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 | { | |
54 | iser_err("got cq event %d \n", cause->event); | |
55 | } | |
56 | ||
57 | static void iser_qp_event_callback(struct ib_event *cause, void *context) | |
58 | { | |
59 | iser_err("got qp event %d\n",cause->event); | |
60 | } | |
61 | ||
2110f9bf OG |
62 | static void iser_event_handler(struct ib_event_handler *handler, |
63 | struct ib_event *event) | |
64 | { | |
65 | iser_err("async event %d on device %s port %d\n", event->event, | |
66 | event->device->name, event->element.port_num); | |
67 | } | |
68 | ||
1cfa0a75 OG |
69 | /** |
70 | * iser_create_device_ib_res - creates Protection Domain (PD), Completion | |
71 | * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with | |
72 | * the adapator. | |
73 | * | |
74 | * returns 0 on success, -1 on failure | |
75 | */ | |
76 | static int iser_create_device_ib_res(struct iser_device *device) | |
77 | { | |
65198d6b | 78 | struct ib_device_attr *dev_attr = &device->dev_attr; |
f4641ef7 | 79 | int ret, i, max_cqe; |
5a33a669 | 80 | |
65198d6b SG |
81 | ret = ib_query_device(device->ib_device, dev_attr); |
82 | if (ret) { | |
5587856c | 83 | pr_warn("Query device failed for %s\n", device->ib_device->name); |
65198d6b | 84 | return ret; |
5587856c SG |
85 | } |
86 | ||
87 | /* Assign function handles - based on FMR support */ | |
88 | if (device->ib_device->alloc_fmr && device->ib_device->dealloc_fmr && | |
89 | device->ib_device->map_phys_fmr && device->ib_device->unmap_fmr) { | |
90 | iser_info("FMR supported, using FMR for registration\n"); | |
91 | device->iser_alloc_rdma_reg_res = iser_create_fmr_pool; | |
92 | device->iser_free_rdma_reg_res = iser_free_fmr_pool; | |
93 | device->iser_reg_rdma_mem = iser_reg_rdma_mem_fmr; | |
94 | device->iser_unreg_rdma_mem = iser_unreg_mem_fmr; | |
95 | } else | |
96 | if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { | |
7306b8fa SG |
97 | iser_info("FastReg supported, using FastReg for registration\n"); |
98 | device->iser_alloc_rdma_reg_res = iser_create_fastreg_pool; | |
99 | device->iser_free_rdma_reg_res = iser_free_fastreg_pool; | |
100 | device->iser_reg_rdma_mem = iser_reg_rdma_mem_fastreg; | |
101 | device->iser_unreg_rdma_mem = iser_unreg_mem_fastreg; | |
5587856c | 102 | } else { |
7306b8fa | 103 | iser_err("IB device does not support FMRs nor FastRegs, can't register memory\n"); |
65198d6b | 104 | return -1; |
5587856c | 105 | } |
b4e155ff | 106 | |
da64bdb2 | 107 | device->comps_used = min_t(int, num_online_cpus(), |
bf175540 | 108 | device->ib_device->num_comp_vectors); |
f4641ef7 | 109 | |
da64bdb2 SG |
110 | device->comps = kcalloc(device->comps_used, sizeof(*device->comps), |
111 | GFP_KERNEL); | |
112 | if (!device->comps) | |
113 | goto comps_err; | |
114 | ||
f4641ef7 MT |
115 | max_cqe = min(ISER_MAX_CQ_LEN, dev_attr->max_cqe); |
116 | ||
117 | iser_info("using %d CQs, device %s supports %d vectors max_cqe %d\n", | |
bf175540 | 118 | device->comps_used, device->ib_device->name, |
f4641ef7 | 119 | device->ib_device->num_comp_vectors, max_cqe); |
5a33a669 | 120 | |
1cfa0a75 OG |
121 | device->pd = ib_alloc_pd(device->ib_device); |
122 | if (IS_ERR(device->pd)) | |
123 | goto pd_err; | |
124 | ||
bf175540 SG |
125 | for (i = 0; i < device->comps_used; i++) { |
126 | struct iser_comp *comp = &device->comps[i]; | |
127 | ||
128 | comp->device = device; | |
6aabfa76 SG |
129 | comp->cq = ib_create_cq(device->ib_device, |
130 | iser_cq_callback, | |
131 | iser_cq_event_callback, | |
132 | (void *)comp, | |
f4641ef7 | 133 | max_cqe, i); |
6aabfa76 SG |
134 | if (IS_ERR(comp->cq)) { |
135 | comp->cq = NULL; | |
5a33a669 | 136 | goto cq_err; |
c33b15f0 | 137 | } |
1cfa0a75 | 138 | |
6aabfa76 | 139 | if (ib_req_notify_cq(comp->cq, IB_CQ_NEXT_COMP)) |
5a33a669 | 140 | goto cq_err; |
1cfa0a75 | 141 | |
bf175540 SG |
142 | tasklet_init(&comp->tasklet, iser_cq_tasklet_fn, |
143 | (unsigned long)comp); | |
5a33a669 | 144 | } |
1cfa0a75 | 145 | |
d8111028 EZ |
146 | device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE | |
147 | IB_ACCESS_REMOTE_WRITE | | |
148 | IB_ACCESS_REMOTE_READ); | |
1cfa0a75 OG |
149 | if (IS_ERR(device->mr)) |
150 | goto dma_mr_err; | |
151 | ||
2110f9bf OG |
152 | INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device, |
153 | iser_event_handler); | |
154 | if (ib_register_event_handler(&device->event_handler)) | |
155 | goto handler_err; | |
156 | ||
1cfa0a75 OG |
157 | return 0; |
158 | ||
2110f9bf OG |
159 | handler_err: |
160 | ib_dereg_mr(device->mr); | |
1cfa0a75 | 161 | dma_mr_err: |
bf175540 SG |
162 | for (i = 0; i < device->comps_used; i++) |
163 | tasklet_kill(&device->comps[i].tasklet); | |
5a33a669 | 164 | cq_err: |
bf175540 SG |
165 | for (i = 0; i < device->comps_used; i++) { |
166 | struct iser_comp *comp = &device->comps[i]; | |
167 | ||
6aabfa76 SG |
168 | if (comp->cq) |
169 | ib_destroy_cq(comp->cq); | |
5a33a669 | 170 | } |
1cfa0a75 OG |
171 | ib_dealloc_pd(device->pd); |
172 | pd_err: | |
da64bdb2 SG |
173 | kfree(device->comps); |
174 | comps_err: | |
1cfa0a75 OG |
175 | iser_err("failed to allocate an IB resource\n"); |
176 | return -1; | |
177 | } | |
178 | ||
179 | /** | |
38dc732f | 180 | * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR, |
1cfa0a75 OG |
181 | * CQ and PD created with the device associated with the adapator. |
182 | */ | |
183 | static void iser_free_device_ib_res(struct iser_device *device) | |
184 | { | |
5a33a669 | 185 | int i; |
1cfa0a75 OG |
186 | BUG_ON(device->mr == NULL); |
187 | ||
bf175540 SG |
188 | for (i = 0; i < device->comps_used; i++) { |
189 | struct iser_comp *comp = &device->comps[i]; | |
190 | ||
191 | tasklet_kill(&comp->tasklet); | |
6aabfa76 SG |
192 | ib_destroy_cq(comp->cq); |
193 | comp->cq = NULL; | |
5a33a669 AT |
194 | } |
195 | ||
2110f9bf | 196 | (void)ib_unregister_event_handler(&device->event_handler); |
1cfa0a75 | 197 | (void)ib_dereg_mr(device->mr); |
1cfa0a75 OG |
198 | (void)ib_dealloc_pd(device->pd); |
199 | ||
da64bdb2 SG |
