projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/mlx4: Implement IB_QP_CREATE_USE_GFP_NOIO
[linux-2.6-block.git] / drivers / net / ethernet / mellanox / mlx4 / mr.c
CommitLineData
225c7b1f
RD		 1/*
2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
51a379d0 3 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
225c7b1f
RD		 4 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
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.
33 */
34
225c7b1f 35#include <linux/errno.h>
ee40fa06 36#include <linux/export.h>
5a0e3ad6 37#include <linux/slab.h>
ea51b377 38#include <linux/kernel.h>
89dd86db 39#include <linux/vmalloc.h>
225c7b1f
RD		 40
41#include <linux/mlx4/cmd.h>
42
43#include "mlx4.h"
44#include "icm.h"
45
225c7b1f
RD		 46static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order)
47{
48 int o;
49 int m;
50 u32 seg;
51
52 spin_lock(&buddy->lock);
53
e4044cfc
RD		 54 for (o = order; o <= buddy->max_order; ++o)
55 if (buddy->num_free[o]) {
56 m = 1 << (buddy->max_order - o);
57 seg = find_first_bit(buddy->bits[o], m);
58 if (seg < m)
59 goto found;
60 }
225c7b1f
RD		 61
62 spin_unlock(&buddy->lock);
63 return -1;
64
65 found:
66 clear_bit(seg, buddy->bits[o]);
e4044cfc 67 --buddy->num_free[o];
225c7b1f
RD		 68
69 while (o > order) {
70 --o;
71 seg <<= 1;
72 set_bit(seg ^ 1, buddy->bits[o]);
e4044cfc 73 ++buddy->num_free[o];
225c7b1f
RD		 74 }
75
76 spin_unlock(&buddy->lock);
77
78 seg <<= order;
79
80 return seg;
81}
82
83static void mlx4_buddy_free(struct mlx4_buddy *buddy, u32 seg, int order)
84{
85 seg >>= order;
86
87 spin_lock(&buddy->lock);
88
89 while (test_bit(seg ^ 1, buddy->bits[order])) {
90 clear_bit(seg ^ 1, buddy->bits[order]);
e4044cfc 91 --buddy->num_free[order];
225c7b1f
RD		 92 seg >>= 1;
93 ++order;
94 }
95
96 set_bit(seg, buddy->bits[order]);
e4044cfc 97 ++buddy->num_free[order];
225c7b1f
RD		 98
99 spin_unlock(&buddy->lock);
100}
101
e8f9b2ed 102static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order)
225c7b1f
RD		 103{
104 int i, s;
105
106 buddy->max_order = max_order;
107 spin_lock_init(&buddy->lock);
108
96f17d59 109 buddy->bits = kcalloc(buddy->max_order + 1, sizeof (long *),
225c7b1f 110 GFP_KERNEL);
a8312755 111 buddy->num_free = kcalloc((buddy->max_order + 1), sizeof *buddy->num_free,
e4044cfc
RD		 112 GFP_KERNEL);
113 if (!buddy->bits || !buddy->num_free)
225c7b1f
RD		 114 goto err_out;
115
116 for (i = 0; i <= buddy->max_order; ++i) {
117 s = BITS_TO_LONGS(1 << (buddy->max_order - i));
96f17d59 118 buddy->bits[i] = kcalloc(s, sizeof (long), GFP_KERNEL | __GFP_NOWARN);
89dd86db 119 if (!buddy->bits[i]) {
96f17d59 120 buddy->bits[i] = vzalloc(s * sizeof(long));
89dd86db
YH		 121 if (!buddy->bits[i])
122 goto err_out_free;
123 }
225c7b1f
RD		 124 }
125
126 set_bit(0, buddy->bits[buddy->max_order]);
e4044cfc 127 buddy->num_free[buddy->max_order] = 1;
225c7b1f
RD		 128
129 return 0;
130
131err_out_free:
132 for (i = 0; i <= buddy->max_order; ++i)
89dd86db
YH		 133 if (buddy->bits[i] && is_vmalloc_addr(buddy->bits[i]))
134 vfree(buddy->bits[i]);
135 else
136 kfree(buddy->bits[i]);
225c7b1f 137
e4044cfc 138err_out:
225c7b1f 139 kfree(buddy->bits);
e4044cfc 140 kfree(buddy->num_free);
225c7b1f 141
225c7b1f
RD		 142 return -ENOMEM;
143}
144
145static void mlx4_buddy_cleanup(struct mlx4_buddy *buddy)
146{
147 int i;
148
149 for (i = 0; i <= buddy->max_order; ++i)
89dd86db
YH		 150 if (is_vmalloc_addr(buddy->bits[i]))
151 vfree(buddy->bits[i]);
152 else
