Commit | Line | Data |
---|---|---|
1cfa0a75 OG |
1 | /* |
2 | * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. | |
3 | * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. | |
3ee07d27 | 4 | * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved. |
1cfa0a75 OG |
5 | * |
6 | * This software is available to you under a choice of one of two | |
7 | * licenses. You may choose to be licensed under the terms of the GNU | |
8 | * General Public License (GPL) Version 2, available from the file | |
9 | * COPYING in the main directory of this source tree, or the | |
10 | * OpenIB.org BSD license below: | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or | |
13 | * without modification, are permitted provided that the following | |
14 | * conditions are met: | |
15 | * | |
16 | * - Redistributions of source code must retain the above | |
17 | * copyright notice, this list of conditions and the following | |
18 | * disclaimer. | |
19 | * | |
20 | * - Redistributions in binary form must reproduce the above | |
21 | * copyright notice, this list of conditions and the following | |
22 | * disclaimer in the documentation and/or other materials | |
23 | * provided with the distribution. | |
24 | * | |
25 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
26 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
27 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
28 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
29 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
30 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
31 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
32 | * SOFTWARE. | |
1cfa0a75 | 33 | */ |
1cfa0a75 OG |
34 | #include <linux/kernel.h> |
35 | #include <linux/module.h> | |
5a0e3ad6 | 36 | #include <linux/slab.h> |
1cfa0a75 | 37 | #include <linux/delay.h> |
1cfa0a75 OG |
38 | |
39 | #include "iscsi_iser.h" | |
40 | ||
41 | #define ISCSI_ISER_MAX_CONN 8 | |
6aabfa76 SG |
42 | #define ISER_MAX_RX_LEN (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN) |
43 | #define ISER_MAX_TX_LEN (ISER_QP_MAX_REQ_DTOS * ISCSI_ISER_MAX_CONN) | |
ff3dd52d SG |
44 | #define ISER_MAX_CQ_LEN (ISER_MAX_RX_LEN + ISER_MAX_TX_LEN + \ |
45 | ISCSI_ISER_MAX_CONN) | |
1cfa0a75 | 46 | |
183cfa43 SG |
47 | static int iser_cq_poll_limit = 512; |
48 | ||
1cfa0a75 OG |
49 | static void iser_cq_tasklet_fn(unsigned long data); |
50 | static void iser_cq_callback(struct ib_cq *cq, void *cq_context); | |
1cfa0a75 OG |
51 | |
52 | static void iser_cq_event_callback(struct ib_event *cause, void *context) | |
53 | { | |
871e00af SG |
54 | iser_err("cq event %s (%d)\n", |
55 | ib_event_msg(cause->event), cause->event); | |
1cfa0a75 OG |
56 | } |
57 | ||
58 | static void iser_qp_event_callback(struct ib_event *cause, void *context) | |
59 | { | |
871e00af SG |
60 | iser_err("qp event %s (%d)\n", |
61 | ib_event_msg(cause->event), cause->event); | |
1cfa0a75 OG |
62 | } |
63 | ||
2110f9bf OG |
64 | static void iser_event_handler(struct ib_event_handler *handler, |
65 | struct ib_event *event) | |
66 | { | |
871e00af SG |
67 | iser_err("async event %s (%d) on device %s port %d\n", |
68 | ib_event_msg(event->event), event->event, | |
69 | event->device->name, event->element.port_num); | |
2110f9bf OG |
70 | } |
71 | ||
1cfa0a75 OG |
72 | /** |
73 | * iser_create_device_ib_res - creates Protection Domain (PD), Completion | |
74 | * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with | |
75 | * the adapator. | |
76 | * | |
77 | * returns 0 on success, -1 on failure | |
78 | */ | |
79 | static int iser_create_device_ib_res(struct iser_device *device) | |
80 | { | |
65198d6b | 81 | struct ib_device_attr *dev_attr = &device->dev_attr; |
f4641ef7 | 82 | int ret, i, max_cqe; |
5a33a669 | 83 | |
65198d6b SG |
84 | ret = ib_query_device(device->ib_device, dev_attr); |
85 | if (ret) { | |
5587856c | 86 | pr_warn("Query device failed for %s\n", device->ib_device->name); |
65198d6b | 87 | return ret; |
5587856c SG |
88 | } |
89 | ||
48afbff6 SG |
90 | ret = iser_assign_reg_ops(device); |
91 | if (ret) | |
92 | return ret; | |
b4e155ff | 93 | |
da64bdb2 | 94 | device->comps_used = min_t(int, num_online_cpus(), |
bf175540 | 95 | device->ib_device->num_comp_vectors); |
f4641ef7 | 96 | |
da64bdb2 SG |
97 | device->comps = kcalloc(device->comps_used, sizeof(*device->comps), |
98 | GFP_KERNEL); | |
99 | if (!device->comps) | |
100 | goto comps_err; | |
101 | ||
f4641ef7 MT |
102 | max_cqe = min(ISER_MAX_CQ_LEN, dev_attr->max_cqe); |
103 | ||
104 | iser_info("using %d CQs, device %s supports %d vectors max_cqe %d\n", | |
bf175540 | 105 | device->comps_used, device->ib_device->name, |
f4641ef7 | 106 | device->ib_device->num_comp_vectors, max_cqe); |
5a33a669 | 107 | |
1cfa0a75 OG |
108 | device->pd = ib_alloc_pd(device->ib_device); |
109 | if (IS_ERR(device->pd)) | |
110 | goto pd_err; | |
111 | ||
bf175540 | 112 | for (i = 0; i < device->comps_used; i++) { |
8e37210b | 113 | struct ib_cq_init_attr cq_attr = {}; |
bf175540 SG |
114 | struct iser_comp *comp = &device->comps[i]; |
115 | ||
116 | comp->device = device; | |
8e37210b MB |
117 | cq_attr.cqe = max_cqe; |
118 | cq_attr.comp_vector = i; | |
6aabfa76 SG |
119 | comp->cq = ib_create_cq(device->ib_device, |
120 | iser_cq_callback, | |
121 | iser_cq_event_callback, | |
122 | (void *)comp, | |
8e37210b | 123 | &cq_attr); |
6aabfa76 SG |
124 | if (IS_ERR(comp->cq)) { |
125 | comp->cq = NULL; | |
5a33a669 | 126 | goto cq_err; |
c33b15f0 | 127 | } |
1cfa0a75 | 128 | |
6aabfa76 | 129 | if (ib_req_notify_cq(comp->cq, IB_CQ_NEXT_COMP)) |
5a33a669 | 130 | goto cq_err; |
1cfa0a75 | 131 | |
bf175540 SG |
132 | tasklet_init(&comp->tasklet, iser_cq_tasklet_fn, |
133 | (unsigned long)comp); | |
5a33a669 | 134 | } |
1cfa0a75 | 135 | |
3cffd930 SG |
136 | if (!iser_always_reg) { |
137 | int access = IB_ACCESS_LOCAL_WRITE | | |
138 | IB_ACCESS_REMOTE_WRITE | | |
139 | IB_ACCESS_REMOTE_READ; | |
140 | ||
141 | device->mr = ib_get_dma_mr(device->pd, access); | |
142 | if (IS_ERR(device->mr)) | |
143 | goto dma_mr_err; | |
144 | } | |
1cfa0a75 | 145 | |
2110f9bf OG |
146 | INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device, |
147 | iser_event_handler); | |
148 | if (ib_register_event_handler(&device->event_handler)) | |
149 | goto handler_err; | |
150 | ||
1cfa0a75 OG |
151 | return 0; |
152 | ||
2110f9bf | 153 | handler_err: |
3cffd930 SG |
154 | if (device->mr) |
155 | ib_dereg_mr(device->mr); | |
1cfa0a75 | 156 | dma_mr_err: |
