projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/mthca: Fix printk format used for firmware version in warning
[linux-2.6-block.git] / drivers / infiniband / ulp / iser / iser_verbs.c
CommitLineData
1cfa0a75
OG		 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>
1cfa0a75
OG		 38#include <linux/delay.h>
39#include <linux/version.h>
40
41#include "iscsi_iser.h"
42
43#define ISCSI_ISER_MAX_CONN 8
44#define ISER_MAX_CQ_LEN ((ISER_QP_MAX_RECV_DTOS + \
45 ISER_QP_MAX_REQ_DTOS) * \
46 ISCSI_ISER_MAX_CONN)
47
48static void iser_cq_tasklet_fn(unsigned long data);
49static void iser_cq_callback(struct ib_cq *cq, void *cq_context);
1cfa0a75
OG		 50
51static void iser_cq_event_callback(struct ib_event *cause, void *context)
52{
53 iser_err("got cq event %d \n", cause->event);
54}
55
56static void iser_qp_event_callback(struct ib_event *cause, void *context)
57{
58 iser_err("got qp event %d\n",cause->event);
59}
60
61/**
62 * iser_create_device_ib_res - creates Protection Domain (PD), Completion
63 * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
64 * the adapator.
65 *
66 * returns 0 on success, -1 on failure
67 */
68static int iser_create_device_ib_res(struct iser_device *device)
69{
70 device->pd = ib_alloc_pd(device->ib_device);
71 if (IS_ERR(device->pd))
72 goto pd_err;
73
74 device->cq = ib_create_cq(device->ib_device,
75 iser_cq_callback,
76 iser_cq_event_callback,
77 (void *)device,
f4fd0b22 78 ISER_MAX_CQ_LEN, 0);
1cfa0a75
OG		 79 if (IS_ERR(device->cq))
80 goto cq_err;
81
82 if (ib_req_notify_cq(device->cq, IB_CQ_NEXT_COMP))
83 goto cq_arm_err;
84
85 tasklet_init(&device->cq_tasklet,
86 iser_cq_tasklet_fn,
87 (unsigned long)device);
88
d8111028
EZ		 89 device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE |
90 IB_ACCESS_REMOTE_WRITE |
91 IB_ACCESS_REMOTE_READ);
1cfa0a75
OG		 92 if (IS_ERR(device->mr))
93 goto dma_mr_err;
94
95 return 0;
96
97dma_mr_err:
98 tasklet_kill(&device->cq_tasklet);
99cq_arm_err:
100 ib_destroy_cq(device->cq);
101cq_err:
102 ib_dealloc_pd(device->pd);
103pd_err:
104 iser_err("failed to allocate an IB resource\n");
105 return -1;
106}
107
108/**
109 * iser_free_device_ib_res - destory/dealloc/dereg the DMA MR,
110 * CQ and PD created with the device associated with the adapator.
111 */
112static void iser_free_device_ib_res(struct iser_device *device)
113{
114 BUG_ON(device->mr == NULL);
115
116 tasklet_kill(&device->cq_tasklet);
117
118 (void)ib_dereg_mr(device->mr);
119 (void)ib_destroy_cq(device->cq);
120 (void)ib_dealloc_pd(device->pd);
121
122 device->mr = NULL;
123 device->cq = NULL;
124 device->pd = NULL;
125}
126
127/**
128 * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP)
129 *
130 * returns 0 on success, -1 on failure
131 */
132static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
133{
134 struct iser_device *device;
135 struct ib_qp_init_attr init_attr;
136 int ret;
137 struct ib_fmr_pool_param params;
138
139 BUG_ON(ib_conn->device == NULL);
140
141 device = ib_conn->device;
142
143 ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
144 (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
145 GFP_KERNEL);
146 if (!ib_conn->page_vec) {
147 ret = -ENOMEM;
148 goto alloc_err;
149 }
150 ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
151
8dfa0876 152 params.page_shift = SHIFT_4K;
1cfa0a75
OG		 153 /* when the first/last SG element are not start/end *
154 * page aligned, the map whould be of N+1 pages */
155 params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
156 /* make the pool size twice the max number of SCSI commands *
157 * the ML is expected to queue, watermark for unmap at 50% */
1548271e
MC		 158 params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
159 params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
1cfa0a75
OG		 160 params.cache = 0;