200 | kfree(device->comps); |
201 | device->comps = NULL; | |
202 | ||
1cfa0a75 | 203 | device->mr = NULL; |
1cfa0a75 OG |
204 | device->pd = NULL; |
205 | } | |
206 | ||
207 | /** | |
986db0d6 | 208 | * iser_create_fmr_pool - Creates FMR pool and page_vector |
1cfa0a75 | 209 | * |
986db0d6 | 210 | * returns 0 on success, or errno code on failure |
1cfa0a75 | 211 | */ |
a4ee3539 | 212 | int iser_create_fmr_pool(struct ib_conn *ib_conn, unsigned cmds_max) |
1cfa0a75 | 213 | { |
a4ee3539 | 214 | struct iser_device *device = ib_conn->device; |
1cfa0a75 | 215 | struct ib_fmr_pool_param params; |
986db0d6 | 216 | int ret = -ENOMEM; |
bcc60c38 | 217 | |
a4ee3539 | 218 | ib_conn->fmr.page_vec = kmalloc(sizeof(*ib_conn->fmr.page_vec) + |
7306b8fa SG |
219 | (sizeof(u64)*(ISCSI_ISER_SG_TABLESIZE + 1)), |
220 | GFP_KERNEL); | |
a4ee3539 | 221 | if (!ib_conn->fmr.page_vec) |
986db0d6 | 222 | return ret; |
9fda1ac5 | 223 | |
a4ee3539 | 224 | ib_conn->fmr.page_vec->pages = (u64 *)(ib_conn->fmr.page_vec + 1); |
1cfa0a75 | 225 | |
8dfa0876 | 226 | params.page_shift = SHIFT_4K; |
1cfa0a75 OG |
227 | /* when the first/last SG element are not start/end * |
228 | * page aligned, the map whould be of N+1 pages */ | |
229 | params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1; | |
230 | /* make the pool size twice the max number of SCSI commands * | |
231 | * the ML is expected to queue, watermark for unmap at 50% */ | |
b7f04513 SP |
232 | params.pool_size = cmds_max * 2; |
233 | params.dirty_watermark = cmds_max; | |
1cfa0a75 OG |
234 | params.cache = 0; |
235 | params.flush_function = NULL; | |
236 | params.access = (IB_ACCESS_LOCAL_WRITE | | |
237 | IB_ACCESS_REMOTE_WRITE | | |
238 | IB_ACCESS_REMOTE_READ); | |
239 | ||
a4ee3539 SG |
240 | ib_conn->fmr.pool = ib_create_fmr_pool(device->pd, ¶ms); |
241 | if (!IS_ERR(ib_conn->fmr.pool)) | |
986db0d6 SP |
242 | return 0; |
243 | ||
244 | /* no FMR => no need for page_vec */ | |
a4ee3539 SG |
245 | kfree(ib_conn->fmr.page_vec); |
246 | ib_conn->fmr.page_vec = NULL; | |
986db0d6 | 247 | |
a4ee3539 SG |
248 | ret = PTR_ERR(ib_conn->fmr.pool); |
249 | ib_conn->fmr.pool = NULL; | |
986db0d6 SP |
250 | if (ret != -ENOSYS) { |
251 | iser_err("FMR allocation failed, err %d\n", ret); | |
252 | return ret; | |
253 | } else { | |
5525d210 | 254 | iser_warn("FMRs are not supported, using unaligned mode\n"); |
986db0d6 | 255 | return 0; |
1cfa0a75 | 256 | } |
986db0d6 SP |
257 | } |
258 | ||
259 | /** | |
260 | * iser_free_fmr_pool - releases the FMR pool and page vec | |
261 | */ | |
a4ee3539 | 262 | void iser_free_fmr_pool(struct ib_conn *ib_conn) |
986db0d6 SP |
263 | { |
264 | iser_info("freeing conn %p fmr pool %p\n", | |
a4ee3539 | 265 | ib_conn, ib_conn->fmr.pool); |
986db0d6 | 266 | |
a4ee3539 SG |
267 | if (ib_conn->fmr.pool != NULL) |
268 | ib_destroy_fmr_pool(ib_conn->fmr.pool); | |
986db0d6 | 269 | |
a4ee3539 | 270 | ib_conn->fmr.pool = NULL; |
986db0d6 | 271 | |
a4ee3539 SG |
272 | kfree(ib_conn->fmr.page_vec); |
273 | ib_conn->fmr.page_vec = NULL; | |
986db0d6 SP |
274 | } |
275 | ||
310b347c SG |
276 | static int |
277 | iser_create_fastreg_desc(struct ib_device *ib_device, struct ib_pd *pd, | |
6b5a8fb0 | 278 | bool pi_enable, struct fast_reg_descriptor *desc) |
310b347c SG |
279 | { |
280 | int ret; | |
281 | ||
282 | desc->data_frpl = ib_alloc_fast_reg_page_list(ib_device, | |
283 | ISCSI_ISER_SG_TABLESIZE + 1); | |
284 | if (IS_ERR(desc->data_frpl)) { | |
285 | ret = PTR_ERR(desc->data_frpl); | |
286 | iser_err("Failed to allocate ib_fast_reg_page_list err=%d\n", | |
287 | ret); | |
288 | return PTR_ERR(desc->data_frpl); | |
289 | } | |
290 | ||
291 | desc->data_mr = ib_alloc_fast_reg_mr(pd, ISCSI_ISER_SG_TABLESIZE + 1); | |
292 | if (IS_ERR(desc->data_mr)) { | |
293 | ret = PTR_ERR(desc->data_mr); | |
294 | iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret); | |
295 | goto fast_reg_mr_failure; | |
296 | } | |
6b5a8fb0 AT |
297 | desc->reg_indicators |= ISER_DATA_KEY_VALID; |
298 | ||
299 | if (pi_enable) { | |
300 | struct ib_mr_init_attr mr_init_attr = {0}; | |
301 | struct iser_pi_context *pi_ctx = NULL; | |
302 | ||
303 | desc->pi_ctx = kzalloc(sizeof(*desc->pi_ctx), GFP_KERNEL); | |
304 | if (!desc->pi_ctx) { | |
305 | iser_err("Failed to allocate pi context\n"); | |
306 | ret = -ENOMEM; | |
307 | goto pi_ctx_alloc_failure; | |
308 | } | |
309 | pi_ctx = desc->pi_ctx; | |
310 | ||
311 | pi_ctx->prot_frpl = ib_alloc_fast_reg_page_list(ib_device, | |
312 | ISCSI_ISER_SG_TABLESIZE); | |
313 | if (IS_ERR(pi_ctx->prot_frpl)) { | |
314 | ret = PTR_ERR(pi_ctx->prot_frpl); | |
315 | iser_err("Failed to allocate prot frpl ret=%d\n", | |
316 | ret); | |
317 | goto prot_frpl_failure; | |
318 | } | |
319 | ||
320 | pi_ctx->prot_mr = ib_alloc_fast_reg_mr(pd, | |
321 | ISCSI_ISER_SG_TABLESIZE + 1); | |
322 | if (IS_ERR(pi_ctx->prot_mr)) { | |
323 | ret = PTR_ERR(pi_ctx->prot_mr); | |
324 | iser_err("Failed to allocate prot frmr ret=%d\n", | |
325 | ret); | |
326 | goto prot_mr_failure; | |
327 | } | |
328 | desc->reg_indicators |= ISER_PROT_KEY_VALID; | |
329 | ||
330 | mr_init_attr.max_reg_descriptors = 2; | |
331 | mr_init_attr.flags |= IB_MR_SIGNATURE_EN; | |
332 | pi_ctx->sig_mr = ib_create_mr(pd, &mr_init_attr); | |
333 | if (IS_ERR(pi_ctx->sig_mr)) { | |
334 | ret = PTR_ERR(pi_ctx->sig_mr); | |
335 | iser_err("Failed to allocate signature enabled mr err=%d\n", | |
336 | ret); | |
337 | goto sig_mr_failure; | |
338 | } | |
339 | desc->reg_indicators |= ISER_SIG_KEY_VALID; | |
340 | } | |
341 | desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; | |
342 | ||
4f9208ad OG |
343 | iser_dbg("Create fr_desc %p page_list %p\n", |
344 | desc, desc->data_frpl->page_list); | |
310b347c SG |
345 | |
346 | return 0; | |
6b5a8fb0 AT |
347 | sig_mr_failure: |
348 | ib_dereg_mr(desc->pi_ctx->prot_mr); | |
349 | prot_mr_failure: | |
350 | ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); | |
351 | prot_frpl_failure: | |
352 | kfree(desc->pi_ctx); | |
353 | pi_ctx_alloc_failure: | |
