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 | |
78ad0a34 OG |
42 | #define ISER_MAX_RX_CQ_LEN (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN) |
43 | #define ISER_MAX_TX_CQ_LEN (ISER_QP_MAX_REQ_DTOS * ISCSI_ISER_MAX_CONN) | |
1cfa0a75 | 44 | |
183cfa43 SG |
45 | static int iser_cq_poll_limit = 512; |
46 | ||
1cfa0a75 OG |
47 | static void iser_cq_tasklet_fn(unsigned long data); |
48 | static void iser_cq_callback(struct ib_cq *cq, void *cq_context); | |
bf175540 | 49 | static int iser_drain_tx_cq(struct iser_comp *comp); |
1cfa0a75 OG |
50 | |
51 | static void iser_cq_event_callback(struct ib_event *cause, void *context) | |
52 | { | |
53 | iser_err("got cq event %d \n", cause->event); | |
54 | } | |
55 | ||
56 | static void iser_qp_event_callback(struct ib_event *cause, void *context) | |
57 | { | |
58 | iser_err("got qp event %d\n",cause->event); | |
59 | } | |
60 | ||
2110f9bf OG |
61 | static void iser_event_handler(struct ib_event_handler *handler, |
62 | struct ib_event *event) | |
63 | { | |
64 | iser_err("async event %d on device %s port %d\n", event->event, | |
65 | event->device->name, event->element.port_num); | |
66 | } | |
67 | ||
1cfa0a75 OG |
68 | /** |
69 | * iser_create_device_ib_res - creates Protection Domain (PD), Completion | |
70 | * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with | |
71 | * the adapator. | |
72 | * | |
73 | * returns 0 on success, -1 on failure | |
74 | */ | |
75 | static int iser_create_device_ib_res(struct iser_device *device) | |
76 | { | |
65198d6b | 77 | struct ib_device_attr *dev_attr = &device->dev_attr; |
c33b15f0 | 78 | int ret, i; |
5a33a669 | 79 | |
65198d6b SG |
80 | ret = ib_query_device(device->ib_device, dev_attr); |
81 | if (ret) { | |
5587856c | 82 | pr_warn("Query device failed for %s\n", device->ib_device->name); |
65198d6b | 83 | return ret; |
5587856c SG |
84 | } |
85 | ||
86 | /* Assign function handles - based on FMR support */ | |
87 | if (device->ib_device->alloc_fmr && device->ib_device->dealloc_fmr && | |
88 | device->ib_device->map_phys_fmr && device->ib_device->unmap_fmr) { | |
89 | iser_info("FMR supported, using FMR for registration\n"); | |
90 | device->iser_alloc_rdma_reg_res = iser_create_fmr_pool; | |
91 | device->iser_free_rdma_reg_res = iser_free_fmr_pool; | |
92 | device->iser_reg_rdma_mem = iser_reg_rdma_mem_fmr; | |
93 | device->iser_unreg_rdma_mem = iser_unreg_mem_fmr; | |
94 | } else | |
95 | if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { | |
7306b8fa SG |
96 | iser_info("FastReg supported, using FastReg for registration\n"); |
97 | device->iser_alloc_rdma_reg_res = iser_create_fastreg_pool; | |
98 | device->iser_free_rdma_reg_res = iser_free_fastreg_pool; | |
99 | device->iser_reg_rdma_mem = iser_reg_rdma_mem_fastreg; | |
100 | device->iser_unreg_rdma_mem = iser_unreg_mem_fastreg; | |
5587856c | 101 | } else { |
7306b8fa | 102 | iser_err("IB device does not support FMRs nor FastRegs, can't register memory\n"); |
65198d6b | 103 | return -1; |
5587856c | 104 | } |
b4e155ff | 105 | |
bf175540 SG |
106 | device->comps_used = min(ISER_MAX_CQ, |
107 | device->ib_device->num_comp_vectors); | |
4f363882 | 108 | iser_info("using %d CQs, device %s supports %d vectors\n", |
bf175540 | 109 | device->comps_used, device->ib_device->name, |
4f363882 | 110 | device->ib_device->num_comp_vectors); |
5a33a669 | 111 | |
1cfa0a75 OG |
112 | device->pd = ib_alloc_pd(device->ib_device); |
113 | if (IS_ERR(device->pd)) | |
114 | goto pd_err; | |
115 | ||
bf175540 SG |
116 | for (i = 0; i < device->comps_used; i++) { |
117 | struct iser_comp *comp = &device->comps[i]; | |
118 | ||
119 | comp->device = device; | |
120 | comp->rx_cq = ib_create_cq(device->ib_device, | |
121 | iser_cq_callback, | |
122 | iser_cq_event_callback, | |
123 | (void *)comp, | |
124 | ISER_MAX_RX_CQ_LEN, i); | |
125 | if (IS_ERR(comp->rx_cq)) { | |
126 | comp->rx_cq = NULL; | |
5a33a669 | 127 | goto cq_err; |
c33b15f0 | 128 | } |
1cfa0a75 | 129 | |
bf175540 SG |
130 | comp->tx_cq = ib_create_cq(device->ib_device, NULL, |
131 | iser_cq_event_callback, | |
132 | (void *)comp, | |
133 | ISER_MAX_TX_CQ_LEN, i); | |
134 | if (IS_ERR(comp->tx_cq)) { | |
135 | comp->tx_cq = NULL; | |
5a33a669 | 136 | goto cq_err; |
c33b15f0 | 137 | } |
78ad0a34 | 138 | |
bf175540 | 139 | if (ib_req_notify_cq(comp->rx_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 | ||
168 | if (comp->tx_cq) | |
169 | ib_destroy_cq(comp->tx_cq); | |
170 | if (comp->rx_cq) | |
171 | ib_destroy_cq(comp->rx_cq); | |
5a33a669 | 172 | } |
1cfa0a75 OG |
173 | ib_dealloc_pd(device->pd); |
174 | pd_err: | |
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); | |
192 | ib_destroy_cq(comp->tx_cq); | |
193 | ib_destroy_cq(comp->rx_cq); | |
194 | comp->tx_cq = NULL; | |
195 | comp->rx_cq = NULL; | |
5a33a669 AT |
196 | } |
197 | ||
2110f9bf | 198 | (void)ib_unregister_event_handler(&device->event_handler); |
1cfa0a75 | 199 | (void)ib_dereg_mr(device->mr); |
1cfa0a75 OG |
200 | (void)ib_dealloc_pd(device->pd); |
201 | ||
202 | device->mr = NULL; | |
1cfa0a75 OG |
203 | device->pd = NULL; |
204 | } | |
205 | ||
206 | /** | |
986db0d6 | 207 | * iser_create_fmr_pool - Creates FMR pool and page_vector |
1cfa0a75 | 208 | * |
986db0d6 | 209 | * returns 0 on success, or errno code on failure |
1cfa0a75 | 210 | */ |
a4ee3539 | 211 | int iser_create_fmr_pool(struct ib_conn *ib_conn, unsigned cmds_max) |
1cfa0a75 | 212 | { |
a4ee3539 | 213 | struct iser_device *device = ib_conn->device; |
1cfa0a75 | 214 | struct ib_fmr_pool_param params; |
986db0d6 | 215 | int ret = -ENOMEM; |
bcc60c38 | 216 | |
a4ee3539 | 217 | ib_conn->fmr.page_vec = kmalloc(sizeof(*ib_conn->fmr.page_vec) + |
7306b8fa SG |
218 | (sizeof(u64)*(ISCSI_ISER_SG_TABLESIZE + 1)), |
219 | GFP_KERNEL); | |
a4ee3539 | 220 | if (!ib_conn->fmr.page_vec) |
986db0d6 | 221 | return ret; |
9fda1ac5 | 222 | |
a4ee3539 | 223 | ib_conn->fmr.page_vec->pages = (u64 *)(ib_conn->fmr.page_vec + 1); |
1cfa0a75 | 224 | |
8dfa0876 | 225 | params.page_shift = SHIFT_4K; |
1cfa0a75 OG |
226 | /* when the first/last SG element are not start/end * |
