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/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
41void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc)
42{
43 iser_err_comp(wc, "memreg");
44}
45
46static struct iser_fr_desc *iser_reg_desc_get_fr(struct ib_conn *ib_conn)
47{
48 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
49 struct iser_fr_desc *desc;
50 unsigned long flags;
51
52 spin_lock_irqsave(&fr_pool->lock, flags);
53 desc = list_first_entry(&fr_pool->list,
54 struct iser_fr_desc, list);
55 list_del(&desc->list);
56 spin_unlock_irqrestore(&fr_pool->lock, flags);
57
58 return desc;
59}
60
61static void iser_reg_desc_put_fr(struct ib_conn *ib_conn,
62 struct iser_fr_desc *desc)
63{
64 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
65 unsigned long flags;
66
67 spin_lock_irqsave(&fr_pool->lock, flags);
68 list_add(&desc->list, &fr_pool->list);
69 spin_unlock_irqrestore(&fr_pool->lock, flags);
70}
71
72int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
73 enum iser_data_dir iser_dir,
74 enum dma_data_direction dma_dir)
75{
76 struct iser_data_buf *data = &iser_task->data[iser_dir];
77 struct ib_device *dev;
78
79 iser_task->dir[iser_dir] = 1;
80 dev = iser_task->iser_conn->ib_conn.device->ib_device;
81
82 data->dma_nents = ib_dma_map_sg(dev, data->sg, data->size, dma_dir);
83 if (unlikely(data->dma_nents == 0)) {
84 iser_err("dma_map_sg failed!!!\n");
85 return -EINVAL;
86 }
87
88 if (scsi_prot_sg_count(iser_task->sc)) {
89 struct iser_data_buf *pdata = &iser_task->prot[iser_dir];
90
91 pdata->dma_nents = ib_dma_map_sg(dev, pdata->sg, pdata->size, dma_dir);
92 if (unlikely(pdata->dma_nents == 0)) {
93 iser_err("protection dma_map_sg failed!!!\n");
94 goto out_unmap;
95 }
96 }
97
98 return 0;
99
100out_unmap:
101 ib_dma_unmap_sg(dev, data->sg, data->size, dma_dir);
102 return -EINVAL;
103}
104
105
106void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task,
107 enum iser_data_dir iser_dir,
108 enum dma_data_direction dma_dir)
109{
110 struct iser_data_buf *data = &iser_task->data[iser_dir];
111 struct ib_device *dev;
112
113 dev = iser_task->iser_conn->ib_conn.device->ib_device;
114 ib_dma_unmap_sg(dev, data->sg, data->size, dma_dir);
115
116 if (scsi_prot_sg_count(iser_task->sc)) {
117 struct iser_data_buf *pdata = &iser_task->prot[iser_dir];
118
119 ib_dma_unmap_sg(dev, pdata->sg, pdata->size, dma_dir);
120 }
121}
122
123static int iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem,
124 struct iser_mem_reg *reg)
125{
126 struct scatterlist *sg = mem->sg;
127
128 reg->sge.lkey = device->pd->local_dma_lkey;
129 /*
130 * FIXME: rework the registration code path to differentiate
131 * rkey/lkey use cases
132 */
133
134 if (device->pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY)
135 reg->rkey = device->pd->unsafe_global_rkey;
136 else
137 reg->rkey = 0;
138 reg->sge.addr = sg_dma_address(&sg[0]);
139 reg->sge.length = sg_dma_len(&sg[0]);
140
141 iser_dbg("Single DMA entry: lkey=0x%x, rkey=0x%x, addr=0x%llx,"
142 " length=0x%x\n", reg->sge.lkey, reg->rkey,
143 reg->sge.addr, reg->sge.length);
144
145 return 0;
146}
147
148void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task,
149 enum iser_data_dir cmd_dir)
150{
151 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
152 struct iser_fr_desc *desc;
153 struct ib_mr_status mr_status;
154
155 desc = reg->desc;
156 if (!desc)
157 return;
158
159 /*
160 * The signature MR cannot be invalidated and reused without checking.
161 * libiscsi calls the check_protection transport handler only if
162 * SCSI-Response is received. And the signature MR is not checked if
163 * the task is completed for some other reason like a timeout or error
164 * handling. That's why we must check the signature MR here before
165 * putting it to the free pool.
