aboutsummaryrefslogtreecommitdiffstats
path: root/drivers
diff options
context:
space:
mode:
authorOr Gerlitz <ogerlitz@voltaire.com>2006-05-11 03:02:46 -0400
committerRoland Dreier <rolandd@cisco.com>2006-06-22 10:51:11 -0400
commit1cfa0a75dbef1d5bf687aacafabb023288f6b36a (patch)
tree8296842d0f7afb479e1437330d54bd31a0272c95 /drivers
parente85b24b5e7de9f507c6253183d089370f37618c5 (diff)
IB/iser: iSER RDMA CM (CMA) and IB verbs interaction
This file contains the low level interaction with the RDMA CM and the IB verbs, where iSER is consumer of both. Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com> Signed-off-by: Roland Dreier <rolandd@cisco.com>
Diffstat (limited to 'drivers')
-rw-r--r--drivers/infiniband/ulp/iser/iser_verbs.c827
1 files changed, 827 insertions, 0 deletions
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
new file mode 100644
index 000000000000..ff117bbf81b4
--- /dev/null
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -0,0 +1,827 @@
1/*
2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 *
33 * $Id: iser_verbs.c 7051 2006-05-10 12:29:11Z ogerlitz $
34 */
35#include <asm/io.h>
36#include <linux/kernel.h>
37#include <linux/module.h>
38#include <linux/smp_lock.h>
39#include <linux/delay.h>
40#include <linux/version.h>
41
42#include "iscsi_iser.h"
43
44#define ISCSI_ISER_MAX_CONN 8
45#define ISER_MAX_CQ_LEN ((ISER_QP_MAX_RECV_DTOS + \
46 ISER_QP_MAX_REQ_DTOS) * \
47 ISCSI_ISER_MAX_CONN)
48
49static void iser_cq_tasklet_fn(unsigned long data);
50static void iser_cq_callback(struct ib_cq *cq, void *cq_context);
51static void iser_comp_error_worker(void *data);
52
53static void iser_cq_event_callback(struct ib_event *cause, void *context)
54{
55 iser_err("got cq event %d \n", cause->event);
56}
57
58static void iser_qp_event_callback(struct ib_event *cause, void *context)
59{
60 iser_err("got qp event %d\n",cause->event);
61}
62
63/**
64 * iser_create_device_ib_res - creates Protection Domain (PD), Completion
65 * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
66 * the adapator.
67 *
68 * returns 0 on success, -1 on failure
69 */
70static int iser_create_device_ib_res(struct iser_device *device)
71{
72 device->pd = ib_alloc_pd(device->ib_device);
73 if (IS_ERR(device->pd))
74 goto pd_err;
75
76 device->cq = ib_create_cq(device->ib_device,
77 iser_cq_callback,
78 iser_cq_event_callback,
79 (void *)device,
80 ISER_MAX_CQ_LEN);
81 if (IS_ERR(device->cq))
82 goto cq_err;
83
84 if (ib_req_notify_cq(device->cq, IB_CQ_NEXT_COMP))
85 goto cq_arm_err;
86
87 tasklet_init(&device->cq_tasklet,
88 iser_cq_tasklet_fn,
89 (unsigned long)device);
90
91 device->mr = ib_get_dma_mr(device->pd,
92 IB_ACCESS_LOCAL_WRITE);
93 if (IS_ERR(device->mr))
94 goto dma_mr_err;
95
96 return 0;
97
98dma_mr_err:
99 tasklet_kill(&device->cq_tasklet);
100cq_arm_err:
101 ib_destroy_cq(device->cq);
102cq_err:
103 ib_dealloc_pd(device->pd);
104pd_err:
105 iser_err("failed to allocate an IB resource\n");
106 return -1;
107}
108
109/**
110 * iser_free_device_ib_res - destory/dealloc/dereg the DMA MR,
111 * CQ and PD created with the device associated with the adapator.