bf175540 SG |
157 | for (i = 0; i < device->comps_used; i++) |
158 | tasklet_kill(&device->comps[i].tasklet); | |
5a33a669 | 159 | cq_err: |
bf175540 SG |
160 | for (i = 0; i < device->comps_used; i++) { |
161 | struct iser_comp *comp = &device->comps[i]; | |
162 | ||
6aabfa76 SG |
163 | if (comp->cq) |
164 | ib_destroy_cq(comp->cq); | |
5a33a669 | 165 | } |
1cfa0a75 OG |
166 | ib_dealloc_pd(device->pd); |
167 | pd_err: | |
da64bdb2 SG |
168 | kfree(device->comps); |
169 | comps_err: | |
1cfa0a75 OG |
170 | iser_err("failed to allocate an IB resource\n"); |
171 | return -1; | |
172 | } | |
173 | ||
174 | /** | |
38dc732f | 175 | * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR, |
1cfa0a75 OG |
176 | * CQ and PD created with the device associated with the adapator. |
177 | */ | |
178 | static void iser_free_device_ib_res(struct iser_device *device) | |
179 | { | |
5a33a669 | 180 | int i; |
1cfa0a75 | 181 | |
bf175540 SG |
182 | for (i = 0; i < device->comps_used; i++) { |
183 | struct iser_comp *comp = &device->comps[i]; | |
184 | ||
185 | tasklet_kill(&comp->tasklet); | |
6aabfa76 SG |
186 | ib_destroy_cq(comp->cq); |
187 | comp->cq = NULL; | |
5a33a669 AT |
188 | } |
189 | ||
2110f9bf | 190 | (void)ib_unregister_event_handler(&device->event_handler); |
3cffd930 SG |
191 | if (device->mr) |
192 | (void)ib_dereg_mr(device->mr); | |
7dd78647 | 193 | ib_dealloc_pd(device->pd); |
1cfa0a75 | 194 | |
da64bdb2 SG |
195 | kfree(device->comps); |
196 | device->comps = NULL; | |
197 | ||
1cfa0a75 | 198 | device->mr = NULL; |
1cfa0a75 OG |
199 | device->pd = NULL; |
200 | } | |
201 | ||
202 | /** | |
48afbff6 | 203 | * iser_alloc_fmr_pool - Creates FMR pool and page_vector |
1cfa0a75 | 204 | * |
986db0d6 | 205 | * returns 0 on success, or errno code on failure |
1cfa0a75 | 206 | */ |
f8db651d SG |
207 | int iser_alloc_fmr_pool(struct ib_conn *ib_conn, |
208 | unsigned cmds_max, | |
209 | unsigned int size) | |
1cfa0a75 | 210 | { |
a4ee3539 | 211 | struct iser_device *device = ib_conn->device; |
385ad87d SG |
212 | struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; |
213 | struct iser_page_vec *page_vec; | |
2b3bf958 | 214 | struct iser_fr_desc *desc; |
385ad87d | 215 | struct ib_fmr_pool *fmr_pool; |
1cfa0a75 | 216 | struct ib_fmr_pool_param params; |
2b3bf958 | 217 | int ret; |
bcc60c38 | 218 | |
2b3bf958 | 219 | INIT_LIST_HEAD(&fr_pool->list); |
385ad87d SG |
220 | spin_lock_init(&fr_pool->lock); |
221 | ||
2b3bf958 AL |
222 | desc = kzalloc(sizeof(*desc), GFP_KERNEL); |
223 | if (!desc) | |
224 | return -ENOMEM; | |
225 | ||
f8db651d | 226 | page_vec = kmalloc(sizeof(*page_vec) + (sizeof(u64) * size), |
385ad87d | 227 | GFP_KERNEL); |
2b3bf958 AL |
228 | if (!page_vec) { |
229 | ret = -ENOMEM; | |
230 | goto err_frpl; | |
231 | } | |
9fda1ac5 | 232 | |
385ad87d | 233 | page_vec->pages = (u64 *)(page_vec + 1); |
1cfa0a75 | 234 | |
8dfa0876 | 235 | params.page_shift = SHIFT_4K; |
f8db651d | 236 | params.max_pages_per_fmr = size; |
1cfa0a75 OG |
237 | /* make the pool size twice the max number of SCSI commands * |
238 | * the ML is expected to queue, watermark for unmap at 50% */ | |
b7f04513 SP |
239 | params.pool_size = cmds_max * 2; |
240 | params.dirty_watermark = cmds_max; | |
1cfa0a75 OG |
241 | params.cache = 0; |
242 | params.flush_function = NULL; | |
243 | params.access = (IB_ACCESS_LOCAL_WRITE | | |
244 | IB_ACCESS_REMOTE_WRITE | | |
245 | IB_ACCESS_REMOTE_READ); | |
246 | ||
385ad87d SG |
247 | fmr_pool = ib_create_fmr_pool(device->pd, ¶ms); |
248 | if (IS_ERR(fmr_pool)) { | |
249 | ret = PTR_ERR(fmr_pool); | |
8c18ed03 | 250 | iser_err("FMR allocation failed, err %d\n", ret); |
2b3bf958 | 251 | goto err_fmr; |
8c18ed03 SG |
252 | } |
253 | ||
2b3bf958 AL |
254 | desc->rsc.page_vec = page_vec; |
255 | desc->rsc.fmr_pool = fmr_pool; | |
256 | list_add(&desc->list, &fr_pool->list); | |
385ad87d | 257 | |
8c18ed03 | 258 | return 0; |
986db0d6 | 259 | |
2b3bf958 | 260 | err_fmr: |
385ad87d | 261 | kfree(page_vec); |
2b3bf958 AL |
262 | err_frpl: |
263 | kfree(desc); | |
264 | ||
8c18ed03 | 265 | return ret; |
986db0d6 SP |
266 | } |
267 | ||
268 | /** | |
269 | * iser_free_fmr_pool - releases the FMR pool and page vec | |
270 | */ | |
a4ee3539 | 271 | void iser_free_fmr_pool(struct ib_conn *ib_conn) |
986db0d6 | 272 | { |
385ad87d | 273 | struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; |
2b3bf958 AL |
274 | struct iser_fr_desc *desc; |
275 | ||
276 | desc = list_first_entry(&fr_pool->list, | |
277 | struct iser_fr_desc, list); | |
278 | list_del(&desc->list); | |
986db0d6 | 279 | |
385ad87d | 280 | iser_info("freeing conn %p fmr pool %p\n", |
2b3bf958 | 281 | ib_conn, desc->rsc.fmr_pool); |
986db0d6 | 282 | |
2b3bf958 AL |
283 | ib_destroy_fmr_pool(desc->rsc.fmr_pool); |
284 | kfree(desc->rsc.page_vec); | |
285 | kfree(desc); | |
986db0d6 SP |
286 | } |
287 | ||
d711d81d | 288 | static int |
f8db651d SG |
289 | iser_alloc_reg_res(struct ib_device *ib_device, |
290 | struct ib_pd *pd, | |
291 | struct iser_reg_resources *res, | |
292 | unsigned int size) | |
d711d81d SG |
293 | { |
294 | int ret; | |
295 | ||
f8db651d | 296 | res->mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, size); |
d711d81d SG |
297 | if (IS_ERR(res->mr)) { |
298 | ret = PTR_ERR(res->mr); | |
299 | iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret); | |
39405885 | 300 | return ret; |
d711d81d SG |
301 | } |
302 | res->mr_valid = 1; | |
303 | ||
304 | return 0; | |
d711d81d SG |
305 | } |
306 | ||
307 | static void | |
308 | iser_free_reg_res(struct iser_reg_resources *rsc) | |
309 | { | |
310 | ib_dereg_mr(rsc->mr); | |
d711d81d SG |
311 | } |
312 | ||
4dec2a27 | 313 | static int |
f8db651d SG |
314 | iser_alloc_pi_ctx(struct ib_device *ib_device, |
315 | struct ib_pd *pd, | |
316 | struct iser_fr_desc *desc, | |
317 | unsigned int size) | |
4dec2a27 SG |
318 | { |
319 | struct iser_pi_context *pi_ctx = NULL; | |
9bee178b | 320 | int ret; |
4dec2a27 SG |
321 | |
322 | desc->pi_ctx = kzalloc(sizeof(*desc->pi_ctx), GFP_KERNEL); | |
323 | if (!desc->pi_ctx) | |
324 | return -ENOMEM; | |
325 | ||
326 | pi_ctx = desc->pi_ctx; | |
327 | ||
f8db651d | 328 | ret = iser_alloc_reg_res(ib_device, pd, &pi_ctx->rsc, size); |
d711d81d SG |
329 | if (ret) { |
330 | iser_err("failed to allocate reg_resources\n"); | |
331 | goto alloc_reg_res_err; | |
4dec2a27 | 332 | } |
4dec2a27 | 333 | |
9bee178b | 334 | pi_ctx->sig_mr = ib_alloc_mr(pd, IB_MR_TYPE_SIGNATURE, 2); |
4dec2a27 SG |
