Loading...
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"
41static
42int 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);
46static
47int 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
52static 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
61static 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
70void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc)
71{
72 iser_err_comp(wc, "memreg");
73}
74
75int 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->alloc_fmr && ib_dev->dealloc_fmr &&
81 ib_dev->map_phys_fmr && ib_dev->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
96struct iser_fr_desc *
97iser_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
112void
113iser_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
124struct iser_fr_desc *
125iser_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
133void
134iser_reg_desc_put_fmr(struct ib_conn *ib_conn,
135 struct iser_fr_desc *desc)
136{
137}
138
139static 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)ib_sg_dma_address(ibdev, sg),
149 sg_page(sg), sg->offset,
150 sg->length, ib_sg_dma_len(ibdev, sg));
151}
152
153static void iser_dump_page_vec(struct iser_page_vec *page_vec)
154{
155 int i;
156
157 iser_err("page vec npages %d data length %d\n",
158 page_vec->npages, page_vec->fake_mr.length);
159 for (i = 0; i < page_vec->npages; i++)
160 iser_err("vec[%d]: %llx\n", i, page_vec->pages[i]);
161}
162
163int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
164 struct iser_data_buf *data,
165 enum iser_data_dir iser_dir,
166 enum dma_data_direction dma_dir)
167{
168 struct ib_device *dev;
169
170 iser_task->dir[iser_dir] = 1;
171 dev = iser_task->iser_conn->ib_conn.device->ib_device;
172
173 data->dma_nents = ib_dma_map_sg(dev, data->sg, data->size, dma_dir);
174 if (data->dma_nents == 0) {
175 iser_err("dma_map_sg failed!!!\n");
176 return -EINVAL;
177 }
178 return 0;
179}
180
181void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task,
182 struct iser_data_buf *data,
183 enum dma_data_direction dir)
184{
185 struct ib_device *dev;
186
187 dev = iser_task->iser_conn->ib_conn.device->ib_device;
188 ib_dma_unmap_sg(dev, data->sg, data->size, dir);
189}
190
191static int
192iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem,
193 struct iser_mem_reg *reg)
194{
195 struct scatterlist *sg = mem->sg;
196
197 reg->sge.lkey = device->pd->local_dma_lkey;
198 /*
199 * FIXME: rework the registration code path to differentiate
200 * rkey/lkey use cases
201 */
202
203 if (device->pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)
204 reg->rkey = device->pd->unsafe_global_rkey;
205 else
206 reg->rkey = 0;
207 reg->sge.addr = ib_sg_dma_address(device->ib_device, &sg[0]);
208 reg->sge.length = ib_sg_dma_len(device->ib_device, &sg[0]);
209
210 iser_dbg("Single DMA entry: lkey=0x%x, rkey=0x%x, addr=0x%llx,"
211 " length=0x%x\n", reg->sge.lkey, reg->rkey,
212 reg->sge.addr, reg->sge.length);
213
214 return 0;
215}
216
217static int iser_set_page(struct ib_mr *mr, u64 addr)
218{
219 struct iser_page_vec *page_vec =
220 container_of(mr, struct iser_page_vec, fake_mr);
221
222 page_vec->pages[page_vec->npages++] = addr;
223
224 return 0;
225}
226
227static
228int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task,
229 struct iser_data_buf *mem,
230 struct iser_reg_resources *rsc,
231 struct iser_mem_reg *reg)
232{
233 struct ib_conn *ib_conn = &iser_task->iser_conn->ib_conn;
234 struct iser_device *device = ib_conn->device;
235 struct iser_page_vec *page_vec = rsc->page_vec;
236 struct ib_fmr_pool *fmr_pool = rsc->fmr_pool;
237 struct ib_pool_fmr *fmr;
238 int ret, plen;
239
240 page_vec->npages = 0;
241 page_vec->fake_mr.page_size = SIZE_4K;
242 plen = ib_sg_to_pages(&page_vec->fake_mr, mem->sg,
243 mem->size, NULL, iser_set_page);
244 if (unlikely(plen < mem->size)) {
245 iser_err("page vec too short to hold this SG\n");
246 iser_data_buf_dump(mem, device->ib_device);
247 iser_dump_page_vec(page_vec);
248 return -EINVAL;
249 }
250
251 fmr = ib_fmr_pool_map_phys(fmr_pool, page_vec->pages,
252 page_vec->npages, page_vec->pages[0]);
253 if (IS_ERR(fmr)) {
254 ret = PTR_ERR(fmr);
255 iser_err("ib_fmr_pool_map_phys failed: %d\n", ret);
256 return ret;
257 }
258
259 reg->sge.lkey = fmr->fmr->lkey;
260 reg->rkey = fmr->fmr->rkey;
261 reg->sge.addr = page_vec->fake_mr.iova;
262 reg->sge.length = page_vec->fake_mr.length;
263 reg->mem_h = fmr;
264
265 iser_dbg("fmr reg: lkey=0x%x, rkey=0x%x, addr=0x%llx,"
266 " length=0x%x\n", reg->sge.lkey, reg->rkey,
267 reg->sge.addr, reg->sge.length);
268
269 return 0;
270}
271
272/**
273 * Unregister (previosuly registered using FMR) memory.
