1 /*
2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
3 * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33 #include <linux/module.h>
34 #include <linux/kernel.h>
35 #include <linux/slab.h>
36 #include <linux/mm.h>
37 #include <linux/highmem.h>
38 #include <linux/scatterlist.h>
39
40 #include "iscsi_iser.h"
41
iser_reg_comp(struct ib_cq * cq,struct ib_wc * wc)42 void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc)
43 {
44 iser_err_comp(wc, "memreg");
45 }
46
47 static struct iser_fr_desc *
iser_reg_desc_get_fr(struct ib_conn * ib_conn)48 iser_reg_desc_get_fr(struct ib_conn *ib_conn)
49 {
50 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
51 struct iser_fr_desc *desc;
52 unsigned long flags;
53
54 spin_lock_irqsave(&fr_pool->lock, flags);
55 desc = list_first_entry(&fr_pool->list,
56 struct iser_fr_desc, list);
57 list_del(&desc->list);
58 spin_unlock_irqrestore(&fr_pool->lock, flags);
59
60 return desc;
61 }
62
63 static void
iser_reg_desc_put_fr(struct ib_conn * ib_conn,struct iser_fr_desc * desc)64 iser_reg_desc_put_fr(struct ib_conn *ib_conn,
65 struct iser_fr_desc *desc)
66 {
67 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
68 unsigned long flags;
69
70 spin_lock_irqsave(&fr_pool->lock, flags);
71 list_add(&desc->list, &fr_pool->list);
72 spin_unlock_irqrestore(&fr_pool->lock, flags);
73 }
74
iser_dma_map_task_data(struct iscsi_iser_task * iser_task,struct iser_data_buf * data,enum iser_data_dir iser_dir,enum dma_data_direction dma_dir)75 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
76 struct iser_data_buf *data,
77 enum iser_data_dir iser_dir,
78 enum dma_data_direction dma_dir)
79 {
80 struct ib_device *dev;
81
82 iser_task->dir[iser_dir] = 1;
83 dev = iser_task->iser_conn->ib_conn.device->ib_device;
84
85 data->dma_nents = ib_dma_map_sg(dev, data->sg, data->size, dma_dir);
86 if (unlikely(data->dma_nents == 0)) {
87 iser_err("dma_map_sg failed!!!\n");
88 return -EINVAL;
89 }
90 return 0;
91 }
92
iser_dma_unmap_task_data(struct iscsi_iser_task * iser_task,struct iser_data_buf * data,enum dma_data_direction dir)93 void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task,
94 struct iser_data_buf *data,
95 enum dma_data_direction dir)
96 {
97 struct ib_device *dev;
98
99 dev = iser_task->iser_conn->ib_conn.device->ib_device;
100 ib_dma_unmap_sg(dev, data->sg, data->size, dir);
101 }
102
103 static int
iser_reg_dma(struct iser_device * device,struct iser_data_buf * mem,struct iser_mem_reg * reg)104 iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem,
105 struct iser_mem_reg *reg)
106 {
107 struct scatterlist *sg = mem->sg;
108
109 reg->sge.lkey = device->pd->local_dma_lkey;
110 /*
111 * FIXME: rework the registration code path to differentiate
112 * rkey/lkey use cases
113 */
114
115 if (device->pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)
116 reg->rkey = device->pd->unsafe_global_rkey;
117 else
118 reg->rkey = 0;
119 reg->sge.addr = sg_dma_address(&sg[0]);
120 reg->sge.length = sg_dma_len(&sg[0]);
121
122 iser_dbg("Single DMA entry: lkey=0x%x, rkey=0x%x, addr=0x%llx,"
123 " length=0x%x\n", reg->sge.lkey, reg->rkey,
124 reg->sge.addr, reg->sge.length);
125
126 return 0;
127 }
128
iser_unreg_mem_fastreg(struct iscsi_iser_task * iser_task,enum iser_data_dir cmd_dir)129 void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task,
130 enum iser_data_dir cmd_dir)
131 {
132 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
133 struct iser_fr_desc *desc;
134 struct ib_mr_status mr_status;
135
136 desc = reg->mem_h;
137 if (!desc)
138 return;
139
140 /*
141 * The signature MR cannot be invalidated and reused without checking.
142 * libiscsi calls the check_protection transport handler only if
143 * SCSI-Response is received. And the signature MR is not checked if
144 * the task is completed for some other reason like a timeout or error
145 * handling. That's why we must check the signature MR here before
146 * putting it to the free pool.
147 */
148 if (unlikely(desc->sig_protected)) {
149 desc->sig_protected = false;
150 ib_check_mr_status(desc->rsc.sig_mr, IB_MR_CHECK_SIG_STATUS,
151 &mr_status);
152 }
153 iser_reg_desc_put_fr(&iser_task->iser_conn->ib_conn, reg->mem_h);
154 reg->mem_h = NULL;
155 }
156
157 static void
iser_set_dif_domain(struct scsi_cmnd * sc,struct ib_sig_domain * domain)158 iser_set_dif_domain(struct scsi_cmnd *sc, struct ib_sig_domain *domain)
159 {
160 domain->sig_type = IB_SIG_TYPE_T10_DIF;
161 domain->sig.dif.pi_interval = scsi_prot_interval(sc);
162 domain->sig.dif.ref_tag = t10_pi_ref_tag(sc->request);
163 /*
164 * At the moment we hard code those, but in the future
165 * we will take them from sc.