161 params.flush_function = NULL;
162 params.access = (IB_ACCESS_LOCAL_WRITE |
163 IB_ACCESS_REMOTE_WRITE |
164 IB_ACCESS_REMOTE_READ);
165
166 ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, &params);
167 if (IS_ERR(ib_conn->fmr_pool)) {
168 ret = PTR_ERR(ib_conn->fmr_pool);
169 goto fmr_pool_err;
170 }
171
172 memset(&init_attr, 0, sizeof init_attr);
173
174 init_attr.event_handler = iser_qp_event_callback;
175 init_attr.qp_context = (void *)ib_conn;
176 init_attr.send_cq = device->cq;
177 init_attr.recv_cq = device->cq;
178 init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
179 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
180 init_attr.cap.max_send_sge = MAX_REGD_BUF_VECTOR_LEN;
181 init_attr.cap.max_recv_sge = 2;
182 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
183 init_attr.qp_type = IB_QPT_RC;
184
185 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
186 if (ret)
187 goto qp_err;
188
189 ib_conn->qp = ib_conn->cma_id->qp;
190 iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
191 ib_conn, ib_conn->cma_id,
192 ib_conn->fmr_pool, ib_conn->cma_id->qp);
193 return ret;
194
195qp_err:
196 (void)ib_destroy_fmr_pool(ib_conn->fmr_pool);
197fmr_pool_err:
198 kfree(ib_conn->page_vec);
199alloc_err:
200 iser_err("unable to alloc mem or create resource, err %d\n", ret);
201 return ret;
202}
203
204/**
205 * releases the FMR pool, QP and CMA ID objects, returns 0 on success,
206 * -1 on failure
207 */
208static int iser_free_ib_conn_res(struct iser_conn *ib_conn)
209{
210 BUG_ON(ib_conn == NULL);
211
212 iser_err("freeing conn %p cma_id %p fmr pool %p qp %p\n",
213 ib_conn, ib_conn->cma_id,
214 ib_conn->fmr_pool, ib_conn->qp);
215
216 /* qp is created only once both addr & route are resolved */
217 if (ib_conn->fmr_pool != NULL)
218 ib_destroy_fmr_pool(ib_conn->fmr_pool);
219
220 if (ib_conn->qp != NULL)
221 rdma_destroy_qp(ib_conn->cma_id);
222
223 if (ib_conn->cma_id != NULL)
224 rdma_destroy_id(ib_conn->cma_id);
225
226 ib_conn->fmr_pool = NULL;
227 ib_conn->qp = NULL;
228 ib_conn->cma_id = NULL;
229 kfree(ib_conn->page_vec);
230
231 return 0;
232}
233
234/**
235 * based on the resolved device node GUID see if there already allocated
236 * device for this device. If there's no such, create one.
237 */
238static
239struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
240{
241 struct list_head *p_list;
242 struct iser_device *device = NULL;
243
244 mutex_lock(&ig.device_list_mutex);
245
246 p_list = ig.device_list.next;
247 while (p_list != &ig.device_list) {
248 device = list_entry(p_list, struct iser_device, ig_list);
249 /* find if there's a match using the node GUID */
250 if (device->ib_device->node_guid == cma_id->device->node_guid)
251 break;
252 }
253
254 if (device == NULL) {
255 device = kzalloc(sizeof *device, GFP_KERNEL);
256 if (device == NULL)
257 goto out;
258 /* assign this device to the device */
259 device->ib_device = cma_id->device;
260 /* init the device and link it into ig device list */
261 if (iser_create_device_ib_res(device)) {
262 kfree(device);
263 device = NULL;
264 goto out;
265 }
266 list_add(&device->ig_list, &ig.device_list);
267 }
268out:
269 BUG_ON(device == NULL);
270 device->refcount++;
271 mutex_unlock(&ig.device_list_mutex);
272 return device;
273}
274
275/* if there's no demand for this device, release it */
276static void iser_device_try_release(struct iser_device *device)
277{
278 mutex_lock(&ig.device_list_mutex);
279 device->refcount--;
280 iser_err("device %p refcount %d\n",device,device->refcount);
281 if (!device->refcount) {
282 iser_free_device_ib_res(device);
283 list_del(&device->ig_list);
284 kfree(device);
285 }
286 mutex_unlock(&ig.device_list_mutex);
287}
288
289int iser_conn_state_comp(struct iser_conn *ib_conn,
290 enum iser_ib_conn_state comp)
