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 | ||
90 | /* Assign function handles - based on FMR support */ | |
91 | if (device->ib_device->alloc_fmr && device->ib_device->dealloc_fmr && | |
92 | device->ib_device->map_phys_fmr && device->ib_device->unmap_fmr) { | |
93 | iser_info("FMR supported, using FMR for registration\n"); | |
94 | device->iser_alloc_rdma_reg_res = iser_create_fmr_pool; | |
95 | device->iser_free_rdma_reg_res = iser_free_fmr_pool; | |
96 | device->iser_reg_rdma_mem = iser_reg_rdma_mem_fmr; | |
97 | device->iser_unreg_rdma_mem = iser_unreg_mem_fmr; | |
98 | } else | |
99 | if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { | |
7306b8fa SG |
100 | iser_info("FastReg supported, using FastReg for registration\n"); |
101 | device->iser_alloc_rdma_reg_res = iser_create_fastreg_pool; | |
102 | device->iser_free_rdma_reg_res = iser_free_fastreg_pool; | |
103 | device->iser_reg_rdma_mem = iser_reg_rdma_mem_fastreg; | |
104 | device->iser_unreg_rdma_mem = iser_unreg_mem_fastreg; | |
5587856c | 105 | } else { |
7306b8fa | 106 | iser_err("IB device does not support FMRs nor FastRegs, can't register memory\n"); |
65198d6b | 107 | return -1; |
5587856c | 108 | } |
b4e155ff | 109 | |
da64bdb2 | 110 | device->comps_used = min_t(int, num_online_cpus(), |
bf175540 | 111 | device->ib_device->num_comp_vectors); |
f4641ef7 | 112 | |
da64bdb2 SG |
113 | device->comps = kcalloc(device->comps_used, sizeof(*device->comps), |
114 | GFP_KERNEL); | |
115 | if (!device->comps) | |
116 | goto comps_err; | |
117 | ||
f4641ef7 MT |
118 | max_cqe = min(ISER_MAX_CQ_LEN, dev_attr->max_cqe); |
119 | ||
120 | iser_info("using %d CQs, device %s supports %d vectors max_cqe %d\n", | |
bf175540 | 121 | device->comps_used, device->ib_device->name, |
f4641ef7 | 122 | device->ib_device->num_comp_vectors, max_cqe); |
5a33a669 | 123 | |
1cfa0a75 OG |
124 | device->pd = ib_alloc_pd(device->ib_device); |
125 | if (IS_ERR(device->pd)) | |
126 | goto pd_err; | |
127 | ||
bf175540 | 128 | for (i = 0; i < device->comps_used; i++) { |
8e37210b | 129 | struct ib_cq_init_attr cq_attr = {}; |
bf175540 SG |
130 | struct iser_comp *comp = &device->comps[i]; |
131 | ||
132 | comp->device = device; | |
8e37210b MB |
133 | cq_attr.cqe = max_cqe; |
134 | cq_attr.comp_vector = i; | |
6aabfa76 SG |
135 | comp->cq = ib_create_cq(device->ib_device, |
136 | iser_cq_callback, | |
137 | iser_cq_event_callback, | |
138 | (void *)comp, | |
8e37210b | 139 | &cq_attr); |
6aabfa76 SG |
140 | if (IS_ERR(comp->cq)) { |
141 | comp->cq = NULL; | |
5a33a669 | 142 | goto cq_err; |
c33b15f0 | 143 | } |
1cfa0a75 | 144 | |
6aabfa76 | 145 | if (ib_req_notify_cq(comp->cq, IB_CQ_NEXT_COMP)) |
5a33a669 | 146 | goto cq_err; |
1cfa0a75 | 147 | |
bf175540 SG |
148 | tasklet_init(&comp->tasklet, iser_cq_tasklet_fn, |
149 | (unsigned long)comp); | |
5a33a669 | 150 | } |
1cfa0a75 | 151 | |
d8111028 EZ |
152 | device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE | |
153 | IB_ACCESS_REMOTE_WRITE | | |
154 | IB_ACCESS_REMOTE_READ); | |
1cfa0a75 OG |
155 | if (IS_ERR(device->mr)) |
156 | goto dma_mr_err; | |
157 | ||
2110f9bf OG |
158 | INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device, |
159 | iser_event_handler); | |
160 | if (ib_register_event_handler(&device->event_handler)) | |
161 | goto handler_err; | |
162 | ||
1cfa0a75 OG |
163 | return 0; |
164 | ||
2110f9bf OG |
165 | handler_err: |
166 | ib_dereg_mr(device->mr); | |
1cfa0a75 | 167 | dma_mr_err: |
bf175540 SG |
168 | for (i = 0; i < device->comps_used; i++) |
169 | tasklet_kill(&device->comps[i].tasklet); | |
5a33a669 | 170 | cq_err: |
bf175540 SG |
171 | for (i = 0; i < device->comps_used; i++) { |
172 | struct iser_comp *comp = &device->comps[i]; | |
173 | ||
6aabfa76 SG |
174 | if (comp->cq) |
175 | ib_destroy_cq(comp->cq); | |
5a33a669 | 176 | } |
1cfa0a75 OG |
177 | ib_dealloc_pd(device->pd); |
178 | pd_err: | |
da64bdb2 SG |
179 | kfree(device->comps); |
180 | comps_err: | |
1cfa0a75 OG |
181 | iser_err("failed to allocate an IB resource\n"); |
182 | return -1; | |
183 | } | |
184 | ||
185 | /** | |
38dc732f | 186 | * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR, |
1cfa0a75 OG |
187 | * CQ and PD created with the device associated with the adapator. |
188 | */ | |
189 | static void iser_free_device_ib_res(struct iser_device *device) | |
190 | { | |
5a33a669 | 191 | int i; |
1cfa0a75 OG |
192 | BUG_ON(device->mr == NULL); |
193 | ||
bf175540 SG |
194 | for (i = 0; i < device->comps_used; i++) { |
195 | struct iser_comp *comp = &device->comps[i]; | |
196 | ||
197 | tasklet_kill(&comp->tasklet); | |
6aabfa76 SG |
198 | ib_destroy_cq(comp->cq); |
199 | comp->cq = NULL; | |
5a33a669 AT |
200 | } |
201 | ||
2110f9bf | 202 | (void)ib_unregister_event_handler(&device->event_handler); |
1cfa0a75 | 203 | (void)ib_dereg_mr(device->mr); |
1cfa0a75 OG |
204 | (void)ib_dealloc_pd(device->pd); |
205 | ||
da64bdb2 SG |
206 | kfree(device->comps); |
207 | device->comps = NULL; | |
208 | ||
1cfa0a75 | 209 | device->mr = NULL; |
1cfa0a75 OG |
210 | device->pd = NULL; |
211 | } | |
212 | ||
213 | /** | |
986db0d6 | 214 | * iser_create_fmr_pool - Creates FMR pool and page_vector |
1cfa0a75 | 215 | * |
986db0d6 | 216 | * returns 0 on success, or errno code on failure |
1cfa0a75 | 217 | */ |
a4ee3539 | 218 | int iser_create_fmr_pool(struct ib_conn *ib_conn, unsigned cmds_max) |
1cfa0a75 | 219 | { |
a4ee3539 | 220 | struct iser_device *device = ib_conn->device; |
1cfa0a75 | 221 | struct ib_fmr_pool_param params; |
986db0d6 | 222 | int ret = -ENOMEM; |
bcc60c38 | 223 | |
a4ee3539 | 224 | ib_conn->fmr.page_vec = kmalloc(sizeof(*ib_conn->fmr.page_vec) + |
7306b8fa SG |
225 | (sizeof(u64)*(ISCSI_ISER_SG_TABLESIZE + 1)), |
226 | GFP_KERNEL); | |
a4ee3539 | 227 | if (!ib_conn->fmr.page_vec) |
986db0d6 | 228 | return ret; |
9fda1ac5 | 229 | |
a4ee3539 | 230 | ib_conn->fmr.page_vec->pages = (u64 *)(ib_conn->fmr.page_vec + 1); |
