]> git.proxmox.com Git - mirror_ubuntu-kernels.git/blob - drivers/infiniband/ulp/iser/iser_memory.c
Merge branches 'for-5.1/upstream-fixes', 'for-5.2/core', 'for-5.2/ish', 'for-5.2...
[mirror_ubuntu-kernels.git] / drivers / infiniband / ulp / iser / iser_memory.c
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 static
42 int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task,
43 struct iser_data_buf *mem,
44 struct iser_reg_resources *rsc,
45 struct iser_mem_reg *mem_reg);
46 static
47 int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
48 struct iser_data_buf *mem,
49 struct iser_reg_resources *rsc,
50 struct iser_mem_reg *mem_reg);
51
52 static const struct iser_reg_ops fastreg_ops = {
53 .alloc_reg_res = iser_alloc_fastreg_pool,
54 .free_reg_res = iser_free_fastreg_pool,
55 .reg_mem = iser_fast_reg_mr,
56 .unreg_mem = iser_unreg_mem_fastreg,
57 .reg_desc_get = iser_reg_desc_get_fr,
58 .reg_desc_put = iser_reg_desc_put_fr,
59 };
60
61 static const struct iser_reg_ops fmr_ops = {
62 .alloc_reg_res = iser_alloc_fmr_pool,
63 .free_reg_res = iser_free_fmr_pool,
64 .reg_mem = iser_fast_reg_fmr,
65 .unreg_mem = iser_unreg_mem_fmr,
66 .reg_desc_get = iser_reg_desc_get_fmr,
67 .reg_desc_put = iser_reg_desc_put_fmr,
68 };
69
70 void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc)
71 {
72 iser_err_comp(wc, "memreg");
73 }
74
75 int iser_assign_reg_ops(struct iser_device *device)
76 {
77 struct ib_device *ib_dev = device->ib_device;
78
79 /* Assign function handles - based on FMR support */
80 if (ib_dev->ops.alloc_fmr && ib_dev->ops.dealloc_fmr &&
81 ib_dev->ops.map_phys_fmr && ib_dev->ops.unmap_fmr) {
82 iser_info("FMR supported, using FMR for registration\n");
83 device->reg_ops = &fmr_ops;
84 } else if (ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) {
85 iser_info("FastReg supported, using FastReg for registration\n");
86 device->reg_ops = &fastreg_ops;
87 device->remote_inv_sup = iser_always_reg;
88 } else {
89 iser_err("IB device does not support FMRs nor FastRegs, can't register memory\n");
90 return -1;
91 }
92
93 return 0;
94 }
95
96 struct iser_fr_desc *
97 iser_reg_desc_get_fr(struct ib_conn *ib_conn)
98 {
99 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
100 struct iser_fr_desc *desc;
101 unsigned long flags;
102
103 spin_lock_irqsave(&fr_pool->lock, flags);
104 desc = list_first_entry(&fr_pool->list,
105 struct iser_fr_desc, list);
106 list_del(&desc->list);
107 spin_unlock_irqrestore(&fr_pool->lock, flags);
108
109 return desc;
110 }
111
112 void
113 iser_reg_desc_put_fr(struct ib_conn *ib_conn,
114 struct iser_fr_desc *desc)
115 {
116 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
117 unsigned long flags;
118
119 spin_lock_irqsave(&fr_pool->lock, flags);
120 list_add(&desc->list, &fr_pool->list);
121 spin_unlock_irqrestore(&fr_pool->lock, flags);
122 }
123
124 struct iser_fr_desc *
125 iser_reg_desc_get_fmr(struct ib_conn *ib_conn)
126 {
127 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
128
129 return list_first_entry(&fr_pool->list,
130 struct iser_fr_desc, list);
131 }
132
133 void
134 iser_reg_desc_put_fmr(struct ib_conn *ib_conn,
135 struct iser_fr_desc *desc)
136 {
137 }
138
139 static void iser_data_buf_dump(struct iser_data_buf *data,
140 struct ib_device *ibdev)
141 {
142 struct scatterlist *sg;
143 int i;
144
145 for_each_sg(data->sg, sg, data->dma_nents, i)
146 iser_dbg("sg[%d] dma_addr:0x%lX page:0x%p "
147 "off:0x%x sz:0x%x dma_len:0x%x\n",
148 i, (unsigned long)sg_dma_address(sg),
149 sg_page(sg), sg->offset, sg->length, sg_dma_len(sg));