291{
292 int ret;
293
294 spin_lock_bh(&ib_conn->lock);
295 ret = (ib_conn->state == comp);
296 spin_unlock_bh(&ib_conn->lock);
297 return ret;
298}
299
300static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
301 enum iser_ib_conn_state comp,
302 enum iser_ib_conn_state exch)
303{
304 int ret;
305
306 spin_lock_bh(&ib_conn->lock);
307 if ((ret = (ib_conn->state == comp)))
308 ib_conn->state = exch;
309 spin_unlock_bh(&ib_conn->lock);
310 return ret;
311}
312
313/**
314 * triggers start of the disconnect procedures and wait for them to be done
315 */
316void iser_conn_terminate(struct iser_conn *ib_conn)
317{
318 int err = 0;
319
320 /* change the ib conn state only if the conn is UP, however always call
321 * rdma_disconnect since this is the only way to cause the CMA to change
322 * the QP state to ERROR
323 */
324
325 iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING);
326 err = rdma_disconnect(ib_conn->cma_id);
327 if (err)
328 iser_err("Failed to disconnect, conn: 0x%p err %d\n",
329 ib_conn,err);
330
331 wait_event_interruptible(ib_conn->wait,
332 ib_conn->state == ISER_CONN_DOWN);
333
334 iser_conn_release(ib_conn);
335}
336
337static void iser_connect_error(struct rdma_cm_id *cma_id)
338{
339 struct iser_conn *ib_conn;
340 ib_conn = (struct iser_conn *)cma_id->context;
341
342 ib_conn->state = ISER_CONN_DOWN;
343 wake_up_interruptible(&ib_conn->wait);
344}
345
346static void iser_addr_handler(struct rdma_cm_id *cma_id)
347{
348 struct iser_device *device;
349 struct iser_conn *ib_conn;
350 int ret;
351
352 device = iser_device_find_by_ib_device(cma_id);
353 ib_conn = (struct iser_conn *)cma_id->context;
354 ib_conn->device = device;
355
356 ret = rdma_resolve_route(cma_id, 1000);
357 if (ret) {
358 iser_err("resolve route failed: %d\n", ret);
359 iser_connect_error(cma_id);
360 }
361 return;
362}
363
364static void iser_route_handler(struct rdma_cm_id *cma_id)
365{
366 struct rdma_conn_param conn_param;
367 int ret;
368
369 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context);
370 if (ret)
371 goto failure;
372
373 iser_dbg("path.mtu is %d setting it to %d\n",
374 cma_id->route.path_rec->mtu, IB_MTU_1024);
375
376 /* we must set the MTU to 1024 as this is what the target is assuming */
377 if (cma_id->route.path_rec->mtu > IB_MTU_1024)
378 cma_id->route.path_rec->mtu = IB_MTU_1024;
379
380 memset(&conn_param, 0, sizeof conn_param);
381 conn_param.responder_resources = 4;
382 conn_param.initiator_depth = 1;
383 conn_param.retry_count = 7;
384 conn_param.rnr_retry_count = 6;
385
386 ret = rdma_connect(cma_id, &conn_param);
387 if (ret) {
388 iser_err("failure connecting: %d\n", ret);
389 goto failure;
390 }
391
392 return;
393failure:
394 iser_connect_error(cma_id);
395}
396
397static void iser_connected_handler(struct rdma_cm_id *cma_id)
398{
399 struct iser_conn *ib_conn;
400
401 ib_conn = (struct iser_conn *)cma_id->context;
402 ib_conn->state = ISER_CONN_UP;
403 wake_up_interruptible(&ib_conn->wait);
404}
405
406static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
407{
408 struct iser_conn *ib_conn;
409
410 ib_conn = (struct iser_conn *)cma_id->context;
411 ib_conn->disc_evt_flag = 1;
412
413 /* getting here when the state is UP means that the conn is being *
414 * terminated asynchronously from the iSCSI layer's perspective. */
415 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
416 ISER_CONN_TERMINATING))
417 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
418 ISCSI_ERR_CONN_FAILED);
419
420 /* Complete the termination process if no posts are pending */
421 if ((atomic_read(&ib_conn->post_recv_buf_count) == 0) &&
422 (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
423 ib_conn->state = ISER_CONN_DOWN;
424 wake_up_interruptible(&ib_conn->wait);
425 }
426}
427
