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
32 * $Id: ib_srp.c 3932 2005-11-01 17:19:29Z roland $
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>
43 #include <asm/atomic.h>
45 #include <scsi/scsi.h>
46 #include <scsi/scsi_device.h>
47 #include <scsi/scsi_dbg.h>
50 #include <rdma/ib_cache.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 int topspin_workarounds = 1;
66 module_param(topspin_workarounds, int, 0444);
67 MODULE_PARM_DESC(topspin_workarounds,
68 "Enable workarounds for Topspin/Cisco SRP target bugs if != 0");
70 static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad };
72 static void srp_add_one(struct ib_device *device);
73 static void srp_remove_one(struct ib_device *device);
74 static void srp_completion(struct ib_cq *cq, void *target_ptr);
75 static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event);
77 static struct ib_client srp_client = {
80 .remove = srp_remove_one
83 static inline struct srp_target_port *host_to_target(struct Scsi_Host *host)
85 return (struct srp_target_port *) host->hostdata;
88 static const char *srp_target_info(struct Scsi_Host *host)
90 return host_to_target(host)->target_name;
93 static struct srp_iu *srp_alloc_iu(struct srp_host *host, size_t size,
95 enum dma_data_direction direction)
99 iu = kmalloc(sizeof *iu, gfp_mask);
103 iu->buf = kzalloc(size, gfp_mask);
107 iu->dma = dma_map_single(host->dev->dma_device, iu->buf, size, direction);
108 if (dma_mapping_error(iu->dma))
112 iu->direction = direction;
124 static void srp_free_iu(struct srp_host *host, struct srp_iu *iu)
129 dma_unmap_single(host->dev->dma_device, iu->dma, iu->size, iu->direction);
134 static void srp_qp_event(struct ib_event *event, void *context)
136 printk(KERN_ERR PFX "QP event %d\n", event->event);
139 static int srp_init_qp(struct srp_target_port *target,
142 struct ib_qp_attr *attr;
145 attr = kmalloc(sizeof *attr, GFP_KERNEL);
149 ret = ib_find_cached_pkey(target->srp_host->dev,
150 target->srp_host->port,
151 be16_to_cpu(target->path.pkey),
156 attr->qp_state = IB_QPS_INIT;
157 attr->qp_access_flags = (IB_ACCESS_REMOTE_READ |
158 IB_ACCESS_REMOTE_WRITE);
159 attr->port_num = target->srp_host->port;
161 ret = ib_modify_qp(qp, attr,
172 static int srp_create_target_ib(struct srp_target_port *target)
174 struct ib_qp_init_attr *init_attr;
177 init_attr = kzalloc(sizeof *init_attr, GFP_KERNEL);
181 target->cq = ib_create_cq(target->srp_host->dev, srp_completion,
182 NULL, target, SRP_CQ_SIZE);
183 if (IS_ERR(target->cq)) {
184 ret = PTR_ERR(target->cq);
188 ib_req_notify_cq(target->cq, IB_CQ_NEXT_COMP);
190 init_attr->event_handler = srp_qp_event;
191 init_attr->cap.max_send_wr = SRP_SQ_SIZE;
192 init_attr->cap.max_recv_wr = SRP_RQ_SIZE;
193 init_attr->cap.max_recv_sge = 1;
194 init_attr->cap.max_send_sge = 1;
195 init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
196 init_attr->qp_type = IB_QPT_RC;
197 init_attr->send_cq = target->cq;
198 init_attr->recv_cq = target->cq;
200 target->qp = ib_create_qp(target->srp_host->pd, init_attr);
201 if (IS_ERR(target->qp)) {
202 ret = PTR_ERR(target->qp);
203 ib_destroy_cq(target->cq);
207 ret = srp_init_qp(target, target->qp);
209 ib_destroy_qp(target->qp);
210 ib_destroy_cq(target->cq);
219 static void srp_free_target_ib(struct srp_target_port *target)
223 ib_destroy_qp(target->qp);
224 ib_destroy_cq(target->cq);
226 for (i = 0; i < SRP_RQ_SIZE; ++i)
227 srp_free_iu(target->srp_host, target->rx_ring[i]);
228 for (i = 0; i < SRP_SQ_SIZE + 1; ++i)
229 srp_free_iu(target->srp_host, target->tx_ring[i]);
232 static void srp_path_rec_completion(int status,
233 struct ib_sa_path_rec *pathrec,
236 struct srp_target_port *target = target_ptr;
238 target->status = status;
240 printk(KERN_ERR PFX "Got failed path rec status %d\n", status);
242 target->path = *pathrec;
243 complete(&target->done);
246 static int srp_lookup_path(struct srp_target_port *target)
248 target->path.numb_path = 1;
250 init_completion(&target->done);
252 target->path_query_id = ib_sa_path_rec_get(target->srp_host->dev,
253 target->srp_host->port,
255 IB_SA_PATH_REC_DGID |
256 IB_SA_PATH_REC_SGID |
257 IB_SA_PATH_REC_NUMB_PATH |
259 SRP_PATH_REC_TIMEOUT_MS,
261 srp_path_rec_completion,
262 target, &target->path_query);
263 if (target->path_query_id < 0)
264 return target->path_query_id;
266 wait_for_completion(&target->done);
268 if (target->status < 0)
269 printk(KERN_WARNING PFX "Path record query failed\n");
271 return target->status;
274 static int srp_send_req(struct srp_target_port *target)
277 struct ib_cm_req_param param;
278 struct srp_login_req priv;
282 req = kzalloc(sizeof *req, GFP_KERNEL);
286 req->param.primary_path = &target->path;
287 req->param.alternate_path = NULL;
288 req->param.service_id = target->service_id;
289 req->param.qp_num = target->qp->qp_num;
290 req->param.qp_type = target->qp->qp_type;
291 req->param.private_data = &req->priv;
292 req->param.private_data_len = sizeof req->priv;
293 req->param.flow_control = 1;
295 get_random_bytes(&req->param.starting_psn, 4);
296 req->param.starting_psn &= 0xffffff;
299 * Pick some arbitrary defaults here; we could make these
300 * module parameters if anyone cared about setting them.