274 * If memory is non-FMR does nothing.
275 */
276void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task,
277 enum iser_data_dir cmd_dir)
278{
279 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
280 int ret;
281
282 if (!reg->mem_h)
283 return;
284
285 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n", reg->mem_h);
286
287 ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h);
288 if (ret)
289 iser_err("ib_fmr_pool_unmap failed %d\n", ret);
290
291 reg->mem_h = NULL;
292}
293
294void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task,
295 enum iser_data_dir cmd_dir)
296{
297 struct iser_device *device = iser_task->iser_conn->ib_conn.device;
298 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
299
300 if (!reg->mem_h)
301 return;
302
303 device->reg_ops->reg_desc_put(&iser_task->iser_conn->ib_conn,
304 reg->mem_h);
305 reg->mem_h = NULL;
306}
307
308static void
309iser_set_dif_domain(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs,
310 struct ib_sig_domain *domain)
311{
312 domain->sig_type = IB_SIG_TYPE_T10_DIF;
313 domain->sig.dif.pi_interval = scsi_prot_interval(sc);
314 domain->sig.dif.ref_tag = scsi_prot_ref_tag(sc);
315 /*
316 * At the moment we hard code those, but in the future
317 * we will take them from sc.
318 */
319 domain->sig.dif.apptag_check_mask = 0xffff;
320 domain->sig.dif.app_escape = true;
321 domain->sig.dif.ref_escape = true;
322 if (sc->prot_flags & SCSI_PROT_REF_INCREMENT)
323 domain->sig.dif.ref_remap = true;
324};
325
326static int
327iser_set_sig_attrs(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs)
328{
329 switch (scsi_get_prot_op(sc)) {
330 case SCSI_PROT_WRITE_INSERT:
331 case SCSI_PROT_READ_STRIP:
332 sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
333 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->wire);
334 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
335 break;
336 case SCSI_PROT_READ_INSERT:
337 case SCSI_PROT_WRITE_STRIP:
338 sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
339 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem);
340 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
341 IB_T10DIF_CSUM : IB_T10DIF_CRC;
342 break;
343 case SCSI_PROT_READ_PASS:
344 case SCSI_PROT_WRITE_PASS:
345 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->wire);
346 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
347 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem);
348 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
349 IB_T10DIF_CSUM : IB_T10DIF_CRC;
350 break;
351 default:
352 iser_err("Unsupported PI operation %d\n",
353 scsi_get_prot_op(sc));
354 return -EINVAL;
355 }
356
357 return 0;
358}
359
360static inline void
361iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask)
362{
363 *mask = 0;
364 if (sc->prot_flags & SCSI_PROT_REF_CHECK)
365 *mask |= ISER_CHECK_REFTAG;
366 if (sc->prot_flags & SCSI_PROT_GUARD_CHECK)
367 *mask |= ISER_CHECK_GUARD;
368}
369
370static inline void
371iser_inv_rkey(struct ib_send_wr *inv_wr,
372 struct ib_mr *mr,
373 struct ib_cqe *cqe)
374{
375 inv_wr->opcode = IB_WR_LOCAL_INV;
376 inv_wr->wr_cqe = cqe;
377 inv_wr->ex.invalidate_rkey = mr->rkey;
378 inv_wr->send_flags = 0;
379 inv_wr->num_sge = 0;
380}
381
382static int
383iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
384 struct iser_pi_context *pi_ctx,
385 struct iser_mem_reg *data_reg,
386 struct iser_mem_reg *prot_reg,
387 struct iser_mem_reg *sig_reg)
388{
389 struct iser_tx_desc *tx_desc = &iser_task->desc;