335 | if (IS_ERR(pi_ctx->sig_mr)) { |
336 | ret = PTR_ERR(pi_ctx->sig_mr); | |
337 | goto sig_mr_failure; | |
338 | } | |
d711d81d SG |
339 | pi_ctx->sig_mr_valid = 1; |
340 | desc->pi_ctx->sig_protected = 0; | |
4dec2a27 SG |
341 | |
342 | return 0; | |
343 | ||
344 | sig_mr_failure: | |
d711d81d SG |
345 | iser_free_reg_res(&pi_ctx->rsc); |
346 | alloc_reg_res_err: | |
4dec2a27 SG |
347 | kfree(desc->pi_ctx); |
348 | ||
349 | return ret; | |
350 | } | |
351 | ||
352 | static void | |
353 | iser_free_pi_ctx(struct iser_pi_context *pi_ctx) | |
354 | { | |
d711d81d | 355 | iser_free_reg_res(&pi_ctx->rsc); |
8b91ffc1 | 356 | ib_dereg_mr(pi_ctx->sig_mr); |
4dec2a27 SG |
357 | kfree(pi_ctx); |
358 | } | |
359 | ||
eb6ea8c3 | 360 | static struct iser_fr_desc * |
f8db651d SG |
361 | iser_create_fastreg_desc(struct ib_device *ib_device, |
362 | struct ib_pd *pd, | |
363 | bool pi_enable, | |
364 | unsigned int size) | |
310b347c | 365 | { |
eb6ea8c3 | 366 | struct iser_fr_desc *desc; |
310b347c SG |
367 | int ret; |
368 | ||
eb6ea8c3 SG |
369 | desc = kzalloc(sizeof(*desc), GFP_KERNEL); |
370 | if (!desc) | |
371 | return ERR_PTR(-ENOMEM); | |
372 | ||
f8db651d | 373 | ret = iser_alloc_reg_res(ib_device, pd, &desc->rsc, size); |
eb6ea8c3 SG |
374 | if (ret) |
375 | goto reg_res_alloc_failure; | |
6b5a8fb0 AT |
376 | |
377 | if (pi_enable) { | |
f8db651d | 378 | ret = iser_alloc_pi_ctx(ib_device, pd, desc, size); |
4dec2a27 | 379 | if (ret) |
6b5a8fb0 | 380 | goto pi_ctx_alloc_failure; |
6b5a8fb0 | 381 | } |
310b347c | 382 | |
eb6ea8c3 | 383 | return desc; |
d711d81d | 384 | |
6b5a8fb0 | 385 | pi_ctx_alloc_failure: |
d711d81d | 386 | iser_free_reg_res(&desc->rsc); |
eb6ea8c3 SG |
387 | reg_res_alloc_failure: |
388 | kfree(desc); | |
310b347c | 389 | |
eb6ea8c3 | 390 | return ERR_PTR(ret); |
310b347c SG |
391 | } |
392 | ||
5587856c | 393 | /** |
48afbff6 | 394 | * iser_alloc_fastreg_pool - Creates pool of fast_reg descriptors |
5587856c SG |
395 | * for fast registration work requests. |
396 | * returns 0 on success, or errno code on failure | |
397 | */ | |
f8db651d SG |
398 | int iser_alloc_fastreg_pool(struct ib_conn *ib_conn, |
399 | unsigned cmds_max, | |
400 | unsigned int size) | |
5587856c | 401 | { |
a4ee3539 | 402 | struct iser_device *device = ib_conn->device; |
385ad87d | 403 | struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; |
5190cc26 | 404 | struct iser_fr_desc *desc; |
5587856c SG |
405 | int i, ret; |
406 | ||
2b3bf958 | 407 | INIT_LIST_HEAD(&fr_pool->list); |
385ad87d | 408 | spin_lock_init(&fr_pool->lock); |
2b3bf958 | 409 | fr_pool->size = 0; |
5587856c | 410 | for (i = 0; i < cmds_max; i++) { |
eb6ea8c3 | 411 | desc = iser_create_fastreg_desc(device->ib_device, device->pd, |
f8db651d | 412 | ib_conn->pi_support, size); |
eb6ea8c3 SG |
413 | if (IS_ERR(desc)) { |
414 | ret = PTR_ERR(desc); | |
310b347c | 415 | goto err; |
5587856c SG |
416 | } |
417 | ||
2b3bf958 AL |
418 | list_add_tail(&desc->list, &fr_pool->list); |
419 | fr_pool->size++; | |
5587856c SG |
420 | } |
421 | ||
422 | return 0; | |
27ae2d1e | 423 | |
5587856c | 424 | err: |
a4ee3539 | 425 | iser_free_fastreg_pool(ib_conn); |
5587856c SG |
426 | return ret; |
427 | } | |
428 | ||
429 | /** | |
7306b8fa | 430 | * iser_free_fastreg_pool - releases the pool of fast_reg descriptors |
5587856c | 431 | */ |
a4ee3539 | 432 | void iser_free_fastreg_pool(struct ib_conn *ib_conn) |
5587856c | 433 | { |
385ad87d | 434 | struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; |
5190cc26 | 435 | struct iser_fr_desc *desc, *tmp; |
5587856c SG |
436 | int i = 0; |
437 | ||
2b3bf958 | 438 | if (list_empty(&fr_pool->list)) |
5587856c SG |
439 | return; |
440 | ||
a4ee3539 | 441 | iser_info("freeing conn %p fr pool\n", ib_conn); |
5587856c | 442 | |
2b3bf958 | 443 | list_for_each_entry_safe(desc, tmp, &fr_pool->list, list) { |
5587856c | 444 | list_del(&desc->list); |
d711d81d | 445 | iser_free_reg_res(&desc->rsc); |
4dec2a27 SG |
446 | if (desc->pi_ctx) |
447 | iser_free_pi_ctx(desc->pi_ctx); | |
5587856c SG |
448 | kfree(desc); |
449 | ++i; | |
450 | } | |
451 | ||
2b3bf958 | 452 | if (i < fr_pool->size) |
5587856c | 453 | iser_warn("pool still has %d regions registered\n", |
2b3bf958 | 454 | fr_pool->size - i); |
5587856c SG |
455 | } |
456 | ||
986db0d6 SP |
457 | /** |
458 | * iser_create_ib_conn_res - Queue-Pair (QP) | |
459 | * | |
460 | * returns 0 on success, -1 on failure | |
461 | */ | |
a4ee3539 | 462 | static int iser_create_ib_conn_res(struct ib_conn *ib_conn) |
986db0d6 | 463 | { |
7edc5a99 | 464 | struct iser_conn *iser_conn = to_iser_conn(ib_conn); |
986db0d6 | 465 | struct iser_device *device; |
f4641ef7 | 466 | struct ib_device_attr *dev_attr; |
986db0d6 SP |
467 | struct ib_qp_init_attr init_attr; |
468 | int ret = -ENOMEM; | |
469 | int index, min_index = 0; | |
470 | ||
a4ee3539 | 471 | BUG_ON(ib_conn->device == NULL); |
986db0d6 | 472 | |
a4ee3539 | 473 | device = ib_conn->device; |
f4641ef7 | 474 | dev_attr = &device->dev_attr; |
1cfa0a75 OG |
475 | |
476 | memset(&init_attr, 0, sizeof init_attr); | |
477 | ||
5a33a669 AT |
478 | mutex_lock(&ig.connlist_mutex); |
479 | /* select the CQ with the minimal number of usages */ | |
bf175540 SG |
480 | for (index = 0; index < device->comps_used; index++) { |
481 | if (device->comps[index].active_qps < | |
482 | device->comps[min_index].active_qps) | |
5a33a669 | 483 | min_index = index; |
bf175540 SG |
484 | } |
485 | ib_conn->comp = &device->comps[min_index]; | |
486 | ib_conn->comp->active_qps++; | |
5a33a669 | 487 | mutex_unlock(&ig.connlist_mutex); |
a4ee3539 | 488 | iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn); |
5a33a669 | 489 | |
1cfa0a75 | 490 | init_attr.event_handler = iser_qp_event_callback; |
a4ee3539 | 491 | init_attr.qp_context = (void *)ib_conn; |
6aabfa76 SG |
492 | init_attr.send_cq = ib_conn->comp->cq; |
493 | init_attr.recv_cq = ib_conn->comp->cq; | |
1cfa0a75 | 494 | init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS; |
f19624aa | 495 | init_attr.cap.max_send_sge = 2; |
bcc60c38 | 496 | init_attr.cap.max_recv_sge = 1; |
1cfa0a75 OG |
497 | init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; |
498 | init_attr.qp_type = IB_QPT_RC; | |
a4ee3539 | 499 | if (ib_conn->pi_support) { |
ff3dd52d | 500 | init_attr.cap.max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1; |
6b5a8fb0 | 501 | init_attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN; |
f4641ef7 MT |