150 }
151
152 static void iser_dump_page_vec(struct iser_page_vec *page_vec)
153 {
154 int i;
155
156 iser_err("page vec npages %d data length %lld\n",
157 page_vec->npages, page_vec->fake_mr.length);
158 for (i = 0; i < page_vec->npages; i++)
159 iser_err("vec[%d]: %llx\n", i, page_vec->pages[i]);
160 }
161
162 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
163 struct iser_data_buf *data,
164 enum iser_data_dir iser_dir,
165 enum dma_data_direction dma_dir)
166 {
167 struct ib_device *dev;
168
169 iser_task->dir[iser_dir] = 1;
170 dev = iser_task->iser_conn->ib_conn.device->ib_device;
171
172 data->dma_nents = ib_dma_map_sg(dev, data->sg, data->size, dma_dir);
173 if (data->dma_nents == 0) {
174 iser_err("dma_map_sg failed!!!\n");
175 return -EINVAL;
176 }
177 return 0;
178 }
179
180 void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task,
181 struct iser_data_buf *data,
182 enum dma_data_direction dir)
183 {
184 struct ib_device *dev;
185
186 dev = iser_task->iser_conn->ib_conn.device->ib_device;
187 ib_dma_unmap_sg(dev, data->sg, data->size, dir);
188 }
189
190 static int
191 iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem,
192 struct iser_mem_reg *reg)
193 {
194 struct scatterlist *sg = mem->sg;
195
196 reg->sge.lkey = device->pd->local_dma_lkey;
197 /*
198 * FIXME: rework the registration code path to differentiate
199 * rkey/lkey use cases
200 */
201
202 if (device->pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)
203 reg->rkey = device->pd->unsafe_global_rkey;
204 else
205 reg->rkey = 0;
206 reg->sge.addr = sg_dma_address(&sg[0]);
207 reg->sge.length = sg_dma_len(&sg[0]);
208
209 iser_dbg("Single DMA entry: lkey=0x%x, rkey=0x%x, addr=0x%llx,"
210 " length=0x%x\n", reg->sge.lkey, reg->rkey,
211 reg->sge.addr, reg->sge.length);
212
213 return 0;
214 }
215
216 static int iser_set_page(struct ib_mr *mr, u64 addr)
217 {
218 struct iser_page_vec *page_vec =
219 container_of(mr, struct iser_page_vec, fake_mr);
220
221 page_vec->pages[page_vec->npages++] = addr;
222
223 return 0;
224 }
225
226 static
227 int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task,
228 struct iser_data_buf *mem,
229 struct iser_reg_resources *rsc,
230 struct iser_mem_reg *reg)
231 {
232 struct ib_conn *ib_conn = &iser_task->iser_conn->ib_conn;
233 struct iser_device *device = ib_conn->device;
234 struct iser_page_vec *page_vec = rsc->page_vec;
235 struct ib_fmr_pool *fmr_pool = rsc->fmr_pool;
236 struct ib_pool_fmr *fmr;
237 int ret, plen;
238
239 page_vec->npages = 0;
240 page_vec->fake_mr.page_size = SIZE_4K;
241 plen = ib_sg_to_pages(&page_vec->fake_mr, mem->sg,
242 mem->dma_nents, NULL, iser_set_page);
243 if (unlikely(plen < mem->dma_nents)) {
244 iser_err("page vec too short to hold this SG\n");
245 iser_data_buf_dump(mem, device->ib_device);
246 iser_dump_page_vec(page_vec);
247 return -EINVAL;
248 }
249
250 fmr = ib_fmr_pool_map_phys(fmr_pool, page_vec->pages,
251 page_vec->npages, page_vec->pages[0]);
252 if (IS_ERR(fmr)) {
253 ret = PTR_ERR(fmr);
254 iser_err("ib_fmr_pool_map_phys failed: %d\n", ret);
255 return ret;
256 }
257
258 reg->sge.lkey = fmr->fmr->lkey;
259 reg->rkey = fmr->fmr->rkey;
260 reg->sge.addr = page_vec->fake_mr.iova;
261 reg->sge.length = page_vec->fake_mr.length;
262 reg->mem_h = fmr;
263
264 iser_dbg("fmr reg: lkey=0x%x, rkey=0x%x, addr=0x%llx,"
265 " length=0x%x\n", reg->sge.lkey, reg->rkey,
266 reg->sge.addr, reg->sge.length);
267
268 return 0;
269 }
270
271 /**
272 * Unregister (previosuly registered using FMR) memory.