1cfa0a75 | 231 | |
8dfa0876 | 232 | params.page_shift = SHIFT_4K; |
1cfa0a75 OG |
233 | /* when the first/last SG element are not start/end * |
234 | * page aligned, the map whould be of N+1 pages */ | |
235 | params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1; | |
236 | /* make the pool size twice the max number of SCSI commands * | |
237 | * the ML is expected to queue, watermark for unmap at 50% */ | |
b7f04513 SP |
238 | params.pool_size = cmds_max * 2; |
239 | params.dirty_watermark = cmds_max; | |
1cfa0a75 OG |
240 | params.cache = 0; |
241 | params.flush_function = NULL; | |
242 | params.access = (IB_ACCESS_LOCAL_WRITE | | |
243 | IB_ACCESS_REMOTE_WRITE | | |
244 | IB_ACCESS_REMOTE_READ); | |
245 | ||
a4ee3539 SG |
246 | ib_conn->fmr.pool = ib_create_fmr_pool(device->pd, ¶ms); |
247 | if (!IS_ERR(ib_conn->fmr.pool)) | |
986db0d6 SP |
248 | return 0; |
249 | ||
250 | /* no FMR => no need for page_vec */ | |
a4ee3539 SG |
251 | kfree(ib_conn->fmr.page_vec); |
252 | ib_conn->fmr.page_vec = NULL; | |
986db0d6 | 253 | |
a4ee3539 SG |
254 | ret = PTR_ERR(ib_conn->fmr.pool); |
255 | ib_conn->fmr.pool = NULL; | |
986db0d6 SP |
256 | if (ret != -ENOSYS) { |
257 | iser_err("FMR allocation failed, err %d\n", ret); | |
258 | return ret; | |
259 | } else { | |
5525d210 | 260 | iser_warn("FMRs are not supported, using unaligned mode\n"); |
986db0d6 | 261 | return 0; |
1cfa0a75 | 262 | } |
986db0d6 SP |
263 | } |
264 | ||
265 | /** | |
266 | * iser_free_fmr_pool - releases the FMR pool and page vec | |
267 | */ | |
a4ee3539 | 268 | void iser_free_fmr_pool(struct ib_conn *ib_conn) |
986db0d6 SP |
269 | { |
270 | iser_info("freeing conn %p fmr pool %p\n", | |
a4ee3539 | 271 | ib_conn, ib_conn->fmr.pool); |
986db0d6 | 272 | |
a4ee3539 SG |
273 | if (ib_conn->fmr.pool != NULL) |
274 | ib_destroy_fmr_pool(ib_conn->fmr.pool); | |
986db0d6 | 275 | |
a4ee3539 | 276 | ib_conn->fmr.pool = NULL; |
986db0d6 | 277 | |
a4ee3539 SG |
278 | kfree(ib_conn->fmr.page_vec); |
279 | ib_conn->fmr.page_vec = NULL; | |
986db0d6 SP |
280 | } |
281 | ||
4dec2a27 SG |
282 | static int |
283 | iser_alloc_pi_ctx(struct ib_device *ib_device, struct ib_pd *pd, | |
284 | struct fast_reg_descriptor *desc) | |
285 | { | |
286 | struct iser_pi_context *pi_ctx = NULL; | |
9bee178b | 287 | int ret; |
4dec2a27 SG |
288 | |
289 | desc->pi_ctx = kzalloc(sizeof(*desc->pi_ctx), GFP_KERNEL); | |
290 | if (!desc->pi_ctx) | |
291 | return -ENOMEM; | |
292 | ||
293 | pi_ctx = desc->pi_ctx; | |
294 | ||
295 | pi_ctx->prot_frpl = ib_alloc_fast_reg_page_list(ib_device, | |
296 | ISCSI_ISER_SG_TABLESIZE); | |
297 | if (IS_ERR(pi_ctx->prot_frpl)) { | |
298 | ret = PTR_ERR(pi_ctx->prot_frpl); | |
299 | goto prot_frpl_failure; | |
300 | } | |
301 | ||
34780f01 SG |
302 | pi_ctx->prot_mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, |
303 | ISCSI_ISER_SG_TABLESIZE + 1); | |
4dec2a27 SG |
304 | if (IS_ERR(pi_ctx->prot_mr)) { |
305 | ret = PTR_ERR(pi_ctx->prot_mr); | |
306 | goto prot_mr_failure; | |
307 | } | |
308 | desc->reg_indicators |= ISER_PROT_KEY_VALID; | |
309 | ||
9bee178b | 310 | pi_ctx->sig_mr = ib_alloc_mr(pd, IB_MR_TYPE_SIGNATURE, 2); |
4dec2a27 SG |
311 | if (IS_ERR(pi_ctx->sig_mr)) { |
312 | ret = PTR_ERR(pi_ctx->sig_mr); | |
313 | goto sig_mr_failure; | |
314 | } | |
315 | desc->reg_indicators |= ISER_SIG_KEY_VALID; | |
316 | desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; | |
317 | ||
318 | return 0; | |
319 | ||
320 | sig_mr_failure: | |
321 | ib_dereg_mr(desc->pi_ctx->prot_mr); | |
322 | prot_mr_failure: | |
323 | ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); | |
324 | prot_frpl_failure: | |
325 | kfree(desc->pi_ctx); | |
326 | ||
327 | return ret; | |
328 | } | |
329 | ||
330 | static void | |
331 | iser_free_pi_ctx(struct iser_pi_context *pi_ctx) | |
332 | { | |
333 | ib_free_fast_reg_page_list(pi_ctx->prot_frpl); | |
334 | ib_dereg_mr(pi_ctx->prot_mr); | |
8b91ffc1 | 335 | ib_dereg_mr(pi_ctx->sig_mr); |
4dec2a27 SG |
336 | kfree(pi_ctx); |
337 | } | |
338 | ||
310b347c SG |
339 | static int |
340 | iser_create_fastreg_desc(struct ib_device *ib_device, struct ib_pd *pd, | |
6b5a8fb0 | 341 | bool pi_enable, struct fast_reg_descriptor *desc) |
310b347c SG |
342 | { |
343 | int ret; | |
344 | ||
345 | desc->data_frpl = ib_alloc_fast_reg_page_list(ib_device, | |
346 | ISCSI_ISER_SG_TABLESIZE + 1); | |
347 | if (IS_ERR(desc->data_frpl)) { | |
348 | ret = PTR_ERR(desc->data_frpl); | |
349 | iser_err("Failed to allocate ib_fast_reg_page_list err=%d\n", | |
350 | ret); | |
351 | return PTR_ERR(desc->data_frpl); | |
352 | } | |
353 | ||
34780f01 SG |
354 | desc->data_mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, |
355 | ISCSI_ISER_SG_TABLESIZE + 1); | |
310b347c SG |
356 | if (IS_ERR(desc->data_mr)) { |
357 | ret = PTR_ERR(desc->data_mr); | |
358 | iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret); | |
359 | goto fast_reg_mr_failure; | |
360 | } | |
6b5a8fb0 AT |
361 | desc->reg_indicators |= ISER_DATA_KEY_VALID; |
362 | ||
363 | if (pi_enable) { | |
4dec2a27 SG |
364 | ret = iser_alloc_pi_ctx(ib_device, pd, desc); |
365 | if (ret) | |
6b5a8fb0 | 366 | goto pi_ctx_alloc_failure; |
6b5a8fb0 | 367 | } |
310b347c SG |
368 | |
369 | return 0; | |
6b5a8fb0 AT |
370 | pi_ctx_alloc_failure: |
371 | ib_dereg_mr(desc->data_mr); | |
310b347c SG |
372 | fast_reg_mr_failure: |
373 | ib_free_fast_reg_page_list(desc->data_frpl); | |
374 | ||
375 | return ret; | |
376 | } | |
377 | ||
5587856c | 378 | /** |
7306b8fa | 379 | * iser_create_fastreg_pool - Creates pool of fast_reg descriptors |
5587856c SG |
380 | * for fast registration work requests. |
381 | * returns 0 on success, or errno code on failure | |
382 | */ | |
a4ee3539 | 383 | int iser_create_fastreg_pool(struct ib_conn *ib_conn, unsigned cmds_max) |
5587856c | 384 | { |
a4ee3539 SG |
385 | struct iser_device *device = ib_conn->device; |
386 | struct fast_reg_descriptor *desc; | |
5587856c SG |
