev_attr
= kmalloc(sizeof *dev_attr
, GFP_KERNEL
);
2427 if (ib_query_device(device
, dev_attr
)) {
2428 pr_warn("Query device failed for %s\n", device
->name
);
2432 srp_dev
= kmalloc(sizeof *srp_dev
, GFP_KERNEL
);
2437 * Use the smallest page size supported by the HCA, down to a
2438 * minimum of 4096 bytes. We're unlikely to build large sglists
2439 * out of smaller entries.
2441 fmr_page_shift
= max(12, ffs(dev_attr
->page_size_cap
) - 1);
2442 srp_dev
->fmr_page_size
= 1 << fmr_page_shift
;
2443 srp_dev
->fmr_page_mask
= ~((u64
) srp_dev
->fmr_page_size
- 1);
2444 srp_dev
->fmr_max_size
= srp_dev
->fmr_page_size
* SRP_FMR_SIZE
;
2446 INIT_LIST_HEAD(&srp_dev
->dev_list
);
2448 srp_dev
->dev
= device
;
2449 srp_dev
->pd
= ib_alloc_pd(device
);
2450 if (IS_ERR(srp_dev
->pd
))
2453 srp_dev
->mr
= ib_get_dma_mr(srp_dev
->pd
,
2454 IB_ACCESS_LOCAL_WRITE
|
2455 IB_ACCESS_REMOTE_READ
|
2456 IB_ACCESS_REMOTE_WRITE
);
2457 if (IS_ERR(srp_dev
->mr
))
2460 for (max_pages_per_fmr
= SRP_FMR_SIZE
;
2461 max_pages_per_fmr
>= SRP_FMR_MIN_SIZE
;
2462 max_pages_per_fmr
/= 2, srp_dev
->fmr_max_size
/= 2) {
2463 memset(&fmr_param
, 0, sizeof fmr_param
);
2464 fmr_param
.pool_size
= SRP_FMR_POOL_SIZE
;
2465 fmr_param
.dirty_watermark
= SRP_FMR_DIRTY_SIZE
;
2466 fmr_param
.cache
= 1;
2467 fmr_param
.max_pages_per_fmr
= max_pages_per_fmr
;
2468 fmr_param
.page_shift
= fmr_page_shift
;
2469 fmr_param
.access
= (IB_ACCESS_LOCAL_WRITE
|
2470 IB_ACCESS_REMOTE_WRITE
|
2471 IB_ACCESS_REMOTE_READ
);
2473 srp_dev
->fmr_pool
= ib_create_fmr_pool(srp_dev
->pd
, &fmr_param
);
2474 if (!IS_ERR(srp_dev
->fmr_pool
))
2478 if (IS_ERR(srp_dev
->fmr_pool
))
2479 srp_dev
->fmr_pool
= NULL
;
2481 if (device
->node_type
== RDMA_NODE_IB_SWITCH
) {
2486 e
= device
->phys_port_cnt
;
2489 for (p
= s
; p
<= e
; ++p
) {
2490 host
= srp_add_port(srp_dev
, p
);
2492 list_add_tail(&host
->list
, &srp_dev
->dev_list
);
2495 ib_set_client_data(device
, &srp_client
, srp_dev
);
2500 ib_dealloc_pd(srp_dev
->pd
);
2509 static void srp_remove_one(struct ib_device
*device
)
2511 struct srp_device
*srp_dev
;
2512 struct srp_host
*host
, *tmp_host
;
2513 struct srp_target_port
*target
;
2515 srp_dev
= ib_get_client_data(device
, &srp_client
);
2517 list_for_each_entry_safe(host
, tmp_host
, &srp_dev
->dev_list
, list
) {
2518 device_unregister(&host
->dev
);
2520 * Wait for the sysfs entry to go away, so that no new
2521 * target ports can be created.
2523 wait_for_completion(&host
->released
);
2526 * Remove all target ports.
2528 spin_lock(&host
->target_lock
);
2529 list_for_each_entry(target
, &host
->target_list
, list
)
2530 srp_queue_remove_work(target
);
2531 spin_unlock(&host
->target_lock
);
2534 * Wait for tl_err and target port removal tasks.
2536 flush_workqueue(system_long_wq
);
2537 flush_workqueue(srp_remove_wq
);
2542 if (srp_dev
->fmr_pool
)
2543 ib_destroy_fmr_pool(srp_dev
->fmr_pool
);
2544 ib_dereg_mr(srp_dev
->mr
);
2545 ib_dealloc_pd(srp_dev
->pd
);
2550 static struct srp_function_template ib_srp_transport_functions
= {
2551 .rport_delete
= srp_rport_delete
,
2554 static int __init
srp_init_module(void)
2558 BUILD_BUG_ON(FIELD_SIZEOF(struct ib_wc
, wr_id
) < sizeof(void *));
2560 if (srp_sg_tablesize
) {
2561 pr_warn("srp_sg_tablesize is deprecated, please use cmd_sg_entries\n");
2562 if (!cmd_sg_entries
)
2563 cmd_sg_entries
= srp_sg_tablesize
;
2566 if (!cmd_sg_entries
)
2567 cmd_sg_entries
= SRP_DEF_SG_TABLESIZE
;
2569 if (cmd_sg_entries
> 255) {
2570 pr_warn("Clamping cmd_sg_entries to 255\n");
2571 cmd_sg_entries
= 255;
2574 if (!indirect_sg_entries
)
2575 indirect_sg_entries
= cmd_sg_entries
;
2576 else if (indirect_sg_entries
< cmd_sg_entries
) {
2577 pr_warn("Bumping up indirect_sg_entries to match cmd_sg_entries (%u)\n",
2579 indirect_sg_entries
= cmd_sg_entries
;
2582 srp_remove_wq
= create_workqueue("srp_remove");
2583 if (IS_ERR(srp_remove_wq
)) {
2584 ret
= PTR_ERR(srp_remove_wq
);
2589 ib_srp_transport_template
=
2590 srp_attach_transport(&ib_srp_transport_functions
);
2591 if (!ib_srp_transport_template
)
2594 ret
= class_register(&srp_class
);
2596 pr_err("couldn't register class infiniband_srp\n");
2600 ib_sa_register_client(&srp_sa_client
);
2602 ret
= ib_register_client(&srp_client
);
2604 pr_err("couldn't register IB client\n");
2612 ib_sa_unregister_client(&srp_sa_client
);
2613 class_unregister(&srp_class
);
2616 srp_release_transport(ib_srp_transport_template
);
2619 destroy_workqueue(srp_remove_wq
);
2623 static void __exit
srp_cleanup_module(void)
2625 ib_unregister_client(&srp_client
);
2626 ib_sa_unregister_client(&srp_sa_client
);
2627 class_unregister(&srp_class
);
2628 srp_release_transport(ib_srp_transport_template
);
2629 destroy_workqueue(srp_remove_wq
);
2632 module_init(srp_init_module
);
2633 module_exit(srp_cleanup_module
);