354 | ib_dereg_mr(desc->data_mr); | |
310b347c SG |
355 | fast_reg_mr_failure: |
356 | ib_free_fast_reg_page_list(desc->data_frpl); | |
357 | ||
358 | return ret; | |
359 | } | |
360 | ||
5587856c | 361 | /** |
7306b8fa | 362 | * iser_create_fastreg_pool - Creates pool of fast_reg descriptors |
5587856c SG |
363 | * for fast registration work requests. |
364 | * returns 0 on success, or errno code on failure | |
365 | */ | |
a4ee3539 | 366 | int iser_create_fastreg_pool(struct ib_conn *ib_conn, unsigned cmds_max) |
5587856c | 367 | { |
a4ee3539 SG |
368 | struct iser_device *device = ib_conn->device; |
369 | struct fast_reg_descriptor *desc; | |
5587856c SG |
370 | int i, ret; |
371 | ||
a4ee3539 SG |
372 | INIT_LIST_HEAD(&ib_conn->fastreg.pool); |
373 | ib_conn->fastreg.pool_size = 0; | |
5587856c | 374 | for (i = 0; i < cmds_max; i++) { |
6b5a8fb0 | 375 | desc = kzalloc(sizeof(*desc), GFP_KERNEL); |
5587856c SG |
376 | if (!desc) { |
377 | iser_err("Failed to allocate a new fast_reg descriptor\n"); | |
378 | ret = -ENOMEM; | |
379 | goto err; | |
380 | } | |
381 | ||
6b5a8fb0 | 382 | ret = iser_create_fastreg_desc(device->ib_device, device->pd, |
a4ee3539 | 383 | ib_conn->pi_support, desc); |
310b347c SG |
384 | if (ret) { |
385 | iser_err("Failed to create fastreg descriptor err=%d\n", | |
386 | ret); | |
387 | kfree(desc); | |
388 | goto err; | |
5587856c SG |
389 | } |
390 | ||
a4ee3539 SG |
391 | list_add_tail(&desc->list, &ib_conn->fastreg.pool); |
392 | ib_conn->fastreg.pool_size++; | |
5587856c SG |
393 | } |
394 | ||
395 | return 0; | |
27ae2d1e | 396 | |
5587856c | 397 | err: |
a4ee3539 | 398 | iser_free_fastreg_pool(ib_conn); |
5587856c SG |
399 | return ret; |
400 | } | |
401 | ||
402 | /** | |
7306b8fa | 403 | * iser_free_fastreg_pool - releases the pool of fast_reg descriptors |
5587856c | 404 | */ |
a4ee3539 | 405 | void iser_free_fastreg_pool(struct ib_conn *ib_conn) |
5587856c SG |
406 | { |
407 | struct fast_reg_descriptor *desc, *tmp; | |
408 | int i = 0; | |
409 | ||
a4ee3539 | 410 | if (list_empty(&ib_conn->fastreg.pool)) |
5587856c SG |
411 | return; |
412 | ||
a4ee3539 | 413 | iser_info("freeing conn %p fr pool\n", ib_conn); |
5587856c | 414 | |
a4ee3539 | 415 | list_for_each_entry_safe(desc, tmp, &ib_conn->fastreg.pool, list) { |
5587856c SG |
416 | list_del(&desc->list); |
417 | ib_free_fast_reg_page_list(desc->data_frpl); | |
418 | ib_dereg_mr(desc->data_mr); | |
6b5a8fb0 AT |
419 | if (desc->pi_ctx) { |
420 | ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); | |
421 | ib_dereg_mr(desc->pi_ctx->prot_mr); | |
422 | ib_destroy_mr(desc->pi_ctx->sig_mr); | |
423 | kfree(desc->pi_ctx); | |
424 | } | |
5587856c SG |
425 | kfree(desc); |
426 | ++i; | |
427 | } | |
428 | ||
a4ee3539 | 429 | if (i < ib_conn->fastreg.pool_size) |
5587856c | 430 | iser_warn("pool still has %d regions registered\n", |
a4ee3539 | 431 | ib_conn->fastreg.pool_size - i); |
5587856c SG |
432 | } |
433 | ||
986db0d6 SP |
434 | /** |
435 | * iser_create_ib_conn_res - Queue-Pair (QP) | |
436 | * | |
437 | * returns 0 on success, -1 on failure | |
438 | */ | |
a4ee3539 | 439 | static int iser_create_ib_conn_res(struct ib_conn *ib_conn) |
986db0d6 | 440 | { |
f4641ef7 MT |
441 | struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn, |
442 | ib_conn); | |
986db0d6 | 443 | struct iser_device *device; |
f4641ef7 | 444 | struct ib_device_attr *dev_attr; |
986db0d6 SP |
445 | struct ib_qp_init_attr init_attr; |
446 | int ret = -ENOMEM; | |
447 | int index, min_index = 0; | |
448 | ||
a4ee3539 | 449 | BUG_ON(ib_conn->device == NULL); |
986db0d6 | 450 | |
a4ee3539 | 451 | device = ib_conn->device; |
f4641ef7 | 452 | dev_attr = &device->dev_attr; |
1cfa0a75 OG |
453 | |
454 | memset(&init_attr, 0, sizeof init_attr); | |
455 | ||
5a33a669 AT |
456 | mutex_lock(&ig.connlist_mutex); |
457 | /* select the CQ with the minimal number of usages */ | |
bf175540 SG |
458 | for (index = 0; index < device->comps_used; index++) { |
459 | if (device->comps[index].active_qps < | |
460 | device->comps[min_index].active_qps) | |
5a33a669 | 461 | min_index = index; |
bf175540 SG |
462 | } |
463 | ib_conn->comp = &device->comps[min_index]; | |
464 | ib_conn->comp->active_qps++; | |
5a33a669 | 465 | mutex_unlock(&ig.connlist_mutex); |
a4ee3539 | 466 | iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn); |
5a33a669 | 467 | |
1cfa0a75 | 468 | init_attr.event_handler = iser_qp_event_callback; |
a4ee3539 | 469 | init_attr.qp_context = (void *)ib_conn; |
6aabfa76 SG |
470 | init_attr.send_cq = ib_conn->comp->cq; |
471 | init_attr.recv_cq = ib_conn->comp->cq; | |
1cfa0a75 | 472 | init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS; |
f19624aa | 473 | init_attr.cap.max_send_sge = 2; |
bcc60c38 | 474 | init_attr.cap.max_recv_sge = 1; |
1cfa0a75 OG |
475 | init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; |
476 | init_attr.qp_type = IB_QPT_RC; | |
a4ee3539 | 477 | if (ib_conn->pi_support) { |
ff3dd52d | 478 | init_attr.cap.max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1; |
6b5a8fb0 | 479 | init_attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN; |
f4641ef7 MT |
480 | iser_conn->max_cmds = |
481 | ISER_GET_MAX_XMIT_CMDS(ISER_QP_SIG_MAX_REQ_DTOS); | |
6b5a8fb0 | 482 | } else { |
f4641ef7 MT |
483 | if (dev_attr->max_qp_wr > ISER_QP_MAX_REQ_DTOS) { |
484 | init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS + 1; | |
485 | iser_conn->max_cmds = | |
486 | ISER_GET_MAX_XMIT_CMDS(ISER_QP_MAX_REQ_DTOS); | |
487 | } else { | |
488 | init_attr.cap.max_send_wr = dev_attr->max_qp_wr; | |
489 | iser_conn->max_cmds = | |
490 | ISER_GET_MAX_XMIT_CMDS(dev_attr->max_qp_wr); | |
491 | iser_dbg("device %s supports max_send_wr %d\n", | |
492 | device->ib_device->name, dev_attr->max_qp_wr); | |
493 | } | |
6b5a8fb0 | 494 | } |
1cfa0a75 | 495 | |
a4ee3539 | 496 | ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); |
1cfa0a75 | 497 | if (ret) |
9fda1ac5 | 498 | goto out_err; |
1cfa0a75 | 499 | |
a4ee3539 | 500 | ib_conn->qp = ib_conn->cma_id->qp; |