387 | int i, ret; |
388 | ||
a4ee3539 SG |
389 | INIT_LIST_HEAD(&ib_conn->fastreg.pool); |
390 | ib_conn->fastreg.pool_size = 0; | |
5587856c | 391 | for (i = 0; i < cmds_max; i++) { |
6b5a8fb0 | 392 | desc = kzalloc(sizeof(*desc), GFP_KERNEL); |
5587856c SG |
393 | if (!desc) { |
394 | iser_err("Failed to allocate a new fast_reg descriptor\n"); | |
395 | ret = -ENOMEM; | |
396 | goto err; | |
397 | } | |
398 | ||
6b5a8fb0 | 399 | ret = iser_create_fastreg_desc(device->ib_device, device->pd, |
a4ee3539 | 400 | ib_conn->pi_support, desc); |
310b347c SG |
401 | if (ret) { |
402 | iser_err("Failed to create fastreg descriptor err=%d\n", | |
403 | ret); | |
404 | kfree(desc); | |
405 | goto err; | |
5587856c SG |
406 | } |
407 | ||
a4ee3539 SG |
408 | list_add_tail(&desc->list, &ib_conn->fastreg.pool); |
409 | ib_conn->fastreg.pool_size++; | |
5587856c SG |
410 | } |
411 | ||
412 | return 0; | |
27ae2d1e | 413 | |
5587856c | 414 | err: |
a4ee3539 | 415 | iser_free_fastreg_pool(ib_conn); |
5587856c SG |
416 | return ret; |
417 | } | |
418 | ||
419 | /** | |
7306b8fa | 420 | * iser_free_fastreg_pool - releases the pool of fast_reg descriptors |
5587856c | 421 | */ |
a4ee3539 | 422 | void iser_free_fastreg_pool(struct ib_conn *ib_conn) |
5587856c SG |
423 | { |
424 | struct fast_reg_descriptor *desc, *tmp; | |
425 | int i = 0; | |
426 | ||
a4ee3539 | 427 | if (list_empty(&ib_conn->fastreg.pool)) |
5587856c SG |
428 | return; |
429 | ||
a4ee3539 | 430 | iser_info("freeing conn %p fr pool\n", ib_conn); |
5587856c | 431 | |
a4ee3539 | 432 | list_for_each_entry_safe(desc, tmp, &ib_conn->fastreg.pool, list) { |
5587856c SG |
433 | list_del(&desc->list); |
434 | ib_free_fast_reg_page_list(desc->data_frpl); | |
435 | ib_dereg_mr(desc->data_mr); | |
4dec2a27 SG |
436 | if (desc->pi_ctx) |
437 | iser_free_pi_ctx(desc->pi_ctx); | |
5587856c SG |
438 | kfree(desc); |
439 | ++i; | |
440 | } | |
441 | ||
a4ee3539 | 442 | if (i < ib_conn->fastreg.pool_size) |
5587856c | 443 | iser_warn("pool still has %d regions registered\n", |
a4ee3539 | 444 | ib_conn->fastreg.pool_size - i); |
5587856c SG |
445 | } |
446 | ||
986db0d6 SP |
447 | /** |
448 | * iser_create_ib_conn_res - Queue-Pair (QP) | |
449 | * | |
450 | * returns 0 on success, -1 on failure | |
451 | */ | |
a4ee3539 | 452 | static int iser_create_ib_conn_res(struct ib_conn *ib_conn) |
986db0d6 | 453 | { |
f4641ef7 MT |
454 | struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn, |
455 | ib_conn); | |
986db0d6 | 456 | struct iser_device *device; |
f4641ef7 | 457 | struct ib_device_attr *dev_attr; |
986db0d6 SP |
458 | struct ib_qp_init_attr init_attr; |
459 | int ret = -ENOMEM; | |
460 | int index, min_index = 0; | |
461 | ||
a4ee3539 | 462 | BUG_ON(ib_conn->device == NULL); |
986db0d6 | 463 | |
a4ee3539 | 464 | device = ib_conn->device; |
f4641ef7 | 465 | dev_attr = &device->dev_attr; |
1cfa0a75 OG |
466 | |
467 | memset(&init_attr, 0, sizeof init_attr); | |
468 | ||
5a33a669 AT |
469 | mutex_lock(&ig.connlist_mutex); |
470 | /* select the CQ with the minimal number of usages */ | |
bf175540 SG |
471 | for (index = 0; index < device->comps_used; index++) { |
472 | if (device->comps[index].active_qps < | |
473 | device->comps[min_index].active_qps) | |
5a33a669 | 474 | min_index = index; |
bf175540 SG |
475 | } |
476 | ib_conn->comp = &device->comps[min_index]; | |
477 | ib_conn->comp->active_qps++; | |
5a33a669 | 478 | mutex_unlock(&ig.connlist_mutex); |
a4ee3539 | 479 | iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn); |
5a33a669 | 480 | |
1cfa0a75 | 481 | init_attr.event_handler = iser_qp_event_callback; |
a4ee3539 | 482 | init_attr.qp_context = (void *)ib_conn; |
6aabfa76 SG |
483 | init_attr.send_cq = ib_conn->comp->cq; |
484 | init_attr.recv_cq = ib_conn->comp->cq; | |
1cfa0a75 | 485 | init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS; |
f19624aa | 486 | init_attr.cap.max_send_sge = 2; |
bcc60c38 | 487 | init_attr.cap.max_recv_sge = 1; |
1cfa0a75 OG |
488 | init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; |
489 | init_attr.qp_type = IB_QPT_RC; | |
a4ee3539 | 490 | if (ib_conn->pi_support) { |
ff3dd52d | 491 | init_attr.cap.max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1; |
6b5a8fb0 | 492 | init_attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN; |
f4641ef7 MT |
493 | iser_conn->max_cmds = |
494 | ISER_GET_MAX_XMIT_CMDS(ISER_QP_SIG_MAX_REQ_DTOS); | |
6b5a8fb0 | 495 | } else { |
f4641ef7 MT |
496 | if (dev_attr->max_qp_wr > ISER_QP_MAX_REQ_DTOS) { |
497 | init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS + 1; | |
498 | iser_conn->max_cmds = | |
499 | ISER_GET_MAX_XMIT_CMDS(ISER_QP_MAX_REQ_DTOS); | |
500 | } else { | |
501 | init_attr.cap.max_send_wr = dev_attr->max_qp_wr; | |
502 | iser_conn->max_cmds = | |
503 | ISER_GET_MAX_XMIT_CMDS(dev_attr->max_qp_wr); | |
504 | iser_dbg("device %s supports max_send_wr %d\n", | |
505 | device->ib_device->name, dev_attr->max_qp_wr); | |
506 | } | |
6b5a8fb0 | 507 | } |
1cfa0a75 | 508 | |
a4ee3539 | 509 | ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); |
1cfa0a75 | 510 | if (ret) |
9fda1ac5 | 511 | goto out_err; |
1cfa0a75 | 512 | |
a4ee3539 | 513 | ib_conn->qp = ib_conn->cma_id->qp; |
986db0d6 | 514 | iser_info("setting conn %p cma_id %p qp %p\n", |
a4ee3539 SG |
515 | ib_conn, ib_conn->cma_id, |
516 | ib_conn->cma_id->qp); | |
1cfa0a75 OG |
517 | return ret; |
518 | ||
9fda1ac5 | 519 | out_err: |
93acb7bb SG |
520 | mutex_lock(&ig.connlist_mutex); |
521 | ib_conn->comp->active_qps--; | |
522 | mutex_unlock(&ig.connlist_mutex); | |
1cfa0a75 | 523 | iser_err("unable to alloc mem or create resource, err %d\n", ret); |
93acb7bb | 524 | |
1cfa0a75 OG |
525 | return ret; |
526 | } | |
527 | ||
1cfa0a75 OG |
528 | /** |
529 | * based on the resolved device node GUID see if there already allocated | |
530 | * device for this device. If there's no such, create one. | |
531 | */ | |
532 | static | |