428static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
429{
430 int ret = 0;
431
432 iser_err("event %d conn %p id %p\n",event->event,cma_id->context,cma_id);
433
434 switch (event->event) {
435 case RDMA_CM_EVENT_ADDR_RESOLVED:
436 iser_addr_handler(cma_id);
437 break;
438 case RDMA_CM_EVENT_ROUTE_RESOLVED:
439 iser_route_handler(cma_id);
440 break;
441 case RDMA_CM_EVENT_ESTABLISHED:
442 iser_connected_handler(cma_id);
443 break;
444 case RDMA_CM_EVENT_ADDR_ERROR:
445 case RDMA_CM_EVENT_ROUTE_ERROR:
446 case RDMA_CM_EVENT_CONNECT_ERROR:
447 case RDMA_CM_EVENT_UNREACHABLE:
448 case RDMA_CM_EVENT_REJECTED:
449 iser_err("event: %d, error: %d\n", event->event, event->status);
450 iser_connect_error(cma_id);
451 break;
452 case RDMA_CM_EVENT_DISCONNECTED:
453 iser_disconnected_handler(cma_id);
454 break;
455 case RDMA_CM_EVENT_DEVICE_REMOVAL:
456 BUG();
457 break;
458 case RDMA_CM_EVENT_CONNECT_RESPONSE:
459 BUG();
460 break;
461 case RDMA_CM_EVENT_CONNECT_REQUEST:
462 default:
463 break;
464 }
465 return ret;
466}
467
468int iser_conn_init(struct iser_conn **ibconn)
469{
470 struct iser_conn *ib_conn;
471
472 ib_conn = kzalloc(sizeof *ib_conn, GFP_KERNEL);
473 if (!ib_conn) {
474 iser_err("can't alloc memory for struct iser_conn\n");
475 return -ENOMEM;
476 }
477 ib_conn->state = ISER_CONN_INIT;
478 init_waitqueue_head(&ib_conn->wait);
479 atomic_set(&ib_conn->post_recv_buf_count, 0);
480 atomic_set(&ib_conn->post_send_buf_count, 0);
1cfa0a75
OG		 481 INIT_LIST_HEAD(&ib_conn->conn_list);
482 spin_lock_init(&ib_conn->lock);
483
484 *ibconn = ib_conn;
485 return 0;
486}
487
488 /**
489 * starts the process of connecting to the target
490 * sleeps untill the connection is established or rejected
491 */
492int iser_connect(struct iser_conn *ib_conn,
493 struct sockaddr_in *src_addr,
494 struct sockaddr_in *dst_addr,
495 int non_blocking)
496{
497 struct sockaddr *src, *dst;
498 int err = 0;
499
500 sprintf(ib_conn->name,"%d.%d.%d.%d:%d",
501 NIPQUAD(dst_addr->sin_addr.s_addr), dst_addr->sin_port);
502
503 /* the device is known only --after-- address resolution */
504 ib_conn->device = NULL;
505
506 iser_err("connecting to: %d.%d.%d.%d, port 0x%x\n",
507 NIPQUAD(dst_addr->sin_addr), dst_addr->sin_port);
508
509 ib_conn->state = ISER_CONN_PENDING;
510
511 ib_conn->cma_id = rdma_create_id(iser_cma_handler,
512 (void *)ib_conn,
513 RDMA_PS_TCP);
514 if (IS_ERR(ib_conn->cma_id)) {
515 err = PTR_ERR(ib_conn->cma_id);
516 iser_err("rdma_create_id failed: %d\n", err);
517 goto id_failure;
518 }
519
520 src = (struct sockaddr *)src_addr;
521 dst = (struct sockaddr *)dst_addr;
522 err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000);
523 if (err) {
524 iser_err("rdma_resolve_addr failed: %d\n", err);
525 goto addr_failure;
526 }
527
528 if (!non_blocking) {
529 wait_event_interruptible(ib_conn->wait,
530 (ib_conn->state != ISER_CONN_PENDING));
531
532 if (ib_conn->state != ISER_CONN_UP) {
533 err = -EIO;
534 goto connect_failure;
535 }
536 }
537
538 mutex_lock(&ig.connlist_mutex);
539 list_add(&ib_conn->conn_list, &ig.connlist);
540 mutex_unlock(&ig.connlist_mutex);
541 return 0;
542
543id_failure:
544 ib_conn->cma_id = NULL;
545addr_failure:
546 ib_conn->state = ISER_CONN_DOWN;
547connect_failure:
548 iser_conn_release(ib_conn);
549 return err;
550}
551
552/**
553 * Frees all conn objects and deallocs conn descriptor
554 */
555void iser_conn_release(struct iser_conn *ib_conn)
556{
557 struct iser_device *device = ib_conn->device;
558
559 BUG_ON(ib_conn->state != ISER_CONN_DOWN);
560
561 mutex_lock(&ig.connlist_mutex);
562 list_del(&ib_conn->conn_list);
563 mutex_unlock(&ig.connlist_mutex);
564
565 iser_free_ib_conn_res(ib_conn);
566 ib_conn->device = NULL;
567 /* on EVENT_ADDR_ERROR there's no device yet for this conn */
568 if (device != NULL)