302 req->param.responder_resources = 4;
303 req->param.remote_cm_response_timeout = 20;
304 req->param.local_cm_response_timeout = 20;
305 req->param.retry_count = 7;
306 req->param.rnr_retry_count = 7;
307 req->param.max_cm_retries = 15;
309 req->priv.opcode = SRP_LOGIN_REQ;
311 req->priv.req_it_iu_len = cpu_to_be32(SRP_MAX_IU_LEN);
312 req->priv.req_buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
313 SRP_BUF_FORMAT_INDIRECT);
314 memcpy(req->priv.initiator_port_id, target->srp_host->initiator_port_id, 16);
316 * Topspin/Cisco SRP targets will reject our login unless we
317 * zero out the first 8 bytes of our initiator port ID. The
318 * second 8 bytes must be our local node GUID, but we always
321 if (topspin_workarounds && !memcmp(&target->ioc_guid, topspin_oui, 3)) {
322 printk(KERN_DEBUG PFX "Topspin/Cisco initiator port ID workaround "
323 "activated for target GUID %016llx\n",
324 (unsigned long long) be64_to_cpu(target->ioc_guid));
325 memset(req->priv.initiator_port_id, 0, 8);
327 memcpy(req->priv.target_port_id, &target->id_ext, 8);
328 memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8);
330 status = ib_send_cm_req(target->cm_id, &req->param);
337 static void srp_disconnect_target(struct srp_target_port *target)
339 /* XXX should send SRP_I_LOGOUT request */
341 init_completion(&target->done);
342 ib_send_cm_dreq(target->cm_id, NULL, 0);
343 wait_for_completion(&target->done);
346 static void srp_remove_work(void *target_ptr)
348 struct srp_target_port *target = target_ptr;
350 spin_lock_irq(target->scsi_host->host_lock);
351 if (target->state != SRP_TARGET_DEAD) {
352 spin_unlock_irq(target->scsi_host->host_lock);
353 scsi_host_put(target->scsi_host);
356 target->state = SRP_TARGET_REMOVED;
357 spin_unlock_irq(target->scsi_host->host_lock);
359 down(&target->srp_host->target_mutex);
360 list_del(&target->list);
361 up(&target->srp_host->target_mutex);
363 scsi_remove_host(target->scsi_host);
364 ib_destroy_cm_id(target->cm_id);
365 srp_free_target_ib(target);
366 scsi_host_put(target->scsi_host);
367 /* And another put to really free the target port... */
368 scsi_host_put(target->scsi_host);
371 static int srp_connect_target(struct srp_target_port *target)
375 ret = srp_lookup_path(target);
380 init_completion(&target->done);
381 ret = srp_send_req(target);
384 wait_for_completion(&target->done);
387 * The CM event handling code will set status to
388 * SRP_PORT_REDIRECT if we get a port redirect REJ
389 * back, or SRP_DLID_REDIRECT if we get a lid/qp
392 switch (target->status) {
396 case SRP_PORT_REDIRECT:
397 ret = srp_lookup_path(target);
402 case SRP_DLID_REDIRECT:
406 return target->status;
411 static int srp_reconnect_target(struct srp_target_port *target)
413 struct ib_cm_id *new_cm_id;
414 struct ib_qp_attr qp_attr;
415 struct srp_request *req;
420 spin_lock_irq(target->scsi_host->host_lock);
421 if (target->state != SRP_TARGET_LIVE) {
422 spin_unlock_irq(target->scsi_host->host_lock);
425 target->state = SRP_TARGET_CONNECTING;
426 spin_unlock_irq(target->scsi_host->host_lock);
428 srp_disconnect_target(target);
430 * Now get a new local CM ID so that we avoid confusing the
431 * target in case things are really fouled up.
433 new_cm_id = ib_create_cm_id(target->srp_host->dev,
434 srp_cm_handler, target);
435 if (IS_ERR(new_cm_id)) {
436 ret = PTR_ERR(new_cm_id);
439 ib_destroy_cm_id(target->cm_id);
440 target->cm_id = new_cm_id;
442 qp_attr.qp_state = IB_QPS_RESET;
443 ret = ib_modify_qp(target->qp, &qp_attr, IB_QP_STATE);
447 ret = srp_init_qp(target, target->qp);
451 while (ib_poll_cq(target->cq, 1, &wc) > 0)
454 list_for_each_entry(req, &target->req_queue, list) {
455 req->scmnd->result = DID_RESET << 16;
456 req->scmnd->scsi_done(req->scmnd);
462 target->req_head = 0;
463 for (i = 0; i < SRP_SQ_SIZE - 1; ++i)
464 target->req_ring[i].next = i + 1;
465 target->req_ring[SRP_SQ_SIZE - 1].next = -1;
466 INIT_LIST_HEAD(&target->req_queue);
468 ret = srp_connect_target(target);
472 spin_lock_irq(target->scsi_host->host_lock);
473 if (target->state == SRP_TARGET_CONNECTING) {
475 target->state = SRP_TARGET_LIVE;
478 spin_unlock_irq(target->scsi_host->host_lock);
483 printk(KERN_ERR PFX "reconnect failed (%d), removing target port.\n", ret);
486 * We couldn't reconnect, so kill our target port off.