533 | struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id) | |
534 | { | |
9a378270 | 535 | struct iser_device *device; |
1cfa0a75 OG |
536 | |
537 | mutex_lock(&ig.device_list_mutex); | |
538 | ||
9a378270 | 539 | list_for_each_entry(device, &ig.device_list, ig_list) |
1cfa0a75 OG |
540 | /* find if there's a match using the node GUID */ |
541 | if (device->ib_device->node_guid == cma_id->device->node_guid) | |
d33ed425 | 542 | goto inc_refcnt; |
9a378270 AR |
543 | |
544 | device = kzalloc(sizeof *device, GFP_KERNEL); | |
545 | if (device == NULL) | |
546 | goto out; | |
547 | ||
548 | /* assign this device to the device */ | |
549 | device->ib_device = cma_id->device; | |
550 | /* init the device and link it into ig device list */ | |
551 | if (iser_create_device_ib_res(device)) { | |
552 | kfree(device); | |
553 | device = NULL; | |
554 | goto out; | |
1cfa0a75 | 555 | } |
9a378270 AR |
556 | list_add(&device->ig_list, &ig.device_list); |
557 | ||
d33ed425 | 558 | inc_refcnt: |
1cfa0a75 | 559 | device->refcount++; |
d33ed425 | 560 | out: |
1cfa0a75 OG |
561 | mutex_unlock(&ig.device_list_mutex); |
562 | return device; | |
563 | } | |
564 | ||
565 | /* if there's no demand for this device, release it */ | |
566 | static void iser_device_try_release(struct iser_device *device) | |
567 | { | |
568 | mutex_lock(&ig.device_list_mutex); | |
569 | device->refcount--; | |
4f363882 | 570 | iser_info("device %p refcount %d\n", device, device->refcount); |
1cfa0a75 OG |
571 | if (!device->refcount) { |
572 | iser_free_device_ib_res(device); | |
573 | list_del(&device->ig_list); | |
574 | kfree(device); | |
575 | } | |
576 | mutex_unlock(&ig.device_list_mutex); | |
577 | } | |
578 | ||
504130c0 AN |
579 | /** |
580 | * Called with state mutex held | |
581 | **/ | |
5716af6e SG |
582 | static int iser_conn_state_comp_exch(struct iser_conn *iser_conn, |
583 | enum iser_conn_state comp, | |
584 | enum iser_conn_state exch) | |
1cfa0a75 OG |
585 | { |
586 | int ret; | |
587 | ||
5716af6e SG |
588 | ret = (iser_conn->state == comp); |
589 | if (ret) | |
590 | iser_conn->state = exch; | |
591 | ||
1cfa0a75 OG |
592 | return ret; |
593 | } | |
594 | ||
b73c3ada AN |
595 | void iser_release_work(struct work_struct *work) |
596 | { | |
5716af6e | 597 | struct iser_conn *iser_conn; |
b73c3ada | 598 | |
5716af6e | 599 | iser_conn = container_of(work, struct iser_conn, release_work); |
b73c3ada | 600 | |
c107a6c0 SG |
601 | /* Wait for conn_stop to complete */ |
602 | wait_for_completion(&iser_conn->stop_completion); | |
603 | /* Wait for IB resouces cleanup to complete */ | |
604 | wait_for_completion(&iser_conn->ib_completion); | |
b73c3ada | 605 | |
5716af6e SG |
606 | mutex_lock(&iser_conn->state_mutex); |
607 | iser_conn->state = ISER_CONN_DOWN; | |
608 | mutex_unlock(&iser_conn->state_mutex); | |
504130c0 | 609 | |
5716af6e | 610 | iser_conn_release(iser_conn); |
b73c3ada AN |
611 | } |
612 | ||
96f15198 SG |
613 | /** |
614 | * iser_free_ib_conn_res - release IB related resources | |
615 | * @iser_conn: iser connection struct | |
6606e6a2 SG |
616 | * @destroy: indicator if we need to try to release the |
617 | * iser device and memory regoins pool (only iscsi | |
618 | * shutdown and DEVICE_REMOVAL will use this). | |
96f15198 SG |
619 | * |
620 | * This routine is called with the iser state mutex held | |
621 | * so the cm_id removal is out of here. It is Safe to | |
622 | * be invoked multiple times. | |
623 | */ | |
c47a3c9e | 624 | static void iser_free_ib_conn_res(struct iser_conn *iser_conn, |
6606e6a2 | 625 | bool destroy) |
96f15198 SG |
626 | { |
627 | struct ib_conn *ib_conn = &iser_conn->ib_conn; | |
628 | struct iser_device *device = ib_conn->device; | |
629 | ||
630 | iser_info("freeing conn %p cma_id %p qp %p\n", | |
631 | iser_conn, ib_conn->cma_id, ib_conn->qp); | |
632 | ||
96f15198 | 633 | if (ib_conn->qp != NULL) { |
bf175540 | 634 | ib_conn->comp->active_qps--; |
96f15198 SG |
635 | rdma_destroy_qp(ib_conn->cma_id); |
636 | ib_conn->qp = NULL; | |
637 | } | |
638 | ||
6606e6a2 SG |
639 | if (destroy) { |
640 | if (iser_conn->rx_descs) | |
641 | iser_free_rx_descriptors(iser_conn); | |
642 | ||
643 | if (device != NULL) { | |
644 | iser_device_try_release(device); | |
645 | ib_conn->device = NULL; | |
646 | } | |
96f15198 SG |
647 | } |
648 | } | |
649 | ||
41179e2d RD |
650 | /** |
651 | * Frees all conn objects and deallocs conn descriptor | |
652 | */ | |
5716af6e | 653 | void iser_conn_release(struct iser_conn *iser_conn) |
41179e2d | 654 | { |
a4ee3539 | 655 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
41179e2d | 656 | |
41179e2d | 657 | mutex_lock(&ig.connlist_mutex); |
5716af6e | 658 | list_del(&iser_conn->conn_list); |
41179e2d | 659 | mutex_unlock(&ig.connlist_mutex); |
504130c0 | 660 | |
5716af6e | 661 | mutex_lock(&iser_conn->state_mutex); |
9a3119e4 | 662 | /* In case we endup here without ep_disconnect being invoked. */ |
5426b171 | 663 | if (iser_conn->state != ISER_CONN_DOWN) { |
aea8f4df AN |
664 | iser_warn("iser conn %p state %d, expected state down.\n", |
665 | iser_conn, iser_conn->state); | |
9a3119e4 | 666 | iscsi_destroy_endpoint(iser_conn->ep); |
5426b171 AN |
667 | iser_conn->state = ISER_CONN_DOWN; |
668 | } | |
c47a3c9e SG |
669 | /* |
670 | * In case we never got to bind stage, we still need to | |
671 | * release IB resources (which is safe to call more than once). | |
672 | */ | |
673 | iser_free_ib_conn_res(iser_conn, true); | |
5716af6e | 674 | mutex_unlock(&iser_conn->state_mutex); |
504130c0 | 675 | |
a4ee3539 SG |
676 | if (ib_conn->cma_id != NULL) { |
677 | rdma_destroy_id(ib_conn->cma_id); | |
678 | ib_conn->cma_id = NULL; | |
5b61ff43 | 679 | } |
a4ee3539 | 680 | |
5716af6e | 681 | kfree(iser_conn); |
41179e2d RD |
682 | } |
683 | ||
1cfa0a75 OG |
684 | /** |
685 | * triggers start of the disconnect procedures and wait for them to be done | |
c47a3c9e | 686 | * Called with state mutex held |