227 | * page aligned, the map whould be of N+1 pages */ | |
228 | params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1; | |
229 | /* make the pool size twice the max number of SCSI commands * | |
230 | * the ML is expected to queue, watermark for unmap at 50% */ | |
b7f04513 SP |
231 | params.pool_size = cmds_max * 2; |
232 | params.dirty_watermark = cmds_max; | |
1cfa0a75 OG |
233 | params.cache = 0; |
234 | params.flush_function = NULL; | |
235 | params.access = (IB_ACCESS_LOCAL_WRITE | | |
236 | IB_ACCESS_REMOTE_WRITE | | |
237 | IB_ACCESS_REMOTE_READ); | |
238 | ||
a4ee3539 SG |
239 | ib_conn->fmr.pool = ib_create_fmr_pool(device->pd, ¶ms); |
240 | if (!IS_ERR(ib_conn->fmr.pool)) | |
986db0d6 SP |
241 | return 0; |
242 | ||
243 | /* no FMR => no need for page_vec */ | |
a4ee3539 SG |
244 | kfree(ib_conn->fmr.page_vec); |
245 | ib_conn->fmr.page_vec = NULL; | |
986db0d6 | 246 | |
a4ee3539 SG |
247 | ret = PTR_ERR(ib_conn->fmr.pool); |
248 | ib_conn->fmr.pool = NULL; | |
986db0d6 SP |
249 | if (ret != -ENOSYS) { |
250 | iser_err("FMR allocation failed, err %d\n", ret); | |
251 | return ret; | |
252 | } else { | |
5525d210 | 253 | iser_warn("FMRs are not supported, using unaligned mode\n"); |
986db0d6 | 254 | return 0; |
1cfa0a75 | 255 | } |
986db0d6 SP |
256 | } |
257 | ||
258 | /** | |
259 | * iser_free_fmr_pool - releases the FMR pool and page vec | |
260 | */ | |
a4ee3539 | 261 | void iser_free_fmr_pool(struct ib_conn *ib_conn) |
986db0d6 SP |
262 | { |
263 | iser_info("freeing conn %p fmr pool %p\n", | |
a4ee3539 | 264 | ib_conn, ib_conn->fmr.pool); |
986db0d6 | 265 | |
a4ee3539 SG |
266 | if (ib_conn->fmr.pool != NULL) |
267 | ib_destroy_fmr_pool(ib_conn->fmr.pool); | |
986db0d6 | 268 | |
a4ee3539 | 269 | ib_conn->fmr.pool = NULL; |
986db0d6 | 270 | |
a4ee3539 SG |
271 | kfree(ib_conn->fmr.page_vec); |
272 | ib_conn->fmr.page_vec = NULL; | |
986db0d6 SP |
273 | } |
274 | ||
310b347c SG |
275 | static int |
276 | iser_create_fastreg_desc(struct ib_device *ib_device, struct ib_pd *pd, | |
6b5a8fb0 | 277 | bool pi_enable, struct fast_reg_descriptor *desc) |
310b347c SG |
278 | { |
279 | int ret; | |
280 | ||
281 | desc->data_frpl = ib_alloc_fast_reg_page_list(ib_device, | |
282 | ISCSI_ISER_SG_TABLESIZE + 1); | |
283 | if (IS_ERR(desc->data_frpl)) { | |
284 | ret = PTR_ERR(desc->data_frpl); | |
285 | iser_err("Failed to allocate ib_fast_reg_page_list err=%d\n", | |
286 | ret); | |
287 | return PTR_ERR(desc->data_frpl); | |
288 | } | |
289 | ||
290 | desc->data_mr = ib_alloc_fast_reg_mr(pd, ISCSI_ISER_SG_TABLESIZE + 1); | |
291 | if (IS_ERR(desc->data_mr)) { | |
292 | ret = PTR_ERR(desc->data_mr); | |
293 | iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret); | |
294 | goto fast_reg_mr_failure; | |
295 | } | |
6b5a8fb0 AT |
296 | desc->reg_indicators |= ISER_DATA_KEY_VALID; |
297 | ||
298 | if (pi_enable) { | |
299 | struct ib_mr_init_attr mr_init_attr = {0}; | |
300 | struct iser_pi_context *pi_ctx = NULL; | |
301 | ||
302 | desc->pi_ctx = kzalloc(sizeof(*desc->pi_ctx), GFP_KERNEL); | |
303 | if (!desc->pi_ctx) { | |
304 | iser_err("Failed to allocate pi context\n"); | |
305 | ret = -ENOMEM; | |
306 | goto pi_ctx_alloc_failure; | |
307 | } | |
308 | pi_ctx = desc->pi_ctx; | |
309 | ||
310 | pi_ctx->prot_frpl = ib_alloc_fast_reg_page_list(ib_device, | |
311 | ISCSI_ISER_SG_TABLESIZE); | |
312 | if (IS_ERR(pi_ctx->prot_frpl)) { | |
313 | ret = PTR_ERR(pi_ctx->prot_frpl); | |
314 | iser_err("Failed to allocate prot frpl ret=%d\n", | |
315 | ret); | |
316 | goto prot_frpl_failure; | |
317 | } | |
318 | ||
319 | pi_ctx->prot_mr = ib_alloc_fast_reg_mr(pd, | |
320 | ISCSI_ISER_SG_TABLESIZE + 1); | |
321 | if (IS_ERR(pi_ctx->prot_mr)) { | |
322 | ret = PTR_ERR(pi_ctx->prot_mr); | |
323 | iser_err("Failed to allocate prot frmr ret=%d\n", | |
324 | ret); | |
325 | goto prot_mr_failure; | |
326 | } | |
327 | desc->reg_indicators |= ISER_PROT_KEY_VALID; | |
328 | ||
329 | mr_init_attr.max_reg_descriptors = 2; | |
330 | mr_init_attr.flags |= IB_MR_SIGNATURE_EN; | |
331 | pi_ctx->sig_mr = ib_create_mr(pd, &mr_init_attr); | |
332 | if (IS_ERR(pi_ctx->sig_mr)) { | |
333 | ret = PTR_ERR(pi_ctx->sig_mr); | |
334 | iser_err("Failed to allocate signature enabled mr err=%d\n", | |
335 | ret); | |
336 | goto sig_mr_failure; | |
337 | } | |
338 | desc->reg_indicators |= ISER_SIG_KEY_VALID; | |
339 | } | |
340 | desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; | |
341 | ||
4f9208ad OG |
342 | iser_dbg("Create fr_desc %p page_list %p\n", |
343 | desc, desc->data_frpl->page_list); | |
310b347c SG |
344 | |
345 | return 0; | |
6b5a8fb0 AT |
346 | sig_mr_failure: |
347 | ib_dereg_mr(desc->pi_ctx->prot_mr); | |
348 | prot_mr_failure: | |
349 | ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); | |
350 | prot_frpl_failure: | |
351 | kfree(desc->pi_ctx); | |
352 | pi_ctx_alloc_failure: | |
353 | ib_dereg_mr(desc->data_mr); | |
310b347c SG |
354 | fast_reg_mr_failure: |
355 | ib_free_fast_reg_page_list(desc->data_frpl); | |
356 | ||
357 | return ret; | |
358 | } | |
359 | ||
5587856c | 360 | /** |
7306b8fa | 361 | * iser_create_fastreg_pool - Creates pool of fast_reg descriptors |
5587856c SG |
362 | * for fast registration work requests. |
363 | * returns 0 on success, or errno code on failure | |
364 | */ | |
a4ee3539 | 365 | int iser_create_fastreg_pool(struct ib_conn *ib_conn, unsigned cmds_max) |
5587856c | 366 | { |
a4ee3539 SG |
367 | struct iser_device *device = ib_conn->device; |
368 | struct fast_reg_descriptor *desc; | |
5587856c SG |
369 | int i, ret; |
370 | ||
a4ee3539 SG |
371 | INIT_LIST_HEAD(&ib_conn->fastreg.pool); |
372 | ib_conn->fastreg.pool_size = 0; | |
5587856c | 373 | for (i = 0; i < cmds_max; i++) { |
6b5a8fb0 | 374 | desc = kzalloc(sizeof(*desc), GFP_KERNEL); |
5587856c SG |
375 | if (!desc) { |
376 | iser_err("Failed to allocate a new fast_reg descriptor\n"); | |
377 | ret = -ENOMEM; | |
378 | goto err; | |
379 | } | |
380 | ||
6b5a8fb0 | 381 | ret = iser_create_fastreg_desc(device->ib_device, device->pd, |
a4ee3539 | 382 | ib_conn->pi_support, desc); |
310b347c SG |
383 | if (ret) { |