166 */
167 domain->sig.dif.apptag_check_mask = 0xffff;
168 domain->sig.dif.app_escape = true;
169 domain->sig.dif.ref_escape = true;
170 if (sc->prot_flags & SCSI_PROT_REF_INCREMENT)
171 domain->sig.dif.ref_remap = true;
172 };
173
174 static int
iser_set_sig_attrs(struct scsi_cmnd * sc,struct ib_sig_attrs * sig_attrs)175 iser_set_sig_attrs(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs)
176 {
177 switch (scsi_get_prot_op(sc)) {
178 case SCSI_PROT_WRITE_INSERT:
179 case SCSI_PROT_READ_STRIP:
180 sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
181 iser_set_dif_domain(sc, &sig_attrs->wire);
182 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
183 break;
184 case SCSI_PROT_READ_INSERT:
185 case SCSI_PROT_WRITE_STRIP:
186 sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
187 iser_set_dif_domain(sc, &sig_attrs->mem);
188 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
189 IB_T10DIF_CSUM : IB_T10DIF_CRC;
190 break;
191 case SCSI_PROT_READ_PASS:
192 case SCSI_PROT_WRITE_PASS:
193 iser_set_dif_domain(sc, &sig_attrs->wire);
194 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
195 iser_set_dif_domain(sc, &sig_attrs->mem);
196 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
197 IB_T10DIF_CSUM : IB_T10DIF_CRC;
198 break;
199 default:
200 iser_err("Unsupported PI operation %d\n",
201 scsi_get_prot_op(sc));
202 return -EINVAL;
203 }
204
205 return 0;
206 }
207
208 static inline void
iser_set_prot_checks(struct scsi_cmnd * sc,u8 * mask)209 iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask)
210 {
211 *mask = 0;
212 if (sc->prot_flags & SCSI_PROT_REF_CHECK)
213 *mask |= IB_SIG_CHECK_REFTAG;
214 if (sc->prot_flags & SCSI_PROT_GUARD_CHECK)
215 *mask |= IB_SIG_CHECK_GUARD;
216 }
217
218 static inline void
iser_inv_rkey(struct ib_send_wr * inv_wr,struct ib_mr * mr,struct ib_cqe * cqe,struct ib_send_wr * next_wr)219 iser_inv_rkey(struct ib_send_wr *inv_wr,
220 struct ib_mr *mr,
221 struct ib_cqe *cqe,
222 struct ib_send_wr *next_wr)
223 {
224 inv_wr->opcode = IB_WR_LOCAL_INV;
225 inv_wr->wr_cqe = cqe;
226 inv_wr->ex.invalidate_rkey = mr->rkey;
227 inv_wr->send_flags = 0;
228 inv_wr->num_sge = 0;
229 inv_wr->next = next_wr;
230 }
231
232 static int
iser_reg_sig_mr(struct iscsi_iser_task * iser_task,struct iser_data_buf * mem,struct iser_data_buf * sig_mem,struct iser_reg_resources * rsc,struct iser_mem_reg * sig_reg)233 iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
234 struct iser_data_buf *mem,
235 struct iser_data_buf *sig_mem,
236 struct iser_reg_resources *rsc,
237 struct iser_mem_reg *sig_reg)
238 {
239 struct iser_tx_desc *tx_desc = &iser_task->desc;
240 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
241 struct ib_mr *mr = rsc->sig_mr;
242 struct ib_sig_attrs *sig_attrs = mr->sig_attrs;
243 struct ib_reg_wr *wr = &tx_desc->reg_wr;
244 int ret;
245
246 memset(sig_attrs, 0, sizeof(*sig_attrs));
247 ret = iser_set_sig_attrs(iser_task->sc, sig_attrs);
248 if (ret)
249 goto err;
250
251 iser_set_prot_checks(iser_task->sc, &sig_attrs->check_mask);
252
253 if (rsc->mr_valid)
254 iser_inv_rkey(&tx_desc->inv_wr, mr, cqe, &wr->wr);
255
256 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
257
258 ret = ib_map_mr_sg_pi(mr, mem->sg, mem->dma_nents, NULL,
259 sig_mem->sg, sig_mem->dma_nents, NULL, SZ_4K);
260 if (unlikely(ret)) {
261 iser_err("failed to map PI sg (%d)\n",
262 mem->dma_nents + sig_mem->dma_nents);
263 goto err;
264 }
265
266 memset(wr, 0, sizeof(*wr));
267 wr->wr.next = &tx_desc->send_wr;
268 wr->wr.opcode = IB_WR_REG_MR_INTEGRITY;
269 wr->wr.wr_cqe = cqe;
270 wr->wr.num_sge = 0;
271 wr->wr.send_flags = 0;