1cfa0a75 | 687 | */ |
c47a3c9e | 688 | int iser_conn_terminate(struct iser_conn *iser_conn) |
1cfa0a75 | 689 | { |
a4ee3539 | 690 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
ff3dd52d | 691 | struct ib_send_wr *bad_wr; |
1cfa0a75 OG |
692 | int err = 0; |
693 | ||
c47a3c9e SG |
694 | /* terminate the iser conn only if the conn state is UP */ |
695 | if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP, | |
696 | ISER_CONN_TERMINATING)) | |
697 | return 0; | |
698 | ||
699 | iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state); | |
700 | ||
701 | /* suspend queuing of new iscsi commands */ | |
702 | if (iser_conn->iscsi_conn) | |
703 | iscsi_suspend_queue(iser_conn->iscsi_conn); | |
704 | ||
705 | /* | |
706 | * In case we didn't already clean up the cma_id (peer initiated | |
707 | * a disconnection), we need to Cause the CMA to change the QP | |
708 | * state to ERROR. | |
1cfa0a75 | 709 | */ |
c47a3c9e SG |
710 | if (ib_conn->cma_id) { |
711 | err = rdma_disconnect(ib_conn->cma_id); | |
712 | if (err) | |
713 | iser_err("Failed to disconnect, conn: 0x%p err %d\n", | |
714 | iser_conn, err); | |
715 | ||
ff3dd52d SG |
716 | /* post an indication that all flush errors were consumed */ |
717 | err = ib_post_send(ib_conn->qp, &ib_conn->beacon, &bad_wr); | |
16df2a26 | 718 | if (err) { |
ff3dd52d | 719 | iser_err("conn %p failed to post beacon", ib_conn); |
16df2a26 SG |
720 | return 1; |
721 | } | |
ff3dd52d | 722 | |
6aabfa76 | 723 | wait_for_completion(&ib_conn->flush_comp); |
c47a3c9e | 724 | } |
1cfa0a75 | 725 | |
c47a3c9e | 726 | return 1; |
1cfa0a75 OG |
727 | } |
728 | ||
504130c0 AN |
729 | /** |
730 | * Called with state mutex held | |
731 | **/ | |
b73c3ada | 732 | static void iser_connect_error(struct rdma_cm_id *cma_id) |
1cfa0a75 | 733 | { |
5716af6e | 734 | struct iser_conn *iser_conn; |
b73c3ada | 735 | |
5716af6e | 736 | iser_conn = (struct iser_conn *)cma_id->context; |
c4de4663 | 737 | iser_conn->state = ISER_CONN_TERMINATING; |
1cfa0a75 OG |
738 | } |
739 | ||
504130c0 AN |
740 | /** |
741 | * Called with state mutex held | |
742 | **/ | |
b73c3ada | 743 | static void iser_addr_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
744 | { |
745 | struct iser_device *device; | |
5716af6e | 746 | struct iser_conn *iser_conn; |
a4ee3539 | 747 | struct ib_conn *ib_conn; |
1cfa0a75 OG |
748 | int ret; |
749 | ||
5716af6e SG |
750 | iser_conn = (struct iser_conn *)cma_id->context; |
751 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
752 | /* bailout */ |
753 | return; | |
754 | ||
a4ee3539 | 755 | ib_conn = &iser_conn->ib_conn; |
1cfa0a75 | 756 | device = iser_device_find_by_ib_device(cma_id); |
d33ed425 AR |
757 | if (!device) { |
758 | iser_err("device lookup/creation failed\n"); | |
b73c3ada AN |
759 | iser_connect_error(cma_id); |
760 | return; | |
d33ed425 AR |
761 | } |
762 | ||
a4ee3539 | 763 | ib_conn->device = device; |
1cfa0a75 | 764 | |
7f733847 AT |
765 | /* connection T10-PI support */ |
766 | if (iser_pi_enable) { | |
767 | if (!(device->dev_attr.device_cap_flags & | |
768 | IB_DEVICE_SIGNATURE_HANDOVER)) { | |
769 | iser_warn("T10-PI requested but not supported on %s, " | |
770 | "continue without T10-PI\n", | |
a4ee3539 SG |
771 | ib_conn->device->ib_device->name); |
772 | ib_conn->pi_support = false; | |
7f733847 | 773 | } else { |
a4ee3539 | 774 | ib_conn->pi_support = true; |
7f733847 AT |
775 | } |
776 | } | |
777 | ||
1cfa0a75 OG |
778 | ret = rdma_resolve_route(cma_id, 1000); |
779 | if (ret) { | |
780 | iser_err("resolve route failed: %d\n", ret); | |
b73c3ada AN |
781 | iser_connect_error(cma_id); |
782 | return; | |
1cfa0a75 | 783 | } |
1cfa0a75 OG |
784 | } |
785 | ||
504130c0 AN |
786 | /** |
787 | * Called with state mutex held | |
788 | **/ | |
b73c3ada | 789 | static void iser_route_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
790 | { |
791 | struct rdma_conn_param conn_param; | |
792 | int ret; | |
8d8399de | 793 | struct iser_cm_hdr req_hdr; |
5716af6e | 794 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
a4ee3539 SG |
795 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
796 | struct iser_device *device = ib_conn->device; | |
1cfa0a75 | 797 | |
5716af6e | 798 | if (iser_conn->state != ISER_CONN_PENDING) |
504130c0 AN |
799 | /* bailout */ |
800 | return; | |
801 | ||
a4ee3539 | 802 | ret = iser_create_ib_conn_res(ib_conn); |
1cfa0a75 OG |
803 | if (ret) |
804 | goto failure; | |
805 | ||
1cfa0a75 | 806 | memset(&conn_param, 0, sizeof conn_param); |
2ea32938 | 807 | conn_param.responder_resources = device->dev_attr.max_qp_rd_atom; |
1cfa0a75 OG |
808 | conn_param.initiator_depth = 1; |
809 | conn_param.retry_count = 7; | |
810 | conn_param.rnr_retry_count = 6; | |
811 | ||
8d8399de OG |
812 | memset(&req_hdr, 0, sizeof(req_hdr)); |
813 | req_hdr.flags = (ISER_ZBVA_NOT_SUPPORTED | | |
814 | ISER_SEND_W_INV_NOT_SUPPORTED); | |
815 | conn_param.private_data = (void *)&req_hdr; | |
816 | conn_param.private_data_len = sizeof(struct iser_cm_hdr); | |
817 | ||
1cfa0a75 OG |
818 | ret = rdma_connect(cma_id, &conn_param); |
819 | if (ret) { | |
820 | iser_err("failure connecting: %d\n", ret); | |
821 | goto failure; | |
822 | } | |
823 | ||
b73c3ada | 824 | return; |
1cfa0a75 | 825 | failure: |
b73c3ada | 826 | iser_connect_error(cma_id); |
1cfa0a75 OG |
827 | } |
828 | ||
829 | static void iser_connected_handler(struct rdma_cm_id *cma_id) | |
830 | { | |
5716af6e | 831 | struct iser_conn *iser_conn; |
4f9208ad OG |
832 | struct ib_qp_attr attr; |
833 | struct ib_qp_init_attr init_attr; | |
834 | ||
5716af6e SG |
835 | iser_conn = (struct iser_conn *)cma_id->context; |
836 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
837 | /* bailout */ |
838 | return; | |
839 | ||
4f9208ad OG |
840 | (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr); |
841 | iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num); | |
1cfa0a75 | 842 | |
5716af6e SG |