569 iser_device_try_release(device);
87e8df7a
EZ		 570 if (ib_conn->iser_conn)
571 ib_conn->iser_conn->ib_conn = NULL;
1cfa0a75
OG		 572 kfree(ib_conn);
573}
574
575
576/**
577 * iser_reg_page_vec - Register physical memory
578 *
579 * returns: 0 on success, errno code on failure
580 */
581int iser_reg_page_vec(struct iser_conn *ib_conn,
582 struct iser_page_vec *page_vec,
583 struct iser_mem_reg *mem_reg)
584{
585 struct ib_pool_fmr *mem;
586 u64 io_addr;
587 u64 *page_list;
588 int status;
589
590 page_list = page_vec->pages;
591 io_addr = page_list[0];
592
593 mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool,
594 page_list,
595 page_vec->length,
adfaa888 596 io_addr);
1cfa0a75
OG		 597
598 if (IS_ERR(mem)) {
599 status = (int)PTR_ERR(mem);
600 iser_err("ib_fmr_pool_map_phys failed: %d\n", status);
601 return status;
602 }
603
604 mem_reg->lkey = mem->fmr->lkey;
605 mem_reg->rkey = mem->fmr->rkey;
8dfa0876 606 mem_reg->len = page_vec->length * SIZE_4K;
1cfa0a75 607 mem_reg->va = io_addr;
d8111028 608 mem_reg->is_fmr = 1;
1cfa0a75
OG		 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 */
87e8df7a 695 ib_conn = recv_dto->ib_conn;
1cfa0a75
OG		 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
87e8df7a 728 ib_conn = dto->ib_conn;
1cfa0a75
OG		 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
1cfa0a75
OG		 753static void iser_handle_comp_error(struct iser_desc *desc)
754{
755 struct iser_dto *dto = &desc->dto;
87e8df7a 756 struct iser_conn *ib_conn = dto->ib_conn;
1cfa0a75
OG		 757
758 iser_dto_buffs_release(dto);
759
760 if (desc->type == ISCSI_RX) {
761 kfree(desc->data);
762 kmem_cache_free(ig.desc_cache, desc);
763 atomic_dec(&ib_conn->post_recv_buf_count);
764 } else { /* type is TX control/command/dataout */
765 if (desc->type == ISCSI_TX_DATAOUT)
766 kmem_cache_free(ig.desc_cache, desc);
767 atomic_dec(&ib_conn->post_send_buf_count);
768 }
769
770 if (atomic_read(&ib_conn->post_recv_buf_count) == 0 &&
1d426d64
EZ		 771 atomic_read(&ib_conn->post_send_buf_count) == 0) {
772 /* getting here when the state is UP means that the conn is *
773 * being terminated asynchronously from the iSCSI layer's *
774 * perspective. */
775 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
776 ISER_CONN_TERMINATING))
777 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
778 ISCSI_ERR_CONN_FAILED);
779
780 /* complete the termination process if disconnect event was delivered *
781 * note there are no more non completed posts to the QP */
782 if (ib_conn->disc_evt_flag) {
783 ib_conn->state = ISER_CONN_DOWN;
784 wake_up_interruptible(&ib_conn->wait);
785 }
786 }
1cfa0a75
OG		 787}
788
789static void iser_cq_tasklet_fn(unsigned long data)
790{
791 struct iser_device *device = (struct iser_device *)data;
792 struct ib_cq *cq = device->cq;
793 struct ib_wc wc;
794 struct iser_desc *desc;
795 unsigned long xfer_len;
796
797 while (ib_poll_cq(cq, 1, &wc) == 1) {
798 desc = (struct iser_desc *) (unsigned long) wc.wr_id;
799 BUG_ON(desc == NULL);
800
801 if (wc.status == IB_WC_SUCCESS) {
802 if (desc->type == ISCSI_RX) {
803 xfer_len = (unsigned long)wc.byte_len;
804 iser_rcv_completion(desc, xfer_len);
805 } else /* type == ISCSI_TX_CONTROL/SCSI_CMD/DOUT */
806 iser_snd_completion(desc);
807 } else {
808 iser_err("comp w. error op %d status %d\n",desc->type,wc.status);
809 iser_handle_comp_error(desc);
810 }
811 }
812 /* #warning "it is assumed here that arming CQ only once its empty" *
813 * " would not cause interrupts to be missed" */
814 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
815}
816
817static void iser_cq_callback(struct ib_cq *cq, void *cq_context)
818{
819 struct iser_device *device = (struct iser_device *)cq_context;
820
821 tasklet_schedule(&device->cq_tasklet);
822}
Linux 6.x block layer and io_uring tree(s)