112 */
113static void iser_free_device_ib_res(struct iser_device *device)
114{
115 BUG_ON(device->mr == NULL);
116
117 tasklet_kill(&device->cq_tasklet);
118
119 (void)ib_dereg_mr(device->mr);
120 (void)ib_destroy_cq(device->cq);
121 (void)ib_dealloc_pd(device->pd);
122
123 device->mr = NULL;
124 device->cq = NULL;
125 device->pd = NULL;
126}
127
128/**
129 * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP)
130 *
131 * returns 0 on success, -1 on failure
132 */
133static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
134{
135 struct iser_device *device;
136 struct ib_qp_init_attr init_attr;
137 int ret;
138 struct ib_fmr_pool_param params;
139
140 BUG_ON(ib_conn->device == NULL);
141
142 device = ib_conn->device;
143
144 ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
145 (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
146 GFP_KERNEL);
147 if (!ib_conn->page_vec) {
148 ret = -ENOMEM;
149 goto alloc_err;
150 }
151 ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
152
153 params.page_shift = PAGE_SHIFT;
154 /* when the first/last SG element are not start/end *
155 * page aligned, the map whould be of N+1 pages */
156 params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
157 /* make the pool size twice the max number of SCSI commands *
158 * the ML is expected to queue, watermark for unmap at 50% */
159 params.pool_size = ISCSI_XMIT_CMDS_MAX * 2;
160 params.dirty_watermark = ISCSI_XMIT_CMDS_MAX;
161 params.cache = 0;
162 params.flush_function = NULL;
163 params.access = (IB_ACCESS_LOCAL_WRITE |
164 IB_ACCESS_REMOTE_WRITE |
165 IB_ACCESS_REMOTE_READ);
166
167 ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, &params);
168 if (IS_ERR(ib_conn->fmr_pool)) {
169 ret = PTR_ERR(ib_conn->fmr_pool);
170 goto fmr_pool_err;
171 }
172
173 memset(&init_attr, 0, sizeof init_attr);
174
175 init_attr.event_handler = iser_qp_event_callback;
176 init_attr.qp_context = (void *)ib_conn;
177 init_attr.send_cq = device->cq;
178 init_attr.recv_cq = device->cq;
179 init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
180 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
181 init_attr.cap.max_send_sge = MAX_REGD_BUF_VECTOR_LEN;
182 init_attr.cap.max_recv_sge = 2;
183 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
184 init_attr.qp_type = IB_QPT_RC;
185
186 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
187 if (ret)
188 goto qp_err;
189
190 ib_conn->qp = ib_conn->cma_id->qp;
191 iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
192 ib_conn, ib_conn->cma_id,
193 ib_conn->fmr_pool, ib_conn->cma_id->qp);
194 return ret;
195
196qp_err:
197 (void)ib_destroy_fmr_pool(ib_conn->fmr_pool);
198fmr_pool_err:
199 kfree(ib_conn->page_vec);
200alloc_err:
201 iser_err("unable to alloc mem or create resource, err %d\n", ret);
202 return ret;
203}
204
205/**
206 * releases the FMR pool, QP and CMA ID objects, returns 0 on success,
207 * -1 on failure
208 */
209static int iser_free_ib_conn_res(struct iser_conn *ib_conn)
210{
211 BUG_ON(ib_conn == NULL);
212
213 iser_err("freeing conn %p cma_id %p fmr pool %p qp %p\n",
214 ib_conn, ib_conn->cma_id,
215 ib_conn->fmr_pool, ib_conn->qp);
216
217 /* qp is created only once both addr & route are resolved */
218 if (ib_conn->fmr_pool != NULL)
219 ib_destroy_fmr_pool(ib_conn->fmr_pool);
220
221 if (ib_conn->qp != NULL)
222 rdma_destroy_qp(ib_conn->cma_id);
223
224 if (ib_conn->cma_id != NULL)
225 rdma_destroy_id(ib_conn->cma_id);
226
227 ib_conn->fmr_pool = NULL;
228 ib_conn->qp = NULL;
229 ib_conn->cma_id = NULL;
230 kfree(ib_conn->page_vec);
231
232 return 0;
233}
234
235/**
236 * based on the resolved device node GUID see if there already allocated
237 * device for this device. If there's no such, create one.