384 | iser_err("Failed to create fastreg descriptor err=%d\n", | |
385 | ret); | |
386 | kfree(desc); | |
387 | goto err; | |
5587856c SG |
388 | } |
389 | ||
a4ee3539 SG |
390 | list_add_tail(&desc->list, &ib_conn->fastreg.pool); |
391 | ib_conn->fastreg.pool_size++; | |
5587856c SG |
392 | } |
393 | ||
394 | return 0; | |
27ae2d1e | 395 | |
5587856c | 396 | err: |
a4ee3539 | 397 | iser_free_fastreg_pool(ib_conn); |
5587856c SG |
398 | return ret; |
399 | } | |
400 | ||
401 | /** | |
7306b8fa | 402 | * iser_free_fastreg_pool - releases the pool of fast_reg descriptors |
5587856c | 403 | */ |
a4ee3539 | 404 | void iser_free_fastreg_pool(struct ib_conn *ib_conn) |
5587856c SG |
405 | { |
406 | struct fast_reg_descriptor *desc, *tmp; | |
407 | int i = 0; | |
408 | ||
a4ee3539 | 409 | if (list_empty(&ib_conn->fastreg.pool)) |
5587856c SG |
410 | return; |
411 | ||
a4ee3539 | 412 | iser_info("freeing conn %p fr pool\n", ib_conn); |
5587856c | 413 | |
a4ee3539 | 414 | list_for_each_entry_safe(desc, tmp, &ib_conn->fastreg.pool, list) { |
5587856c SG |
415 | list_del(&desc->list); |
416 | ib_free_fast_reg_page_list(desc->data_frpl); | |
417 | ib_dereg_mr(desc->data_mr); | |
6b5a8fb0 AT |
418 | if (desc->pi_ctx) { |
419 | ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); | |
420 | ib_dereg_mr(desc->pi_ctx->prot_mr); | |
421 | ib_destroy_mr(desc->pi_ctx->sig_mr); | |
422 | kfree(desc->pi_ctx); | |
423 | } | |
5587856c SG |
424 | kfree(desc); |
425 | ++i; | |
426 | } | |
427 | ||
a4ee3539 | 428 | if (i < ib_conn->fastreg.pool_size) |
5587856c | 429 | iser_warn("pool still has %d regions registered\n", |
a4ee3539 | 430 | ib_conn->fastreg.pool_size - i); |
5587856c SG |
431 | } |
432 | ||
986db0d6 SP |
433 | /** |
434 | * iser_create_ib_conn_res - Queue-Pair (QP) | |
435 | * | |
436 | * returns 0 on success, -1 on failure | |
437 | */ | |
a4ee3539 | 438 | static int iser_create_ib_conn_res(struct ib_conn *ib_conn) |
986db0d6 SP |
439 | { |
440 | struct iser_device *device; | |
441 | struct ib_qp_init_attr init_attr; | |
442 | int ret = -ENOMEM; | |
443 | int index, min_index = 0; | |
444 | ||
a4ee3539 | 445 | BUG_ON(ib_conn->device == NULL); |
986db0d6 | 446 | |
a4ee3539 | 447 | device = ib_conn->device; |
1cfa0a75 OG |
448 | |
449 | memset(&init_attr, 0, sizeof init_attr); | |
450 | ||
5a33a669 AT |
451 | mutex_lock(&ig.connlist_mutex); |
452 | /* select the CQ with the minimal number of usages */ | |
bf175540 SG |
453 | for (index = 0; index < device->comps_used; index++) { |
454 | if (device->comps[index].active_qps < | |
455 | device->comps[min_index].active_qps) | |
5a33a669 | 456 | min_index = index; |
bf175540 SG |
457 | } |
458 | ib_conn->comp = &device->comps[min_index]; | |
459 | ib_conn->comp->active_qps++; | |
5a33a669 | 460 | mutex_unlock(&ig.connlist_mutex); |
a4ee3539 | 461 | iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn); |
5a33a669 | 462 | |
1cfa0a75 | 463 | init_attr.event_handler = iser_qp_event_callback; |
a4ee3539 | 464 | init_attr.qp_context = (void *)ib_conn; |
bf175540 SG |
465 | init_attr.send_cq = ib_conn->comp->tx_cq; |
466 | init_attr.recv_cq = ib_conn->comp->rx_cq; | |
1cfa0a75 | 467 | init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS; |
f19624aa | 468 | init_attr.cap.max_send_sge = 2; |
bcc60c38 | 469 | init_attr.cap.max_recv_sge = 1; |
1cfa0a75 OG |
470 | init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; |
471 | init_attr.qp_type = IB_QPT_RC; | |
a4ee3539 | 472 | if (ib_conn->pi_support) { |
6b5a8fb0 AT |
473 | init_attr.cap.max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS; |
474 | init_attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN; | |
475 | } else { | |
476 | init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS; | |
477 | } | |
1cfa0a75 | 478 | |
a4ee3539 | 479 | ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); |
1cfa0a75 | 480 | if (ret) |
9fda1ac5 | 481 | goto out_err; |
1cfa0a75 | 482 | |
a4ee3539 | 483 | ib_conn->qp = ib_conn->cma_id->qp; |
986db0d6 | 484 | iser_info("setting conn %p cma_id %p qp %p\n", |
a4ee3539 SG |
485 | ib_conn, ib_conn->cma_id, |
486 | ib_conn->cma_id->qp); | |
1cfa0a75 OG |
487 | return ret; |
488 | ||
9fda1ac5 | 489 | out_err: |
1cfa0a75 OG |
490 | iser_err("unable to alloc mem or create resource, err %d\n", ret); |
491 | return ret; | |
492 | } | |
493 | ||
1cfa0a75 OG |
494 | /** |
495 | * based on the resolved device node GUID see if there already allocated | |
496 | * device for this device. If there's no such, create one. | |
497 | */ | |
498 | static | |
499 | struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id) | |
500 | { | |
9a378270 | 501 | struct iser_device *device; |
1cfa0a75 OG |
502 | |
503 | mutex_lock(&ig.device_list_mutex); | |
504 | ||
9a378270 | 505 | list_for_each_entry(device, &ig.device_list, ig_list) |
1cfa0a75 OG |
506 | /* find if there's a match using the node GUID */ |
507 | if (device->ib_device->node_guid == cma_id->device->node_guid) | |
d33ed425 | 508 | goto inc_refcnt; |
9a378270 AR |
509 | |
510 | device = kzalloc(sizeof *device, GFP_KERNEL); | |
511 | if (device == NULL) | |
512 | goto out; | |
513 | ||
514 | /* assign this device to the device */ | |
515 | device->ib_device = cma_id->device; | |
516 | /* init the device and link it into ig device list */ | |
517 | if (iser_create_device_ib_res(device)) { | |
518 | kfree(device); | |
519 | device = NULL; | |
520 | goto out; | |
1cfa0a75 | 521 | } |
9a378270 AR |
522 | list_add(&device->ig_list, &ig.device_list); |
523 | ||
d33ed425 | 524 | inc_refcnt: |
1cfa0a75 | 525 | device->refcount++; |
d33ed425 | 526 | out: |
1cfa0a75 OG |
527 | mutex_unlock(&ig.device_list_mutex); |
528 | return device; | |
529 | } | |
530 | ||
531 | /* if there's no demand for this device, release it */ | |
532 | static void iser_device_try_release(struct iser_device *device) | |
533 | { | |
534 | mutex_lock(&ig.device_list_mutex); | |
535 | device->refcount--; | |
4f363882 | 536 | iser_info("device %p refcount %d\n", device, device->refcount); |
1cfa0a75 OG |
537 | if (!device->refcount) { |
538 | iser_free_device_ib_res(device); | |
539 | list_del(&device->ig_list); | |