153 kfree(buddy->bits[i]);
225c7b1f
RD		 154
155 kfree(buddy->bits);
e4044cfc 156 kfree(buddy->num_free);
225c7b1f
RD		 157}
158
c82e9aa0 159u32 __mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order)
225c7b1f
RD		 160{
161 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
162 u32 seg;
2b8fb286
MA		 163 int seg_order;
164 u32 offset;
225c7b1f 165
2b8fb286
MA		 166 seg_order = max_t(int, order - log_mtts_per_seg, 0);
167
168 seg = mlx4_buddy_alloc(&mr_table->mtt_buddy, seg_order);
225c7b1f
RD		 169 if (seg == -1)
170 return -1;
171
2b8fb286
MA		 172 offset = seg * (1 << log_mtts_per_seg);
173
174 if (mlx4_table_get_range(dev, &mr_table->mtt_table, offset,
175 offset + (1 << order) - 1)) {
176 mlx4_buddy_free(&mr_table->mtt_buddy, seg, seg_order);
225c7b1f
RD		 177 return -1;
178 }
179
2b8fb286 180 return offset;
225c7b1f
RD		 181}
182
ea51b377
JM		 183static u32 mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order)
184{
e7dbeba8 185 u64 in_param = 0;
ea51b377
JM		 186 u64 out_param;
187 int err;
188
189 if (mlx4_is_mfunc(dev)) {
190 set_param_l(&in_param, order);
191 err = mlx4_cmd_imm(dev, in_param, &out_param, RES_MTT,
192 RES_OP_RESERVE_AND_MAP,
193 MLX4_CMD_ALLOC_RES,
194 MLX4_CMD_TIME_CLASS_A,
195 MLX4_CMD_WRAPPED);
196 if (err)
197 return -1;
198 return get_param_l(&out_param);
199 }
200 return __mlx4_alloc_mtt_range(dev, order);
201}
202
225c7b1f
RD		 203int mlx4_mtt_init(struct mlx4_dev *dev, int npages, int page_shift,
204 struct mlx4_mtt *mtt)
205{
206 int i;
207
208 if (!npages) {
209 mtt->order = -1;
210 mtt->page_shift = MLX4_ICM_PAGE_SHIFT;
211 return 0;
212 } else
213 mtt->page_shift = page_shift;
214
2b8fb286 215 for (mtt->order = 0, i = 1; i < npages; i <<= 1)
225c7b1f
RD		 216 ++mtt->order;
217
2b8fb286
MA		 218 mtt->offset = mlx4_alloc_mtt_range(dev, mtt->order);
219 if (mtt->offset == -1)
225c7b1f
RD		 220 return -ENOMEM;
221
222 return 0;
223}
224EXPORT_SYMBOL_GPL(mlx4_mtt_init);
225
2b8fb286 226void __mlx4_free_mtt_range(struct mlx4_dev *dev, u32 offset, int order)
225c7b1f 227{
2b8fb286
MA		 228 u32 first_seg;
229 int seg_order;
225c7b1f
RD		 230 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
231
2b8fb286
MA		 232 seg_order = max_t(int, order - log_mtts_per_seg, 0);
233 first_seg = offset / (1 << log_mtts_per_seg);
234
235 mlx4_buddy_free(&mr_table->mtt_buddy, first_seg, seg_order);
1e27ca69
MA		 236 mlx4_table_put_range(dev, &mr_table->mtt_table, offset,
237 offset + (1 << order) - 1);
ea51b377
JM		 238}
239
2b8fb286 240static void mlx4_free_mtt_range(struct mlx4_dev *dev, u32 offset, int order)
ea51b377 241{
e7dbeba8 242 u64 in_param = 0;
ea51b377
JM		 243 int err;
244
245 if (mlx4_is_mfunc(dev)) {
2b8fb286 246 set_param_l(&in_param, offset);
ea51b377
JM		 247 set_param_h(&in_param, order);
248 err = mlx4_cmd(dev, in_param, RES_MTT, RES_OP_RESERVE_AND_MAP,
249 MLX4_CMD_FREE_RES,
250 MLX4_CMD_TIME_CLASS_A,
251 MLX4_CMD_WRAPPED);
252 if (err)
2b8fb286
MA		 253 mlx4_warn(dev, "Failed to free mtt range at:"
254 "%d order:%d\n", offset, order);
ea51b377
JM		 255 return;
256 }
2b8fb286 257 __mlx4_free_mtt_range(dev, offset, order);
ea51b377
JM		 258}
259
260void mlx4_mtt_cleanup(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
261{
225c7b1f
RD		 262 if (mtt->order < 0)
263 return;
264
2b8fb286 265 mlx4_free_mtt_range(dev, mtt->offset, mtt->order);
225c7b1f
RD		 266}
267EXPORT_SYMBOL_GPL(mlx4_mtt_cleanup);
268
269u64 mlx4_mtt_addr(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