502 | iser_conn->max_cmds = |
503 | ISER_GET_MAX_XMIT_CMDS(ISER_QP_SIG_MAX_REQ_DTOS); | |
6b5a8fb0 | 504 | } else { |
f4641ef7 MT |
505 | if (dev_attr->max_qp_wr > ISER_QP_MAX_REQ_DTOS) { |
506 | init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS + 1; | |
507 | iser_conn->max_cmds = | |
508 | ISER_GET_MAX_XMIT_CMDS(ISER_QP_MAX_REQ_DTOS); | |
509 | } else { | |
510 | init_attr.cap.max_send_wr = dev_attr->max_qp_wr; | |
511 | iser_conn->max_cmds = | |
512 | ISER_GET_MAX_XMIT_CMDS(dev_attr->max_qp_wr); | |
513 | iser_dbg("device %s supports max_send_wr %d\n", | |
514 | device->ib_device->name, dev_attr->max_qp_wr); | |
515 | } | |
6b5a8fb0 | 516 | } |
1cfa0a75 | 517 | |
a4ee3539 | 518 | ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); |
1cfa0a75 | 519 | if (ret) |
9fda1ac5 | 520 | goto out_err; |
1cfa0a75 | 521 | |
a4ee3539 | 522 | ib_conn->qp = ib_conn->cma_id->qp; |
986db0d6 | 523 | iser_info("setting conn %p cma_id %p qp %p\n", |
a4ee3539 SG |
524 | ib_conn, ib_conn->cma_id, |
525 | ib_conn->cma_id->qp); | |
1cfa0a75 OG |
526 | return ret; |
527 | ||
9fda1ac5 | 528 | out_err: |
93acb7bb SG |
529 | mutex_lock(&ig.connlist_mutex); |
530 | ib_conn->comp->active_qps--; | |
531 | mutex_unlock(&ig.connlist_mutex); | |
1cfa0a75 | 532 | iser_err("unable to alloc mem or create resource, err %d\n", ret); |
93acb7bb | 533 | |
1cfa0a75 OG |
534 | return ret; |
535 | } | |
536 | ||
1cfa0a75 OG |
537 | /** |
538 | * based on the resolved device node GUID see if there already allocated | |
539 | * device for this device. If there's no such, create one. | |
540 | */ | |
541 | static | |
542 | struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id) | |
543 | { | |
9a378270 | 544 | struct iser_device *device; |
1cfa0a75 OG |
545 | |
546 | mutex_lock(&ig.device_list_mutex); | |
547 | ||
9a378270 | 548 | list_for_each_entry(device, &ig.device_list, ig_list) |
1cfa0a75 OG |
549 | /* find if there's a match using the node GUID */ |
550 | if (device->ib_device->node_guid == cma_id->device->node_guid) | |
d33ed425 | 551 | goto inc_refcnt; |
9a378270 AR |
552 | |
553 | device = kzalloc(sizeof *device, GFP_KERNEL); | |
554 | if (device == NULL) | |
555 | goto out; | |
556 | ||
557 | /* assign this device to the device */ | |
558 | device->ib_device = cma_id->device; | |
559 | /* init the device and link it into ig device list */ | |
560 | if (iser_create_device_ib_res(device)) { | |
561 | kfree(device); | |
562 | device = NULL; | |
563 | goto out; | |
1cfa0a75 | 564 | } |
9a378270 AR |
565 | list_add(&device->ig_list, &ig.device_list); |
566 | ||
d33ed425 | 567 | inc_refcnt: |
1cfa0a75 | 568 | device->refcount++; |
d33ed425 | 569 | out: |
1cfa0a75 OG |
570 | mutex_unlock(&ig.device_list_mutex); |
571 | return device; | |
572 | } | |
573 | ||
574 | /* if there's no demand for this device, release it */ | |
575 | static void iser_device_try_release(struct iser_device *device) | |
576 | { | |
577 | mutex_lock(&ig.device_list_mutex); | |
578 | device->refcount--; | |
4f363882 | 579 | iser_info("device %p refcount %d\n", device, device->refcount); |
1cfa0a75 OG |
580 | if (!device->refcount) { |
581 | iser_free_device_ib_res(device); | |
582 | list_del(&device->ig_list); | |
583 | kfree(device); | |
584 | } | |
585 | mutex_unlock(&ig.device_list_mutex); | |
586 | } | |
587 | ||
504130c0 AN |
588 | /** |
589 | * Called with state mutex held | |
590 | **/ | |
5716af6e SG |
591 | static int iser_conn_state_comp_exch(struct iser_conn *iser_conn, |
592 | enum iser_conn_state comp, | |
593 | enum iser_conn_state exch) | |
1cfa0a75 OG |
594 | { |
595 | int ret; | |
596 | ||
5716af6e SG |
597 | ret = (iser_conn->state == comp); |
598 | if (ret) | |
599 | iser_conn->state = exch; | |
600 | ||
1cfa0a75 OG |
601 | return ret; |
602 | } | |
603 | ||
b73c3ada AN |
604 | void iser_release_work(struct work_struct *work) |
605 | { | |
5716af6e | 606 | struct iser_conn *iser_conn; |
b73c3ada | 607 | |
5716af6e | 608 | iser_conn = container_of(work, struct iser_conn, release_work); |
b73c3ada | 609 | |
c107a6c0 SG |
610 | /* Wait for conn_stop to complete */ |
611 | wait_for_completion(&iser_conn->stop_completion); | |
612 | /* Wait for IB resouces cleanup to complete */ | |
613 | wait_for_completion(&iser_conn->ib_completion); | |
b73c3ada | 614 | |
5716af6e SG |
615 | mutex_lock(&iser_conn->state_mutex); |
616 | iser_conn->state = ISER_CONN_DOWN; | |
617 | mutex_unlock(&iser_conn->state_mutex); | |
504130c0 | 618 | |
5716af6e | 619 | iser_conn_release(iser_conn); |
b73c3ada AN |
620 | } |
621 | ||
96f15198 SG |
622 | /** |
623 | * iser_free_ib_conn_res - release IB related resources | |
624 | * @iser_conn: iser connection struct | |
6606e6a2 SG |
625 | * @destroy: indicator if we need to try to release the |
626 | * iser device and memory regoins pool (only iscsi | |
627 | * shutdown and DEVICE_REMOVAL will use this). | |
96f15198 SG |
628 | * |
629 | * This routine is called with the iser state mutex held | |
630 | * so the cm_id removal is out of here. It is Safe to | |
631 | * be invoked multiple times. | |
632 | */ | |
c47a3c9e | 633 | static void iser_free_ib_conn_res(struct iser_conn *iser_conn, |
6606e6a2 | 634 | bool destroy) |
96f15198 SG |
635 | { |
636 | struct ib_conn *ib_conn = &iser_conn->ib_conn; | |
637 | struct iser_device *device = ib_conn->device; | |
638 | ||
639 | iser_info("freeing conn %p cma_id %p qp %p\n", | |
640 | iser_conn, ib_conn->cma_id, ib_conn->qp); | |
641 | ||
96f15198 | 642 | if (ib_conn->qp != NULL) { |
bf175540 | 643 | ib_conn->comp->active_qps--; |
96f15198 SG |
644 | rdma_destroy_qp(ib_conn->cma_id); |
645 | ib_conn->qp = NULL; | |
646 | } | |
647 | ||
6606e6a2 SG |
648 | if (destroy) { |
649 | if (iser_conn->rx_descs) | |
650 | iser_free_rx_descriptors(iser_conn); | |
651 | ||
652 | if (device != NULL) { | |
653 | iser_device_try_release(device); | |
654 | ib_conn->device = NULL; | |
655 | } | |
96f15198 SG |
656 | } |
657 | } | |
658 | ||
41179e2d RD |
659 | /** |
660 | * Frees all conn objects and deallocs conn descriptor | |
661 | */ | |
5716af6e | 662 | void iser_conn_release(struct iser_conn *iser_conn) |
41179e2d | 663 | { |
a4ee3539 | 664 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
41179e2d | 665 | |
41179e2d | 666 | mutex_lock(&ig.connlist_mutex); |
5716af6e | 667 | list_del(&iser_conn->conn_list); |
41179e2d | 668 | mutex_unlock(&ig.connlist_mutex); |
504130c0 | 669 | |
5716af6e | 670 | mutex_lock(&iser_conn->state_mutex); |