273 * If memory is non-FMR does nothing.
274 */
275 void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task,
276 enum iser_data_dir cmd_dir)
277 {
278 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
279
280 if (!reg->mem_h)
281 return;
282
283 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n", reg->mem_h);
284
285 ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h);
286
287 reg->mem_h = NULL;
288 }
289
290 void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task,
291 enum iser_data_dir cmd_dir)
292 {
293 struct iser_device *device = iser_task->iser_conn->ib_conn.device;
294 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
295
296 if (!reg->mem_h)
297 return;
298
299 device->reg_ops->reg_desc_put(&iser_task->iser_conn->ib_conn,
300 reg->mem_h);
301 reg->mem_h = NULL;
302 }
303
304 static void
305 iser_set_dif_domain(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs,
306 struct ib_sig_domain *domain)
307 {
308 domain->sig_type = IB_SIG_TYPE_T10_DIF;
309 domain->sig.dif.pi_interval = scsi_prot_interval(sc);
310 domain->sig.dif.ref_tag = t10_pi_ref_tag(sc->request);
311 /*
312 * At the moment we hard code those, but in the future
313 * we will take them from sc.
314 */
315 domain->sig.dif.apptag_check_mask = 0xffff;
316 domain->sig.dif.app_escape = true;
317 domain->sig.dif.ref_escape = true;
318 if (sc->prot_flags & SCSI_PROT_REF_INCREMENT)
319 domain->sig.dif.ref_remap = true;
320 };
321
322 static int
323 iser_set_sig_attrs(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs)
324 {
325 switch (scsi_get_prot_op(sc)) {
326 case SCSI_PROT_WRITE_INSERT:
327 case SCSI_PROT_READ_STRIP:
328 sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
329 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->wire);
330 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
331 break;
332 case SCSI_PROT_READ_INSERT:
333 case SCSI_PROT_WRITE_STRIP:
334 sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
335 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem);
336 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
337 IB_T10DIF_CSUM : IB_T10DIF_CRC;
338 break;
339 case SCSI_PROT_READ_PASS:
340 case SCSI_PROT_WRITE_PASS:
341 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->wire);
342 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
343 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem);
344 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
345 IB_T10DIF_CSUM : IB_T10DIF_CRC;
346 break;
347 default:
348 iser_err("Unsupported PI operation %d\n",
349 scsi_get_prot_op(sc));
350 return -EINVAL;
351 }
352
353 return 0;
354 }
355
356 static inline void
357 iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask)
358 {
359 *mask = 0;
360 if (sc->prot_flags & SCSI_PROT_REF_CHECK)
361 *mask |= IB_SIG_CHECK_REFTAG;
362 if (sc->prot_flags & SCSI_PROT_GUARD_CHECK)
363 *mask |= IB_SIG_CHECK_GUARD;
364 }
365
366 static inline void