487 * However, we have to defer the real removal because we might
488 * be in the context of the SCSI error handler now, which
489 * would deadlock if we call scsi_remove_host().
491 spin_lock_irq(target->scsi_host->host_lock);
492 if (target->state == SRP_TARGET_CONNECTING) {
493 target->state = SRP_TARGET_DEAD;
494 INIT_WORK(&target->work, srp_remove_work, target);
495 schedule_work(&target->work);
497 spin_unlock_irq(target->scsi_host->host_lock);
502 static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target,
503 struct srp_request *req)
505 struct srp_cmd *cmd = req->cmd->buf;
509 if (!scmnd->request_buffer || scmnd->sc_data_direction == DMA_NONE)
510 return sizeof (struct srp_cmd);
512 if (scmnd->sc_data_direction != DMA_FROM_DEVICE &&
513 scmnd->sc_data_direction != DMA_TO_DEVICE) {
514 printk(KERN_WARNING PFX "Unhandled data direction %d\n",
515 scmnd->sc_data_direction);
520 struct scatterlist *scat = scmnd->request_buffer;
524 n = dma_map_sg(target->srp_host->dev->dma_device,
525 scat, scmnd->use_sg, scmnd->sc_data_direction);
528 struct srp_direct_buf *buf = (void *) cmd->add_data;
530 fmt = SRP_DATA_DESC_DIRECT;
532 buf->va = cpu_to_be64(sg_dma_address(scat));
533 buf->key = cpu_to_be32(target->srp_host->mr->rkey);
534 buf->len = cpu_to_be32(sg_dma_len(scat));
536 len = sizeof (struct srp_cmd) +
537 sizeof (struct srp_direct_buf);
539 struct srp_indirect_buf *buf = (void *) cmd->add_data;
542 fmt = SRP_DATA_DESC_INDIRECT;
544 if (scmnd->sc_data_direction == DMA_TO_DEVICE)
545 cmd->data_out_desc_cnt = n;
547 cmd->data_in_desc_cnt = n;
549 buf->table_desc.va = cpu_to_be64(req->cmd->dma +
552 buf->table_desc.key =
553 cpu_to_be32(target->srp_host->mr->rkey);
554 buf->table_desc.len =
555 cpu_to_be32(n * sizeof (struct srp_direct_buf));
557 for (i = 0; i < n; ++i) {
558 buf->desc_list[i].va = cpu_to_be64(sg_dma_address(&scat[i]));
559 buf->desc_list[i].key =
560 cpu_to_be32(target->srp_host->mr->rkey);
561 buf->desc_list[i].len = cpu_to_be32(sg_dma_len(&scat[i]));
563 datalen += sg_dma_len(&scat[i]);
566 buf->len = cpu_to_be32(datalen);
568 len = sizeof (struct srp_cmd) +
569 sizeof (struct srp_indirect_buf) +
570 n * sizeof (struct srp_direct_buf);
573 struct srp_direct_buf *buf = (void *) cmd->add_data;
576 dma = dma_map_single(target->srp_host->dev->dma_device,
577 scmnd->request_buffer, scmnd->request_bufflen,
578 scmnd->sc_data_direction);
579 if (dma_mapping_error(dma)) {
580 printk(KERN_WARNING PFX "unable to map %p/%d (dir %d)\n",
581 scmnd->request_buffer, (int) scmnd->request_bufflen,
582 scmnd->sc_data_direction);
586 pci_unmap_addr_set(req, direct_mapping, dma);
588 buf->va = cpu_to_be64(dma);
589 buf->key = cpu_to_be32(target->srp_host->mr->rkey);
590 buf->len = cpu_to_be32(scmnd->request_bufflen);
592 fmt = SRP_DATA_DESC_DIRECT;
594 len = sizeof (struct srp_cmd) + sizeof (struct srp_direct_buf);
597 if (scmnd->sc_data_direction == DMA_TO_DEVICE)
598 cmd->buf_fmt = fmt << 4;
606 static void srp_unmap_data(struct scsi_cmnd *scmnd,
607 struct srp_target_port *target,
608 struct srp_request *req)
610 if (!scmnd->request_buffer ||
611 (scmnd->sc_data_direction != DMA_TO_DEVICE &&
612 scmnd->sc_data_direction != DMA_FROM_DEVICE))
616 dma_unmap_sg(target->srp_host->dev->dma_device,
617 (struct scatterlist *) scmnd->request_buffer,
618 scmnd->use_sg, scmnd->sc_data_direction);
620 dma_unmap_single(target->srp_host->dev->dma_device,
621 pci_unmap_addr(req, direct_mapping),
622 scmnd->request_bufflen,
623 scmnd->sc_data_direction);
626 static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
628 struct srp_request *req;
629 struct scsi_cmnd *scmnd;
633 delta = (s32) be32_to_cpu(rsp->req_lim_delta);