272 wr->mr = mr;
273 wr->key = mr->rkey;
274 wr->access = IB_ACCESS_LOCAL_WRITE |
275 IB_ACCESS_REMOTE_READ |
276 IB_ACCESS_REMOTE_WRITE;
277 rsc->mr_valid = 1;
278
279 sig_reg->sge.lkey = mr->lkey;
280 sig_reg->rkey = mr->rkey;
281 sig_reg->sge.addr = mr->iova;
282 sig_reg->sge.length = mr->length;
283
284 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=%u\n",
285 sig_reg->sge.lkey, sig_reg->rkey, sig_reg->sge.addr,
286 sig_reg->sge.length);
287 err:
288 return ret;
289 }
290
iser_fast_reg_mr(struct iscsi_iser_task * iser_task,struct iser_data_buf * mem,struct iser_reg_resources * rsc,struct iser_mem_reg * reg)291 static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
292 struct iser_data_buf *mem,
293 struct iser_reg_resources *rsc,
294 struct iser_mem_reg *reg)
295 {
296 struct iser_tx_desc *tx_desc = &iser_task->desc;
297 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
298 struct ib_mr *mr = rsc->mr;
299 struct ib_reg_wr *wr = &tx_desc->reg_wr;
300 int n;
301
302 if (rsc->mr_valid)
303 iser_inv_rkey(&tx_desc->inv_wr, mr, cqe, &wr->wr);
304
305 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
306
307 n = ib_map_mr_sg(mr, mem->sg, mem->dma_nents, NULL, SZ_4K);
308 if (unlikely(n != mem->dma_nents)) {
309 iser_err("failed to map sg (%d/%d)\n",
310 n, mem->dma_nents);
311 return n < 0 ? n : -EINVAL;
312 }
313
314 wr->wr.next = &tx_desc->send_wr;
315 wr->wr.opcode = IB_WR_REG_MR;
316 wr->wr.wr_cqe = cqe;
317 wr->wr.send_flags = 0;
318 wr->wr.num_sge = 0;
319 wr->mr = mr;
320 wr->key = mr->rkey;
321 wr->access = IB_ACCESS_LOCAL_WRITE |
322 IB_ACCESS_REMOTE_WRITE |
323 IB_ACCESS_REMOTE_READ;
324
325 rsc->mr_valid = 1;
326
327 reg->sge.lkey = mr->lkey;
328 reg->rkey = mr->rkey;
329 reg->sge.addr = mr->iova;
330 reg->sge.length = mr->length;
331
332 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=0x%x\n",
333 reg->sge.lkey, reg->rkey, reg->sge.addr, reg->sge.length);
334
335 return 0;
336 }
337
338 static int
iser_reg_data_sg(struct iscsi_iser_task * task,struct iser_data_buf * mem,struct iser_fr_desc * desc,bool use_dma_key,struct iser_mem_reg * reg)339 iser_reg_data_sg(struct iscsi_iser_task *task,
340 struct iser_data_buf *mem,
341 struct iser_fr_desc *desc,
342 bool use_dma_key,
343 struct iser_mem_reg *reg)
344 {
345 struct iser_device *device = task->iser_conn->ib_conn.device;
346
347 if (use_dma_key)
348 return iser_reg_dma(device, mem, reg);
349
350 return iser_fast_reg_mr(task, mem, &desc->rsc, reg);
351 }
352
iser_reg_mem_fastreg(struct iscsi_iser_task * task,enum iser_data_dir dir,bool all_imm)353 int iser_reg_mem_fastreg(struct iscsi_iser_task *task,
354 enum iser_data_dir dir,
355 bool all_imm)
356 {
357 struct ib_conn *ib_conn = &task->iser_conn->ib_conn;
358 struct iser_data_buf *mem = &task->data[dir];
359 struct iser_mem_reg *reg = &task->rdma_reg[dir];
360 struct iser_fr_desc *desc = NULL;
361 bool use_dma_key;
362 int err;
363
364 use_dma_key = mem->dma_nents == 1 && (all_imm || !iser_always_reg) &&
365 scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL;
366
367 if (!use_dma_key) {
368 desc = iser_reg_desc_get_fr(ib_conn);
369 reg->mem_h = desc;
370 }
371
372 if (scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL) {
373 err = iser_reg_data_sg(task, mem, desc, use_dma_key, reg);
374 if (unlikely(err))
375 goto err_reg;
376 } else {
377 err = iser_reg_sig_mr(task, mem, &task->prot[dir],
378 &desc->rsc, reg);
379 if (unlikely(err))
380 goto err_reg;
381
382 desc->sig_protected = true;
383 }
384
385 return 0;
386
387 err_reg:
388 if (desc)
389 iser_reg_desc_put_fr(ib_conn, desc);
390
391 return err;
392 }
393
394