367 iser_inv_rkey(struct ib_send_wr *inv_wr,
368 struct ib_mr *mr,
369 struct ib_cqe *cqe)
370 {
371 inv_wr->opcode = IB_WR_LOCAL_INV;
372 inv_wr->wr_cqe = cqe;
373 inv_wr->ex.invalidate_rkey = mr->rkey;
374 inv_wr->send_flags = 0;
375 inv_wr->num_sge = 0;
376 }
377
378 static int
379 iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
380 struct iser_pi_context *pi_ctx,
381 struct iser_mem_reg *data_reg,
382 struct iser_mem_reg *prot_reg,
383 struct iser_mem_reg *sig_reg)
384 {
385 struct iser_tx_desc *tx_desc = &iser_task->desc;
386 struct ib_sig_attrs *sig_attrs = &tx_desc->sig_attrs;
387 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
388 struct ib_sig_handover_wr *wr;
389 struct ib_mr *mr = pi_ctx->sig_mr;
390 int ret;
391
392 memset(sig_attrs, 0, sizeof(*sig_attrs));
393 ret = iser_set_sig_attrs(iser_task->sc, sig_attrs);
394 if (ret)
395 goto err;
396
397 iser_set_prot_checks(iser_task->sc, &sig_attrs->check_mask);
398
399 if (pi_ctx->sig_mr_valid)
400 iser_inv_rkey(iser_tx_next_wr(tx_desc), mr, cqe);
401
402 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
403
404 wr = container_of(iser_tx_next_wr(tx_desc), struct ib_sig_handover_wr,
405 wr);
406 wr->wr.opcode = IB_WR_REG_SIG_MR;
407 wr->wr.wr_cqe = cqe;
408 wr->wr.sg_list = &data_reg->sge;
409 wr->wr.num_sge = 1;
410 wr->wr.send_flags = 0;
411 wr->sig_attrs = sig_attrs;
412 wr->sig_mr = mr;
413 if (scsi_prot_sg_count(iser_task->sc))
414 wr->prot = &prot_reg->sge;
415 else
416 wr->prot = NULL;
417 wr->access_flags = IB_ACCESS_LOCAL_WRITE |
418 IB_ACCESS_REMOTE_READ |
419 IB_ACCESS_REMOTE_WRITE;
420 pi_ctx->sig_mr_valid = 1;
421
422 sig_reg->sge.lkey = mr->lkey;
423 sig_reg->rkey = mr->rkey;
424 sig_reg->sge.addr = 0;
425 sig_reg->sge.length = scsi_transfer_length(iser_task->sc);
426
427 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=%u\n",
428 sig_reg->sge.lkey, sig_reg->rkey, sig_reg->sge.addr,
429 sig_reg->sge.length);
430 err:
431 return ret;
432 }
433
434 static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
435 struct iser_data_buf *mem,
436 struct iser_reg_resources *rsc,
437 struct iser_mem_reg *reg)
438 {
439 struct iser_tx_desc *tx_desc = &iser_task->desc;
440 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
441 struct ib_mr *mr = rsc->mr;
442 struct ib_reg_wr *wr;
443 int n;
444
445 if (rsc->mr_valid)
446 iser_inv_rkey(iser_tx_next_wr(tx_desc), mr, cqe);
447
448 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
449
450 n = ib_map_mr_sg(mr, mem->sg, mem->dma_nents, NULL, SIZE_4K);
451 if (unlikely(n != mem->dma_nents)) {
452 iser_err("failed to map sg (%d/%d)\n",
453 n, mem->dma_nents);
454 return n < 0 ? n : -EINVAL;
455 }
456
457 wr = container_of(iser_tx_next_wr(tx_desc), struct ib_reg_wr, wr);
458 wr->wr.opcode = IB_WR_REG_MR;
459 wr->wr.wr_cqe = cqe;
460 wr->wr.send_flags = 0;
461 wr->wr.num_sge = 0;
462 wr->mr = mr;
463 wr->key = mr->rkey;
464 wr->access = IB_ACCESS_LOCAL_WRITE |