986db0d6 | 501 | iser_info("setting conn %p cma_id %p qp %p\n", |
a4ee3539 SG |
502 | ib_conn, ib_conn->cma_id, |
503 | ib_conn->cma_id->qp); | |
1cfa0a75 OG |
504 | return ret; |
505 | ||
9fda1ac5 | 506 | out_err: |
93acb7bb SG |
507 | mutex_lock(&ig.connlist_mutex); |
508 | ib_conn->comp->active_qps--; | |
509 | mutex_unlock(&ig.connlist_mutex); | |
1cfa0a75 | 510 | iser_err("unable to alloc mem or create resource, err %d\n", ret); |
93acb7bb | 511 | |
1cfa0a75 OG |
512 | return ret; |
513 | } | |
514 | ||
1cfa0a75 OG |
515 | /** |
516 | * based on the resolved device node GUID see if there already allocated | |
517 | * device for this device. If there's no such, create one. | |
518 | */ | |
519 | static | |
520 | struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id) | |
521 | { | |
9a378270 | 522 | struct iser_device *device; |
1cfa0a75 OG |
523 | |
524 | mutex_lock(&ig.device_list_mutex); | |
525 | ||
9a378270 | 526 | list_for_each_entry(device, &ig.device_list, ig_list) |
1cfa0a75 OG |
527 | /* find if there's a match using the node GUID */ |
528 | if (device->ib_device->node_guid == cma_id->device->node_guid) | |
d33ed425 | 529 | goto inc_refcnt; |
9a378270 AR |
530 | |
531 | device = kzalloc(sizeof *device, GFP_KERNEL); | |
532 | if (device == NULL) | |
533 | goto out; | |
534 | ||
535 | /* assign this device to the device */ | |
536 | device->ib_device = cma_id->device; | |
537 | /* init the device and link it into ig device list */ | |
538 | if (iser_create_device_ib_res(device)) { | |
539 | kfree(device); | |
540 | device = NULL; | |
541 | goto out; | |
1cfa0a75 | 542 | } |
9a378270 AR |
543 | list_add(&device->ig_list, &ig.device_list); |
544 | ||
d33ed425 | 545 | inc_refcnt: |
1cfa0a75 | 546 | device->refcount++; |
d33ed425 | 547 | out: |
1cfa0a75 OG |
548 | mutex_unlock(&ig.device_list_mutex); |
549 | return device; | |
550 | } | |
551 | ||
552 | /* if there's no demand for this device, release it */ | |
553 | static void iser_device_try_release(struct iser_device *device) | |
554 | { | |
555 | mutex_lock(&ig.device_list_mutex); | |
556 | device->refcount--; | |
4f363882 | 557 | iser_info("device %p refcount %d\n", device, device->refcount); |
1cfa0a75 OG |
558 | if (!device->refcount) { |
559 | iser_free_device_ib_res(device); | |
560 | list_del(&device->ig_list); | |
561 | kfree(device); | |
562 | } | |
563 | mutex_unlock(&ig.device_list_mutex); | |
564 | } | |
565 | ||
504130c0 AN |
566 | /** |
567 | * Called with state mutex held | |
568 | **/ | |
5716af6e SG |
569 | static int iser_conn_state_comp_exch(struct iser_conn *iser_conn, |
570 | enum iser_conn_state comp, | |
571 | enum iser_conn_state exch) | |
1cfa0a75 OG |
572 | { |
573 | int ret; | |
574 | ||
5716af6e SG |
575 | ret = (iser_conn->state == comp); |
576 | if (ret) | |
577 | iser_conn->state = exch; | |
578 | ||
1cfa0a75 OG |
579 | return ret; |
580 | } | |
581 | ||
b73c3ada AN |
582 | void iser_release_work(struct work_struct *work) |
583 | { | |
5716af6e | 584 | struct iser_conn *iser_conn; |
b73c3ada | 585 | |
5716af6e | 586 | iser_conn = container_of(work, struct iser_conn, release_work); |
b73c3ada | 587 | |
c107a6c0 SG |
588 | /* Wait for conn_stop to complete */ |
589 | wait_for_completion(&iser_conn->stop_completion); | |
590 | /* Wait for IB resouces cleanup to complete */ | |
591 | wait_for_completion(&iser_conn->ib_completion); | |
b73c3ada | 592 | |
5716af6e SG |
593 | mutex_lock(&iser_conn->state_mutex); |
594 | iser_conn->state = ISER_CONN_DOWN; | |
595 | mutex_unlock(&iser_conn->state_mutex); | |
504130c0 | 596 | |
5716af6e | 597 | iser_conn_release(iser_conn); |
b73c3ada AN |
598 | } |
599 | ||
96f15198 SG |
600 | /** |
601 | * iser_free_ib_conn_res - release IB related resources | |
602 | * @iser_conn: iser connection struct | |
6606e6a2 SG |
603 | * @destroy: indicator if we need to try to release the |
604 | * iser device and memory regoins pool (only iscsi | |
605 | * shutdown and DEVICE_REMOVAL will use this). | |
96f15198 SG |
606 | * |
607 | * This routine is called with the iser state mutex held | |
608 | * so the cm_id removal is out of here. It is Safe to | |
609 | * be invoked multiple times. | |
610 | */ | |
c47a3c9e | 611 | static void iser_free_ib_conn_res(struct iser_conn *iser_conn, |
6606e6a2 | 612 | bool destroy) |
96f15198 SG |
613 | { |
614 | struct ib_conn *ib_conn = &iser_conn->ib_conn; | |
615 | struct iser_device *device = ib_conn->device; | |
616 | ||
617 | iser_info("freeing conn %p cma_id %p qp %p\n", | |
618 | iser_conn, ib_conn->cma_id, ib_conn->qp); | |
619 | ||
96f15198 | 620 | if (ib_conn->qp != NULL) { |
bf175540 | 621 | ib_conn->comp->active_qps--; |
96f15198 SG |
622 | rdma_destroy_qp(ib_conn->cma_id); |
623 | ib_conn->qp = NULL; | |
624 | } | |
625 | ||
6606e6a2 SG |
626 | if (destroy) { |
627 | if (iser_conn->rx_descs) | |
628 | iser_free_rx_descriptors(iser_conn); | |
629 | ||
630 | if (device != NULL) { | |
631 | iser_device_try_release(device); | |
632 | ib_conn->device = NULL; | |
633 | } | |
96f15198 SG |
634 | } |
635 | } | |
636 | ||
41179e2d RD |
637 | /** |
638 | * Frees all conn objects and deallocs conn descriptor | |
639 | */ | |
5716af6e | 640 | void iser_conn_release(struct iser_conn *iser_conn) |
41179e2d | 641 | { |
a4ee3539 | 642 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
41179e2d | 643 | |
41179e2d | 644 | mutex_lock(&ig.connlist_mutex); |
5716af6e | 645 | list_del(&iser_conn->conn_list); |
41179e2d | 646 | mutex_unlock(&ig.connlist_mutex); |
504130c0 | 647 | |
5716af6e | 648 | mutex_lock(&iser_conn->state_mutex); |
9a3119e4 | 649 | /* In case we endup here without ep_disconnect being invoked. */ |
5426b171 | 650 | if (iser_conn->state != ISER_CONN_DOWN) { |
aea8f4df AN |
651 | iser_warn("iser conn %p state %d, expected state down.\n", |
652 | iser_conn, iser_conn->state); | |
9a3119e4 | 653 | iscsi_destroy_endpoint(iser_conn->ep); |
5426b171 AN |
654 | iser_conn->state = ISER_CONN_DOWN; |
655 | } | |
c47a3c9e SG |
656 | /* |
657 | * In case we never got to bind stage, we still need to | |