843 | iser_conn->state = ISER_CONN_UP; |
844 | complete(&iser_conn->up_completion); | |
1cfa0a75 OG |
845 | } |
846 | ||
b73c3ada | 847 | static void iser_disconnected_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 | 848 | { |
c47a3c9e | 849 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
1cfa0a75 | 850 | |
c47a3c9e | 851 | if (iser_conn_terminate(iser_conn)) { |
5716af6e | 852 | if (iser_conn->iscsi_conn) |
c47a3c9e SG |
853 | iscsi_conn_failure(iser_conn->iscsi_conn, |
854 | ISCSI_ERR_CONN_FAILED); | |
7d9eacf9 RD |
855 | else |
856 | iser_err("iscsi_iser connection isn't bound\n"); | |
857 | } | |
c47a3c9e SG |
858 | } |
859 | ||
860 | static void iser_cleanup_handler(struct rdma_cm_id *cma_id, | |
6606e6a2 | 861 | bool destroy) |
c47a3c9e SG |
862 | { |
863 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; | |
1cfa0a75 | 864 | |
c47a3c9e SG |
865 | /* |
866 | * We are not guaranteed that we visited disconnected_handler | |
867 | * by now, call it here to be safe that we handle CM drep | |
868 | * and flush errors. | |
8d4aca7f | 869 | */ |
c47a3c9e | 870 | iser_disconnected_handler(cma_id); |
6606e6a2 | 871 | iser_free_ib_conn_res(iser_conn, destroy); |
c47a3c9e SG |
872 | complete(&iser_conn->ib_completion); |
873 | }; | |
1cfa0a75 OG |
874 | |
875 | static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) | |
876 | { | |
5716af6e | 877 | struct iser_conn *iser_conn; |
c47a3c9e | 878 | int ret = 0; |
504130c0 | 879 | |
5716af6e | 880 | iser_conn = (struct iser_conn *)cma_id->context; |
871e00af SG |
881 | iser_info("%s (%d): status %d conn %p id %p\n", |
882 | rdma_event_msg(event->event), event->event, | |
883 | event->status, cma_id->context, cma_id); | |
1cfa0a75 | 884 | |
5716af6e | 885 | mutex_lock(&iser_conn->state_mutex); |
1cfa0a75 OG |
886 | switch (event->event) { |
887 | case RDMA_CM_EVENT_ADDR_RESOLVED: | |
b73c3ada | 888 | iser_addr_handler(cma_id); |
1cfa0a75 OG |
889 | break; |
890 | case RDMA_CM_EVENT_ROUTE_RESOLVED: | |
b73c3ada | 891 | iser_route_handler(cma_id); |
1cfa0a75 OG |
892 | break; |
893 | case RDMA_CM_EVENT_ESTABLISHED: | |
894 | iser_connected_handler(cma_id); | |
895 | break; | |
896 | case RDMA_CM_EVENT_ADDR_ERROR: | |
897 | case RDMA_CM_EVENT_ROUTE_ERROR: | |
898 | case RDMA_CM_EVENT_CONNECT_ERROR: | |
899 | case RDMA_CM_EVENT_UNREACHABLE: | |
900 | case RDMA_CM_EVENT_REJECTED: | |
b73c3ada | 901 | iser_connect_error(cma_id); |
1cfa0a75 OG |
902 | break; |
903 | case RDMA_CM_EVENT_DISCONNECTED: | |
2f5de151 | 904 | case RDMA_CM_EVENT_ADDR_CHANGE: |
5426b171 AN |
905 | case RDMA_CM_EVENT_TIMEWAIT_EXIT: |
906 | iser_cleanup_handler(cma_id, false); | |
1cfa0a75 | 907 | break; |
c47a3c9e SG |
908 | case RDMA_CM_EVENT_DEVICE_REMOVAL: |
909 | /* | |
910 | * we *must* destroy the device as we cannot rely | |
911 | * on iscsid to be around to initiate error handling. | |
5426b171 AN |
912 | * also if we are not in state DOWN implicitly destroy |
913 | * the cma_id. | |
c47a3c9e SG |
914 | */ |
915 | iser_cleanup_handler(cma_id, true); | |
5426b171 AN |
916 | if (iser_conn->state != ISER_CONN_DOWN) { |
917 | iser_conn->ib_conn.cma_id = NULL; | |
918 | ret = 1; | |
919 | } | |
c47a3c9e | 920 | break; |
1cfa0a75 | 921 | default: |
871e00af SG |
922 | iser_err("Unexpected RDMA CM event: %s (%d)\n", |
923 | rdma_event_msg(event->event), event->event); | |
1cfa0a75 OG |
924 | break; |
925 | } | |
5716af6e | 926 | mutex_unlock(&iser_conn->state_mutex); |
c47a3c9e SG |
927 | |
928 | return ret; | |
1cfa0a75 OG |
929 | } |
930 | ||
5716af6e | 931 | void iser_conn_init(struct iser_conn *iser_conn) |
1cfa0a75 | 932 | { |
5716af6e | 933 | iser_conn->state = ISER_CONN_INIT; |
a4ee3539 | 934 | iser_conn->ib_conn.post_recv_buf_count = 0; |
6aabfa76 | 935 | init_completion(&iser_conn->ib_conn.flush_comp); |
5716af6e | 936 | init_completion(&iser_conn->stop_completion); |
c47a3c9e | 937 | init_completion(&iser_conn->ib_completion); |
5716af6e SG |
938 | init_completion(&iser_conn->up_completion); |
939 | INIT_LIST_HEAD(&iser_conn->conn_list); | |
a4ee3539 | 940 | spin_lock_init(&iser_conn->ib_conn.lock); |
5716af6e | 941 | mutex_init(&iser_conn->state_mutex); |
1cfa0a75 OG |
942 | } |
943 | ||
944 | /** | |
945 | * starts the process of connecting to the target | |
94e2bd68 | 946 | * sleeps until the connection is established or rejected |
1cfa0a75 | 947 | */ |
5716af6e | 948 | int iser_connect(struct iser_conn *iser_conn, |
96ed02d4 RD |
949 | struct sockaddr *src_addr, |
950 | struct sockaddr *dst_addr, | |
1cfa0a75 OG |
951 | int non_blocking) |
952 | { | |
a4ee3539 | 953 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
1cfa0a75 OG |
954 | int err = 0; |
955 | ||
5716af6e | 956 | mutex_lock(&iser_conn->state_mutex); |
504130c0 | 957 | |
5716af6e | 958 | sprintf(iser_conn->name, "%pISp", dst_addr); |
96ed02d4 | 959 | |
5716af6e | 960 | iser_info("connecting to: %s\n", iser_conn->name); |
1cfa0a75 OG |
961 | |
962 | /* the device is known only --after-- address resolution */ | |
a4ee3539 | 963 | ib_conn->device = NULL; |
1cfa0a75 | 964 | |
5716af6e | 965 | iser_conn->state = ISER_CONN_PENDING; |
1cfa0a75 | 966 | |
ff3dd52d SG |
967 | ib_conn->beacon.wr_id = ISER_BEACON_WRID; |
968 | ib_conn->beacon.opcode = IB_WR_SEND; | |
969 | ||
a4ee3539 SG |
970 | ib_conn->cma_id = rdma_create_id(iser_cma_handler, |
971 | (void *)iser_conn, | |
972 | RDMA_PS_TCP, IB_QPT_RC); | |
973 | if (IS_ERR(ib_conn->cma_id)) { | |
974 | err = PTR_ERR(ib_conn->cma_id); | |
1cfa0a75 OG |
975 | iser_err("rdma_create_id failed: %d\n", err); |
976 | goto id_failure; | |
977 | } | |
978 | ||
a4ee3539 | 979 | err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000); |
1cfa0a75 OG |
980 | if (err) { |
981 | iser_err("rdma_resolve_addr failed: %d\n", err); | |
982 | goto addr_failure; | |
983 | } | |
984 | ||
985 | if (!non_blocking) { | |
5716af6e | 986 | wait_for_completion_interruptible(&iser_conn->up_completion); |
1cfa0a75 | 987 | |