9a3119e4 | 671 | /* In case we endup here without ep_disconnect being invoked. */ |
5426b171 | 672 | if (iser_conn->state != ISER_CONN_DOWN) { |
aea8f4df AN |
673 | iser_warn("iser conn %p state %d, expected state down.\n", |
674 | iser_conn, iser_conn->state); | |
9a3119e4 | 675 | iscsi_destroy_endpoint(iser_conn->ep); |
5426b171 AN |
676 | iser_conn->state = ISER_CONN_DOWN; |
677 | } | |
c47a3c9e SG |
678 | /* |
679 | * In case we never got to bind stage, we still need to | |
680 | * release IB resources (which is safe to call more than once). | |
681 | */ | |
682 | iser_free_ib_conn_res(iser_conn, true); | |
5716af6e | 683 | mutex_unlock(&iser_conn->state_mutex); |
504130c0 | 684 | |
a4ee3539 SG |
685 | if (ib_conn->cma_id != NULL) { |
686 | rdma_destroy_id(ib_conn->cma_id); | |
687 | ib_conn->cma_id = NULL; | |
5b61ff43 | 688 | } |
a4ee3539 | 689 | |
5716af6e | 690 | kfree(iser_conn); |
41179e2d RD |
691 | } |
692 | ||
1cfa0a75 OG |
693 | /** |
694 | * triggers start of the disconnect procedures and wait for them to be done | |
c47a3c9e | 695 | * Called with state mutex held |
1cfa0a75 | 696 | */ |
c47a3c9e | 697 | int iser_conn_terminate(struct iser_conn *iser_conn) |
1cfa0a75 | 698 | { |
a4ee3539 | 699 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
ff3dd52d | 700 | struct ib_send_wr *bad_wr; |
1cfa0a75 OG |
701 | int err = 0; |
702 | ||
c47a3c9e SG |
703 | /* terminate the iser conn only if the conn state is UP */ |
704 | if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP, | |
705 | ISER_CONN_TERMINATING)) | |
706 | return 0; | |
707 | ||
708 | iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state); | |
709 | ||
710 | /* suspend queuing of new iscsi commands */ | |
711 | if (iser_conn->iscsi_conn) | |
712 | iscsi_suspend_queue(iser_conn->iscsi_conn); | |
713 | ||
714 | /* | |
715 | * In case we didn't already clean up the cma_id (peer initiated | |
716 | * a disconnection), we need to Cause the CMA to change the QP | |
717 | * state to ERROR. | |
1cfa0a75 | 718 | */ |
c47a3c9e SG |
719 | if (ib_conn->cma_id) { |
720 | err = rdma_disconnect(ib_conn->cma_id); | |
721 | if (err) | |
722 | iser_err("Failed to disconnect, conn: 0x%p err %d\n", | |
723 | iser_conn, err); | |
724 | ||
ff3dd52d SG |
725 | /* post an indication that all flush errors were consumed */ |
726 | err = ib_post_send(ib_conn->qp, &ib_conn->beacon, &bad_wr); | |
16df2a26 | 727 | if (err) { |
ff3dd52d | 728 | iser_err("conn %p failed to post beacon", ib_conn); |
16df2a26 SG |
729 | return 1; |
730 | } | |
ff3dd52d | 731 | |
6aabfa76 | 732 | wait_for_completion(&ib_conn->flush_comp); |
c47a3c9e | 733 | } |
1cfa0a75 | 734 | |
c47a3c9e | 735 | return 1; |
1cfa0a75 OG |
736 | } |
737 | ||
504130c0 AN |
738 | /** |
739 | * Called with state mutex held | |
740 | **/ | |
b73c3ada | 741 | static void iser_connect_error(struct rdma_cm_id *cma_id) |
1cfa0a75 | 742 | { |
5716af6e | 743 | struct iser_conn *iser_conn; |
b73c3ada | 744 | |
5716af6e | 745 | iser_conn = (struct iser_conn *)cma_id->context; |
c4de4663 | 746 | iser_conn->state = ISER_CONN_TERMINATING; |
1cfa0a75 OG |
747 | } |
748 | ||
df749cdc SG |
749 | static void |
750 | iser_calc_scsi_params(struct iser_conn *iser_conn, | |
751 | unsigned int max_sectors) | |
752 | { | |
753 | struct iser_device *device = iser_conn->ib_conn.device; | |
754 | unsigned short sg_tablesize, sup_sg_tablesize; | |
755 | ||
756 | sg_tablesize = DIV_ROUND_UP(max_sectors * 512, SIZE_4K); | |
757 | sup_sg_tablesize = min_t(unsigned, ISCSI_ISER_MAX_SG_TABLESIZE, | |
758 | device->dev_attr.max_fast_reg_page_list_len); | |
759 | ||
760 | if (sg_tablesize > sup_sg_tablesize) { | |
761 | sg_tablesize = sup_sg_tablesize; | |
762 | iser_conn->scsi_max_sectors = sg_tablesize * SIZE_4K / 512; | |
763 | } else { | |
764 | iser_conn->scsi_max_sectors = max_sectors; | |
765 | } | |
766 | ||
767 | iser_conn->scsi_sg_tablesize = sg_tablesize; | |
768 | ||
769 | iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n", | |
770 | iser_conn, iser_conn->scsi_sg_tablesize, | |
771 | iser_conn->scsi_max_sectors); | |
772 | } | |
773 | ||
504130c0 AN |
774 | /** |
775 | * Called with state mutex held | |
776 | **/ | |
b73c3ada | 777 | static void iser_addr_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
778 | { |
779 | struct iser_device *device; | |
5716af6e | 780 | struct iser_conn *iser_conn; |
a4ee3539 | 781 | struct ib_conn *ib_conn; |
1cfa0a75 OG |
782 | int ret; |
783 | ||
5716af6e SG |
784 | iser_conn = (struct iser_conn *)cma_id->context; |
785 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
786 | /* bailout */ |
787 | return; | |
788 | ||
a4ee3539 | 789 | ib_conn = &iser_conn->ib_conn; |
1cfa0a75 | 790 | device = iser_device_find_by_ib_device(cma_id); |
d33ed425 AR |
791 | if (!device) { |
792 | iser_err("device lookup/creation failed\n"); | |
b73c3ada AN |
793 | iser_connect_error(cma_id); |
794 | return; | |
d33ed425 AR |
795 | } |
796 | ||
a4ee3539 | 797 | ib_conn->device = device; |
1cfa0a75 | 798 | |
7f733847 AT |
799 | /* connection T10-PI support */ |
800 | if (iser_pi_enable) { | |
801 | if (!(device->dev_attr.device_cap_flags & | |
802 | IB_DEVICE_SIGNATURE_HANDOVER)) { | |
803 | iser_warn("T10-PI requested but not supported on %s, " | |
804 | "continue without T10-PI\n", | |
a4ee3539 SG |
805 | ib_conn->device->ib_device->name); |
806 | ib_conn->pi_support = false; | |
7f733847 | 807 | } else { |
a4ee3539 | 808 | ib_conn->pi_support = true; |
7f733847 AT |
809 | } |
810 | } | |
811 | ||
df749cdc SG |
812 | iser_calc_scsi_params(iser_conn, iser_max_sectors); |
813 | ||
1cfa0a75 OG |
814 | ret = rdma_resolve_route(cma_id, 1000); |
815 | if (ret) { | |
816 | iser_err("resolve route failed: %d\n", ret); | |
b73c3ada AN |
817 | iser_connect_error(cma_id); |
818 | return; | |
1cfa0a75 | 819 | } |
1cfa0a75 OG |
820 | } |
821 | ||
504130c0 AN |
822 | /** |
823 | * Called with state mutex held | |
824 | **/ | |
b73c3ada | 825 | static void iser_route_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
826 | { |
827 | struct rdma_conn_param conn_param; | |
828 | int ret; | |
8d8399de | 829 | struct iser_cm_hdr req_hdr; |
5716af6e | 830 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
a4ee3539 SG |
831 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
832 | struct iser_device *device = ib_conn->device; | |
1cfa0a75 | 833 | |
5716af6e | 834 | if (iser_conn->state != ISER_CONN_PENDING) |
504130c0 AN |
835 | /* bailout */ |
836 | return; | |
837 | ||