658 | * release IB resources (which is safe to call more than once). | |
659 | */ | |
660 | iser_free_ib_conn_res(iser_conn, true); | |
5716af6e | 661 | mutex_unlock(&iser_conn->state_mutex); |
504130c0 | 662 | |
a4ee3539 SG |
663 | if (ib_conn->cma_id != NULL) { |
664 | rdma_destroy_id(ib_conn->cma_id); | |
665 | ib_conn->cma_id = NULL; | |
5b61ff43 | 666 | } |
a4ee3539 | 667 | |
5716af6e | 668 | kfree(iser_conn); |
41179e2d RD |
669 | } |
670 | ||
1cfa0a75 OG |
671 | /** |
672 | * triggers start of the disconnect procedures and wait for them to be done | |
c47a3c9e | 673 | * Called with state mutex held |
1cfa0a75 | 674 | */ |
c47a3c9e | 675 | int iser_conn_terminate(struct iser_conn *iser_conn) |
1cfa0a75 | 676 | { |
a4ee3539 | 677 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
ff3dd52d | 678 | struct ib_send_wr *bad_wr; |
1cfa0a75 OG |
679 | int err = 0; |
680 | ||
c47a3c9e SG |
681 | /* terminate the iser conn only if the conn state is UP */ |
682 | if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP, | |
683 | ISER_CONN_TERMINATING)) | |
684 | return 0; | |
685 | ||
686 | iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state); | |
687 | ||
688 | /* suspend queuing of new iscsi commands */ | |
689 | if (iser_conn->iscsi_conn) | |
690 | iscsi_suspend_queue(iser_conn->iscsi_conn); | |
691 | ||
692 | /* | |
693 | * In case we didn't already clean up the cma_id (peer initiated | |
694 | * a disconnection), we need to Cause the CMA to change the QP | |
695 | * state to ERROR. | |
1cfa0a75 | 696 | */ |
c47a3c9e SG |
697 | if (ib_conn->cma_id) { |
698 | err = rdma_disconnect(ib_conn->cma_id); | |
699 | if (err) | |
700 | iser_err("Failed to disconnect, conn: 0x%p err %d\n", | |
701 | iser_conn, err); | |
702 | ||
ff3dd52d SG |
703 | /* post an indication that all flush errors were consumed */ |
704 | err = ib_post_send(ib_conn->qp, &ib_conn->beacon, &bad_wr); | |
16df2a26 | 705 | if (err) { |
ff3dd52d | 706 | iser_err("conn %p failed to post beacon", ib_conn); |
16df2a26 SG |
707 | return 1; |
708 | } | |
ff3dd52d | 709 | |
6aabfa76 | 710 | wait_for_completion(&ib_conn->flush_comp); |
c47a3c9e | 711 | } |
1cfa0a75 | 712 | |
c47a3c9e | 713 | return 1; |
1cfa0a75 OG |
714 | } |
715 | ||
504130c0 AN |
716 | /** |
717 | * Called with state mutex held | |
718 | **/ | |
b73c3ada | 719 | static void iser_connect_error(struct rdma_cm_id *cma_id) |
1cfa0a75 | 720 | { |
5716af6e | 721 | struct iser_conn *iser_conn; |
b73c3ada | 722 | |
5716af6e | 723 | iser_conn = (struct iser_conn *)cma_id->context; |
c4de4663 | 724 | iser_conn->state = ISER_CONN_TERMINATING; |
1cfa0a75 OG |
725 | } |
726 | ||
504130c0 AN |
727 | /** |
728 | * Called with state mutex held | |
729 | **/ | |
b73c3ada | 730 | static void iser_addr_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
731 | { |
732 | struct iser_device *device; | |
5716af6e | 733 | struct iser_conn *iser_conn; |
a4ee3539 | 734 | struct ib_conn *ib_conn; |
1cfa0a75 OG |
735 | int ret; |
736 | ||
5716af6e SG |
737 | iser_conn = (struct iser_conn *)cma_id->context; |
738 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
739 | /* bailout */ |
740 | return; | |
741 | ||
a4ee3539 | 742 | ib_conn = &iser_conn->ib_conn; |
1cfa0a75 | 743 | device = iser_device_find_by_ib_device(cma_id); |
d33ed425 AR |
744 | if (!device) { |
745 | iser_err("device lookup/creation failed\n"); | |
b73c3ada AN |
746 | iser_connect_error(cma_id); |
747 | return; | |
d33ed425 AR |
748 | } |
749 | ||
a4ee3539 | 750 | ib_conn->device = device; |
1cfa0a75 | 751 | |
7f733847 AT |
752 | /* connection T10-PI support */ |
753 | if (iser_pi_enable) { | |
754 | if (!(device->dev_attr.device_cap_flags & | |
755 | IB_DEVICE_SIGNATURE_HANDOVER)) { | |
756 | iser_warn("T10-PI requested but not supported on %s, " | |
757 | "continue without T10-PI\n", | |
a4ee3539 SG |
758 | ib_conn->device->ib_device->name); |
759 | ib_conn->pi_support = false; | |
7f733847 | 760 | } else { |
a4ee3539 | 761 | ib_conn->pi_support = true; |
7f733847 AT |
762 | } |
763 | } | |
764 | ||
1cfa0a75 OG |
765 | ret = rdma_resolve_route(cma_id, 1000); |
766 | if (ret) { | |
767 | iser_err("resolve route failed: %d\n", ret); | |
b73c3ada AN |
768 | iser_connect_error(cma_id); |
769 | return; | |
1cfa0a75 | 770 | } |
1cfa0a75 OG |
771 | } |
772 | ||
504130c0 AN |
773 | /** |
774 | * Called with state mutex held | |
775 | **/ | |
b73c3ada | 776 | static void iser_route_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
777 | { |
778 | struct rdma_conn_param conn_param; | |
779 | int ret; | |
8d8399de | 780 | struct iser_cm_hdr req_hdr; |
5716af6e | 781 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
a4ee3539 SG |
782 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
783 | struct iser_device *device = ib_conn->device; | |
1cfa0a75 | 784 | |
5716af6e | 785 | if (iser_conn->state != ISER_CONN_PENDING) |
504130c0 AN |
786 | /* bailout */ |
787 | return; | |
788 | ||
a4ee3539 | 789 | ret = iser_create_ib_conn_res(ib_conn); |
1cfa0a75 OG |
790 | if (ret) |
791 | goto failure; | |
792 | ||
1cfa0a75 | 793 | memset(&conn_param, 0, sizeof conn_param); |
2ea32938 | 794 | conn_param.responder_resources = device->dev_attr.max_qp_rd_atom; |
1cfa0a75 OG |
795 | conn_param.initiator_depth = 1; |
796 | conn_param.retry_count = 7; | |
797 | conn_param.rnr_retry_count = 6; | |
798 | ||
8d8399de OG |
799 | memset(&req_hdr, 0, sizeof(req_hdr)); |
800 | req_hdr.flags = (ISER_ZBVA_NOT_SUPPORTED | | |
801 | ISER_SEND_W_INV_NOT_SUPPORTED); | |
802 | conn_param.private_data = (void *)&req_hdr; | |
803 | conn_param.private_data_len = sizeof(struct iser_cm_hdr); | |
804 | ||
1cfa0a75 OG |
805 | ret = rdma_connect(cma_id, &conn_param); |
806 | if (ret) { | |
807 | iser_err("failure connecting: %d\n", ret); | |
808 | goto failure; | |
809 | } | |
810 | ||
b73c3ada | 811 | return; |
1cfa0a75 | 812 | failure: |
b73c3ada | 813 | iser_connect_error(cma_id); |
1cfa0a75 OG |
814 | } |
815 | ||
816 | static void iser_connected_handler(struct rdma_cm_id *cma_id) | |