635 spin_lock_irqsave(target->scsi_host->host_lock, flags);
637 target->req_lim += delta;
639 req = &target->req_ring[rsp->tag & ~SRP_TAG_TSK_MGMT];
641 if (unlikely(rsp->tag & SRP_TAG_TSK_MGMT)) {
642 if (be32_to_cpu(rsp->resp_data_len) < 4)
643 req->tsk_status = -1;
645 req->tsk_status = rsp->data[3];
646 complete(&req->done);
650 printk(KERN_ERR "Null scmnd for RSP w/tag %016llx\n",
651 (unsigned long long) rsp->tag);
652 scmnd->result = rsp->status;
654 if (rsp->flags & SRP_RSP_FLAG_SNSVALID) {
655 memcpy(scmnd->sense_buffer, rsp->data +
656 be32_to_cpu(rsp->resp_data_len),
657 min_t(int, be32_to_cpu(rsp->sense_data_len),
658 SCSI_SENSE_BUFFERSIZE));
661 if (rsp->flags & (SRP_RSP_FLAG_DOOVER | SRP_RSP_FLAG_DOUNDER))
662 scmnd->resid = be32_to_cpu(rsp->data_out_res_cnt);
663 else if (rsp->flags & (SRP_RSP_FLAG_DIOVER | SRP_RSP_FLAG_DIUNDER))
664 scmnd->resid = be32_to_cpu(rsp->data_in_res_cnt);
666 srp_unmap_data(scmnd, target, req);
668 if (!req->tsk_mgmt) {
670 scmnd->host_scribble = (void *) -1L;
671 scmnd->scsi_done(scmnd);
673 list_del(&req->list);
674 req->next = target->req_head;
675 target->req_head = rsp->tag & ~SRP_TAG_TSK_MGMT;
680 spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
683 static void srp_reconnect_work(void *target_ptr)
685 struct srp_target_port *target = target_ptr;
687 srp_reconnect_target(target);
690 static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc)
695 iu = target->rx_ring[wc->wr_id & ~SRP_OP_RECV];
697 dma_sync_single_for_cpu(target->srp_host->dev->dma_device, iu->dma,
698 target->max_ti_iu_len, DMA_FROM_DEVICE);
700 opcode = *(u8 *) iu->buf;
705 printk(KERN_ERR PFX "recv completion, opcode 0x%02x\n", opcode);
707 for (i = 0; i < wc->byte_len; ++i) {
709 printk(KERN_ERR " [%02x] ", i);
710 printk(" %02x", ((u8 *) iu->buf)[i]);
711 if ((i + 1) % 8 == 0)
715 if (wc->byte_len % 8)
721 srp_process_rsp(target, iu->buf);
725 /* XXX Handle target logout */
726 printk(KERN_WARNING PFX "Got target logout request\n");
730 printk(KERN_WARNING PFX "Unhandled SRP opcode 0x%02x\n", opcode);
734 dma_sync_single_for_device(target->srp_host->dev->dma_device, iu->dma,
735 target->max_ti_iu_len, DMA_FROM_DEVICE);
738 static void srp_completion(struct ib_cq *cq, void *target_ptr)
740 struct srp_target_port *target = target_ptr;
744 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
745 while (ib_poll_cq(cq, 1, &wc) > 0) {
747 printk(KERN_ERR PFX "failed %s status %d\n",
748 wc.wr_id & SRP_OP_RECV ? "receive" : "send",
750 spin_lock_irqsave(target->scsi_host->host_lock, flags);
751 if (target->state == SRP_TARGET_LIVE)
752 schedule_work(&target->work);
753 spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
757 if (wc.wr_id & SRP_OP_RECV)
758 srp_handle_recv(target, &wc);
764 static int __srp_post_recv(struct srp_target_port *target)
768 struct ib_recv_wr wr, *bad_wr;
772 next = target->rx_head & (SRP_RQ_SIZE - 1);
773 wr.wr_id = next | SRP_OP_RECV;
774 iu = target->rx_ring[next];
777 list.length = iu->size;
778 list.lkey = target->srp_host->mr->lkey;
784 ret = ib_post_recv(target->qp, &wr, &bad_wr);
791 static int srp_post_recv(struct srp_target_port *target)
796 spin_lock_irqsave(target->scsi_host->host_lock, flags);
797 ret = __srp_post_recv(target);
798 spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
804 * Must be called with target->scsi_host->host_lock held to protect
805 * req_lim and tx_head.
807 static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target)
809 if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE)
812 return target->tx_ring[target->tx_head & SRP_SQ_SIZE];
816 * Must be called with target->scsi_host->host_lock held to protect
817 * req_lim and tx_head.