540 | kfree(device); | |
541 | } | |
542 | mutex_unlock(&ig.device_list_mutex); | |
543 | } | |
544 | ||
504130c0 AN |
545 | /** |
546 | * Called with state mutex held | |
547 | **/ | |
5716af6e SG |
548 | static int iser_conn_state_comp_exch(struct iser_conn *iser_conn, |
549 | enum iser_conn_state comp, | |
550 | enum iser_conn_state exch) | |
1cfa0a75 OG |
551 | { |
552 | int ret; | |
553 | ||
5716af6e SG |
554 | ret = (iser_conn->state == comp); |
555 | if (ret) | |
556 | iser_conn->state = exch; | |
557 | ||
1cfa0a75 OG |
558 | return ret; |
559 | } | |
560 | ||
b73c3ada AN |
561 | void iser_release_work(struct work_struct *work) |
562 | { | |
5716af6e | 563 | struct iser_conn *iser_conn; |
b73c3ada | 564 | |
5716af6e | 565 | iser_conn = container_of(work, struct iser_conn, release_work); |
b73c3ada | 566 | |
c107a6c0 SG |
567 | /* Wait for conn_stop to complete */ |
568 | wait_for_completion(&iser_conn->stop_completion); | |
569 | /* Wait for IB resouces cleanup to complete */ | |
570 | wait_for_completion(&iser_conn->ib_completion); | |
b73c3ada | 571 | |
5716af6e SG |
572 | mutex_lock(&iser_conn->state_mutex); |
573 | iser_conn->state = ISER_CONN_DOWN; | |
574 | mutex_unlock(&iser_conn->state_mutex); | |
504130c0 | 575 | |
5716af6e | 576 | iser_conn_release(iser_conn); |
b73c3ada AN |
577 | } |
578 | ||
96f15198 SG |
579 | /** |
580 | * iser_free_ib_conn_res - release IB related resources | |
581 | * @iser_conn: iser connection struct | |
c47a3c9e SG |
582 | * @destroy_device: indicator if we need to try to release |
583 | * the iser device (only iscsi shutdown and DEVICE_REMOVAL | |
584 | * will use this. | |
96f15198 SG |
585 | * |
586 | * This routine is called with the iser state mutex held | |
587 | * so the cm_id removal is out of here. It is Safe to | |
588 | * be invoked multiple times. | |
589 | */ | |
c47a3c9e SG |
590 | static void iser_free_ib_conn_res(struct iser_conn *iser_conn, |
591 | bool destroy_device) | |
96f15198 SG |
592 | { |
593 | struct ib_conn *ib_conn = &iser_conn->ib_conn; | |
594 | struct iser_device *device = ib_conn->device; | |
595 | ||
596 | iser_info("freeing conn %p cma_id %p qp %p\n", | |
597 | iser_conn, ib_conn->cma_id, ib_conn->qp); | |
598 | ||
599 | iser_free_rx_descriptors(iser_conn); | |
600 | ||
601 | if (ib_conn->qp != NULL) { | |
bf175540 | 602 | ib_conn->comp->active_qps--; |
96f15198 SG |
603 | rdma_destroy_qp(ib_conn->cma_id); |
604 | ib_conn->qp = NULL; | |
605 | } | |
606 | ||
c47a3c9e | 607 | if (destroy_device && device != NULL) { |
96f15198 SG |
608 | iser_device_try_release(device); |
609 | ib_conn->device = NULL; | |
610 | } | |
611 | } | |
612 | ||
41179e2d RD |
613 | /** |
614 | * Frees all conn objects and deallocs conn descriptor | |
615 | */ | |
5716af6e | 616 | void iser_conn_release(struct iser_conn *iser_conn) |
41179e2d | 617 | { |
a4ee3539 | 618 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
41179e2d | 619 | |
41179e2d | 620 | mutex_lock(&ig.connlist_mutex); |
5716af6e | 621 | list_del(&iser_conn->conn_list); |
41179e2d | 622 | mutex_unlock(&ig.connlist_mutex); |
504130c0 | 623 | |
5716af6e | 624 | mutex_lock(&iser_conn->state_mutex); |
aea8f4df AN |
625 | if (iser_conn->state != ISER_CONN_DOWN) |
626 | iser_warn("iser conn %p state %d, expected state down.\n", | |
627 | iser_conn, iser_conn->state); | |
c47a3c9e SG |
628 | /* |
629 | * In case we never got to bind stage, we still need to | |
630 | * release IB resources (which is safe to call more than once). | |
631 | */ | |
632 | iser_free_ib_conn_res(iser_conn, true); | |
5716af6e | 633 | mutex_unlock(&iser_conn->state_mutex); |
504130c0 | 634 | |
a4ee3539 SG |
635 | if (ib_conn->cma_id != NULL) { |
636 | rdma_destroy_id(ib_conn->cma_id); | |
637 | ib_conn->cma_id = NULL; | |
5b61ff43 | 638 | } |
a4ee3539 | 639 | |
5716af6e | 640 | kfree(iser_conn); |
41179e2d RD |
641 | } |
642 | ||
c47a3c9e SG |
643 | /** |
644 | * iser_poll_for_flush_errors - Don't settle for less than all. | |
645 | * @struct ib_conn: IB context of the connection | |
646 | * | |
647 | * This routine is called when the QP is in error state | |
648 | * It polls the send CQ until all flush errors are consumed and | |
649 | * returns when all flush errors were processed. | |
650 | */ | |
651 | static void iser_poll_for_flush_errors(struct ib_conn *ib_conn) | |
652 | { | |
c47a3c9e SG |
653 | int count = 0; |
654 | ||
655 | while (ib_conn->post_recv_buf_count > 0 || | |
656 | atomic_read(&ib_conn->post_send_buf_count) > 0) { | |
657 | msleep(100); | |
658 | if (atomic_read(&ib_conn->post_send_buf_count) > 0) | |
bf175540 | 659 | iser_drain_tx_cq(ib_conn->comp); |
c47a3c9e SG |
660 | |
661 | count++; | |
662 | /* Don't flood with prints */ | |
663 | if (count % 30 == 0) | |
664 | iser_dbg("post_recv %d post_send %d", | |
665 | ib_conn->post_recv_buf_count, | |
666 | atomic_read(&ib_conn->post_send_buf_count)); | |
667 | } | |
668 | } | |
669 | ||
1cfa0a75 OG |
670 | /** |
671 | * triggers start of the disconnect procedures and wait for them to be done | |
c47a3c9e | 672 | * Called with state mutex held |
1cfa0a75 | 673 | */ |
c47a3c9e | 674 | int iser_conn_terminate(struct iser_conn *iser_conn) |
1cfa0a75 | 675 | { |
a4ee3539 | 676 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
1cfa0a75 OG |
677 | int err = 0; |
678 | ||
c47a3c9e SG |
679 | /* terminate the iser conn only if the conn state is UP */ |
680 | if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP, | |
681 | ISER_CONN_TERMINATING)) | |
682 | return 0; | |
683 | ||
684 | iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state); | |
685 | ||
686 | /* suspend queuing of new iscsi commands */ | |
687 | if (iser_conn->iscsi_conn) | |
688 | iscsi_suspend_queue(iser_conn->iscsi_conn); | |
689 | ||
690 | /* | |
691 | * In case we didn't already clean up the cma_id (peer initiated | |
692 | * a disconnection), we need to Cause the CMA to change the QP | |
693 | * state to ERROR. | |
1cfa0a75 | 694 | */ |
c47a3c9e SG |
695 | if (ib_conn->cma_id) { |
696 | err = rdma_disconnect(ib_conn->cma_id); | |
697 | if (err) | |
698 | iser_err("Failed to disconnect, conn: 0x%p err %d\n", | |