465 IB_ACCESS_REMOTE_WRITE |
466 IB_ACCESS_REMOTE_READ;
467
468 rsc->mr_valid = 1;
469
470 reg->sge.lkey = mr->lkey;
471 reg->rkey = mr->rkey;
472 reg->sge.addr = mr->iova;
473 reg->sge.length = mr->length;
474
475 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=0x%x\n",
476 reg->sge.lkey, reg->rkey, reg->sge.addr, reg->sge.length);
477
478 return 0;
479 }
480
481 static int
482 iser_reg_prot_sg(struct iscsi_iser_task *task,
483 struct iser_data_buf *mem,
484 struct iser_fr_desc *desc,
485 bool use_dma_key,
486 struct iser_mem_reg *reg)
487 {
488 struct iser_device *device = task->iser_conn->ib_conn.device;
489
490 if (use_dma_key)
491 return iser_reg_dma(device, mem, reg);
492
493 return device->reg_ops->reg_mem(task, mem, &desc->pi_ctx->rsc, reg);
494 }
495
496 static int
497 iser_reg_data_sg(struct iscsi_iser_task *task,
498 struct iser_data_buf *mem,
499 struct iser_fr_desc *desc,
500 bool use_dma_key,
501 struct iser_mem_reg *reg)
502 {
503 struct iser_device *device = task->iser_conn->ib_conn.device;
504
505 if (use_dma_key)
506 return iser_reg_dma(device, mem, reg);
507
508 return device->reg_ops->reg_mem(task, mem, &desc->rsc, reg);
509 }
510
511 int iser_reg_rdma_mem(struct iscsi_iser_task *task,
512 enum iser_data_dir dir,
513 bool all_imm)
514 {
515 struct ib_conn *ib_conn = &task->iser_conn->ib_conn;
516 struct iser_device *device = ib_conn->device;
517 struct iser_data_buf *mem = &task->data[dir];
518 struct iser_mem_reg *reg = &task->rdma_reg[dir];
519 struct iser_mem_reg *data_reg;
520 struct iser_fr_desc *desc = NULL;
521 bool use_dma_key;
522 int err;
523
524 use_dma_key = mem->dma_nents == 1 && (all_imm || !iser_always_reg) &&
525 scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL;
526
527 if (!use_dma_key) {
528 desc = device->reg_ops->reg_desc_get(ib_conn);
529 reg->mem_h = desc;
530 }
531
532 if (scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL)
533 data_reg = reg;
534 else
535 data_reg = &task->desc.data_reg;
536
537 err = iser_reg_data_sg(task, mem, desc, use_dma_key, data_reg);
538 if (unlikely(err))
539 goto err_reg;
540
541 if (scsi_get_prot_op(task->sc) != SCSI_PROT_NORMAL) {
542 struct iser_mem_reg *prot_reg = &task->desc.prot_reg;
543
544 if (scsi_prot_sg_count(task->sc)) {
545 mem = &task->prot[dir];
546 err = iser_reg_prot_sg(task, mem, desc,
547 use_dma_key, prot_reg);
548 if (unlikely(err))
549 goto err_reg;
550 }
551
552 err = iser_reg_sig_mr(task, desc->pi_ctx, data_reg,
553 prot_reg, reg);
554 if (unlikely(err))
555 goto err_reg;
556
557 desc->pi_ctx->sig_protected = 1;
558 }
559
560 return 0;
561
562 err_reg:
563 if (desc)
564 device->reg_ops->reg_desc_put(ib_conn, desc);
565
566 return err;
567 }
568
569 void iser_unreg_rdma_mem(struct iscsi_iser_task *task,
570 enum iser_data_dir dir)
571 {
572 struct iser_device *device = task->iser_conn->ib_conn.device;
573
574 device->reg_ops->unreg_mem(task, dir);
575 }