390 struct ib_sig_attrs *sig_attrs = &tx_desc->sig_attrs;
391 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
392 struct ib_sig_handover_wr *wr;
393 struct ib_mr *mr = pi_ctx->sig_mr;
394 int ret;
395
396 memset(sig_attrs, 0, sizeof(*sig_attrs));
397 ret = iser_set_sig_attrs(iser_task->sc, sig_attrs);
398 if (ret)
399 goto err;
400
401 iser_set_prot_checks(iser_task->sc, &sig_attrs->check_mask);
402
403 if (pi_ctx->sig_mr_valid)
404 iser_inv_rkey(iser_tx_next_wr(tx_desc), mr, cqe);
405
406 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
407
408 wr = sig_handover_wr(iser_tx_next_wr(tx_desc));
409 wr->wr.opcode = IB_WR_REG_SIG_MR;
410 wr->wr.wr_cqe = cqe;
411 wr->wr.sg_list = &data_reg->sge;
412 wr->wr.num_sge = 1;
413 wr->wr.send_flags = 0;
414 wr->sig_attrs = sig_attrs;
415 wr->sig_mr = mr;
416 if (scsi_prot_sg_count(iser_task->sc))
417 wr->prot = &prot_reg->sge;
418 else
419 wr->prot = NULL;
420 wr->access_flags = IB_ACCESS_LOCAL_WRITE |
421 IB_ACCESS_REMOTE_READ |
422 IB_ACCESS_REMOTE_WRITE;
423 pi_ctx->sig_mr_valid = 1;
424
425 sig_reg->sge.lkey = mr->lkey;
426 sig_reg->rkey = mr->rkey;
427 sig_reg->sge.addr = 0;
428 sig_reg->sge.length = scsi_transfer_length(iser_task->sc);
429
430 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=%u\n",
431 sig_reg->sge.lkey, sig_reg->rkey, sig_reg->sge.addr,
432 sig_reg->sge.length);
433err:
434 return ret;
435}
436
437static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
438 struct iser_data_buf *mem,
439 struct iser_reg_resources *rsc,
440 struct iser_mem_reg *reg)
441{
442 struct iser_tx_desc *tx_desc = &iser_task->desc;
443 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
444 struct ib_mr *mr = rsc->mr;
445 struct ib_reg_wr *wr;
446 int n;
447
448 if (rsc->mr_valid)
449 iser_inv_rkey(iser_tx_next_wr(tx_desc), mr, cqe);
450
451 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
452
453 n = ib_map_mr_sg(mr, mem->sg, mem->size, NULL, SIZE_4K);
454 if (unlikely(n != mem->size)) {
455 iser_err("failed to map sg (%d/%d)\n",
456 n, mem->size);
457 return n < 0 ? n : -EINVAL;
458 }
459
460 wr = reg_wr(iser_tx_next_wr(tx_desc));
461 wr->wr.opcode = IB_WR_REG_MR;
462 wr->wr.wr_cqe = cqe;
463 wr->wr.send_flags = 0;
464 wr->wr.num_sge = 0;
465 wr->mr = mr;
466 wr->key = mr->rkey;
467 wr->access = IB_ACCESS_LOCAL_WRITE |
468 IB_ACCESS_REMOTE_WRITE |
469 IB_ACCESS_REMOTE_READ;
470
471 rsc->mr_valid = 1;
472
473 reg->sge.lkey = mr->lkey;
474 reg->rkey = mr->rkey;
475 reg->sge.addr = mr->iova;
476 reg->sge.length = mr->length;
477
478 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=0x%x\n",
479 reg->sge.lkey, reg->rkey, reg->sge.addr, reg->sge.length);
480
481 return 0;
482}
483
484static int
485iser_reg_prot_sg(struct iscsi_iser_task *task,
486 struct iser_data_buf *mem,
487 struct iser_fr_desc *desc,
488 bool use_dma_key,
489 struct iser_mem_reg *reg)
490{
491 struct iser_device *device = task->iser_conn->ib_conn.device;
492
493 if (use_dma_key)
494 return iser_reg_dma(device, mem, reg);
495
496 return device->reg_ops->reg_mem(task, mem, &desc->pi_ctx->rsc, reg);
497}
498
499static int
500iser_reg_data_sg(struct iscsi_iser_task *task,
501 struct iser_data_buf *mem,
502 struct iser_fr_desc *desc,
503 bool use_dma_key,
504 struct iser_mem_reg *reg)
505{
506 struct iser_device *device = task->iser_conn->ib_conn.device;
507
508 if (use_dma_key)
509 return iser_reg_dma(device, mem, reg);
510
511 return device->reg_ops->reg_mem(task, mem, &desc->rsc, reg);
512}
513
514int iser_reg_rdma_mem(struct iscsi_iser_task *task,
515 enum iser_data_dir dir,