a4ee3539 | 838 | ret = iser_create_ib_conn_res(ib_conn); |
1cfa0a75 OG |
839 | if (ret) |
840 | goto failure; | |
841 | ||
1cfa0a75 | 842 | memset(&conn_param, 0, sizeof conn_param); |
2ea32938 | 843 | conn_param.responder_resources = device->dev_attr.max_qp_rd_atom; |
1cfa0a75 OG |
844 | conn_param.initiator_depth = 1; |
845 | conn_param.retry_count = 7; | |
846 | conn_param.rnr_retry_count = 6; | |
847 | ||
8d8399de OG |
848 | memset(&req_hdr, 0, sizeof(req_hdr)); |
849 | req_hdr.flags = (ISER_ZBVA_NOT_SUPPORTED | | |
850 | ISER_SEND_W_INV_NOT_SUPPORTED); | |
851 | conn_param.private_data = (void *)&req_hdr; | |
852 | conn_param.private_data_len = sizeof(struct iser_cm_hdr); | |
853 | ||
1cfa0a75 OG |
854 | ret = rdma_connect(cma_id, &conn_param); |
855 | if (ret) { | |
856 | iser_err("failure connecting: %d\n", ret); | |
857 | goto failure; | |
858 | } | |
859 | ||
b73c3ada | 860 | return; |
1cfa0a75 | 861 | failure: |
b73c3ada | 862 | iser_connect_error(cma_id); |
1cfa0a75 OG |
863 | } |
864 | ||
865 | static void iser_connected_handler(struct rdma_cm_id *cma_id) | |
866 | { | |
5716af6e | 867 | struct iser_conn *iser_conn; |
4f9208ad OG |
868 | struct ib_qp_attr attr; |
869 | struct ib_qp_init_attr init_attr; | |
870 | ||
5716af6e SG |
871 | iser_conn = (struct iser_conn *)cma_id->context; |
872 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
873 | /* bailout */ |
874 | return; | |
875 | ||
4f9208ad OG |
876 | (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr); |
877 | iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num); | |
1cfa0a75 | 878 | |
5716af6e SG |
879 | iser_conn->state = ISER_CONN_UP; |
880 | complete(&iser_conn->up_completion); | |
1cfa0a75 OG |
881 | } |
882 | ||
b73c3ada | 883 | static void iser_disconnected_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 | 884 | { |
c47a3c9e | 885 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
1cfa0a75 | 886 | |
c47a3c9e | 887 | if (iser_conn_terminate(iser_conn)) { |
5716af6e | 888 | if (iser_conn->iscsi_conn) |
c47a3c9e SG |
889 | iscsi_conn_failure(iser_conn->iscsi_conn, |
890 | ISCSI_ERR_CONN_FAILED); | |
7d9eacf9 RD |
891 | else |
892 | iser_err("iscsi_iser connection isn't bound\n"); | |
893 | } | |
c47a3c9e SG |
894 | } |
895 | ||
896 | static void iser_cleanup_handler(struct rdma_cm_id *cma_id, | |
6606e6a2 | 897 | bool destroy) |
c47a3c9e SG |
898 | { |
899 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; | |
1cfa0a75 | 900 | |
c47a3c9e SG |
901 | /* |
902 | * We are not guaranteed that we visited disconnected_handler | |
903 | * by now, call it here to be safe that we handle CM drep | |
904 | * and flush errors. | |
8d4aca7f | 905 | */ |
c47a3c9e | 906 | iser_disconnected_handler(cma_id); |
6606e6a2 | 907 | iser_free_ib_conn_res(iser_conn, destroy); |
c47a3c9e SG |
908 | complete(&iser_conn->ib_completion); |
909 | }; | |
1cfa0a75 OG |
910 | |
911 | static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) | |
912 | { | |
5716af6e | 913 | struct iser_conn *iser_conn; |
c47a3c9e | 914 | int ret = 0; |
504130c0 | 915 | |
5716af6e | 916 | iser_conn = (struct iser_conn *)cma_id->context; |
871e00af SG |
917 | iser_info("%s (%d): status %d conn %p id %p\n", |
918 | rdma_event_msg(event->event), event->event, | |
919 | event->status, cma_id->context, cma_id); | |
1cfa0a75 | 920 | |
5716af6e | 921 | mutex_lock(&iser_conn->state_mutex); |
1cfa0a75 OG |
922 | switch (event->event) { |
923 | case RDMA_CM_EVENT_ADDR_RESOLVED: | |
b73c3ada | 924 | iser_addr_handler(cma_id); |
1cfa0a75 OG |
925 | break; |
926 | case RDMA_CM_EVENT_ROUTE_RESOLVED: | |
b73c3ada | 927 | iser_route_handler(cma_id); |
1cfa0a75 OG |
928 | break; |
929 | case RDMA_CM_EVENT_ESTABLISHED: | |
930 | iser_connected_handler(cma_id); | |
931 | break; | |
932 | case RDMA_CM_EVENT_ADDR_ERROR: | |
933 | case RDMA_CM_EVENT_ROUTE_ERROR: | |
934 | case RDMA_CM_EVENT_CONNECT_ERROR: | |
935 | case RDMA_CM_EVENT_UNREACHABLE: | |
936 | case RDMA_CM_EVENT_REJECTED: | |
b73c3ada | 937 | iser_connect_error(cma_id); |
1cfa0a75 OG |
938 | break; |
939 | case RDMA_CM_EVENT_DISCONNECTED: | |
2f5de151 | 940 | case RDMA_CM_EVENT_ADDR_CHANGE: |
5426b171 AN |
941 | case RDMA_CM_EVENT_TIMEWAIT_EXIT: |
942 | iser_cleanup_handler(cma_id, false); | |
1cfa0a75 | 943 | break; |
c47a3c9e SG |
944 | case RDMA_CM_EVENT_DEVICE_REMOVAL: |
945 | /* | |
946 | * we *must* destroy the device as we cannot rely | |
947 | * on iscsid to be around to initiate error handling. | |
5426b171 AN |
948 | * also if we are not in state DOWN implicitly destroy |
949 | * the cma_id. | |
c47a3c9e SG |
950 | */ |
951 | iser_cleanup_handler(cma_id, true); | |
5426b171 AN |
952 | if (iser_conn->state != ISER_CONN_DOWN) { |
953 | iser_conn->ib_conn.cma_id = NULL; | |
954 | ret = 1; | |
955 | } | |
c47a3c9e | 956 | break; |
1cfa0a75 | 957 | default: |
871e00af SG |
958 | iser_err("Unexpected RDMA CM event: %s (%d)\n", |
959 | rdma_event_msg(event->event), event->event); | |
1cfa0a75 OG |
960 | break; |
961 | } | |
5716af6e | 962 | mutex_unlock(&iser_conn->state_mutex); |
c47a3c9e SG |
963 | |
964 | return ret; | |
1cfa0a75 OG |
965 | } |
966 | ||
5716af6e | 967 | void iser_conn_init(struct iser_conn *iser_conn) |
1cfa0a75 | 968 | { |
5716af6e | 969 | iser_conn->state = ISER_CONN_INIT; |
a4ee3539 | 970 | iser_conn->ib_conn.post_recv_buf_count = 0; |
6aabfa76 | 971 | init_completion(&iser_conn->ib_conn.flush_comp); |
5716af6e | 972 | init_completion(&iser_conn->stop_completion); |
c47a3c9e | 973 | init_completion(&iser_conn->ib_completion); |
5716af6e SG |
974 | init_completion(&iser_conn->up_completion); |
975 | INIT_LIST_HEAD(&iser_conn->conn_list); | |
5716af6e | 976 | mutex_init(&iser_conn->state_mutex); |
1cfa0a75 OG |
977 | } |
978 | ||
979 | /** | |
980 | * starts the process of connecting to the target | |
94e2bd68 | 981 | * sleeps until the connection is established or rejected |
1cfa0a75 | 982 | */ |
5716af6e | 983 | int iser_connect(struct iser_conn *iser_conn, |
96ed02d4 RD |
984 | struct sockaddr *src_addr, |
985 | struct sockaddr *dst_addr, | |
1cfa0a75 OG |
986 | int non_blocking) |
987 | { | |
a4ee3539 | 988 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
1cfa0a75 OG |
989 | int err = 0; |
990 | ||
5716af6e | 991 | mutex_lock(&iser_conn->state_mutex); |
504130c0 | 992 | |
5716af6e | 993 | sprintf(iser_conn->name, "%pISp", dst_addr); |
96ed02d4 | 994 | |
5716af6e | 995 | iser_info("connecting to: %s\n", iser_conn->name); |
1cfa0a75 OG |
996 | |