166 */
167 if (unlikely(desc->sig_protected)) {
168 desc->sig_protected = false;
169 ib_check_mr_status(desc->rsc.sig_mr, IB_MR_CHECK_SIG_STATUS,
170 &mr_status);
171 }
172 iser_reg_desc_put_fr(&iser_task->iser_conn->ib_conn, reg->desc);
173 reg->desc = NULL;
174}
175
176static void iser_set_dif_domain(struct scsi_cmnd *sc,
177 struct ib_sig_domain *domain)
178{
179 domain->sig_type = IB_SIG_TYPE_T10_DIF;
180 domain->sig.dif.pi_interval = scsi_prot_interval(sc);
181 domain->sig.dif.ref_tag = t10_pi_ref_tag(scsi_cmd_to_rq(sc));
182 /*
183 * At the moment we hard code those, but in the future
184 * we will take them from sc.
185 */
186 domain->sig.dif.apptag_check_mask = 0xffff;
187 domain->sig.dif.app_escape = true;
188 domain->sig.dif.ref_escape = true;
189 if (sc->prot_flags & SCSI_PROT_REF_INCREMENT)
190 domain->sig.dif.ref_remap = true;
191}
192
193static int iser_set_sig_attrs(struct scsi_cmnd *sc,
194 struct ib_sig_attrs *sig_attrs)
195{
196 switch (scsi_get_prot_op(sc)) {
197 case SCSI_PROT_WRITE_INSERT:
198 case SCSI_PROT_READ_STRIP:
199 sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
200 iser_set_dif_domain(sc, &sig_attrs->wire);
201 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
202 break;
203 case SCSI_PROT_READ_INSERT:
204 case SCSI_PROT_WRITE_STRIP:
205 sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
206 iser_set_dif_domain(sc, &sig_attrs->mem);
207 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
208 IB_T10DIF_CSUM : IB_T10DIF_CRC;
209 break;
210 case SCSI_PROT_READ_PASS:
211 case SCSI_PROT_WRITE_PASS:
212 iser_set_dif_domain(sc, &sig_attrs->wire);
213 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC;
214 iser_set_dif_domain(sc, &sig_attrs->mem);
215 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ?
216 IB_T10DIF_CSUM : IB_T10DIF_CRC;
217 break;
218 default:
219 iser_err("Unsupported PI operation %d\n",
220 scsi_get_prot_op(sc));
221 return -EINVAL;
222 }
223
224 return 0;
225}
226
227static inline void iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask)
228{
229 *mask = 0;
230 if (sc->prot_flags & SCSI_PROT_REF_CHECK)
231 *mask |= IB_SIG_CHECK_REFTAG;
232 if (sc->prot_flags & SCSI_PROT_GUARD_CHECK)
233 *mask |= IB_SIG_CHECK_GUARD;
234}
235
236static inline void iser_inv_rkey(struct ib_send_wr *inv_wr, struct ib_mr *mr,
237 struct ib_cqe *cqe, struct ib_send_wr *next_wr)
238{
239 inv_wr->opcode = IB_WR_LOCAL_INV;
240 inv_wr->wr_cqe = cqe;
241 inv_wr->ex.invalidate_rkey = mr->rkey;
242 inv_wr->send_flags = 0;
243 inv_wr->num_sge = 0;
244 inv_wr->next = next_wr;
245}
246
247static int iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
248 struct iser_data_buf *mem,
249 struct iser_data_buf *sig_mem,
250 struct iser_reg_resources *rsc,
251 struct iser_mem_reg *sig_reg)
252{
253 struct iser_tx_desc *tx_desc = &iser_task->desc;
254 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
255 struct ib_mr *mr = rsc->sig_mr;
256 struct ib_sig_attrs *sig_attrs = mr->sig_attrs;
257 struct ib_reg_wr *wr = &tx_desc->reg_wr;
258 int ret;
259
260 memset(sig_attrs, 0, sizeof(*sig_attrs));
261 ret = iser_set_sig_attrs(iser_task->sc, sig_attrs);
262 if (ret)
263 goto err;
264
265 iser_set_prot_checks(iser_task->sc, &sig_attrs->check_mask);
266
267 if (rsc->sig_mr->need_inval)
268 iser_inv_rkey(&tx_desc->inv_wr, mr, cqe, &wr->wr);
269
270 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
271
272 ret = ib_map_mr_sg_pi(mr, mem->sg, mem->dma_nents, NULL,