516 bool all_imm)
517{
518 struct ib_conn *ib_conn = &task->iser_conn->ib_conn;
519 struct iser_device *device = ib_conn->device;
520 struct iser_data_buf *mem = &task->data[dir];
521 struct iser_mem_reg *reg = &task->rdma_reg[dir];
522 struct iser_mem_reg *data_reg;
523 struct iser_fr_desc *desc = NULL;
524 bool use_dma_key;
525 int err;
526
527 use_dma_key = mem->dma_nents == 1 && (all_imm || !iser_always_reg) &&
528 scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL;
529
530 if (!use_dma_key) {
531 desc = device->reg_ops->reg_desc_get(ib_conn);
532 reg->mem_h = desc;
533 }
534
535 if (scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL)
536 data_reg = reg;
537 else
538 data_reg = &task->desc.data_reg;
539
540 err = iser_reg_data_sg(task, mem, desc, use_dma_key, data_reg);
541 if (unlikely(err))
542 goto err_reg;
543
544 if (scsi_get_prot_op(task->sc) != SCSI_PROT_NORMAL) {
545 struct iser_mem_reg *prot_reg = &task->desc.prot_reg;
546
547 if (scsi_prot_sg_count(task->sc)) {
548 mem = &task->prot[dir];
549 err = iser_reg_prot_sg(task, mem, desc,
550 use_dma_key, prot_reg);
551 if (unlikely(err))
552 goto err_reg;
553 }
554
555 err = iser_reg_sig_mr(task, desc->pi_ctx, data_reg,
556 prot_reg, reg);
557 if (unlikely(err))
558 goto err_reg;
559
560 desc->pi_ctx->sig_protected = 1;
561 }
562
563 return 0;
564
565err_reg:
566 if (desc)
567 device->reg_ops->reg_desc_put(ib_conn, desc);
568
569 return err;
570}
571
572void iser_unreg_rdma_mem(struct iscsi_iser_task *task,
573 enum iser_data_dir dir)
574{
575 struct iser_device *device = task->iser_conn->ib_conn.device;
576
577 device->reg_ops->unreg_mem(task, dir);
578}
1/*
2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32#include <linux/module.h>
33#include <linux/kernel.h>
34#include <linux/slab.h>
35#include <linux/mm.h>
36#include <linux/highmem.h>
37#include <linux/scatterlist.h>
38
39#include "iscsi_iser.h"
40
41#define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */
42
43/**
44 * iser_start_rdma_unaligned_sg
45 */
46static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
47 enum iser_data_dir cmd_dir)
48{
49 int dma_nents;
50 struct ib_device *dev;
51 char *mem = NULL;
52 struct iser_data_buf *data = &iser_task->data[cmd_dir];
53 unsigned long cmd_data_len = data->data_len;
54
55 if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
56 mem = (void *)__get_free_pages(GFP_ATOMIC,
57 ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
58 else
59 mem = kmalloc(cmd_data_len, GFP_ATOMIC);
60
61 if (mem == NULL) {
62 iser_err("Failed to allocate mem size %d %d for copying sglist\n",
63 data->size,(int)cmd_data_len);
64 return -ENOMEM;
65 }
66
67 if (cmd_dir == ISER_DIR_OUT) {
68 /* copy the unaligned sg the buffer which is used for RDMA */
69 struct scatterlist *sgl = (struct scatterlist *)data->buf;
70 struct scatterlist *sg;
71 int i;
72 char *p, *from;
73
74 p = mem;
75 for_each_sg(sgl, sg, data->size, i) {
76 from = kmap_atomic(sg_page(sg));
77 memcpy(p,
78 from + sg->offset,
79 sg->length);
80 kunmap_atomic(from);
81 p += sg->length;
82 }
83 }
84
85 sg_init_one(&iser_task->data_copy[cmd_dir].sg_single, mem, cmd_data_len);
86 iser_task->data_copy[cmd_dir].buf =
87 &iser_task->data_copy[cmd_dir].sg_single;
88 iser_task->data_copy[cmd_dir].size = 1;
89
90 iser_task->data_copy[cmd_dir].copy_buf = mem;
91
92 dev = iser_task->iser_conn->ib_conn->device->ib_device;
93 dma_nents = ib_dma_map_sg(dev,
94 &iser_task->data_copy[cmd_dir].sg_single,
95 1,
96 (cmd_dir == ISER_DIR_OUT) ?