819 static int __srp_post_send(struct srp_target_port *target,
820 struct srp_iu *iu, int len)
823 struct ib_send_wr wr, *bad_wr;
826 if (target->req_lim < 1) {
827 printk(KERN_ERR PFX "Target has req_lim %d\n", target->req_lim);
833 list.lkey = target->srp_host->mr->lkey;
836 wr.wr_id = target->tx_head & SRP_SQ_SIZE;
839 wr.opcode = IB_WR_SEND;
840 wr.send_flags = IB_SEND_SIGNALED;
842 ret = ib_post_send(target->qp, &wr, &bad_wr);
852 static int srp_queuecommand(struct scsi_cmnd *scmnd,
853 void (*done)(struct scsi_cmnd *))
855 struct srp_target_port *target = host_to_target(scmnd->device->host);
856 struct srp_request *req;
862 if (target->state == SRP_TARGET_CONNECTING)
865 if (target->state == SRP_TARGET_DEAD ||
866 target->state == SRP_TARGET_REMOVED) {
867 scmnd->result = DID_BAD_TARGET << 16;
872 iu = __srp_get_tx_iu(target);
876 dma_sync_single_for_cpu(target->srp_host->dev->dma_device, iu->dma,
877 SRP_MAX_IU_LEN, DMA_TO_DEVICE);
879 req_index = target->req_head;
881 scmnd->scsi_done = done;
883 scmnd->host_scribble = (void *) req_index;
886 memset(cmd, 0, sizeof *cmd);
888 cmd->opcode = SRP_CMD;
889 cmd->lun = cpu_to_be64((u64) scmnd->device->lun << 48);
890 cmd->tag = req_index;
891 memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);
893 req = &target->req_ring[req_index];
898 req->tsk_mgmt = NULL;
900 len = srp_map_data(scmnd, target, req);
902 printk(KERN_ERR PFX "Failed to map data\n");
906 if (__srp_post_recv(target)) {
907 printk(KERN_ERR PFX "Recv failed\n");
911 dma_sync_single_for_device(target->srp_host->dev->dma_device, iu->dma,
912 SRP_MAX_IU_LEN, DMA_TO_DEVICE);
914 if (__srp_post_send(target, iu, len)) {
915 printk(KERN_ERR PFX "Send failed\n");
919 target->req_head = req->next;
920 list_add_tail(&req->list, &target->req_queue);
925 srp_unmap_data(scmnd, target, req);
928 return SCSI_MLQUEUE_HOST_BUSY;
931 static int srp_alloc_iu_bufs(struct srp_target_port *target)
935 for (i = 0; i < SRP_RQ_SIZE; ++i) {
936 target->rx_ring[i] = srp_alloc_iu(target->srp_host,
937 target->max_ti_iu_len,
938 GFP_KERNEL, DMA_FROM_DEVICE);
939 if (!target->rx_ring[i])
943 for (i = 0; i < SRP_SQ_SIZE + 1; ++i) {
944 target->tx_ring[i] = srp_alloc_iu(target->srp_host,
946 GFP_KERNEL, DMA_TO_DEVICE);
947 if (!target->tx_ring[i])
954 for (i = 0; i < SRP_RQ_SIZE; ++i) {
955 srp_free_iu(target->srp_host, target->rx_ring[i]);
956 target->rx_ring[i] = NULL;
959 for (i = 0; i < SRP_SQ_SIZE + 1; ++i) {
960 srp_free_iu(target->srp_host, target->tx_ring[i]);
961 target->tx_ring[i] = NULL;
967 static void srp_cm_rej_handler(struct ib_cm_id *cm_id,
968 struct ib_cm_event *event,
969 struct srp_target_port *target)
971 struct ib_class_port_info *cpi;
974 switch (event->param.rej_rcvd.reason) {
975 case IB_CM_REJ_PORT_CM_REDIRECT:
976 cpi = event->param.rej_rcvd.ari;
977 target->path.dlid = cpi->redirect_lid;
978 target->path.pkey = cpi->redirect_pkey;
979 cm_id->remote_cm_qpn = be32_to_cpu(cpi->redirect_qp) & 0x00ffffff;
980 memcpy(target->path.dgid.raw, cpi->redirect_gid, 16);
982 target->status = target->path.dlid ?
983 SRP_DLID_REDIRECT : SRP_PORT_REDIRECT;
986 case IB_CM_REJ_PORT_REDIRECT:
987 if (topspin_workarounds &&
988 !memcmp(&target->ioc_guid, topspin_oui, 3)) {
990 * Topspin/Cisco SRP gateways incorrectly send
991 * reject reason code 25 when they mean 24
994 memcpy(target->path.dgid.raw,
995 event->param.rej_rcvd.ari, 16);
997 printk(KERN_DEBUG PFX "Topspin/Cisco redirect to target port GID %016llx%016llx\n",
998 (unsigned long long) be64_to_cpu(target->path.dgid.global.subnet_prefix),
999 (unsigned long long) be64_to_cpu(target->path.dgid.global.interface_id));
1001 target->status = SRP_PORT_REDIRECT;
1003 printk(KERN_WARNING " REJ reason: IB_CM_REJ_PORT_REDIRECT\n");
1004 target->status = -ECONNRESET;
1008 case IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID:
1009 printk(KERN_WARNING " REJ reason: IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID\n");
1010 target->status = -ECONNRESET;
1013 case IB_CM_REJ_CONSUMER_DEFINED:
1014 opcode = *(u8 *) event->private_data;
1015 if (opcode == SRP_LOGIN_REJ) {
1016 struct srp_login_rej *rej = event->private_data;
1017 u32 reason = be32_to_cpu(rej->reason);
1019 if (reason == SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE)
1020 printk(KERN_WARNING PFX
1021 "SRP_LOGIN_REJ: requested max_it_iu_len too large\n");
1023 printk(KERN_WARNING PFX
1024 "SRP LOGIN REJECTED, reason 0x%08x\n", reason);
1026 printk(KERN_WARNING " REJ reason: IB_CM_REJ_CONSUMER_DEFINED,"
1027 " opcode 0x%02x\n", opcode);
1028 target->status = -ECONNRESET;
1032 printk(KERN_WARNING " REJ reason 0x%x\n",
1033 event->param.rej_rcvd.reason);
1034 target->status = -ECONNRESET;
1038 static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
1040 struct srp_target_port *target = cm_id->context;
1041 struct ib_qp_attr *qp_attr = NULL;
1046 switch (event->event) {
1047 case IB_CM_REQ_ERROR:
1048 printk(KERN_DEBUG PFX "Sending CM REQ failed\n");
1050 target->status = -ECONNRESET;
1053 case IB_CM_REP_RECEIVED:
1055 opcode = *(u8 *) event->private_data;
1057 if (opcode == SRP_LOGIN_RSP) {
1058 struct srp_login_rsp *rsp = event->private_data;
1060 target->max_ti_iu_len = be32_to_cpu(rsp->max_ti_iu_len);
1061 target->req_lim = be32_to_cpu(rsp->req_lim_delta);
1063 target->scsi_host->can_queue = min(target->req_lim,
1064 target->scsi_host->can_queue);
1066 printk(KERN_WARNING PFX "Unhandled RSP opcode %#x\n", opcode);
1067 target->status = -ECONNRESET;
1071 target->status = srp_alloc_iu_bufs(target);
1075 qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL);
1077 target->status = -ENOMEM;
1081 qp_attr->qp_state = IB_QPS_RTR;
1082 target->status = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
1086 target->status = ib_modify_qp(target->qp, qp_attr, attr_mask);
1090 target->status = srp_post_recv(target);
1094 qp_attr->qp_state = IB_QPS_RTS;
1095 target->status = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
1099 target->status = ib_modify_qp(target->qp, qp_attr, attr_mask);
1103 target->status = ib_send_cm_rtu(cm_id, NULL, 0);
1109 case IB_CM_REJ_RECEIVED:
1110 printk(KERN_DEBUG PFX "REJ received\n");
1113 srp_cm_rej_handler(cm_id, event, target);
1116 case IB_CM_MRA_RECEIVED:
1117 printk(KERN_ERR PFX "MRA received\n");
1120 case IB_CM_DREP_RECEIVED:
1123 case IB_CM_TIMEWAIT_EXIT:
1124 printk(KERN_ERR PFX "connection closed\n");
1131 printk(KERN_WARNING PFX "Unhandled CM event %d\n", event->event);
1136 complete(&target->done);
1143 static int srp_send_tsk_mgmt(struct scsi_cmnd *scmnd, u8 func)
1145 struct srp_target_port *target = host_to_target(scmnd->device->host);
1146 struct srp_request *req;
1148 struct srp_tsk_mgmt *tsk_mgmt;
1152 spin_lock_irq(target->scsi_host->host_lock);
1154 if (scmnd->host_scribble == (void *) -1L)
1157 req_index = (long) scmnd->host_scribble;
1158 printk(KERN_ERR "Abort for req_index %d\n", req_index);
1160 req = &target->req_ring[req_index];
1161 init_completion(&req->done);
1163 iu = __srp_get_tx_iu(target);
1168 memset(tsk_mgmt, 0, sizeof *tsk_mgmt);
1170 tsk_mgmt->opcode = SRP_TSK_MGMT;
1171 tsk_mgmt->lun = cpu_to_be64((u64) scmnd->device->lun << 48);
1172 tsk_mgmt->tag = req_index | SRP_TAG_TSK_MGMT;
1173 tsk_mgmt->tsk_mgmt_func = func;
1174 tsk_mgmt->task_tag = req_index;
1176 if (__srp_post_send(target, iu, sizeof *tsk_mgmt))
1181 spin_unlock_irq(target->scsi_host->host_lock);
1182 if (!wait_for_completion_timeout(&req->done,
1183 msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS)))
1185 spin_lock_irq(target->scsi_host->host_lock);
1187 if (req->cmd_done) {
1188 list_del(&req->list);
1189 req->next = target->req_head;
1190 target->req_head = req_index;
1192 scmnd->scsi_done(scmnd);
1193 } else if (!req->tsk_status) {
1194 scmnd->result = DID_ABORT << 16;
1199 spin_unlock_irq(target->scsi_host->host_lock);
1203 static int srp_abort(struct scsi_cmnd *scmnd)
1205 printk(KERN_ERR "SRP abort called\n");
1207 return srp_send_tsk_mgmt(scmnd, SRP_TSK_ABORT_TASK);
1210 static int srp_reset_device(struct scsi_cmnd *scmnd)
1212 printk(KERN_ERR "SRP reset_device called\n");
1214 return srp_send_tsk_mgmt(scmnd, SRP_TSK_LUN_RESET);
1217 static int srp_reset_host(struct scsi_cmnd *scmnd)
1219 struct srp_target_port *target = host_to_target(scmnd->device->host);
1222 printk(KERN_ERR PFX "SRP reset_host called\n");
1224 if (!srp_reconnect_target(target))
1230 static struct scsi_host_template srp_template = {
1231 .module = THIS_MODULE,
1233 .info = srp_target_info,
1234 .queuecommand = srp_queuecommand,
1235 .eh_abort_handler = srp_abort,
1236 .eh_device_reset_handler = srp_reset_device,
1237 .eh_host_reset_handler = srp_reset_host,
1238 .can_queue = SRP_SQ_SIZE,
1240 .sg_tablesize = SRP_MAX_INDIRECT,
1241 .cmd_per_lun = SRP_SQ_SIZE,
1242 .use_clustering = ENABLE_CLUSTERING
1245 static int srp_add_target(struct srp_host *host, struct srp_target_port *target)
1247 sprintf(target->target_name, "SRP.T10:%016llX",
1248 (unsigned long long) be64_to_cpu(target->id_ext));
1250 if (scsi_add_host(target->scsi_host, host->dev->dma_device))
1253 down(&host->target_mutex);
1254 list_add_tail(&target->list, &host->target_list);
1255 up(&host->target_mutex);
1257 target->state = SRP_TARGET_LIVE;
1259 /* XXX: are we supposed to have a definition of SCAN_WILD_CARD ?? */
1260 scsi_scan_target(&target->scsi_host->shost_gendev,
1261 0, target->scsi_id, ~0, 0);
1266 static void srp_release_class_dev(struct class_device *class_dev)
1268 struct srp_host *host =
1269 container_of(class_dev, struct srp_host, class_dev);
1271 complete(&host->released);
1274 static struct class srp_class = {
1275 .name = "infiniband_srp",
1276 .release = srp_release_class_dev
1280 * Target ports are added by writing
1282 * id_ext=<SRP ID ext>,ioc_guid=<SRP IOC GUID>,dgid=<dest GID>,
1283 * pkey=<P_Key>,service_id=<service ID>
1285 * to the add_target sysfs attribute.