997 | /* the device is known only --after-- address resolution */ | |
a4ee3539 | 998 | ib_conn->device = NULL; |
1cfa0a75 | 999 | |
5716af6e | 1000 | iser_conn->state = ISER_CONN_PENDING; |
1cfa0a75 | 1001 | |
ff3dd52d SG |
1002 | ib_conn->beacon.wr_id = ISER_BEACON_WRID; |
1003 | ib_conn->beacon.opcode = IB_WR_SEND; | |
1004 | ||
fa20105e | 1005 | ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler, |
a4ee3539 SG |
1006 | (void *)iser_conn, |
1007 | RDMA_PS_TCP, IB_QPT_RC); | |
1008 | if (IS_ERR(ib_conn->cma_id)) { | |
1009 | err = PTR_ERR(ib_conn->cma_id); | |
1cfa0a75 OG |
1010 | iser_err("rdma_create_id failed: %d\n", err); |
1011 | goto id_failure; | |
1012 | } | |
1013 | ||
a4ee3539 | 1014 | err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000); |
1cfa0a75 OG |
1015 | if (err) { |
1016 | iser_err("rdma_resolve_addr failed: %d\n", err); | |
1017 | goto addr_failure; | |
1018 | } | |
1019 | ||
1020 | if (!non_blocking) { | |
5716af6e | 1021 | wait_for_completion_interruptible(&iser_conn->up_completion); |
1cfa0a75 | 1022 | |
5716af6e | 1023 | if (iser_conn->state != ISER_CONN_UP) { |
1cfa0a75 OG |
1024 | err = -EIO; |
1025 | goto connect_failure; | |
1026 | } | |
1027 | } | |
5716af6e | 1028 | mutex_unlock(&iser_conn->state_mutex); |
1cfa0a75 OG |
1029 | |
1030 | mutex_lock(&ig.connlist_mutex); | |
5716af6e | 1031 | list_add(&iser_conn->conn_list, &ig.connlist); |
1cfa0a75 OG |
1032 | mutex_unlock(&ig.connlist_mutex); |
1033 | return 0; | |
1034 | ||
1035 | id_failure: | |
a4ee3539 | 1036 | ib_conn->cma_id = NULL; |
1cfa0a75 | 1037 | addr_failure: |
5716af6e | 1038 | iser_conn->state = ISER_CONN_DOWN; |
1cfa0a75 | 1039 | connect_failure: |
5716af6e SG |
1040 | mutex_unlock(&iser_conn->state_mutex); |
1041 | iser_conn_release(iser_conn); | |
1cfa0a75 OG |
1042 | return err; |
1043 | } | |
1044 | ||
5716af6e | 1045 | int iser_post_recvl(struct iser_conn *iser_conn) |
bcc60c38 OG |
1046 | { |
1047 | struct ib_recv_wr rx_wr, *rx_wr_failed; | |
a4ee3539 | 1048 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
0f512b34 | 1049 | struct iser_login_desc *desc = &iser_conn->login_desc; |
bcc60c38 OG |
1050 | int ib_ret; |
1051 | ||
0f512b34 SG |
1052 | desc->sge.addr = desc->rsp_dma; |
1053 | desc->sge.length = ISER_RX_LOGIN_SIZE; | |
1054 | desc->sge.lkey = ib_conn->device->pd->local_dma_lkey; | |
bcc60c38 | 1055 | |
0f512b34 SG |
1056 | rx_wr.wr_id = (uintptr_t)desc; |
1057 | rx_wr.sg_list = &desc->sge; | |
bcc60c38 | 1058 | rx_wr.num_sge = 1; |
0f512b34 | 1059 | rx_wr.next = NULL; |
bcc60c38 | 1060 | |
a4ee3539 SG |
1061 | ib_conn->post_recv_buf_count++; |
1062 | ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1063 | if (ib_ret) { |
1064 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1065 | ib_conn->post_recv_buf_count--; |
bcc60c38 OG |
1066 | } |
1067 | return ib_ret; | |
1068 | } | |
1069 | ||
5716af6e | 1070 | int iser_post_recvm(struct iser_conn *iser_conn, int count) |
bcc60c38 OG |
1071 | { |
1072 | struct ib_recv_wr *rx_wr, *rx_wr_failed; | |
1073 | int i, ib_ret; | |
a4ee3539 | 1074 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
5716af6e | 1075 | unsigned int my_rx_head = iser_conn->rx_desc_head; |
bcc60c38 OG |
1076 | struct iser_rx_desc *rx_desc; |
1077 | ||
a4ee3539 | 1078 | for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { |
5716af6e | 1079 | rx_desc = &iser_conn->rx_descs[my_rx_head]; |
49df2781 | 1080 | rx_wr->wr_id = (uintptr_t)rx_desc; |
bcc60c38 OG |
1081 | rx_wr->sg_list = &rx_desc->rx_sg; |
1082 | rx_wr->num_sge = 1; | |
1083 | rx_wr->next = rx_wr + 1; | |
5716af6e | 1084 | my_rx_head = (my_rx_head + 1) & iser_conn->qp_max_recv_dtos_mask; |
bcc60c38 OG |
1085 | } |
1086 | ||
1087 | rx_wr--; | |
1088 | rx_wr->next = NULL; /* mark end of work requests list */ | |
1089 | ||
a4ee3539 SG |
1090 | ib_conn->post_recv_buf_count += count; |
1091 | ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1092 | if (ib_ret) { |
1093 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1094 | ib_conn->post_recv_buf_count -= count; |
bcc60c38 | 1095 | } else |
5716af6e | 1096 | iser_conn->rx_desc_head = my_rx_head; |
bcc60c38 OG |
1097 | return ib_ret; |
1098 | } | |
1099 | ||
1100 | ||
1cfa0a75 OG |
1101 | /** |
1102 | * iser_start_send - Initiate a Send DTO operation | |
1103 | * | |
1104 | * returns 0 on success, -1 on failure | |
1105 | */ | |
6df5a128 SG |
1106 | int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc, |
1107 | bool signal) | |
1cfa0a75 | 1108 | { |
7332bed0 SG |
1109 | struct ib_send_wr *bad_wr, *wr = iser_tx_next_wr(tx_desc); |
1110 | int ib_ret; | |
1cfa0a75 | 1111 | |
a4ee3539 | 1112 | ib_dma_sync_single_for_device(ib_conn->device->ib_device, |
5716af6e SG |
1113 | tx_desc->dma_addr, ISER_HEADERS_LEN, |
1114 | DMA_TO_DEVICE); | |
1cfa0a75 | 1115 | |
7332bed0 SG |
1116 | wr->next = NULL; |
1117 | wr->wr_id = (uintptr_t)tx_desc; | |
1118 | wr->sg_list = tx_desc->tx_sg; | |
1119 | wr->num_sge = tx_desc->num_sge; | |
1120 | wr->opcode = IB_WR_SEND; | |
1121 | wr->send_flags = signal ? IB_SEND_SIGNALED : 0; | |
1cfa0a75 | 1122 | |
e622f2f4 | 1123 | ib_ret = ib_post_send(ib_conn->qp, &tx_desc->wrs[0].send, &bad_wr); |
ff3dd52d | 1124 | if (ib_ret) |
7332bed0 SG |
1125 | iser_err("ib_post_send failed, ret:%d opcode:%d\n", |
1126 | ib_ret, bad_wr->opcode); | |
ff3dd52d | 1127 | |
f19624aa | 1128 | return ib_ret; |
1cfa0a75 OG |
1129 | } |
1130 | ||
6aabfa76 SG |
1131 | /** |
1132 | * is_iser_tx_desc - Indicate if the completion wr_id | |
1133 | * is a TX descriptor or not. | |
1134 | * @iser_conn: iser connection | |
1135 | * @wr_id: completion WR identifier | |
1136 | * | |
1137 | * Since we cannot rely on wc opcode in FLUSH errors | |
1138 | * we must work around it by checking if the wr_id address | |
1139 | * falls in the iser connection rx_descs buffer. If so | |
1140 | * it is an RX descriptor, otherwize it is a TX. | |
1141 | */ | |
1142 | static inline bool | |
1143 | is_iser_tx_desc(struct iser_conn *iser_conn, void *wr_id) | |
1144 | { | |
1145 | void *start = iser_conn->rx_descs; | |
1146 | int len = iser_conn->num_rx_descs * sizeof(*iser_conn->rx_descs); | |
1147 | ||
1148 | if (wr_id >= start && wr_id < start + len) | |
1149 | return false; | |
1150 | ||
1151 | return true; | |
1152 | } | |
1153 | ||
8c204e69 SG |
1154 | /** |
1155 | * iser_handle_comp_error() - Handle error completion | |
8c204e69 SG |
1156 | * @ib_conn: connection RDMA resources |
1157 | * @wc: work completion | |
1158 | * | |