238 */
239static
240struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
241{
242 struct list_head *p_list;
243 struct iser_device *device = NULL;
244
245 mutex_lock(&ig.device_list_mutex);
246
247 p_list = ig.device_list.next;
248 while (p_list != &ig.device_list) {
249 device = list_entry(p_list, struct iser_device, ig_list);
250 /* find if there's a match using the node GUID */
251 if (device->ib_device->node_guid == cma_id->device->node_guid)
252 break;
253 }
254
255 if (device == NULL) {
256 device = kzalloc(sizeof *device, GFP_KERNEL);
257 if (device == NULL)
258 goto out;
259 /* assign this device to the device */
260 device->ib_device = cma_id->device;
261 /* init the device and link it into ig device list */
262 if (iser_create_device_ib_res(device)) {
263 kfree(device);
264 device = NULL;
265 goto out;
266 }
267 list_add(&device->ig_list, &ig.device_list);
268 }
269out:
270 BUG_ON(device == NULL);
271 device->refcount++;
272 mutex_unlock(&ig.device_list_mutex);
273 return device;
274}
275
276/* if there's no demand for this device, release it */
277static void iser_device_try_release(struct iser_device *device)
278{
279 mutex_lock(&ig.device_list_mutex);
280 device->refcount--;
281 iser_err("device %p refcount %d\n",device,device->refcount);
282 if (!device->refcount) {
283 iser_free_device_ib_res(device);
284 list_del(&device->ig_list);
285 kfree(device);
286 }
287 mutex_unlock(&ig.device_list_mutex);
288}
289
290int iser_conn_state_comp(struct iser_conn *ib_conn,
291 enum iser_ib_conn_state comp)
292{
293 int ret;
294
295 spin_lock_bh(&ib_conn->lock);
296 ret = (ib_conn->state == comp);
297 spin_unlock_bh(&ib_conn->lock);
298 return ret;
299}
300
301static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
302 enum iser_ib_conn_state comp,
303 enum iser_ib_conn_state exch)
304{
305 int ret;
306
307 spin_lock_bh(&ib_conn->lock);
308 if ((ret = (ib_conn->state == comp)))
309 ib_conn->state = exch;
310 spin_unlock_bh(&ib_conn->lock);
311 return ret;
312}
313
314/**
315 * triggers start of the disconnect procedures and wait for them to be done
316 */
317void iser_conn_terminate(struct iser_conn *ib_conn)
318{
319 int err = 0;
320
321 /* change the ib conn state only if the conn is UP, however always call
322 * rdma_disconnect since this is the only way to cause the CMA to change
323 * the QP state to ERROR
324 */
325
326 iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING);
327 err = rdma_disconnect(ib_conn->cma_id);
328 if (err)
329 iser_err("Failed to disconnect, conn: 0x%p err %d\n",
330 ib_conn,err);
331
332 wait_event_interruptible(ib_conn->wait,
333 ib_conn->state == ISER_CONN_DOWN);
334
335 iser_conn_release(ib_conn);
336}
337
338static void iser_connect_error(struct rdma_cm_id *cma_id)
339{
340 struct iser_conn *ib_conn;
341 ib_conn = (struct iser_conn *)cma_id->context;
342
343 ib_conn->state = ISER_CONN_DOWN;
344 wake_up_interruptible(&ib_conn->wait);
345}
346
347static void iser_addr_handler(struct rdma_cm_id *cma_id)
348{
349 struct iser_device *device;
350 struct iser_conn *ib_conn;
351 int ret;
352
353 device = iser_device_find_by_ib_device(cma_id);
354 ib_conn = (struct iser_conn *)cma_id->context;
355 ib_conn->device = device;
356
357 ret = rdma_resolve_route(cma_id, 1000);
358 if (ret) {
359 iser_err("resolve route failed: %d\n", ret);
360 iser_connect_error(cma_id);
361 }
362 return;
363}
364
365static void iser_route_handler(struct rdma_cm_id *cma_id)
366{
367 struct rdma_conn_param conn_param;
368 int ret;
369
370 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context);