817 | { | |
5716af6e | 818 | struct iser_conn *iser_conn; |
4f9208ad OG |
819 | struct ib_qp_attr attr; |
820 | struct ib_qp_init_attr init_attr; | |
821 | ||
5716af6e SG |
822 | iser_conn = (struct iser_conn *)cma_id->context; |
823 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
824 | /* bailout */ |
825 | return; | |
826 | ||
4f9208ad OG |
827 | (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr); |
828 | iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num); | |
1cfa0a75 | 829 | |
5716af6e SG |
830 | iser_conn->state = ISER_CONN_UP; |
831 | complete(&iser_conn->up_completion); | |
1cfa0a75 OG |
832 | } |
833 | ||
b73c3ada | 834 | static void iser_disconnected_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 | 835 | { |
c47a3c9e | 836 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
1cfa0a75 | 837 | |
c47a3c9e | 838 | if (iser_conn_terminate(iser_conn)) { |
5716af6e | 839 | if (iser_conn->iscsi_conn) |
c47a3c9e SG |
840 | iscsi_conn_failure(iser_conn->iscsi_conn, |
841 | ISCSI_ERR_CONN_FAILED); | |
7d9eacf9 RD |
842 | else |
843 | iser_err("iscsi_iser connection isn't bound\n"); | |
844 | } | |
c47a3c9e SG |
845 | } |
846 | ||
847 | static void iser_cleanup_handler(struct rdma_cm_id *cma_id, | |
6606e6a2 | 848 | bool destroy) |
c47a3c9e SG |
849 | { |
850 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; | |
1cfa0a75 | 851 | |
c47a3c9e SG |
852 | /* |
853 | * We are not guaranteed that we visited disconnected_handler | |
854 | * by now, call it here to be safe that we handle CM drep | |
855 | * and flush errors. | |
8d4aca7f | 856 | */ |
c47a3c9e | 857 | iser_disconnected_handler(cma_id); |
6606e6a2 | 858 | iser_free_ib_conn_res(iser_conn, destroy); |
c47a3c9e SG |
859 | complete(&iser_conn->ib_completion); |
860 | }; | |
1cfa0a75 OG |
861 | |
862 | static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) | |
863 | { | |
5716af6e | 864 | struct iser_conn *iser_conn; |
c47a3c9e | 865 | int ret = 0; |
504130c0 | 866 | |
5716af6e | 867 | iser_conn = (struct iser_conn *)cma_id->context; |
4f363882 RD |
868 | iser_info("event %d status %d conn %p id %p\n", |
869 | event->event, event->status, cma_id->context, cma_id); | |
1cfa0a75 | 870 | |
5716af6e | 871 | mutex_lock(&iser_conn->state_mutex); |
1cfa0a75 OG |
872 | switch (event->event) { |
873 | case RDMA_CM_EVENT_ADDR_RESOLVED: | |
b73c3ada | 874 | iser_addr_handler(cma_id); |
1cfa0a75 OG |
875 | break; |
876 | case RDMA_CM_EVENT_ROUTE_RESOLVED: | |
b73c3ada | 877 | iser_route_handler(cma_id); |
1cfa0a75 OG |
878 | break; |
879 | case RDMA_CM_EVENT_ESTABLISHED: | |
880 | iser_connected_handler(cma_id); | |
881 | break; | |
882 | case RDMA_CM_EVENT_ADDR_ERROR: | |
883 | case RDMA_CM_EVENT_ROUTE_ERROR: | |
884 | case RDMA_CM_EVENT_CONNECT_ERROR: | |
885 | case RDMA_CM_EVENT_UNREACHABLE: | |
886 | case RDMA_CM_EVENT_REJECTED: | |
b73c3ada | 887 | iser_connect_error(cma_id); |
1cfa0a75 OG |
888 | break; |
889 | case RDMA_CM_EVENT_DISCONNECTED: | |
2f5de151 | 890 | case RDMA_CM_EVENT_ADDR_CHANGE: |
5426b171 AN |
891 | case RDMA_CM_EVENT_TIMEWAIT_EXIT: |
892 | iser_cleanup_handler(cma_id, false); | |
1cfa0a75 | 893 | break; |
c47a3c9e SG |
894 | case RDMA_CM_EVENT_DEVICE_REMOVAL: |
895 | /* | |
896 | * we *must* destroy the device as we cannot rely | |
897 | * on iscsid to be around to initiate error handling. | |
5426b171 AN |
898 | * also if we are not in state DOWN implicitly destroy |
899 | * the cma_id. | |
c47a3c9e SG |
900 | */ |
901 | iser_cleanup_handler(cma_id, true); | |
5426b171 AN |
902 | if (iser_conn->state != ISER_CONN_DOWN) { |
903 | iser_conn->ib_conn.cma_id = NULL; | |
904 | ret = 1; | |
905 | } | |
c47a3c9e | 906 | break; |
1cfa0a75 | 907 | default: |
a4ef1451 | 908 | iser_err("Unexpected RDMA CM event (%d)\n", event->event); |
1cfa0a75 OG |
909 | break; |
910 | } | |
5716af6e | 911 | mutex_unlock(&iser_conn->state_mutex); |
c47a3c9e SG |
912 | |
913 | return ret; | |
1cfa0a75 OG |
914 | } |
915 | ||
5716af6e | 916 | void iser_conn_init(struct iser_conn *iser_conn) |
1cfa0a75 | 917 | { |
5716af6e | 918 | iser_conn->state = ISER_CONN_INIT; |
a4ee3539 | 919 | iser_conn->ib_conn.post_recv_buf_count = 0; |
6aabfa76 | 920 | init_completion(&iser_conn->ib_conn.flush_comp); |
5716af6e | 921 | init_completion(&iser_conn->stop_completion); |
c47a3c9e | 922 | init_completion(&iser_conn->ib_completion); |
5716af6e SG |
923 | init_completion(&iser_conn->up_completion); |
924 | INIT_LIST_HEAD(&iser_conn->conn_list); | |
a4ee3539 | 925 | spin_lock_init(&iser_conn->ib_conn.lock); |
5716af6e | 926 | mutex_init(&iser_conn->state_mutex); |
1cfa0a75 OG |
927 | } |
928 | ||
929 | /** | |
930 | * starts the process of connecting to the target | |
94e2bd68 | 931 | * sleeps until the connection is established or rejected |
1cfa0a75 | 932 | */ |
5716af6e | 933 | int iser_connect(struct iser_conn *iser_conn, |
96ed02d4 RD |
934 | struct sockaddr *src_addr, |
935 | struct sockaddr *dst_addr, | |
1cfa0a75 OG |
936 | int non_blocking) |
937 | { | |
a4ee3539 | 938 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
1cfa0a75 OG |
939 | int err = 0; |
940 | ||
5716af6e | 941 | mutex_lock(&iser_conn->state_mutex); |
504130c0 | 942 | |
5716af6e | 943 | sprintf(iser_conn->name, "%pISp", dst_addr); |
96ed02d4 | 944 | |
5716af6e | 945 | iser_info("connecting to: %s\n", iser_conn->name); |
1cfa0a75 OG |
946 | |
947 | /* the device is known only --after-- address resolution */ | |
a4ee3539 | 948 | ib_conn->device = NULL; |
1cfa0a75 | 949 | |
5716af6e | 950 | iser_conn->state = ISER_CONN_PENDING; |
1cfa0a75 | 951 | |
ff3dd52d SG |
952 | ib_conn->beacon.wr_id = ISER_BEACON_WRID; |
953 | ib_conn->beacon.opcode = IB_WR_SEND; | |
954 | ||
a4ee3539 SG |
955 | ib_conn->cma_id = rdma_create_id(iser_cma_handler, |
956 | (void *)iser_conn, | |
957 | RDMA_PS_TCP, IB_QPT_RC); | |
958 | if (IS_ERR(ib_conn->cma_id)) { | |
959 | err = PTR_ERR(ib_conn->cma_id); | |
1cfa0a75 OG |
960 | iser_err("rdma_create_id failed: %d\n", err); |
961 | goto id_failure; | |