1289 SRP_OPT_ID_EXT = 1 << 0,
1290 SRP_OPT_IOC_GUID = 1 << 1,
1291 SRP_OPT_DGID = 1 << 2,
1292 SRP_OPT_PKEY = 1 << 3,
1293 SRP_OPT_SERVICE_ID = 1 << 4,
1294 SRP_OPT_MAX_SECT = 1 << 5,
1295 SRP_OPT_ALL = (SRP_OPT_ID_EXT |
1299 SRP_OPT_SERVICE_ID),
1302 static match_table_t srp_opt_tokens = {
1303 { SRP_OPT_ID_EXT, "id_ext=%s" },
1304 { SRP_OPT_IOC_GUID, "ioc_guid=%s" },
1305 { SRP_OPT_DGID, "dgid=%s" },
1306 { SRP_OPT_PKEY, "pkey=%x" },
1307 { SRP_OPT_SERVICE_ID, "service_id=%s" },
1308 { SRP_OPT_MAX_SECT, "max_sect=%d" },
1309 { SRP_OPT_ERR, NULL }
1312 static int srp_parse_options(const char *buf, struct srp_target_port *target)
1314 char *options, *sep_opt;
1317 substring_t args[MAX_OPT_ARGS];
1323 options = kstrdup(buf, GFP_KERNEL);
1328 while ((p = strsep(&sep_opt, ",")) != NULL) {
1332 token = match_token(p, srp_opt_tokens, args);
1336 case SRP_OPT_ID_EXT:
1337 p = match_strdup(args);
1338 target->id_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
1342 case SRP_OPT_IOC_GUID:
1343 p = match_strdup(args);
1344 target->ioc_guid = cpu_to_be64(simple_strtoull(p, NULL, 16));
1349 p = match_strdup(args);
1350 if (strlen(p) != 32) {
1351 printk(KERN_WARNING PFX "bad dest GID parameter '%s'\n", p);
1355 for (i = 0; i < 16; ++i) {
1356 strlcpy(dgid, p + i * 2, 3);
1357 target->path.dgid.raw[i] = simple_strtoul(dgid, NULL, 16);
1362 if (match_hex(args, &token)) {
1363 printk(KERN_WARNING PFX "bad P_Key parameter '%s'\n", p);
1366 target->path.pkey = cpu_to_be16(token);
1369 case SRP_OPT_SERVICE_ID:
1370 p = match_strdup(args);
1371 target->service_id = cpu_to_be64(simple_strtoull(p, NULL, 16));
1375 case SRP_OPT_MAX_SECT:
1376 if (match_int(args, &token)) {
1377 printk(KERN_WARNING PFX "bad max sect parameter '%s'\n", p);
1380 target->scsi_host->max_sectors = token;
1384 printk(KERN_WARNING PFX "unknown parameter or missing value "
1385 "'%s' in target creation request\n", p);
1390 if ((opt_mask & SRP_OPT_ALL) == SRP_OPT_ALL)
1393 for (i = 0; i < ARRAY_SIZE(srp_opt_tokens); ++i)
1394 if ((srp_opt_tokens[i].token & SRP_OPT_ALL) &&
1395 !(srp_opt_tokens[i].token & opt_mask))
1396 printk(KERN_WARNING PFX "target creation request is "
1397 "missing parameter '%s'\n",
1398 srp_opt_tokens[i].pattern);
1405 static ssize_t srp_create_target(struct class_device *class_dev,
1406 const char *buf, size_t count)
1408 struct srp_host *host =
1409 container_of(class_dev, struct srp_host, class_dev);
1410 struct Scsi_Host *target_host;
1411 struct srp_target_port *target;
1415 target_host = scsi_host_alloc(&srp_template,
1416 sizeof (struct srp_target_port));
1420 target = host_to_target(target_host);
1421 memset(target, 0, sizeof *target);
1423 target->scsi_host = target_host;
1424 target->srp_host = host;
1426 INIT_WORK(&target->work, srp_reconnect_work, target);
1428 for (i = 0; i < SRP_SQ_SIZE - 1; ++i)
1429 target->req_ring[i].next = i + 1;
1430 target->req_ring[SRP_SQ_SIZE - 1].next = -1;
1431 INIT_LIST_HEAD(&target->req_queue);
1433 ret = srp_parse_options(buf, target);
1437 ib_get_cached_gid(host->dev, host->port, 0, &target->path.sgid);
1439 printk(KERN_DEBUG PFX "new target: id_ext %016llx ioc_guid %016llx pkey %04x "
1440 "service_id %016llx dgid %04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x\n",
1441 (unsigned long long) be64_to_cpu(target->id_ext),
1442 (unsigned long long) be64_to_cpu(target->ioc_guid),
1443 be16_to_cpu(target->path.pkey),
1444 (unsigned long long) be64_to_cpu(target->service_id),
1445 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[0]),
1446 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[2]),
1447 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[4]),
1448 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[6]),
1449 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[8]),
1450 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[10]),
1451 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[12]),
1452 (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[14]));
1454 ret = srp_create_target_ib(target);
1458 target->cm_id = ib_create_cm_id(host->dev, srp_cm_handler, target);
1459 if (IS_ERR(target->cm_id)) {
1460 ret = PTR_ERR(target->cm_id);
1464 ret = srp_connect_target(target);
1466 printk(KERN_ERR PFX "Connection failed\n");
1470 ret = srp_add_target(host, target);
1472 goto err_disconnect;
1477 srp_disconnect_target(target);
1480 ib_destroy_cm_id(target->cm_id);
1483 srp_free_target_ib(target);
1486 scsi_host_put(target_host);