371 if (ret)
372 goto failure;
373
374 iser_dbg("path.mtu is %d setting it to %d\n",
375 cma_id->route.path_rec->mtu, IB_MTU_1024);
376
377 /* we must set the MTU to 1024 as this is what the target is assuming */
378 if (cma_id->route.path_rec->mtu > IB_MTU_1024)
379 cma_id->route.path_rec->mtu = IB_MTU_1024;
380
381 memset(&conn_param, 0, sizeof conn_param);
382 conn_param.responder_resources = 4;
383 conn_param.initiator_depth = 1;
384 conn_param.retry_count = 7;
385 conn_param.rnr_retry_count = 6;
386
387 ret = rdma_connect(cma_id, &conn_param);
388 if (ret) {
389 iser_err("failure connecting: %d\n", ret);
390 goto failure;
391 }
392
393 return;
394failure:
395 iser_connect_error(cma_id);
396}
397
398static void iser_connected_handler(struct rdma_cm_id *cma_id)
399{
400 struct iser_conn *ib_conn;
401
402 ib_conn = (struct iser_conn *)cma_id->context;
403 ib_conn->state = ISER_CONN_UP;
404 wake_up_interruptible(&ib_conn->wait);
405}
406
407static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
408{
409 struct iser_conn *ib_conn;
410
411 ib_conn = (struct iser_conn *)cma_id->context;
412 ib_conn->disc_evt_flag = 1;
413
414 /* getting here when the state is UP means that the conn is being *
415 * terminated asynchronously from the iSCSI layer's perspective. */
416 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
417 ISER_CONN_TERMINATING))
418 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
419 ISCSI_ERR_CONN_FAILED);
420
421 /* Complete the termination process if no posts are pending */
422 if ((atomic_read(&ib_conn->post_recv_buf_count) == 0) &&
423 (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
424 ib_conn->state = ISER_CONN_DOWN;
425 wake_up_interruptible(&ib_conn->wait);
426 }
427}
428
429static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
430{
431 int ret = 0;
432
433 iser_err("event %d conn %p id %p\n",event->event,cma_id->context,cma_id);
434
435 switch (event->event) {
436 case RDMA_CM_EVENT_ADDR_RESOLVED:
437 iser_addr_handler(cma_id);
438 break;
439 case RDMA_CM_EVENT_ROUTE_RESOLVED:
440 iser_route_handler(cma_id);
441 break;
442 case RDMA_CM_EVENT_ESTABLISHED:
443 iser_connected_handler(cma_id);
444 break;
445 case RDMA_CM_EVENT_ADDR_ERROR:
446 case RDMA_CM_EVENT_ROUTE_ERROR:
447 case RDMA_CM_EVENT_CONNECT_ERROR:
448 case RDMA_CM_EVENT_UNREACHABLE:
449 case RDMA_CM_EVENT_REJECTED:
450 iser_err("event: %d, error: %d\n", event->event, event->status);
451 iser_connect_error(cma_id);
452 break;
453 case RDMA_CM_EVENT_DISCONNECTED:
454 iser_disconnected_handler(cma_id);
455 break;
456 case RDMA_CM_EVENT_DEVICE_REMOVAL:
457 BUG();
458 break;
459 case RDMA_CM_EVENT_CONNECT_RESPONSE:
460 BUG();
461 break;
462 case RDMA_CM_EVENT_CONNECT_REQUEST:
463 default:
464 break;
465 }
466 return ret;
467}
468
469int iser_conn_init(struct iser_conn **ibconn)
470{
471 struct iser_conn *ib_conn;
472
473 ib_conn = kzalloc(sizeof *ib_conn, GFP_KERNEL);
474 if (!ib_conn) {
475 iser_err("can't alloc memory for struct iser_conn\n");
476 return -ENOMEM;
477 }
478 ib_conn->state = ISER_CONN_INIT;
479 init_waitqueue_head(&ib_conn->wait);
480 atomic_set(&ib_conn->post_recv_buf_count, 0);
481 atomic_set(&ib_conn->post_send_buf_count, 0);
482 INIT_WORK(&ib_conn->comperror_work, iser_comp_error_worker,
483 ib_conn);
484 INIT_LIST_HEAD(&ib_conn->conn_list);
485 spin_lock_init(&ib_conn->lock);
486
487 *ibconn = ib_conn;
488 return 0;
489}
490
491 /**
492 * starts the process of connecting to the target
493 * sleeps untill the connection is established or rejected