97 DMA_TO_DEVICE : DMA_FROM_DEVICE);
98 BUG_ON(dma_nents == 0);
99
100 iser_task->data_copy[cmd_dir].dma_nents = dma_nents;
101 return 0;
102}
103
104/**
105 * iser_finalize_rdma_unaligned_sg
106 */
107void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
108 enum iser_data_dir cmd_dir)
109{
110 struct ib_device *dev;
111 struct iser_data_buf *mem_copy;
112 unsigned long cmd_data_len;
113
114 dev = iser_task->iser_conn->ib_conn->device->ib_device;
115 mem_copy = &iser_task->data_copy[cmd_dir];
116
117 ib_dma_unmap_sg(dev, &mem_copy->sg_single, 1,
118 (cmd_dir == ISER_DIR_OUT) ?
119 DMA_TO_DEVICE : DMA_FROM_DEVICE);
120
121 if (cmd_dir == ISER_DIR_IN) {
122 char *mem;
123 struct scatterlist *sgl, *sg;
124 unsigned char *p, *to;
125 unsigned int sg_size;
126 int i;
127
128 /* copy back read RDMA to unaligned sg */
129 mem = mem_copy->copy_buf;
130
131 sgl = (struct scatterlist *)iser_task->data[ISER_DIR_IN].buf;
132 sg_size = iser_task->data[ISER_DIR_IN].size;
133
134 p = mem;
135 for_each_sg(sgl, sg, sg_size, i) {
136 to = kmap_atomic(sg_page(sg));
137 memcpy(to + sg->offset,
138 p,
139 sg->length);
140 kunmap_atomic(to);
141 p += sg->length;
142 }
143 }
144
145 cmd_data_len = iser_task->data[cmd_dir].data_len;
146
147 if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
148 free_pages((unsigned long)mem_copy->copy_buf,
149 ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
150 else
151 kfree(mem_copy->copy_buf);
152
153 mem_copy->copy_buf = NULL;
154}
155
156#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0)
157
158/**
159 * iser_sg_to_page_vec - Translates scatterlist entries to physical addresses
160 * and returns the length of resulting physical address array (may be less than
161 * the original due to possible compaction).
162 *
163 * we build a "page vec" under the assumption that the SG meets the RDMA
164 * alignment requirements. Other then the first and last SG elements, all
165 * the "internal" elements can be compacted into a list whose elements are
166 * dma addresses of physical pages. The code supports also the weird case
167 * where --few fragments of the same page-- are present in the SG as
168 * consecutive elements. Also, it handles one entry SG.