699 | iser_conn, err); | |
700 | ||
701 | iser_poll_for_flush_errors(ib_conn); | |
702 | } | |
1cfa0a75 | 703 | |
c47a3c9e | 704 | return 1; |
1cfa0a75 OG |
705 | } |
706 | ||
504130c0 AN |
707 | /** |
708 | * Called with state mutex held | |
709 | **/ | |
b73c3ada | 710 | static void iser_connect_error(struct rdma_cm_id *cma_id) |
1cfa0a75 | 711 | { |
5716af6e | 712 | struct iser_conn *iser_conn; |
b73c3ada | 713 | |
5716af6e SG |
714 | iser_conn = (struct iser_conn *)cma_id->context; |
715 | iser_conn->state = ISER_CONN_DOWN; | |
1cfa0a75 OG |
716 | } |
717 | ||
504130c0 AN |
718 | /** |
719 | * Called with state mutex held | |
720 | **/ | |
b73c3ada | 721 | static void iser_addr_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
722 | { |
723 | struct iser_device *device; | |
5716af6e | 724 | struct iser_conn *iser_conn; |
a4ee3539 | 725 | struct ib_conn *ib_conn; |
1cfa0a75 OG |
726 | int ret; |
727 | ||
5716af6e SG |
728 | iser_conn = (struct iser_conn *)cma_id->context; |
729 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
730 | /* bailout */ |
731 | return; | |
732 | ||
a4ee3539 | 733 | ib_conn = &iser_conn->ib_conn; |
1cfa0a75 | 734 | device = iser_device_find_by_ib_device(cma_id); |
d33ed425 AR |
735 | if (!device) { |
736 | iser_err("device lookup/creation failed\n"); | |
b73c3ada AN |
737 | iser_connect_error(cma_id); |
738 | return; | |
d33ed425 AR |
739 | } |
740 | ||
a4ee3539 | 741 | ib_conn->device = device; |
1cfa0a75 | 742 | |
7f733847 AT |
743 | /* connection T10-PI support */ |
744 | if (iser_pi_enable) { | |
745 | if (!(device->dev_attr.device_cap_flags & | |
746 | IB_DEVICE_SIGNATURE_HANDOVER)) { | |
747 | iser_warn("T10-PI requested but not supported on %s, " | |
748 | "continue without T10-PI\n", | |
a4ee3539 SG |
749 | ib_conn->device->ib_device->name); |
750 | ib_conn->pi_support = false; | |
7f733847 | 751 | } else { |
a4ee3539 | 752 | ib_conn->pi_support = true; |
7f733847 AT |
753 | } |
754 | } | |
755 | ||
1cfa0a75 OG |
756 | ret = rdma_resolve_route(cma_id, 1000); |
757 | if (ret) { | |
758 | iser_err("resolve route failed: %d\n", ret); | |
b73c3ada AN |
759 | iser_connect_error(cma_id); |
760 | return; | |
1cfa0a75 | 761 | } |
1cfa0a75 OG |
762 | } |
763 | ||
504130c0 AN |
764 | /** |
765 | * Called with state mutex held | |
766 | **/ | |
b73c3ada | 767 | static void iser_route_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 OG |
768 | { |
769 | struct rdma_conn_param conn_param; | |
770 | int ret; | |
8d8399de | 771 | struct iser_cm_hdr req_hdr; |
5716af6e | 772 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
a4ee3539 SG |
773 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
774 | struct iser_device *device = ib_conn->device; | |
1cfa0a75 | 775 | |
5716af6e | 776 | if (iser_conn->state != ISER_CONN_PENDING) |
504130c0 AN |
777 | /* bailout */ |
778 | return; | |
779 | ||
a4ee3539 | 780 | ret = iser_create_ib_conn_res(ib_conn); |
1cfa0a75 OG |
781 | if (ret) |
782 | goto failure; | |
783 | ||
1cfa0a75 | 784 | memset(&conn_param, 0, sizeof conn_param); |
2ea32938 | 785 | conn_param.responder_resources = device->dev_attr.max_qp_rd_atom; |
1cfa0a75 OG |
786 | conn_param.initiator_depth = 1; |
787 | conn_param.retry_count = 7; | |
788 | conn_param.rnr_retry_count = 6; | |
789 | ||
8d8399de OG |
790 | memset(&req_hdr, 0, sizeof(req_hdr)); |
791 | req_hdr.flags = (ISER_ZBVA_NOT_SUPPORTED | | |
792 | ISER_SEND_W_INV_NOT_SUPPORTED); | |
793 | conn_param.private_data = (void *)&req_hdr; | |
794 | conn_param.private_data_len = sizeof(struct iser_cm_hdr); | |
795 | ||
1cfa0a75 OG |
796 | ret = rdma_connect(cma_id, &conn_param); |
797 | if (ret) { | |
798 | iser_err("failure connecting: %d\n", ret); | |
799 | goto failure; | |
800 | } | |
801 | ||
b73c3ada | 802 | return; |
1cfa0a75 | 803 | failure: |
b73c3ada | 804 | iser_connect_error(cma_id); |
1cfa0a75 OG |
805 | } |
806 | ||
807 | static void iser_connected_handler(struct rdma_cm_id *cma_id) | |
808 | { | |
5716af6e | 809 | struct iser_conn *iser_conn; |
4f9208ad OG |
810 | struct ib_qp_attr attr; |
811 | struct ib_qp_init_attr init_attr; | |
812 | ||
5716af6e SG |
813 | iser_conn = (struct iser_conn *)cma_id->context; |
814 | if (iser_conn->state != ISER_CONN_PENDING) | |
504130c0 AN |
815 | /* bailout */ |
816 | return; | |
817 | ||
4f9208ad OG |
818 | (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr); |
819 | iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num); | |
1cfa0a75 | 820 | |
5716af6e SG |
821 | iser_conn->state = ISER_CONN_UP; |
822 | complete(&iser_conn->up_completion); | |
1cfa0a75 OG |
823 | } |
824 | ||
b73c3ada | 825 | static void iser_disconnected_handler(struct rdma_cm_id *cma_id) |
1cfa0a75 | 826 | { |
c47a3c9e | 827 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; |
1cfa0a75 | 828 | |
c47a3c9e | 829 | if (iser_conn_terminate(iser_conn)) { |
5716af6e | 830 | if (iser_conn->iscsi_conn) |
c47a3c9e SG |
831 | iscsi_conn_failure(iser_conn->iscsi_conn, |
832 | ISCSI_ERR_CONN_FAILED); | |
7d9eacf9 RD |
833 | else |
834 | iser_err("iscsi_iser connection isn't bound\n"); | |
835 | } | |
c47a3c9e SG |
836 | } |
837 | ||
838 | static void iser_cleanup_handler(struct rdma_cm_id *cma_id, | |
839 | bool destroy_device) | |
840 | { | |
841 | struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context; | |
1cfa0a75 | 842 | |
c47a3c9e SG |
843 | /* |
844 | * We are not guaranteed that we visited disconnected_handler | |
845 | * by now, call it here to be safe that we handle CM drep | |
846 | * and flush errors. | |
8d4aca7f | 847 | */ |
c47a3c9e SG |
848 | iser_disconnected_handler(cma_id); |
849 | iser_free_ib_conn_res(iser_conn, destroy_device); | |
850 | complete(&iser_conn->ib_completion); | |
851 | }; | |
1cfa0a75 OG |
852 | |
853 | static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) | |
854 | { | |
5716af6e | 855 | struct iser_conn *iser_conn; |
c47a3c9e | 856 | int ret = 0; |
504130c0 | 857 | |
5716af6e | 858 | iser_conn = (struct iser_conn *)cma_id->context; |
4f363882 RD |
859 | iser_info("event %d status %d conn %p id %p\n", |