273 sig_mem->sg, sig_mem->dma_nents, NULL, SZ_4K);
274 if (unlikely(ret)) {
275 iser_err("failed to map PI sg (%d)\n",
276 mem->dma_nents + sig_mem->dma_nents);
277 goto err;
278 }
279
280 memset(wr, 0, sizeof(*wr));
281 wr->wr.next = &tx_desc->send_wr;
282 wr->wr.opcode = IB_WR_REG_MR_INTEGRITY;
283 wr->wr.wr_cqe = cqe;
284 wr->wr.num_sge = 0;
285 wr->wr.send_flags = 0;
286 wr->mr = mr;
287 wr->key = mr->rkey;
288 wr->access = IB_ACCESS_LOCAL_WRITE |
289 IB_ACCESS_REMOTE_READ |
290 IB_ACCESS_REMOTE_WRITE;
291 rsc->sig_mr->need_inval = true;
292
293 sig_reg->sge.lkey = mr->lkey;
294 sig_reg->rkey = mr->rkey;
295 sig_reg->sge.addr = mr->iova;
296 sig_reg->sge.length = mr->length;
297
298 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=%u\n",
299 sig_reg->sge.lkey, sig_reg->rkey, sig_reg->sge.addr,
300 sig_reg->sge.length);
301err:
302 return ret;
303}
304
305static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
306 struct iser_data_buf *mem,
307 struct iser_reg_resources *rsc,
308 struct iser_mem_reg *reg)
309{
310 struct iser_tx_desc *tx_desc = &iser_task->desc;
311 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe;
312 struct ib_mr *mr = rsc->mr;
313 struct ib_reg_wr *wr = &tx_desc->reg_wr;
314 int n;
315
316 if (rsc->mr->need_inval)
317 iser_inv_rkey(&tx_desc->inv_wr, mr, cqe, &wr->wr);
318
319 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
320
321 n = ib_map_mr_sg(mr, mem->sg, mem->dma_nents, NULL, SZ_4K);
322 if (unlikely(n != mem->dma_nents)) {
323 iser_err("failed to map sg (%d/%d)\n",
324 n, mem->dma_nents);
325 return n < 0 ? n : -EINVAL;
326 }
327
328 wr->wr.next = &tx_desc->send_wr;
329 wr->wr.opcode = IB_WR_REG_MR;
330 wr->wr.wr_cqe = cqe;
331 wr->wr.send_flags = 0;
332 wr->wr.num_sge = 0;
333 wr->mr = mr;
334 wr->key = mr->rkey;
335 wr->access = IB_ACCESS_LOCAL_WRITE |
336 IB_ACCESS_REMOTE_WRITE |
337 IB_ACCESS_REMOTE_READ;
338
339 rsc->mr->need_inval = true;
340
341 reg->sge.lkey = mr->lkey;
342 reg->rkey = mr->rkey;
343 reg->sge.addr = mr->iova;
344 reg->sge.length = mr->length;
345
346 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=0x%x\n",
347 reg->sge.lkey, reg->rkey, reg->sge.addr, reg->sge.length);
348
349 return 0;
350}
351
352int iser_reg_mem_fastreg(struct iscsi_iser_task *task,
353 enum iser_data_dir dir,
354 bool all_imm)
355{
356 struct ib_conn *ib_conn = &task->iser_conn->ib_conn;
357 struct iser_device *device = ib_conn->device;
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;
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 if (use_dma_key)
367 return iser_reg_dma(device, mem, reg);
368
369 desc = iser_reg_desc_get_fr(ib_conn);
370 if (scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL) {
371 err = iser_fast_reg_mr(task, mem, &desc->rsc, reg);
372 if (unlikely(err))
373 goto err_reg;
374 } else {
375 err = iser_reg_sig_mr(task, mem, &task->prot[dir],
376 &desc->rsc, reg);
377 if (unlikely(err))
378 goto err_reg;
379
380 desc->sig_protected = true;
381 }
382
383 reg->desc = desc;
384
385 return 0;
386
387err_reg:
388 iser_reg_desc_put_fr(ib_conn, desc);
389
390 return err;
391}
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), KM_USER0);
77 memcpy(p,
78 from + sg->offset,
79 sg->length);
80 kunmap_atomic(from, KM_USER0);
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), KM_SOFTIRQ0);
137 memcpy(to + sg->offset,
138 p,
139 sg->length);
140 kunmap_atomic(to, KM_SOFTIRQ0);
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}