169 */
170
171static int iser_sg_to_page_vec(struct iser_data_buf *data,
172 struct iser_page_vec *page_vec,
173 struct ib_device *ibdev)
174{
175 struct scatterlist *sg, *sgl = (struct scatterlist *)data->buf;
176 u64 start_addr, end_addr, page, chunk_start = 0;
177 unsigned long total_sz = 0;
178 unsigned int dma_len;
179 int i, new_chunk, cur_page, last_ent = data->dma_nents - 1;
180
181 /* compute the offset of first element */
182 page_vec->offset = (u64) sgl[0].offset & ~MASK_4K;
183
184 new_chunk = 1;
185 cur_page = 0;
186 for_each_sg(sgl, sg, data->dma_nents, i) {
187 start_addr = ib_sg_dma_address(ibdev, sg);
188 if (new_chunk)
189 chunk_start = start_addr;
190 dma_len = ib_sg_dma_len(ibdev, sg);
191 end_addr = start_addr + dma_len;
192 total_sz += dma_len;
193
194 /* collect page fragments until aligned or end of SG list */
195 if (!IS_4K_ALIGNED(end_addr) && i < last_ent) {
196 new_chunk = 0;
197 continue;
198 }
199 new_chunk = 1;
200
201 /* address of the first page in the contiguous chunk;
202 masking relevant for the very first SG entry,
203 which might be unaligned */
204 page = chunk_start & MASK_4K;
205 do {
206 page_vec->pages[cur_page++] = page;
207 page += SIZE_4K;
208 } while (page < end_addr);
209 }
210
211 page_vec->data_size = total_sz;
212 iser_dbg("page_vec->data_size:%d cur_page %d\n", page_vec->data_size,cur_page);
213 return cur_page;
214}
215
216
217/**
218 * iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned
219 * for RDMA sub-list of a scatter-gather list of memory buffers, and returns
220 * the number of entries which are aligned correctly. Supports the case where
221 * consecutive SG elements are actually fragments of the same physcial page.
222 */
223static int iser_data_buf_aligned_len(struct iser_data_buf *data,
224 struct ib_device *ibdev)
225{
226 struct scatterlist *sgl, *sg, *next_sg = NULL;
227 u64 start_addr, end_addr;
228 int i, ret_len, start_check = 0;
229
230 if (data->dma_nents == 1)
231 return 1;
232
233 sgl = (struct scatterlist *)data->buf;
234 start_addr = ib_sg_dma_address(ibdev, sgl);
235
236 for_each_sg(sgl, sg, data->dma_nents, i) {
237 if (start_check && !IS_4K_ALIGNED(start_addr))
238 break;
239
240 next_sg = sg_next(sg);
241 if (!next_sg)
242 break;
243
244 end_addr = start_addr + ib_sg_dma_len(ibdev, sg);
245 start_addr = ib_sg_dma_address(ibdev, next_sg);
246
247 if (end_addr == start_addr) {
248 start_check = 0;
249 continue;
250 } else
251 start_check = 1;
252
253 if (!IS_4K_ALIGNED(end_addr))
254 break;
255 }
256 ret_len = (next_sg) ? i : i+1;
257 iser_dbg("Found %d aligned entries out of %d in sg:0x%p\n",
258 ret_len, data->dma_nents, data);
259 return ret_len;
260}
261
262static void iser_data_buf_dump(struct iser_data_buf *data,
263 struct ib_device *ibdev)
264{
265 struct scatterlist *sgl = (struct scatterlist *)data->buf;
266 struct scatterlist *sg;
267 int i;
268
269 if (iser_debug_level == 0)
270 return;
271
272 for_each_sg(sgl, sg, data->dma_nents, i)
273 iser_warn("sg[%d] dma_addr:0x%lX page:0x%p "
274 "off:0x%x sz:0x%x dma_len:0x%x\n",
275 i, (unsigned long)ib_sg_dma_address(ibdev, sg),
276 sg_page(sg), sg->offset,
277 sg->length, ib_sg_dma_len(ibdev, sg));
278}
279
280static void iser_dump_page_vec(struct iser_page_vec *page_vec)
281{
282 int i;
283
284 iser_err("page vec length %d data size %d\n",
285 page_vec->length, page_vec->data_size);
286 for (i = 0; i < page_vec->length; i++)
287 iser_err("%d %lx\n",i,(unsigned long)page_vec->pages[i]);
288}
289
290static void iser_page_vec_build(struct iser_data_buf *data,
291 struct iser_page_vec *page_vec,
292 struct ib_device *ibdev)
293{
294 int page_vec_len = 0;
295
296 page_vec->length = 0;
297 page_vec->offset = 0;
298
299 iser_dbg("Translating sg sz: %d\n", data->dma_nents);
300 page_vec_len = iser_sg_to_page_vec(data, page_vec, ibdev);
301 iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents,page_vec_len);