860 | event->event, event->status, cma_id->context, cma_id); | |
1cfa0a75 | 861 | |
5716af6e | 862 | mutex_lock(&iser_conn->state_mutex); |
1cfa0a75 OG |
863 | switch (event->event) { |
864 | case RDMA_CM_EVENT_ADDR_RESOLVED: | |
b73c3ada | 865 | iser_addr_handler(cma_id); |
1cfa0a75 OG |
866 | break; |
867 | case RDMA_CM_EVENT_ROUTE_RESOLVED: | |
b73c3ada | 868 | iser_route_handler(cma_id); |
1cfa0a75 OG |
869 | break; |
870 | case RDMA_CM_EVENT_ESTABLISHED: | |
871 | iser_connected_handler(cma_id); | |
872 | break; | |
873 | case RDMA_CM_EVENT_ADDR_ERROR: | |
874 | case RDMA_CM_EVENT_ROUTE_ERROR: | |
875 | case RDMA_CM_EVENT_CONNECT_ERROR: | |
876 | case RDMA_CM_EVENT_UNREACHABLE: | |
877 | case RDMA_CM_EVENT_REJECTED: | |
b73c3ada | 878 | iser_connect_error(cma_id); |
1cfa0a75 OG |
879 | break; |
880 | case RDMA_CM_EVENT_DISCONNECTED: | |
2f5de151 | 881 | case RDMA_CM_EVENT_ADDR_CHANGE: |
b73c3ada | 882 | iser_disconnected_handler(cma_id); |
1cfa0a75 | 883 | break; |
c47a3c9e SG |
884 | case RDMA_CM_EVENT_DEVICE_REMOVAL: |
885 | /* | |
886 | * we *must* destroy the device as we cannot rely | |
887 | * on iscsid to be around to initiate error handling. | |
888 | * also implicitly destroy the cma_id. | |
889 | */ | |
890 | iser_cleanup_handler(cma_id, true); | |
891 | iser_conn->ib_conn.cma_id = NULL; | |
892 | ret = 1; | |
893 | break; | |
894 | case RDMA_CM_EVENT_TIMEWAIT_EXIT: | |
895 | iser_cleanup_handler(cma_id, false); | |
896 | break; | |
1cfa0a75 | 897 | default: |
a4ef1451 | 898 | iser_err("Unexpected RDMA CM event (%d)\n", event->event); |
1cfa0a75 OG |
899 | break; |
900 | } | |
5716af6e | 901 | mutex_unlock(&iser_conn->state_mutex); |
c47a3c9e SG |
902 | |
903 | return ret; | |
1cfa0a75 OG |
904 | } |
905 | ||
5716af6e | 906 | void iser_conn_init(struct iser_conn *iser_conn) |
1cfa0a75 | 907 | { |
5716af6e | 908 | iser_conn->state = ISER_CONN_INIT; |
a4ee3539 SG |
909 | iser_conn->ib_conn.post_recv_buf_count = 0; |
910 | atomic_set(&iser_conn->ib_conn.post_send_buf_count, 0); | |
5716af6e | 911 | init_completion(&iser_conn->stop_completion); |
c47a3c9e | 912 | init_completion(&iser_conn->ib_completion); |
5716af6e SG |
913 | init_completion(&iser_conn->up_completion); |
914 | INIT_LIST_HEAD(&iser_conn->conn_list); | |
a4ee3539 | 915 | spin_lock_init(&iser_conn->ib_conn.lock); |
5716af6e | 916 | mutex_init(&iser_conn->state_mutex); |
1cfa0a75 OG |
917 | } |
918 | ||
919 | /** | |
920 | * starts the process of connecting to the target | |
94e2bd68 | 921 | * sleeps until the connection is established or rejected |
1cfa0a75 | 922 | */ |
5716af6e | 923 | int iser_connect(struct iser_conn *iser_conn, |
96ed02d4 RD |
924 | struct sockaddr *src_addr, |
925 | struct sockaddr *dst_addr, | |
1cfa0a75 OG |
926 | int non_blocking) |
927 | { | |
a4ee3539 | 928 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
1cfa0a75 OG |
929 | int err = 0; |
930 | ||
5716af6e | 931 | mutex_lock(&iser_conn->state_mutex); |
504130c0 | 932 | |
5716af6e | 933 | sprintf(iser_conn->name, "%pISp", dst_addr); |
96ed02d4 | 934 | |
5716af6e | 935 | iser_info("connecting to: %s\n", iser_conn->name); |
1cfa0a75 OG |
936 | |
937 | /* the device is known only --after-- address resolution */ | |
a4ee3539 | 938 | ib_conn->device = NULL; |
1cfa0a75 | 939 | |
5716af6e | 940 | iser_conn->state = ISER_CONN_PENDING; |
1cfa0a75 | 941 | |
a4ee3539 SG |
942 | ib_conn->cma_id = rdma_create_id(iser_cma_handler, |
943 | (void *)iser_conn, | |
944 | RDMA_PS_TCP, IB_QPT_RC); | |
945 | if (IS_ERR(ib_conn->cma_id)) { | |
946 | err = PTR_ERR(ib_conn->cma_id); | |
1cfa0a75 OG |
947 | iser_err("rdma_create_id failed: %d\n", err); |
948 | goto id_failure; | |
949 | } | |
950 | ||
a4ee3539 | 951 | err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000); |
1cfa0a75 OG |
952 | if (err) { |
953 | iser_err("rdma_resolve_addr failed: %d\n", err); | |
954 | goto addr_failure; | |
955 | } | |
956 | ||
957 | if (!non_blocking) { | |
5716af6e | 958 | wait_for_completion_interruptible(&iser_conn->up_completion); |
1cfa0a75 | 959 | |
5716af6e | 960 | if (iser_conn->state != ISER_CONN_UP) { |
1cfa0a75 OG |
961 | err = -EIO; |
962 | goto connect_failure; | |
963 | } | |
964 | } | |
5716af6e | 965 | mutex_unlock(&iser_conn->state_mutex); |
1cfa0a75 OG |
966 | |
967 | mutex_lock(&ig.connlist_mutex); | |
5716af6e | 968 | list_add(&iser_conn->conn_list, &ig.connlist); |
1cfa0a75 OG |
969 | mutex_unlock(&ig.connlist_mutex); |
970 | return 0; | |
971 | ||
972 | id_failure: | |
a4ee3539 | 973 | ib_conn->cma_id = NULL; |
1cfa0a75 | 974 | addr_failure: |
5716af6e | 975 | iser_conn->state = ISER_CONN_DOWN; |
1cfa0a75 | 976 | connect_failure: |
5716af6e SG |
977 | mutex_unlock(&iser_conn->state_mutex); |
978 | iser_conn_release(iser_conn); | |
1cfa0a75 OG |
979 | return err; |
980 | } | |
981 | ||
1cfa0a75 OG |
982 | /** |
983 | * iser_reg_page_vec - Register physical memory | |
984 | * | |
985 | * returns: 0 on success, errno code on failure | |
986 | */ | |
a4ee3539 | 987 | int iser_reg_page_vec(struct ib_conn *ib_conn, |
1cfa0a75 OG |
988 | struct iser_page_vec *page_vec, |
989 | struct iser_mem_reg *mem_reg) | |
990 | { | |
991 | struct ib_pool_fmr *mem; | |
992 | u64 io_addr; | |
993 | u64 *page_list; | |
994 | int status; | |
995 | ||
996 | page_list = page_vec->pages; | |
997 | io_addr = page_list[0]; | |
998 | ||
a4ee3539 | 999 | mem = ib_fmr_pool_map_phys(ib_conn->fmr.pool, |
1cfa0a75 OG |
1000 | page_list, |
1001 | page_vec->length, | |
adfaa888 | 1002 | io_addr); |
1cfa0a75 OG |
1003 | |
1004 | if (IS_ERR(mem)) { | |
1005 | status = (int)PTR_ERR(mem); | |
1006 | iser_err("ib_fmr_pool_map_phys failed: %d\n", status); | |
1007 | return status; | |
1008 | } | |
1009 | ||
1010 | mem_reg->lkey = mem->fmr->lkey; | |
1011 | mem_reg->rkey = mem->fmr->rkey; | |
8dfa0876 | 1012 | mem_reg->len = page_vec->length * SIZE_4K; |
1cfa0a75 | 1013 | mem_reg->va = io_addr; |
5587856c | 1014 | mem_reg->is_mr = 1; |
1cfa0a75 OG |
1015 | mem_reg->mem_h = (void *)mem; |
1016 | ||
1017 | mem_reg->va += page_vec->offset; | |
1018 | mem_reg->len = page_vec->data_size; | |
1019 | ||
1020 | iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, " | |