1159 | * Notes: We may handle a FLUSH error completion and in this case | |
1160 | * we only cleanup in case TX type was DATAOUT. For non-FLUSH | |
1161 | * error completion we should also notify iscsi layer that | |
1162 | * connection is failed (in case we passed bind stage). | |
1163 | */ | |
1164 | static void | |
6aabfa76 | 1165 | iser_handle_comp_error(struct ib_conn *ib_conn, |
8c204e69 | 1166 | struct ib_wc *wc) |
1cfa0a75 | 1167 | { |
7edc5a99 | 1168 | struct iser_conn *iser_conn = to_iser_conn(ib_conn); |
49df2781 | 1169 | void *wr_id = (void *)(uintptr_t)wc->wr_id; |
8c204e69 SG |
1170 | |
1171 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
1172 | if (iser_conn->iscsi_conn) | |
1173 | iscsi_conn_failure(iser_conn->iscsi_conn, | |
1174 | ISCSI_ERR_CONN_FAILED); | |
1175 | ||
30bf1d58 SG |
1176 | if (wc->wr_id == ISER_FASTREG_LI_WRID) |
1177 | return; | |
1178 | ||
49df2781 SG |
1179 | if (is_iser_tx_desc(iser_conn, wr_id)) { |
1180 | struct iser_tx_desc *desc = wr_id; | |
6aabfa76 | 1181 | |
6aabfa76 SG |
1182 | if (desc->type == ISCSI_TX_DATAOUT) |
1183 | kmem_cache_free(ig.desc_cache, desc); | |
1184 | } else { | |
1185 | ib_conn->post_recv_buf_count--; | |
1186 | } | |
1cfa0a75 OG |
1187 | } |
1188 | ||
6aabfa76 SG |
1189 | /** |
1190 | * iser_handle_wc - handle a single work completion | |
1191 | * @wc: work completion | |
1192 | * | |
1193 | * Soft-IRQ context, work completion can be either | |
1194 | * SEND or RECV, and can turn out successful or | |
1195 | * with error (or flush error). | |
1196 | */ | |
1197 | static void iser_handle_wc(struct ib_wc *wc) | |
78ad0a34 | 1198 | { |
a4ee3539 | 1199 | struct ib_conn *ib_conn; |
6aabfa76 SG |
1200 | struct iser_tx_desc *tx_desc; |
1201 | struct iser_rx_desc *rx_desc; | |
78ad0a34 | 1202 | |
6aabfa76 | 1203 | ib_conn = wc->qp->qp_context; |
06c7fb67 | 1204 | if (likely(wc->status == IB_WC_SUCCESS)) { |
6aabfa76 | 1205 | if (wc->opcode == IB_WC_RECV) { |
49df2781 | 1206 | rx_desc = (struct iser_rx_desc *)(uintptr_t)wc->wr_id; |
6aabfa76 SG |
1207 | iser_rcv_completion(rx_desc, wc->byte_len, |
1208 | ib_conn); | |
1209 | } else | |
1210 | if (wc->opcode == IB_WC_SEND) { | |
49df2781 | 1211 | tx_desc = (struct iser_tx_desc *)(uintptr_t)wc->wr_id; |
6aabfa76 | 1212 | iser_snd_completion(tx_desc, ib_conn); |
78ad0a34 | 1213 | } else { |
6aabfa76 | 1214 | iser_err("Unknown wc opcode %d\n", wc->opcode); |
78ad0a34 | 1215 | } |
6aabfa76 SG |
1216 | } else { |
1217 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
871e00af SG |
1218 | iser_err("%s (%d): wr id %llx vend_err %x\n", |
1219 | ib_wc_status_msg(wc->status), wc->status, | |
1220 | wc->wr_id, wc->vendor_err); | |
6aabfa76 | 1221 | else |
871e00af SG |
1222 | iser_dbg("%s (%d): wr id %llx\n", |
1223 | ib_wc_status_msg(wc->status), wc->status, | |
1224 | wc->wr_id); | |
6aabfa76 | 1225 | |
ff3dd52d | 1226 | if (wc->wr_id == ISER_BEACON_WRID) |
30bf1d58 | 1227 | /* all flush errors were consumed */ |
6aabfa76 | 1228 | complete(&ib_conn->flush_comp); |
30bf1d58 SG |
1229 | else |
1230 | iser_handle_comp_error(ib_conn, wc); | |
78ad0a34 | 1231 | } |
78ad0a34 OG |
1232 | } |
1233 | ||
6aabfa76 SG |
1234 | /** |
1235 | * iser_cq_tasklet_fn - iSER completion polling loop | |
1236 | * @data: iSER completion context | |
1237 | * | |
1238 | * Soft-IRQ context, polling connection CQ until | |
1239 | * either CQ was empty or we exausted polling budget | |
1240 | */ | |
1cfa0a75 OG |
1241 | static void iser_cq_tasklet_fn(unsigned long data) |
1242 | { | |
bf175540 | 1243 | struct iser_comp *comp = (struct iser_comp *)data; |
6aabfa76 | 1244 | struct ib_cq *cq = comp->cq; |
6e6fe2fb SG |
1245 | struct ib_wc *const wcs = comp->wcs; |
1246 | int i, n, completed = 0; | |
1cfa0a75 | 1247 | |
6e6fe2fb SG |
1248 | while ((n = ib_poll_cq(cq, ARRAY_SIZE(comp->wcs), wcs)) > 0) { |
1249 | for (i = 0; i < n; i++) | |
1250 | iser_handle_wc(&wcs[i]); | |
6aabfa76 | 1251 | |
6e6fe2fb SG |
1252 | completed += n; |
1253 | if (completed >= iser_cq_poll_limit) | |
183cfa43 | 1254 | break; |
1cfa0a75 | 1255 | } |
6aabfa76 SG |
1256 | |
1257 | /* | |
1258 | * It is assumed here that arming CQ only once its empty | |
1259 | * would not cause interrupts to be missed. | |
1260 | */ | |
1cfa0a75 | 1261 | ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); |
78ad0a34 | 1262 | |
6aabfa76 | 1263 | iser_dbg("got %d completions\n", completed); |
1cfa0a75 OG |
1264 | } |
1265 | ||
1266 | static void iser_cq_callback(struct ib_cq *cq, void *cq_context) | |
1267 | { | |
bf175540 | 1268 | struct iser_comp *comp = cq_context; |
1cfa0a75 | 1269 | |
bf175540 | 1270 | tasklet_schedule(&comp->tasklet); |
1cfa0a75 | 1271 | } |
0a7a08ad SG |
1272 | |
1273 | u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, | |
1274 | enum iser_data_dir cmd_dir, sector_t *sector) | |
1275 | { | |
b130eded | 1276 | struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir]; |
5190cc26 | 1277 | struct iser_fr_desc *desc = reg->mem_h; |
0a7a08ad SG |
1278 | unsigned long sector_size = iser_task->sc->device->sector_size; |
1279 | struct ib_mr_status mr_status; | |
1280 | int ret; | |
1281 | ||
d711d81d SG |
1282 | if (desc && desc->pi_ctx->sig_protected) { |
1283 | desc->pi_ctx->sig_protected = 0; | |
0a7a08ad SG |
1284 | ret = ib_check_mr_status(desc->pi_ctx->sig_mr, |
1285 | IB_MR_CHECK_SIG_STATUS, &mr_status); | |
1286 | if (ret) { | |
1287 | pr_err("ib_check_mr_status failed, ret %d\n", ret); | |
1288 | goto err; | |
1289 | } | |
1290 | ||
1291 | if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) { | |
1292 | sector_t sector_off = mr_status.sig_err.sig_err_offset; | |
1293 | ||
1294 | do_div(sector_off, sector_size + 8); | |
1295 | *sector = scsi_get_lba(iser_task->sc) + sector_off; | |
1296 | ||
39c978cd | 1297 | pr_err("PI error found type %d at sector %llx " |
0a7a08ad | 1298 | "expected %x vs actual %x\n", |
39c978cd RD |
1299 | mr_status.sig_err.err_type, |
1300 | (unsigned long long)*sector, | |
0a7a08ad SG |
1301 | mr_status.sig_err.expected, |
1302 | mr_status.sig_err.actual); | |
1303 | ||
1304 | switch (mr_status.sig_err.err_type) { | |
1305 | case IB_SIG_BAD_GUARD: | |
1306 | return 0x1; | |
1307 | case IB_SIG_BAD_REFTAG: | |
1308 | return 0x3; | |
1309 | case IB_SIG_BAD_APPTAG: | |
1310 | return 0x2; | |
1311 | } | |
1312 | } | |
1313 | } | |
1314 | ||
1315 | return 0; | |
1316 | err: | |
1317 | /* Not alot we can do here, return ambiguous guard error */ | |
1318 | return 0x1; | |
1319 | } |