962 | } | |
963 | ||
a4ee3539 | 964 | err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000); |
1cfa0a75 OG |
965 | if (err) { |
966 | iser_err("rdma_resolve_addr failed: %d\n", err); | |
967 | goto addr_failure; | |
968 | } | |
969 | ||
970 | if (!non_blocking) { | |
5716af6e | 971 | wait_for_completion_interruptible(&iser_conn->up_completion); |
1cfa0a75 | 972 | |
5716af6e | 973 | if (iser_conn->state != ISER_CONN_UP) { |
1cfa0a75 OG |
974 | err = -EIO; |
975 | goto connect_failure; | |
976 | } | |
977 | } | |
5716af6e | 978 | mutex_unlock(&iser_conn->state_mutex); |
1cfa0a75 OG |
979 | |
980 | mutex_lock(&ig.connlist_mutex); | |
5716af6e | 981 | list_add(&iser_conn->conn_list, &ig.connlist); |
1cfa0a75 OG |
982 | mutex_unlock(&ig.connlist_mutex); |
983 | return 0; | |
984 | ||
985 | id_failure: | |
a4ee3539 | 986 | ib_conn->cma_id = NULL; |
1cfa0a75 | 987 | addr_failure: |
5716af6e | 988 | iser_conn->state = ISER_CONN_DOWN; |
1cfa0a75 | 989 | connect_failure: |
5716af6e SG |
990 | mutex_unlock(&iser_conn->state_mutex); |
991 | iser_conn_release(iser_conn); | |
1cfa0a75 OG |
992 | return err; |
993 | } | |
994 | ||
5716af6e | 995 | int iser_post_recvl(struct iser_conn *iser_conn) |
bcc60c38 OG |
996 | { |
997 | struct ib_recv_wr rx_wr, *rx_wr_failed; | |
a4ee3539 | 998 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
bcc60c38 OG |
999 | struct ib_sge sge; |
1000 | int ib_ret; | |
1001 | ||
5716af6e | 1002 | sge.addr = iser_conn->login_resp_dma; |
bcc60c38 | 1003 | sge.length = ISER_RX_LOGIN_SIZE; |
a4ee3539 | 1004 | sge.lkey = ib_conn->device->mr->lkey; |
bcc60c38 | 1005 | |
49df2781 | 1006 | rx_wr.wr_id = (uintptr_t)iser_conn->login_resp_buf; |
bcc60c38 OG |
1007 | rx_wr.sg_list = &sge; |
1008 | rx_wr.num_sge = 1; | |
1009 | rx_wr.next = NULL; | |
1010 | ||
a4ee3539 SG |
1011 | ib_conn->post_recv_buf_count++; |
1012 | ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1013 | if (ib_ret) { |
1014 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1015 | ib_conn->post_recv_buf_count--; |
bcc60c38 OG |
1016 | } |
1017 | return ib_ret; | |
1018 | } | |
1019 | ||
5716af6e | 1020 | int iser_post_recvm(struct iser_conn *iser_conn, int count) |
bcc60c38 OG |
1021 | { |
1022 | struct ib_recv_wr *rx_wr, *rx_wr_failed; | |
1023 | int i, ib_ret; | |
a4ee3539 | 1024 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
5716af6e | 1025 | unsigned int my_rx_head = iser_conn->rx_desc_head; |
bcc60c38 OG |
1026 | struct iser_rx_desc *rx_desc; |
1027 | ||
a4ee3539 | 1028 | for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { |
5716af6e | 1029 | rx_desc = &iser_conn->rx_descs[my_rx_head]; |
49df2781 | 1030 | rx_wr->wr_id = (uintptr_t)rx_desc; |
bcc60c38 OG |
1031 | rx_wr->sg_list = &rx_desc->rx_sg; |
1032 | rx_wr->num_sge = 1; | |
1033 | rx_wr->next = rx_wr + 1; | |
5716af6e | 1034 | my_rx_head = (my_rx_head + 1) & iser_conn->qp_max_recv_dtos_mask; |
bcc60c38 OG |
1035 | } |
1036 | ||
1037 | rx_wr--; | |
1038 | rx_wr->next = NULL; /* mark end of work requests list */ | |
1039 | ||
a4ee3539 SG |
1040 | ib_conn->post_recv_buf_count += count; |
1041 | ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1042 | if (ib_ret) { |
1043 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1044 | ib_conn->post_recv_buf_count -= count; |
bcc60c38 | 1045 | } else |
5716af6e | 1046 | iser_conn->rx_desc_head = my_rx_head; |
bcc60c38 OG |
1047 | return ib_ret; |
1048 | } | |
1049 | ||
1050 | ||
1cfa0a75 OG |
1051 | /** |
1052 | * iser_start_send - Initiate a Send DTO operation | |
1053 | * | |
1054 | * returns 0 on success, -1 on failure | |
1055 | */ | |
6df5a128 SG |
1056 | int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc, |
1057 | bool signal) | |
1cfa0a75 | 1058 | { |
f19624aa | 1059 | int ib_ret; |
1cfa0a75 | 1060 | struct ib_send_wr send_wr, *send_wr_failed; |
1cfa0a75 | 1061 | |
a4ee3539 | 1062 | ib_dma_sync_single_for_device(ib_conn->device->ib_device, |
5716af6e SG |
1063 | tx_desc->dma_addr, ISER_HEADERS_LEN, |
1064 | DMA_TO_DEVICE); | |
1cfa0a75 OG |
1065 | |
1066 | send_wr.next = NULL; | |
49df2781 | 1067 | send_wr.wr_id = (uintptr_t)tx_desc; |
f19624aa OG |
1068 | send_wr.sg_list = tx_desc->tx_sg; |
1069 | send_wr.num_sge = tx_desc->num_sge; | |
1cfa0a75 | 1070 | send_wr.opcode = IB_WR_SEND; |
6df5a128 | 1071 | send_wr.send_flags = signal ? IB_SEND_SIGNALED : 0; |
1cfa0a75 | 1072 | |
a4ee3539 | 1073 | ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed); |
ff3dd52d | 1074 | if (ib_ret) |
1cfa0a75 | 1075 | iser_err("ib_post_send failed, ret:%d\n", ib_ret); |
ff3dd52d | 1076 | |
f19624aa | 1077 | return ib_ret; |
1cfa0a75 OG |
1078 | } |
1079 | ||
6aabfa76 SG |
1080 | /** |
1081 | * is_iser_tx_desc - Indicate if the completion wr_id | |
1082 | * is a TX descriptor or not. | |
1083 | * @iser_conn: iser connection | |
1084 | * @wr_id: completion WR identifier | |
1085 | * | |
1086 | * Since we cannot rely on wc opcode in FLUSH errors | |
1087 | * we must work around it by checking if the wr_id address | |
1088 | * falls in the iser connection rx_descs buffer. If so | |
1089 | * it is an RX descriptor, otherwize it is a TX. | |
1090 | */ | |
1091 | static inline bool | |
1092 | is_iser_tx_desc(struct iser_conn *iser_conn, void *wr_id) | |
1093 | { | |
1094 | void *start = iser_conn->rx_descs; | |
1095 | int len = iser_conn->num_rx_descs * sizeof(*iser_conn->rx_descs); | |
1096 | ||
1097 | if (wr_id >= start && wr_id < start + len) | |
1098 | return false; | |
1099 | ||
1100 | return true; | |
1101 | } | |
1102 | ||
8c204e69 SG |
1103 | /** |
1104 | * iser_handle_comp_error() - Handle error completion | |
8c204e69 SG |
1105 | * @ib_conn: connection RDMA resources |
1106 | * @wc: work completion | |
1107 | * | |
1108 | * Notes: We may handle a FLUSH error completion and in this case | |
1109 | * we only cleanup in case TX type was DATAOUT. For non-FLUSH | |
1110 | * error completion we should also notify iscsi layer that | |
1111 | * connection is failed (in case we passed bind stage). | |
1112 | */ | |
1113 | static void | |