1021 | "entry[0]: (0x%08lx,%ld)] -> " | |
1022 | "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n", | |
1023 | page_vec, page_vec->length, | |
1024 | (unsigned long)page_vec->pages[0], | |
1025 | (unsigned long)page_vec->data_size, | |
1026 | (unsigned int)mem_reg->lkey, mem_reg->mem_h, | |
1027 | (unsigned long)mem_reg->va, (unsigned long)mem_reg->len); | |
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | /** | |
e657571b SG |
1032 | * Unregister (previosuly registered using FMR) memory. |
1033 | * If memory is non-FMR does nothing. | |
1cfa0a75 | 1034 | */ |
e657571b SG |
1035 | void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, |
1036 | enum iser_data_dir cmd_dir) | |
1cfa0a75 | 1037 | { |
b4e155ff | 1038 | struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; |
1cfa0a75 OG |
1039 | int ret; |
1040 | ||
5587856c | 1041 | if (!reg->is_mr) |
b4e155ff SG |
1042 | return; |
1043 | ||
1cfa0a75 OG |
1044 | iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h); |
1045 | ||
1046 | ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h); | |
1047 | if (ret) | |
1048 | iser_err("ib_fmr_pool_unmap failed %d\n", ret); | |
1049 | ||
1050 | reg->mem_h = NULL; | |
1051 | } | |
1052 | ||
7306b8fa SG |
1053 | void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task, |
1054 | enum iser_data_dir cmd_dir) | |
5587856c SG |
1055 | { |
1056 | struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; | |
5716af6e | 1057 | struct iser_conn *iser_conn = iser_task->iser_conn; |
a4ee3539 | 1058 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
5587856c SG |
1059 | struct fast_reg_descriptor *desc = reg->mem_h; |
1060 | ||
1061 | if (!reg->is_mr) | |
1062 | return; | |
1063 | ||
1064 | reg->mem_h = NULL; | |
1065 | reg->is_mr = 0; | |
a4ee3539 SG |
1066 | spin_lock_bh(&ib_conn->lock); |
1067 | list_add_tail(&desc->list, &ib_conn->fastreg.pool); | |
1068 | spin_unlock_bh(&ib_conn->lock); | |
5587856c SG |
1069 | } |
1070 | ||
5716af6e | 1071 | int iser_post_recvl(struct iser_conn *iser_conn) |
bcc60c38 OG |
1072 | { |
1073 | struct ib_recv_wr rx_wr, *rx_wr_failed; | |
a4ee3539 | 1074 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
bcc60c38 OG |
1075 | struct ib_sge sge; |
1076 | int ib_ret; | |
1077 | ||
5716af6e | 1078 | sge.addr = iser_conn->login_resp_dma; |
bcc60c38 | 1079 | sge.length = ISER_RX_LOGIN_SIZE; |
a4ee3539 | 1080 | sge.lkey = ib_conn->device->mr->lkey; |
bcc60c38 | 1081 | |
5716af6e | 1082 | rx_wr.wr_id = (unsigned long)iser_conn->login_resp_buf; |
bcc60c38 OG |
1083 | rx_wr.sg_list = &sge; |
1084 | rx_wr.num_sge = 1; | |
1085 | rx_wr.next = NULL; | |
1086 | ||
a4ee3539 SG |
1087 | ib_conn->post_recv_buf_count++; |
1088 | ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1089 | if (ib_ret) { |
1090 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1091 | ib_conn->post_recv_buf_count--; |
bcc60c38 OG |
1092 | } |
1093 | return ib_ret; | |
1094 | } | |
1095 | ||
5716af6e | 1096 | int iser_post_recvm(struct iser_conn *iser_conn, int count) |
bcc60c38 OG |
1097 | { |
1098 | struct ib_recv_wr *rx_wr, *rx_wr_failed; | |
1099 | int i, ib_ret; | |
a4ee3539 | 1100 | struct ib_conn *ib_conn = &iser_conn->ib_conn; |
5716af6e | 1101 | unsigned int my_rx_head = iser_conn->rx_desc_head; |
bcc60c38 OG |
1102 | struct iser_rx_desc *rx_desc; |
1103 | ||
a4ee3539 | 1104 | for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { |
5716af6e | 1105 | rx_desc = &iser_conn->rx_descs[my_rx_head]; |
bcc60c38 OG |
1106 | rx_wr->wr_id = (unsigned long)rx_desc; |
1107 | rx_wr->sg_list = &rx_desc->rx_sg; | |
1108 | rx_wr->num_sge = 1; | |
1109 | rx_wr->next = rx_wr + 1; | |
5716af6e | 1110 | my_rx_head = (my_rx_head + 1) & iser_conn->qp_max_recv_dtos_mask; |
bcc60c38 OG |
1111 | } |
1112 | ||
1113 | rx_wr--; | |
1114 | rx_wr->next = NULL; /* mark end of work requests list */ | |
1115 | ||
a4ee3539 SG |
1116 | ib_conn->post_recv_buf_count += count; |
1117 | ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed); | |
bcc60c38 OG |
1118 | if (ib_ret) { |
1119 | iser_err("ib_post_recv failed ret=%d\n", ib_ret); | |
a4ee3539 | 1120 | ib_conn->post_recv_buf_count -= count; |
bcc60c38 | 1121 | } else |
5716af6e | 1122 | iser_conn->rx_desc_head = my_rx_head; |
bcc60c38 OG |
1123 | return ib_ret; |
1124 | } | |
1125 | ||
1126 | ||
1cfa0a75 OG |
1127 | /** |
1128 | * iser_start_send - Initiate a Send DTO operation | |
1129 | * | |
1130 | * returns 0 on success, -1 on failure | |
1131 | */ | |
a4ee3539 | 1132 | int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc) |
1cfa0a75 | 1133 | { |
f19624aa | 1134 | int ib_ret; |
1cfa0a75 | 1135 | struct ib_send_wr send_wr, *send_wr_failed; |
1cfa0a75 | 1136 | |
a4ee3539 | 1137 | ib_dma_sync_single_for_device(ib_conn->device->ib_device, |
5716af6e SG |
1138 | tx_desc->dma_addr, ISER_HEADERS_LEN, |
1139 | DMA_TO_DEVICE); | |
1cfa0a75 OG |
1140 | |
1141 | send_wr.next = NULL; | |
1142 | send_wr.wr_id = (unsigned long)tx_desc; | |
f19624aa OG |
1143 | send_wr.sg_list = tx_desc->tx_sg; |
1144 | send_wr.num_sge = tx_desc->num_sge; | |
1cfa0a75 | 1145 | send_wr.opcode = IB_WR_SEND; |
f19624aa | 1146 | send_wr.send_flags = IB_SEND_SIGNALED; |
1cfa0a75 | 1147 | |
a4ee3539 | 1148 | atomic_inc(&ib_conn->post_send_buf_count); |
1cfa0a75 | 1149 | |
a4ee3539 | 1150 | ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed); |
1cfa0a75 | 1151 | if (ib_ret) { |
1cfa0a75 | 1152 | iser_err("ib_post_send failed, ret:%d\n", ib_ret); |
a4ee3539 | 1153 | atomic_dec(&ib_conn->post_send_buf_count); |
1cfa0a75 | 1154 | } |
f19624aa | 1155 | return ib_ret; |
1cfa0a75 OG |
1156 | } |
1157 | ||
8c204e69 SG |
1158 | /** |
1159 | * iser_handle_comp_error() - Handle error completion | |
1160 | * @desc: iser TX descriptor | |
1161 | * @ib_conn: connection RDMA resources | |
1162 | * @wc: work completion | |
1163 | * | |
1164 | * Notes: We may handle a FLUSH error completion and in this case | |
1165 | * we only cleanup in case TX type was DATAOUT. For non-FLUSH | |
1166 | * error completion we should also notify iscsi layer that | |
1167 | * connection is failed (in case we passed bind stage). | |