302
303 page_vec->length = page_vec_len;
304
305 if (page_vec_len * SIZE_4K < page_vec->data_size) {
306 iser_err("page_vec too short to hold this SG\n");
307 iser_data_buf_dump(data, ibdev);
308 iser_dump_page_vec(page_vec);
309 BUG();
310 }
311}
312
313int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
314 struct iser_data_buf *data,
315 enum iser_data_dir iser_dir,
316 enum dma_data_direction dma_dir)
317{
318 struct ib_device *dev;
319
320 iser_task->dir[iser_dir] = 1;
321 dev = iser_task->iser_conn->ib_conn->device->ib_device;
322
323 data->dma_nents = ib_dma_map_sg(dev, data->buf, data->size, dma_dir);
324 if (data->dma_nents == 0) {
325 iser_err("dma_map_sg failed!!!\n");
326 return -EINVAL;
327 }
328 return 0;
329}
330
331void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task)
332{
333 struct ib_device *dev;
334 struct iser_data_buf *data;
335
336 dev = iser_task->iser_conn->ib_conn->device->ib_device;
337
338 if (iser_task->dir[ISER_DIR_IN]) {
339 data = &iser_task->data[ISER_DIR_IN];
340 ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE);
341 }
342
343 if (iser_task->dir[ISER_DIR_OUT]) {
344 data = &iser_task->data[ISER_DIR_OUT];
345 ib_dma_unmap_sg(dev, data->buf, data->size, DMA_TO_DEVICE);
346 }
347}
348
349/**
350 * iser_reg_rdma_mem - Registers memory intended for RDMA,
351 * obtaining rkey and va
352 *
353 * returns 0 on success, errno code on failure
354 */
355int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task,
356 enum iser_data_dir cmd_dir)
357{
358 struct iscsi_conn *iscsi_conn = iser_task->iser_conn->iscsi_conn;
359 struct iser_conn *ib_conn = iser_task->iser_conn->ib_conn;
360 struct iser_device *device = ib_conn->device;
361 struct ib_device *ibdev = device->ib_device;
362 struct iser_data_buf *mem = &iser_task->data[cmd_dir];
363 struct iser_regd_buf *regd_buf;
364 int aligned_len;
365 int err;
366 int i;
367 struct scatterlist *sg;
368
369 regd_buf = &iser_task->rdma_regd[cmd_dir];
370
371 aligned_len = iser_data_buf_aligned_len(mem, ibdev);
372 if (aligned_len != mem->dma_nents) {
373 iscsi_conn->fmr_unalign_cnt++;
374 iser_warn("rdma alignment violation %d/%d aligned\n",
375 aligned_len, mem->size);
376 iser_data_buf_dump(mem, ibdev);
377
378 /* unmap the command data before accessing it */
379 iser_dma_unmap_task_data(iser_task);
380
381 /* allocate copy buf, if we are writing, copy the */
382 /* unaligned scatterlist, dma map the copy */
383 if (iser_start_rdma_unaligned_sg(iser_task, cmd_dir) != 0)
384 return -ENOMEM;
385 mem = &iser_task->data_copy[cmd_dir];
386 }
387
388 /* if there a single dma entry, FMR is not needed */
389 if (mem->dma_nents == 1) {
390 sg = (struct scatterlist *)mem->buf;
391
392 regd_buf->reg.lkey = device->mr->lkey;
393 regd_buf->reg.rkey = device->mr->rkey;
394 regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]);
395 regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]);
396 regd_buf->reg.is_fmr = 0;
397
398 iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X "
399 "va: 0x%08lX sz: %ld]\n",
400 (unsigned int)regd_buf->reg.lkey,
401 (unsigned int)regd_buf->reg.rkey,
402 (unsigned long)regd_buf->reg.va,
403 (unsigned long)regd_buf->reg.len);
404 } else { /* use FMR for multiple dma entries */
405 iser_page_vec_build(mem, ib_conn->page_vec, ibdev);
406 err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, ®d_buf->reg);
407 if (err) {
408 iser_data_buf_dump(mem, ibdev);
409 iser_err("mem->dma_nents = %d (dlength = 0x%x)\n",
410 mem->dma_nents,
411 ntoh24(iser_task->desc.iscsi_header.dlength));
412 iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n",
413 ib_conn->page_vec->data_size, ib_conn->page_vec->length,
414 ib_conn->page_vec->offset);
415 for (i=0 ; i<ib_conn->page_vec->length ; i++)
416 iser_err("page_vec[%d] = 0x%llx\n", i,
417 (unsigned long long) ib_conn->page_vec->pages[i]);
418 return err;
419 }
420 }
421 return 0;
422}