494 */
495int iser_connect(struct iser_conn *ib_conn,
496 struct sockaddr_in *src_addr,
497 struct sockaddr_in *dst_addr,
498 int non_blocking)
499{
500 struct sockaddr *src, *dst;
501 int err = 0;
502
503 sprintf(ib_conn->name,"%d.%d.%d.%d:%d",
504 NIPQUAD(dst_addr->sin_addr.s_addr), dst_addr->sin_port);
505
506 /* the device is known only --after-- address resolution */
507 ib_conn->device = NULL;
508
509 iser_err("connecting to: %d.%d.%d.%d, port 0x%x\n",
510 NIPQUAD(dst_addr->sin_addr), dst_addr->sin_port);
511
512 ib_conn->state = ISER_CONN_PENDING;
513
514 ib_conn->cma_id = rdma_create_id(iser_cma_handler,
515 (void *)ib_conn,
516 RDMA_PS_TCP);
517 if (IS_ERR(ib_conn->cma_id)) {
518 err = PTR_ERR(ib_conn->cma_id);
519 iser_err("rdma_create_id failed: %d\n", err);
520 goto id_failure;
521 }
522
523 src = (struct sockaddr *)src_addr;
524 dst = (struct sockaddr *)dst_addr;
525 err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000);
526 if (err) {
527 iser_err("rdma_resolve_addr failed: %d\n", err);
528 goto addr_failure;
529 }
530
531 if (!non_blocking) {
532 wait_event_interruptible(ib_conn->wait,
533 (ib_conn->state != ISER_CONN_PENDING));
534
535 if (ib_conn->state != ISER_CONN_UP) {
536 err = -EIO;
537 goto connect_failure;
538 }
539 }
540
541 mutex_lock(&ig.connlist_mutex);
542 list_add(&ib_conn->conn_list, &ig.connlist);
543 mutex_unlock(&ig.connlist_mutex);
544 return 0;
545
546id_failure:
547 ib_conn->cma_id = NULL;
548addr_failure:
549 ib_conn->state = ISER_CONN_DOWN;
550connect_failure:
551 iser_conn_release(ib_conn);
552 return err;
553}
554
555/**
556 * Frees all conn objects and deallocs conn descriptor
557 */
558void iser_conn_release(struct iser_conn *ib_conn)
559{
560 struct iser_device *device = ib_conn->device;
561
562 BUG_ON(ib_conn->state != ISER_CONN_DOWN);
563
564 mutex_lock(&ig.connlist_mutex);
565 list_del(&ib_conn->conn_list);
566 mutex_unlock(&ig.connlist_mutex);
567
568 iser_free_ib_conn_res(ib_conn);
569 ib_conn->device = NULL;
570 /* on EVENT_ADDR_ERROR there's no device yet for this conn */
571 if (device != NULL)
572 iser_device_try_release(device);
573 kfree(ib_conn);
574}
575
576
577/**
578 * iser_reg_page_vec - Register physical memory
579 *
580 * returns: 0 on success, errno code on failure
581 */
582int iser_reg_page_vec(struct iser_conn *ib_conn,
583 struct iser_page_vec *page_vec,
584 struct iser_mem_reg *mem_reg)
585{
586 struct ib_pool_fmr *mem;
587 u64 io_addr;
588 u64 *page_list;
589 int status;
590
591 page_list = page_vec->pages;
592 io_addr = page_list[0];
593
594 mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool,
595 page_list,
596 page_vec->length,
597 &io_addr);
598
599 if (IS_ERR(mem)) {
600 status = (int)PTR_ERR(mem);
601 iser_err("ib_fmr_pool_map_phys failed: %d\n", status);
602 return status;
603 }
604
605 mem_reg->lkey = mem->fmr->lkey;
606 mem_reg->rkey = mem->fmr->rkey;
607 mem_reg->len = page_vec->length * PAGE_SIZE;
608 mem_reg->va = io_addr;
609 mem_reg->mem_h = (void *)mem;
610
611 mem_reg->va += page_vec->offset;
612 mem_reg->len = page_vec->data_size;
613
614 iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, "
615 "entry[0]: (0x%08lx,%ld)] -> "
616 "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n",
617 page_vec, page_vec->length,
618 (unsigned long)page_vec->pages[0],
619 (unsigned long)page_vec->data_size,
620 (unsigned int)mem_reg->lkey, mem_reg->mem_h,
621 (unsigned long)mem_reg->va, (unsigned long)mem_reg->len);
622 return 0;
623}
624
625/**
626 * Unregister (previosuly registered) memory.