1168 | */ | |
1169 | static void | |
1170 | iser_handle_comp_error(struct iser_tx_desc *desc, | |
1171 | struct ib_conn *ib_conn, | |
1172 | struct ib_wc *wc) | |
1cfa0a75 | 1173 | { |
8c204e69 SG |
1174 | struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn, |
1175 | ib_conn); | |
1176 | ||
1177 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
1178 | if (iser_conn->iscsi_conn) | |
1179 | iscsi_conn_failure(iser_conn->iscsi_conn, | |
1180 | ISCSI_ERR_CONN_FAILED); | |
1181 | ||
78ad0a34 OG |
1182 | if (desc && desc->type == ISCSI_TX_DATAOUT) |
1183 | kmem_cache_free(ig.desc_cache, desc); | |
1cfa0a75 OG |
1184 | } |
1185 | ||
bf175540 | 1186 | static int iser_drain_tx_cq(struct iser_comp *comp) |
78ad0a34 | 1187 | { |
bf175540 | 1188 | struct ib_cq *cq = comp->tx_cq; |
78ad0a34 | 1189 | struct ib_wc wc; |
f19624aa | 1190 | struct iser_tx_desc *tx_desc; |
a4ee3539 | 1191 | struct ib_conn *ib_conn; |
78ad0a34 OG |
1192 | int completed_tx = 0; |
1193 | ||
1194 | while (ib_poll_cq(cq, 1, &wc) == 1) { | |
f19624aa | 1195 | tx_desc = (struct iser_tx_desc *) (unsigned long) wc.wr_id; |
a4ee3539 | 1196 | ib_conn = wc.qp->qp_context; |
78ad0a34 OG |
1197 | if (wc.status == IB_WC_SUCCESS) { |
1198 | if (wc.opcode == IB_WC_SEND) | |
a4ee3539 | 1199 | iser_snd_completion(tx_desc, ib_conn); |
db523b8d | 1200 | else |
78ad0a34 OG |
1201 | iser_err("expected opcode %d got %d\n", |
1202 | IB_WC_SEND, wc.opcode); | |
1203 | } else { | |
1204 | iser_err("tx id %llx status %d vend_err %x\n", | |
db523b8d | 1205 | wc.wr_id, wc.status, wc.vendor_err); |
7306b8fa | 1206 | if (wc.wr_id != ISER_FASTREG_LI_WRID) { |
a4ee3539 | 1207 | atomic_dec(&ib_conn->post_send_buf_count); |
8c204e69 | 1208 | iser_handle_comp_error(tx_desc, ib_conn, &wc); |
db523b8d | 1209 | } |
78ad0a34 OG |
1210 | } |
1211 | completed_tx++; | |
1212 | } | |
1213 | return completed_tx; | |
1214 | } | |
1215 | ||
1216 | ||
1cfa0a75 OG |
1217 | static void iser_cq_tasklet_fn(unsigned long data) |
1218 | { | |
bf175540 SG |
1219 | struct iser_comp *comp = (struct iser_comp *)data; |
1220 | struct ib_cq *cq = comp->rx_cq; | |
1221 | struct ib_wc wc; | |
1222 | struct iser_rx_desc *desc; | |
1223 | unsigned long xfer_len; | |
a4ee3539 | 1224 | struct ib_conn *ib_conn; |
1d6c2b73 RD |
1225 | int completed_tx, completed_rx = 0; |
1226 | ||
1227 | /* First do tx drain, so in a case where we have rx flushes and a successful | |
1228 | * tx completion we will still go through completion error handling. | |
1229 | */ | |
bf175540 | 1230 | completed_tx = iser_drain_tx_cq(comp); |
1cfa0a75 OG |
1231 | |
1232 | while (ib_poll_cq(cq, 1, &wc) == 1) { | |
78ad0a34 | 1233 | desc = (struct iser_rx_desc *) (unsigned long) wc.wr_id; |
1cfa0a75 | 1234 | BUG_ON(desc == NULL); |
a4ee3539 | 1235 | ib_conn = wc.qp->qp_context; |
1cfa0a75 | 1236 | if (wc.status == IB_WC_SUCCESS) { |
bcc60c38 | 1237 | if (wc.opcode == IB_WC_RECV) { |
1cfa0a75 | 1238 | xfer_len = (unsigned long)wc.byte_len; |
a4ee3539 | 1239 | iser_rcv_completion(desc, xfer_len, ib_conn); |
78ad0a34 OG |
1240 | } else |
1241 | iser_err("expected opcode %d got %d\n", | |
1242 | IB_WC_RECV, wc.opcode); | |
1cfa0a75 | 1243 | } else { |
bcc60c38 | 1244 | if (wc.status != IB_WC_WR_FLUSH_ERR) |
78ad0a34 | 1245 | iser_err("rx id %llx status %d vend_err %x\n", |
bcc60c38 | 1246 | wc.wr_id, wc.status, wc.vendor_err); |
a4ee3539 | 1247 | ib_conn->post_recv_buf_count--; |
8c204e69 | 1248 | iser_handle_comp_error(NULL, ib_conn, &wc); |
1cfa0a75 | 1249 | } |
78ad0a34 OG |
1250 | completed_rx++; |
1251 | if (!(completed_rx & 63)) | |
bf175540 | 1252 | completed_tx += iser_drain_tx_cq(comp); |
183cfa43 SG |
1253 | if (completed_rx >= iser_cq_poll_limit) |
1254 | break; | |
1cfa0a75 OG |
1255 | } |
1256 | /* #warning "it is assumed here that arming CQ only once its empty" * | |
1257 | * " would not cause interrupts to be missed" */ | |
1258 | ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); | |
78ad0a34 | 1259 | |
78ad0a34 | 1260 | iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx); |
1cfa0a75 OG |
1261 | } |
1262 | ||
1263 | static void iser_cq_callback(struct ib_cq *cq, void *cq_context) | |
1264 | { | |
bf175540 | 1265 | struct iser_comp *comp = cq_context; |
1cfa0a75 | 1266 | |
bf175540 | 1267 | tasklet_schedule(&comp->tasklet); |
1cfa0a75 | 1268 | } |
0a7a08ad SG |
1269 | |
1270 | u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, | |
1271 | enum iser_data_dir cmd_dir, sector_t *sector) | |
1272 | { | |
1273 | struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; | |
1274 | struct fast_reg_descriptor *desc = reg->mem_h; | |
1275 | unsigned long sector_size = iser_task->sc->device->sector_size; | |
1276 | struct ib_mr_status mr_status; | |
1277 | int ret; | |
1278 | ||
1279 | if (desc && desc->reg_indicators & ISER_FASTREG_PROTECTED) { | |
1280 | desc->reg_indicators &= ~ISER_FASTREG_PROTECTED; | |
1281 | ret = ib_check_mr_status(desc->pi_ctx->sig_mr, | |
1282 | IB_MR_CHECK_SIG_STATUS, &mr_status); | |
1283 | if (ret) { | |
1284 | pr_err("ib_check_mr_status failed, ret %d\n", ret); | |
1285 | goto err; | |
1286 | } | |
1287 | ||
1288 | if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) { | |
1289 | sector_t sector_off = mr_status.sig_err.sig_err_offset; | |
1290 | ||
1291 | do_div(sector_off, sector_size + 8); | |
1292 | *sector = scsi_get_lba(iser_task->sc) + sector_off; | |
1293 | ||
39c978cd | 1294 | pr_err("PI error found type %d at sector %llx " |
0a7a08ad | 1295 | "expected %x vs actual %x\n", |
39c978cd RD |
1296 | mr_status.sig_err.err_type, |
1297 | (unsigned long long)*sector, | |
0a7a08ad SG |
1298 | mr_status.sig_err.expected, |
1299 | mr_status.sig_err.actual); | |
1300 | ||
1301 | switch (mr_status.sig_err.err_type) { | |
1302 | case IB_SIG_BAD_GUARD: | |
1303 | return 0x1; | |
1304 | case IB_SIG_BAD_REFTAG: | |
1305 | return 0x3; | |
1306 | case IB_SIG_BAD_APPTAG: | |
1307 | return 0x2; | |
1308 | } | |
1309 | } | |
1310 | } | |
1311 | ||
1312 | return 0; | |
1313 | err: | |
1314 | /* Not alot we can do here, return ambiguous guard error */ | |
1315 | return 0x1; | |
1316 | } |