6aabfa76 | 1114 | iser_handle_comp_error(struct ib_conn *ib_conn, |
8c204e69 | 1115 | struct ib_wc *wc) |
1cfa0a75 | 1116 | { |
49df2781 | 1117 | void *wr_id = (void *)(uintptr_t)wc->wr_id; |
8c204e69 SG |
1118 | struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn, |
1119 | ib_conn); | |
1120 | ||
1121 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
1122 | if (iser_conn->iscsi_conn) | |
1123 | iscsi_conn_failure(iser_conn->iscsi_conn, | |
1124 | ISCSI_ERR_CONN_FAILED); | |
1125 | ||
30bf1d58 SG |
1126 | if (wc->wr_id == ISER_FASTREG_LI_WRID) |
1127 | return; | |
1128 | ||
49df2781 SG |
1129 | if (is_iser_tx_desc(iser_conn, wr_id)) { |
1130 | struct iser_tx_desc *desc = wr_id; | |
6aabfa76 | 1131 | |
6aabfa76 SG |
1132 | if (desc->type == ISCSI_TX_DATAOUT) |
1133 | kmem_cache_free(ig.desc_cache, desc); | |
1134 | } else { | |
1135 | ib_conn->post_recv_buf_count--; | |
1136 | } | |
1cfa0a75 OG |
1137 | } |
1138 | ||
6aabfa76 SG |
1139 | /** |
1140 | * iser_handle_wc - handle a single work completion | |
1141 | * @wc: work completion | |
1142 | * | |
1143 | * Soft-IRQ context, work completion can be either | |
1144 | * SEND or RECV, and can turn out successful or | |
1145 | * with error (or flush error). | |
1146 | */ | |
1147 | static void iser_handle_wc(struct ib_wc *wc) | |
78ad0a34 | 1148 | { |
a4ee3539 | 1149 | struct ib_conn *ib_conn; |
6aabfa76 SG |
1150 | struct iser_tx_desc *tx_desc; |
1151 | struct iser_rx_desc *rx_desc; | |
78ad0a34 | 1152 | |
6aabfa76 | 1153 | ib_conn = wc->qp->qp_context; |
06c7fb67 | 1154 | if (likely(wc->status == IB_WC_SUCCESS)) { |
6aabfa76 | 1155 | if (wc->opcode == IB_WC_RECV) { |
49df2781 | 1156 | rx_desc = (struct iser_rx_desc *)(uintptr_t)wc->wr_id; |
6aabfa76 SG |
1157 | iser_rcv_completion(rx_desc, wc->byte_len, |
1158 | ib_conn); | |
1159 | } else | |
1160 | if (wc->opcode == IB_WC_SEND) { | |
49df2781 | 1161 | tx_desc = (struct iser_tx_desc *)(uintptr_t)wc->wr_id; |
6aabfa76 | 1162 | iser_snd_completion(tx_desc, ib_conn); |
78ad0a34 | 1163 | } else { |
6aabfa76 | 1164 | iser_err("Unknown wc opcode %d\n", wc->opcode); |
78ad0a34 | 1165 | } |
6aabfa76 SG |
1166 | } else { |
1167 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
1168 | iser_err("wr id %llx status %d vend_err %x\n", | |
1169 | wc->wr_id, wc->status, wc->vendor_err); | |
1170 | else | |
1171 | iser_dbg("flush error: wr id %llx\n", wc->wr_id); | |
1172 | ||
ff3dd52d | 1173 | if (wc->wr_id == ISER_BEACON_WRID) |
30bf1d58 | 1174 | /* all flush errors were consumed */ |
6aabfa76 | 1175 | complete(&ib_conn->flush_comp); |
30bf1d58 SG |
1176 | else |
1177 | iser_handle_comp_error(ib_conn, wc); | |
78ad0a34 | 1178 | } |
78ad0a34 OG |
1179 | } |
1180 | ||
6aabfa76 SG |
1181 | /** |
1182 | * iser_cq_tasklet_fn - iSER completion polling loop | |
1183 | * @data: iSER completion context | |
1184 | * | |
1185 | * Soft-IRQ context, polling connection CQ until | |
1186 | * either CQ was empty or we exausted polling budget | |
1187 | */ | |
1cfa0a75 OG |
1188 | static void iser_cq_tasklet_fn(unsigned long data) |
1189 | { | |
bf175540 | 1190 | struct iser_comp *comp = (struct iser_comp *)data; |
6aabfa76 | 1191 | struct ib_cq *cq = comp->cq; |
6e6fe2fb SG |
1192 | struct ib_wc *const wcs = comp->wcs; |
1193 | int i, n, completed = 0; | |
1cfa0a75 | 1194 | |
6e6fe2fb SG |
1195 | while ((n = ib_poll_cq(cq, ARRAY_SIZE(comp->wcs), wcs)) > 0) { |
1196 | for (i = 0; i < n; i++) | |
1197 | iser_handle_wc(&wcs[i]); | |
6aabfa76 | 1198 | |
6e6fe2fb SG |
1199 | completed += n; |
1200 | if (completed >= iser_cq_poll_limit) | |
183cfa43 | 1201 | break; |
1cfa0a75 | 1202 | } |
6aabfa76 SG |
1203 | |
1204 | /* | |
1205 | * It is assumed here that arming CQ only once its empty | |
1206 | * would not cause interrupts to be missed. | |
1207 | */ | |
1cfa0a75 | 1208 | ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); |
78ad0a34 | 1209 | |
6aabfa76 | 1210 | iser_dbg("got %d completions\n", completed); |
1cfa0a75 OG |
1211 | } |
1212 | ||
1213 | static void iser_cq_callback(struct ib_cq *cq, void *cq_context) | |
1214 | { | |
bf175540 | 1215 | struct iser_comp *comp = cq_context; |
1cfa0a75 | 1216 | |
bf175540 | 1217 | tasklet_schedule(&comp->tasklet); |
1cfa0a75 | 1218 | } |
0a7a08ad SG |
1219 | |
1220 | u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, | |
1221 | enum iser_data_dir cmd_dir, sector_t *sector) | |
1222 | { | |
1223 | struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; | |
1224 | struct fast_reg_descriptor *desc = reg->mem_h; | |
1225 | unsigned long sector_size = iser_task->sc->device->sector_size; | |
1226 | struct ib_mr_status mr_status; | |
1227 | int ret; | |
1228 | ||
1229 | if (desc && desc->reg_indicators & ISER_FASTREG_PROTECTED) { | |
1230 | desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; | |
1231 | ret = ib_check_mr_status(desc->pi_ctx->sig_mr, | |
1232 | IB_MR_CHECK_SIG_STATUS, &mr_status); | |
1233 | if (ret) { | |
1234 | pr_err("ib_check_mr_status failed, ret %d\n", ret); | |
1235 | goto err; | |
1236 | } | |
1237 | ||
1238 | if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) { | |
1239 | sector_t sector_off = mr_status.sig_err.sig_err_offset; | |
1240 | ||
1241 | do_div(sector_off, sector_size + 8); | |
1242 | *sector = scsi_get_lba(iser_task->sc) + sector_off; | |
1243 | ||
39c978cd | 1244 | pr_err("PI error found type %d at sector %llx " |
0a7a08ad | 1245 | "expected %x vs actual %x\n", |
39c978cd RD |
1246 | mr_status.sig_err.err_type, |
1247 | (unsigned long long)*sector, | |
0a7a08ad SG |
1248 | mr_status.sig_err.expected, |
1249 | mr_status.sig_err.actual); | |
1250 | ||
1251 | switch (mr_status.sig_err.err_type) { | |
1252 | case IB_SIG_BAD_GUARD: | |
1253 | return 0x1; | |
1254 | case IB_SIG_BAD_REFTAG: | |
1255 | return 0x3; | |
1256 | case IB_SIG_BAD_APPTAG: | |
1257 | return 0x2; | |
1258 | } | |
1259 | } | |
1260 | } | |
1261 | ||
1262 | return 0; | |
1263 | err: | |
1264 | /* Not alot we can do here, return ambiguous guard error */ | |
1265 | return 0x1; | |
1266 | } |