627 */
628void iser_unreg_mem(struct iser_mem_reg *reg)
629{
630 int ret;
631
632 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h);
633
634 ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h);
635 if (ret)
636 iser_err("ib_fmr_pool_unmap failed %d\n", ret);
637
638 reg->mem_h = NULL;
639}
640
641/**
642 * iser_dto_to_iov - builds IOV from a dto descriptor
643 */
644static void iser_dto_to_iov(struct iser_dto *dto, struct ib_sge *iov, int iov_len)
645{
646 int i;
647 struct ib_sge *sge;
648 struct iser_regd_buf *regd_buf;
649
650 if (dto->regd_vector_len > iov_len) {
651 iser_err("iov size %d too small for posting dto of len %d\n",
652 iov_len, dto->regd_vector_len);
653 BUG();
654 }
655
656 for (i = 0; i < dto->regd_vector_len; i++) {
657 sge = &iov[i];
658 regd_buf = dto->regd[i];
659
660 sge->addr = regd_buf->reg.va;
661 sge->length = regd_buf->reg.len;
662 sge->lkey = regd_buf->reg.lkey;
663
664 if (dto->used_sz[i] > 0) /* Adjust size */
665 sge->length = dto->used_sz[i];
666
667 /* offset and length should not exceed the regd buf length */
668 if (sge->length + dto->offset[i] > regd_buf->reg.len) {
669 iser_err("Used len:%ld + offset:%d, exceed reg.buf.len:"
670 "%ld in dto:0x%p [%d], va:0x%08lX\n",
671 (unsigned long)sge->length, dto->offset[i],
672 (unsigned long)regd_buf->reg.len, dto, i,
673 (unsigned long)sge->addr);
674 BUG();
675 }
676
677 sge->addr += dto->offset[i]; /* Adjust offset */
678 }
679}
680
681/**
682 * iser_post_recv - Posts a receive buffer.
683 *
684 * returns 0 on success, -1 on failure
685 */
686int iser_post_recv(struct iser_desc *rx_desc)
687{
688 int ib_ret, ret_val = 0;
689 struct ib_recv_wr recv_wr, *recv_wr_failed;
690 struct ib_sge iov[2];
691 struct iser_conn *ib_conn;
692 struct iser_dto *recv_dto = &rx_desc->dto;
693
694 /* Retrieve conn */
695 ib_conn = recv_dto->conn->ib_conn;
696
697 iser_dto_to_iov(recv_dto, iov, 2);
698
699 recv_wr.next = NULL;
700 recv_wr.sg_list = iov;
701 recv_wr.num_sge = recv_dto->regd_vector_len;
702 recv_wr.wr_id = (unsigned long)rx_desc;
703
704 atomic_inc(&ib_conn->post_recv_buf_count);
705 ib_ret = ib_post_recv(ib_conn->qp, &recv_wr, &recv_wr_failed);
706 if (ib_ret) {
707 iser_err("ib_post_recv failed ret=%d\n", ib_ret);
708 atomic_dec(&ib_conn->post_recv_buf_count);
709 ret_val = -1;
710 }
711
712 return ret_val;
713}
714
715/**
716 * iser_start_send - Initiate a Send DTO operation
717 *
718 * returns 0 on success, -1 on failure
719 */
720int iser_post_send(struct iser_desc *tx_desc)
721{
722 int ib_ret, ret_val = 0;
723 struct ib_send_wr send_wr, *send_wr_failed;
724 struct ib_sge iov[MAX_REGD_BUF_VECTOR_LEN];
725 struct iser_conn *ib_conn;
726 struct iser_dto *dto = &tx_desc->dto;
727
728 ib_conn = dto->conn->ib_conn;
729
730 iser_dto_to_iov(dto, iov, MAX_REGD_BUF_VECTOR_LEN);
731
732 send_wr.next = NULL;
733 send_wr.wr_id = (unsigned long)tx_desc;
734 send_wr.sg_list = iov;
735 send_wr.num_sge = dto->regd_vector_len;
736 send_wr.opcode = IB_WR_SEND;