5716af6e | 988 | if (iser_conn->state != ISER_CONN_UP) { |
1cfa0a75 OG |
989 | err = -EIO; |
990 | goto connect_failure; | |
991 | } | |
992 | } | |
5716af6e | 993 | mutex_unlock(&iser_conn->state_mutex); |
1cfa0a75 OG |
994 | |
995 | mutex_lock(&ig.connlist_mutex); | |
5716af6e | 996 | list_add(&iser_conn->conn_list, &ig.connlist); |
1cfa0a75 OG |
997 | mutex_unlock(&ig.connlist_mutex); |
998 | return 0; | |
999 | ||
1000 | id_failure: | |
a4ee3539 | 1001 | ib_conn->cma_id = NULL; |
1cfa0a75 | 1002 | addr_failure: |
5716af6e | 1003 | iser_conn->state = ISER_CONN_DOWN; |
1cfa0a75 | 1004 | connect_failure: |
5716af6e SG |
1005 | mutex_unlock(&iser_conn->state_mutex); |
1006 | iser_conn_release(iser_conn); | |
1cfa0a75 OG |
1007 | return err; |
1008 | } | |
1009 | ||
5716af6e | 1010 | int iser_post_recvl(struct iser_conn *iser_conn) |
bcc60c38 OG |
1011 | { |
1012 | struct ib_recv_wr rx_wr, *rx_wr_failed; | |
a4ee3539 | 1013 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
bcc60c38 OG |
1014 | struct ib_sge sge; |
1015 | int ib_ret; | |
1016 | ||
5716af6e | 1017 | sge.addr = iser_conn->login_resp_dma; |
bcc60c38 | 1018 | sge.length = ISER_RX_LOGIN_SIZE; |
a4ee3539 | 1019 | sge.lkey = ib_conn->device->mr->lkey; |
bcc60c38 | 1020 | |
49df2781 | 1021 | rx_wr.wr_id = (uintptr_t)iser_conn->login_resp_buf; |
bcc60c38 OG |
1022 | rx_wr.sg_list = &sge; |
1023 | rx_wr.num_sge = 1; | |
1024 | rx_wr.next = NULL; | |
1025 | ||
a4ee3539 SG |
1026 | ib_conn->post_recv_buf_count++; |
1027 | ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1028 | if (ib_ret) { |
1029 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1030 | ib_conn->post_recv_buf_count--; |
bcc60c38 OG |
1031 | } |
1032 | return ib_ret; | |
1033 | } | |
1034 | ||
5716af6e | 1035 | int iser_post_recvm(struct iser_conn *iser_conn, int count) |
bcc60c38 OG |
1036 | { |
1037 | struct ib_recv_wr *rx_wr, *rx_wr_failed; | |
1038 | int i, ib_ret; | |
a4ee3539 | 1039 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
5716af6e | 1040 | unsigned int my_rx_head = iser_conn->rx_desc_head; |
bcc60c38 OG |
1041 | struct iser_rx_desc *rx_desc; |
1042 | ||
a4ee3539 | 1043 | for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { |
5716af6e | 1044 | rx_desc = &iser_conn->rx_descs[my_rx_head]; |
49df2781 | 1045 | rx_wr->wr_id = (uintptr_t)rx_desc; |
bcc60c38 OG |
1046 | rx_wr->sg_list = &rx_desc->rx_sg; |
1047 | rx_wr->num_sge = 1; | |
1048 | rx_wr->next = rx_wr + 1; | |
5716af6e | 1049 | my_rx_head = (my_rx_head + 1) & iser_conn->qp_max_recv_dtos_mask; |
bcc60c38 OG |
1050 | } |
1051 | ||
1052 | rx_wr--; | |
1053 | rx_wr->next = NULL; /* mark end of work requests list */ | |
1054 | ||
a4ee3539 SG |
1055 | ib_conn->post_recv_buf_count += count; |
1056 | ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1057 | if (ib_ret) { |
1058 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1059 | ib_conn->post_recv_buf_count -= count; |
bcc60c38 | 1060 | } else |
5716af6e | 1061 | iser_conn->rx_desc_head = my_rx_head; |
bcc60c38 OG |
1062 | return ib_ret; |
1063 | } | |
1064 | ||
1065 | ||
1cfa0a75 OG |
1066 | /** |
1067 | * iser_start_send - Initiate a Send DTO operation | |
1068 | * | |
1069 | * returns 0 on success, -1 on failure | |
1070 | */ | |
6df5a128 SG |
1071 | int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc, |
1072 | bool signal) | |
1cfa0a75 | 1073 | { |
f19624aa | 1074 | int ib_ret; |
1cfa0a75 | 1075 | struct ib_send_wr send_wr, *send_wr_failed; |
1cfa0a75 | 1076 | |
a4ee3539 | 1077 | ib_dma_sync_single_for_device(ib_conn->device->ib_device, |
5716af6e SG |
1078 | tx_desc->dma_addr, ISER_HEADERS_LEN, |
1079 | DMA_TO_DEVICE); | |
1cfa0a75 OG |
1080 | |
1081 | send_wr.next = NULL; | |
49df2781 | 1082 | send_wr.wr_id = (uintptr_t)tx_desc; |
f19624aa OG |
1083 | send_wr.sg_list = tx_desc->tx_sg; |
1084 | send_wr.num_sge = tx_desc->num_sge; | |
1cfa0a75 | 1085 | send_wr.opcode = IB_WR_SEND; |
6df5a128 | 1086 | send_wr.send_flags = signal ? IB_SEND_SIGNALED : 0; |
1cfa0a75 | 1087 | |
a4ee3539 | 1088 | ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed); |
ff3dd52d | 1089 | if (ib_ret) |
1cfa0a75 | 1090 | iser_err("ib_post_send failed, ret:%d\n", ib_ret); |
ff3dd52d | 1091 | |
f19624aa | 1092 | return ib_ret; |
1cfa0a75 OG |
1093 | } |
1094 | ||
6aabfa76 SG |
1095 | /** |
1096 | * is_iser_tx_desc - Indicate if the completion wr_id | |
1097 | * is a TX descriptor or not. | |
1098 | * @iser_conn: iser connection | |
1099 | * @wr_id: completion WR identifier | |
1100 | * | |
1101 | * Since we cannot rely on wc opcode in FLUSH errors | |
1102 | * we must work around it by checking if the wr_id address | |
1103 | * falls in the iser connection rx_descs buffer. If so | |
1104 | * it is an RX descriptor, otherwize it is a TX. | |
1105 | */ | |
1106 | static inline bool | |
1107 | is_iser_tx_desc(struct iser_conn *iser_conn, void *wr_id) | |
1108 | { | |
1109 | void *start = iser_conn->rx_descs; | |
1110 | int len = iser_conn->num_rx_descs * sizeof(*iser_conn->rx_descs); | |
1111 | ||
1112 | if (wr_id >= start && wr_id < start + len) | |
1113 | return false; | |
1114 | ||
1115 | return true; | |
1116 | } | |
1117 | ||
8c204e69 SG |
1118 | /** |
1119 | * iser_handle_comp_error() - Handle error completion | |
8c204e69 SG |
1120 | * @ib_conn: connection RDMA resources |
1121 | * @wc: work completion | |
1122 | * | |
1123 | * Notes: We may handle a FLUSH error completion and in this case | |
1124 | * we only cleanup in case TX type was DATAOUT. For non-FLUSH | |
1125 | * error completion we should also notify iscsi layer that | |
1126 | * connection is failed (in case we passed bind stage). | |
1127 | */ | |
1128 | static void | |
6aabfa76 | 1129 | iser_handle_comp_error(struct ib_conn *ib_conn, |
8c204e69 | 1130 | struct ib_wc *wc) |
1cfa0a75 | 1131 | { |
49df2781 | 1132 | void *wr_id = (void *)(uintptr_t)wc->wr_id; |
8c204e69 SG |
1133 | struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn, |