1491 static CLASS_DEVICE_ATTR(add_target, S_IWUSR, NULL, srp_create_target);
1493 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
1495 struct srp_host *host =
1496 container_of(class_dev, struct srp_host, class_dev);
1498 return sprintf(buf, "%s\n", host->dev->name);
1501 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1503 static ssize_t show_port(struct class_device *class_dev, char *buf)
1505 struct srp_host *host =
1506 container_of(class_dev, struct srp_host, class_dev);
1508 return sprintf(buf, "%d\n", host->port);
1511 static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1513 static struct srp_host *srp_add_port(struct ib_device *device,
1514 __be64 node_guid, u8 port)
1516 struct srp_host *host;
1518 host = kzalloc(sizeof *host, GFP_KERNEL);
1522 INIT_LIST_HEAD(&host->target_list);
1523 init_MUTEX(&host->target_mutex);
1524 init_completion(&host->released);
1528 host->initiator_port_id[7] = port;
1529 memcpy(host->initiator_port_id + 8, &node_guid, 8);
1531 host->pd = ib_alloc_pd(device);
1532 if (IS_ERR(host->pd))
1535 host->mr = ib_get_dma_mr(host->pd,
1536 IB_ACCESS_LOCAL_WRITE |
1537 IB_ACCESS_REMOTE_READ |
1538 IB_ACCESS_REMOTE_WRITE);
1539 if (IS_ERR(host->mr))
1542 host->class_dev.class = &srp_class;
1543 host->class_dev.dev = device->dma_device;
1544 snprintf(host->class_dev.class_id, BUS_ID_SIZE, "srp-%s-%d",
1545 device->name, port);
1547 if (class_device_register(&host->class_dev))
1549 if (class_device_create_file(&host->class_dev, &class_device_attr_add_target))
1551 if (class_device_create_file(&host->class_dev, &class_device_attr_ibdev))
1553 if (class_device_create_file(&host->class_dev, &class_device_attr_port))
1559 class_device_unregister(&host->class_dev);
1562 ib_dereg_mr(host->mr);
1565 ib_dealloc_pd(host->pd);
1573 static void srp_add_one(struct ib_device *device)
1575 struct list_head *dev_list;
1576 struct srp_host *host;
1577 struct ib_device_attr *dev_attr;
1580 dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
1584 if (ib_query_device(device, dev_attr)) {
1585 printk(KERN_WARNING PFX "Couldn't query node GUID for %s.\n",
1590 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1594 INIT_LIST_HEAD(dev_list);
1596 if (device->node_type == IB_NODE_SWITCH) {
1601 e = device->phys_port_cnt;
1604 for (p = s; p <= e; ++p) {
1605 host = srp_add_port(device, dev_attr->node_guid, p);
1607 list_add_tail(&host->list, dev_list);
1610 ib_set_client_data(device, &srp_client, dev_list);
1616 static void srp_remove_one(struct ib_device *device)
1618 struct list_head *dev_list;
1619 struct srp_host *host, *tmp_host;
1620 LIST_HEAD(target_list);
1621 struct srp_target_port *target, *tmp_target;
1622 unsigned long flags;
1624 dev_list = ib_get_client_data(device, &srp_client);
1626 list_for_each_entry_safe(host, tmp_host, dev_list, list) {
1627 class_device_unregister(&host->class_dev);
1629 * Wait for the sysfs entry to go away, so that no new
1630 * target ports can be created.
1632 wait_for_completion(&host->released);
1635 * Mark all target ports as removed, so we stop queueing
1636 * commands and don't try to reconnect.
1638 down(&host->target_mutex);
1639 list_for_each_entry_safe(target, tmp_target,
1640 &host->target_list, list) {
1641 spin_lock_irqsave(target->scsi_host->host_lock, flags);
1642 if (target->state != SRP_TARGET_REMOVED)
1643 target->state = SRP_TARGET_REMOVED;
1644 spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
1646 up(&host->target_mutex);
1649 * Wait for any reconnection tasks that may have
1650 * started before we marked our target ports as
1651 * removed, and any target port removal tasks.
1653 flush_scheduled_work();
1655 list_for_each_entry_safe(target, tmp_target,
1656 &host->target_list, list) {
1657 scsi_remove_host(target->scsi_host);
1658 srp_disconnect_target(target);
1659 ib_destroy_cm_id(target->cm_id);
1660 srp_free_target_ib(target);
1661 scsi_host_put(target->scsi_host);
1664 ib_dereg_mr(host->mr);
1665 ib_dealloc_pd(host->pd);
1672 static int __init srp_init_module(void)
1676 ret = class_register(&srp_class);
1678 printk(KERN_ERR PFX "couldn't register class infiniband_srp\n");
1682 ret = ib_register_client(&srp_client);
1684 printk(KERN_ERR PFX "couldn't register IB client\n");
1685 class_unregister(&srp_class);
1692 static void __exit srp_cleanup_module(void)
1694 ib_unregister_client(&srp_client);
1695 class_unregister(&srp_class);
1698 module_init(srp_init_module);
1699 module_exit(srp_cleanup_module);