737 send_wr.send_flags = dto->notify_enable ? IB_SEND_SIGNALED : 0;
738
739 atomic_inc(&ib_conn->post_send_buf_count);
740
741 ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed);
742 if (ib_ret) {
743 iser_err("Failed to start SEND DTO, dto: 0x%p, IOV len: %d\n",
744 dto, dto->regd_vector_len);
745 iser_err("ib_post_send failed, ret:%d\n", ib_ret);
746 atomic_dec(&ib_conn->post_send_buf_count);
747 ret_val = -1;
748 }
749
750 return ret_val;
751}
752
753static void iser_comp_error_worker(void *data)
754{
755 struct iser_conn *ib_conn = data;
756
757 /* getting here when the state is UP means that the conn is being *
758 * terminated asynchronously from the iSCSI layer's perspective. */
759 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
760 ISER_CONN_TERMINATING))
761 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
762 ISCSI_ERR_CONN_FAILED);
763
764 /* complete the termination process if disconnect event was delivered *
765 * note there are no more non completed posts to the QP */
766 if (ib_conn->disc_evt_flag) {
767 ib_conn->state = ISER_CONN_DOWN;
768 wake_up_interruptible(&ib_conn->wait);
769 }
770}
771
772static void iser_handle_comp_error(struct iser_desc *desc)
773{
774 struct iser_dto *dto = &desc->dto;
775 struct iser_conn *ib_conn = dto->conn->ib_conn;
776
777 iser_dto_buffs_release(dto);
778
779 if (desc->type == ISCSI_RX) {
780 kfree(desc->data);
781 kmem_cache_free(ig.desc_cache, desc);
782 atomic_dec(&ib_conn->post_recv_buf_count);
783 } else { /* type is TX control/command/dataout */
784 if (desc->type == ISCSI_TX_DATAOUT)
785 kmem_cache_free(ig.desc_cache, desc);
786 atomic_dec(&ib_conn->post_send_buf_count);
787 }
788
789 if (atomic_read(&ib_conn->post_recv_buf_count) == 0 &&
790 atomic_read(&ib_conn->post_send_buf_count) == 0)
791 schedule_work(&ib_conn->comperror_work);
792}
793
794static void iser_cq_tasklet_fn(unsigned long data)
795{
796 struct iser_device *device = (struct iser_device *)data;
797 struct ib_cq *cq = device->cq;
798 struct ib_wc wc;
799 struct iser_desc *desc;
800 unsigned long xfer_len;
801
802 while (ib_poll_cq(cq, 1, &wc) == 1) {
803 desc = (struct iser_desc *) (unsigned long) wc.wr_id;
804 BUG_ON(desc == NULL);
805
806 if (wc.status == IB_WC_SUCCESS) {
807 if (desc->type == ISCSI_RX) {
808 xfer_len = (unsigned long)wc.byte_len;
809 iser_rcv_completion(desc, xfer_len);
810 } else /* type == ISCSI_TX_CONTROL/SCSI_CMD/DOUT */
811 iser_snd_completion(desc);
812 } else {
813 iser_err("comp w. error op %d status %d\n",desc->type,wc.status);
814 iser_handle_comp_error(desc);
815 }
816 }
817 /* #warning "it is assumed here that arming CQ only once its empty" *
818 * " would not cause interrupts to be missed" */
819 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
820}
821
822static void iser_cq_callback(struct ib_cq *cq, void *cq_context)
823{
824 struct iser_device *device = (struct iser_device *)cq_context;
825
826 tasklet_schedule(&device->cq_tasklet);
827}