1134 | ib_conn); | |
1135 | ||
1136 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
1137 | if (iser_conn->iscsi_conn) | |
1138 | iscsi_conn_failure(iser_conn->iscsi_conn, | |
1139 | ISCSI_ERR_CONN_FAILED); | |
1140 | ||
30bf1d58 SG |
1141 | if (wc->wr_id == ISER_FASTREG_LI_WRID) |
1142 | return; | |
1143 | ||
49df2781 SG |
1144 | if (is_iser_tx_desc(iser_conn, wr_id)) { |
1145 | struct iser_tx_desc *desc = wr_id; | |
6aabfa76 | 1146 | |
6aabfa76 SG |
1147 | if (desc->type == ISCSI_TX_DATAOUT) |
1148 | kmem_cache_free(ig.desc_cache, desc); | |
1149 | } else { | |
1150 | ib_conn->post_recv_buf_count--; | |
1151 | } | |
1cfa0a75 OG |
1152 | } |
1153 | ||
6aabfa76 SG |
1154 | /** |
1155 | * iser_handle_wc - handle a single work completion | |
1156 | * @wc: work completion | |
1157 | * | |
1158 | * Soft-IRQ context, work completion can be either | |
1159 | * SEND or RECV, and can turn out successful or | |
1160 | * with error (or flush error). | |
1161 | */ | |
1162 | static void iser_handle_wc(struct ib_wc *wc) | |
78ad0a34 | 1163 | { |
a4ee3539 | 1164 | struct ib_conn *ib_conn; |
6aabfa76 SG |
1165 | struct iser_tx_desc *tx_desc; |
1166 | struct iser_rx_desc *rx_desc; | |
78ad0a34 | 1167 | |
6aabfa76 | 1168 | ib_conn = wc->qp->qp_context; |
06c7fb67 | 1169 | if (likely(wc->status == IB_WC_SUCCESS)) { |
6aabfa76 | 1170 | if (wc->opcode == IB_WC_RECV) { |
49df2781 | 1171 | rx_desc = (struct iser_rx_desc *)(uintptr_t)wc->wr_id; |
6aabfa76 SG |
1172 | iser_rcv_completion(rx_desc, wc->byte_len, |
1173 | ib_conn); | |
1174 | } else | |
1175 | if (wc->opcode == IB_WC_SEND) { | |
49df2781 | 1176 | tx_desc = (struct iser_tx_desc *)(uintptr_t)wc->wr_id; |
6aabfa76 | 1177 | iser_snd_completion(tx_desc, ib_conn); |
78ad0a34 | 1178 | } else { |
6aabfa76 | 1179 | iser_err("Unknown wc opcode %d\n", wc->opcode); |
78ad0a34 | 1180 | } |
6aabfa76 SG |
1181 | } else { |
1182 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
871e00af SG |
1183 | iser_err("%s (%d): wr id %llx vend_err %x\n", |
1184 | ib_wc_status_msg(wc->status), wc->status, | |
1185 | wc->wr_id, wc->vendor_err); | |
6aabfa76 | 1186 | else |
871e00af SG |
1187 | iser_dbg("%s (%d): wr id %llx\n", |
1188 | ib_wc_status_msg(wc->status), wc->status, | |
1189 | wc->wr_id); | |
6aabfa76 | 1190 | |
ff3dd52d | 1191 | if (wc->wr_id == ISER_BEACON_WRID) |
30bf1d58 | 1192 | /* all flush errors were consumed */ |
6aabfa76 | 1193 | complete(&ib_conn->flush_comp); |
30bf1d58 SG |
1194 | else |
1195 | iser_handle_comp_error(ib_conn, wc); | |
78ad0a34 | 1196 | } |
78ad0a34 OG |
1197 | } |
1198 | ||
6aabfa76 SG |
1199 | /** |
1200 | * iser_cq_tasklet_fn - iSER completion polling loop | |
1201 | * @data: iSER completion context | |
1202 | * | |
1203 | * Soft-IRQ context, polling connection CQ until | |
1204 | * either CQ was empty or we exausted polling budget | |
1205 | */ | |
1cfa0a75 OG |
1206 | static void iser_cq_tasklet_fn(unsigned long data) |
1207 | { | |
bf175540 | 1208 | struct iser_comp *comp = (struct iser_comp *)data; |
6aabfa76 | 1209 | struct ib_cq *cq = comp->cq; |
6e6fe2fb SG |
1210 | struct ib_wc *const wcs = comp->wcs; |
1211 | int i, n, completed = 0; | |
1cfa0a75 | 1212 | |
6e6fe2fb SG |
1213 | while ((n = ib_poll_cq(cq, ARRAY_SIZE(comp->wcs), wcs)) > 0) { |
1214 | for (i = 0; i < n; i++) | |
1215 | iser_handle_wc(&wcs[i]); | |
6aabfa76 | 1216 | |
6e6fe2fb SG |
1217 | completed += n; |
1218 | if (completed >= iser_cq_poll_limit) | |
183cfa43 | 1219 | break; |
1cfa0a75 | 1220 | } |
6aabfa76 SG |
1221 | |
1222 | /* | |
1223 | * It is assumed here that arming CQ only once its empty | |
1224 | * would not cause interrupts to be missed. | |
1225 | */ | |
1cfa0a75 | 1226 | ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); |
78ad0a34 | 1227 | |
6aabfa76 | 1228 | iser_dbg("got %d completions\n", completed); |
1cfa0a75 OG |
1229 | } |
1230 | ||
1231 | static void iser_cq_callback(struct ib_cq *cq, void *cq_context) | |
1232 | { | |
bf175540 | 1233 | struct iser_comp *comp = cq_context; |
1cfa0a75 | 1234 | |
bf175540 | 1235 | tasklet_schedule(&comp->tasklet); |
1cfa0a75 | 1236 | } |
0a7a08ad SG |
1237 | |
1238 | u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, | |
1239 | enum iser_data_dir cmd_dir, sector_t *sector) | |
1240 | { | |
b130eded | 1241 | struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir]; |
0a7a08ad SG |
1242 | struct fast_reg_descriptor *desc = reg->mem_h; |
1243 | unsigned long sector_size = iser_task->sc->device->sector_size; | |
1244 | struct ib_mr_status mr_status; | |
1245 | int ret; | |
1246 | ||
1247 | if (desc && desc->reg_indicators & ISER_FASTREG_PROTECTED) { | |
1248 | desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; | |
1249 | ret = ib_check_mr_status(desc->pi_ctx->sig_mr, | |
1250 | IB_MR_CHECK_SIG_STATUS, &mr_status); | |
1251 | if (ret) { | |
1252 | pr_err("ib_check_mr_status failed, ret %d\n", ret); | |
1253 | goto err; | |
1254 | } | |
1255 | ||
1256 | if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) { | |
1257 | sector_t sector_off = mr_status.sig_err.sig_err_offset; | |
1258 | ||
1259 | do_div(sector_off, sector_size + 8); | |
1260 | *sector = scsi_get_lba(iser_task->sc) + sector_off; | |
1261 | ||
39c978cd | 1262 | pr_err("PI error found type %d at sector %llx " |
0a7a08ad | 1263 | "expected %x vs actual %x\n", |
39c978cd RD |
1264 | mr_status.sig_err.err_type, |
1265 | (unsigned long long)*sector, | |
0a7a08ad SG |
1266 | mr_status.sig_err.expected, |
1267 | mr_status.sig_err.actual); | |
1268 | ||
1269 | switch (mr_status.sig_err.err_type) { | |
1270 | case IB_SIG_BAD_GUARD: | |
1271 | return 0x1; | |
1272 | case IB_SIG_BAD_REFTAG: | |
1273 | return 0x3; | |
1274 | case IB_SIG_BAD_APPTAG: | |
1275 | return 0x2; | |
1276 | } | |
1277 | } | |
1278 | } | |
1279 | ||
1280 | return 0; | |
1281 | err: | |
1282 | /* Not alot we can do here, return ambiguous guard error */ | |
1283 | return 0x1; | |
1284 | } |