270{
2b8fb286 271 return (u64) mtt->offset * dev->caps.mtt_entry_sz;
225c7b1f
RD		 272}
273EXPORT_SYMBOL_GPL(mlx4_mtt_addr);
274
275static u32 hw_index_to_key(u32 ind)
276{
277 return (ind >> 24) | (ind << 8);
278}
279
280static u32 key_to_hw_index(u32 key)
281{
282 return (key << 24) | (key >> 8);
283}
284
285static int mlx4_SW2HW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
286 int mpt_index)
287{
eb41049f 288 return mlx4_cmd(dev, mailbox->dma, mpt_index,
ea51b377
JM		 289 0, MLX4_CMD_SW2HW_MPT, MLX4_CMD_TIME_CLASS_B,
290 MLX4_CMD_WRAPPED);
225c7b1f
RD		 291}
292
293static int mlx4_HW2SW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
294 int mpt_index)
295{
296 return mlx4_cmd_box(dev, 0, mailbox ? mailbox->dma : 0, mpt_index,
f9baff50
JM		 297 !mailbox, MLX4_CMD_HW2SW_MPT,
298 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
225c7b1f
RD		 299}
300
66431a7d 301static int mlx4_mr_alloc_reserved(struct mlx4_dev *dev, u32 mridx, u32 pd,
ea51b377
JM		 302 u64 iova, u64 size, u32 access, int npages,
303 int page_shift, struct mlx4_mr *mr)
304{
225c7b1f
RD		 305 mr->iova = iova;
306 mr->size = size;
307 mr->pd = pd;
308 mr->access = access;
b20e519a 309 mr->enabled = MLX4_MPT_DISABLED;
ea51b377
JM		 310 mr->key = hw_index_to_key(mridx);
311
312 return mlx4_mtt_init(dev, npages, page_shift, &mr->mtt);
313}
ea51b377
JM		 314
315static int mlx4_WRITE_MTT(struct mlx4_dev *dev,
316 struct mlx4_cmd_mailbox *mailbox,
317 int num_entries)
318{
319 return mlx4_cmd(dev, mailbox->dma, num_entries, 0, MLX4_CMD_WRITE_MTT,
320 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
321}
322
b20e519a 323int __mlx4_mpt_reserve(struct mlx4_dev *dev)
ea51b377
JM		 324{
325 struct mlx4_priv *priv = mlx4_priv(dev);
326
327 return mlx4_bitmap_alloc(&priv->mr_table.mpt_bitmap);
328}
225c7b1f 329
b20e519a 330static int mlx4_mpt_reserve(struct mlx4_dev *dev)
ea51b377
JM		 331{
332 u64 out_param;
333
334 if (mlx4_is_mfunc(dev)) {
335 if (mlx4_cmd_imm(dev, 0, &out_param, RES_MPT, RES_OP_RESERVE,
336 MLX4_CMD_ALLOC_RES,
337 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED))
338 return -1;
339 return get_param_l(&out_param);
340 }
b20e519a 341 return __mlx4_mpt_reserve(dev);
ea51b377
JM		 342}
343
b20e519a 344void __mlx4_mpt_release(struct mlx4_dev *dev, u32 index)
ea51b377
JM		 345{
346 struct mlx4_priv *priv = mlx4_priv(dev);
347
7c6d74d2 348 mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, index, MLX4_NO_RR);
ea51b377
JM		 349}
350
b20e519a 351static void mlx4_mpt_release(struct mlx4_dev *dev, u32 index)
ea51b377 352{
e7dbeba8 353 u64 in_param = 0;
ea51b377
JM		 354
355 if (mlx4_is_mfunc(dev)) {
356 set_param_l(&in_param, index);
357 if (mlx4_cmd(dev, in_param, RES_MPT, RES_OP_RESERVE,
358 MLX4_CMD_FREE_RES,
359 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED))
360 mlx4_warn(dev, "Failed to release mr index:%d\n",
361 index);
362 return;
363 }
b20e519a 364 __mlx4_mpt_release(dev, index);
ea51b377
JM		 365}
366
40f2287b 367int __mlx4_mpt_alloc_icm(struct mlx4_dev *dev, u32 index, gfp_t gfp)
ea51b377
JM		 368{
369 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
370
40f2287b 371 return mlx4_table_get(dev, &mr_table->dmpt_table, index, gfp);
ea51b377
JM		 372}
373
40f2287b 374static int mlx4_mpt_alloc_icm(struct mlx4_dev *dev, u32 index, gfp_t gfp)
ea51b377 375{
e7dbeba8 376 u64 param = 0;
ea51b377
JM		 377
378 if (mlx4_is_mfunc(dev)) {
379 set_param_l(&param, index);
380 return mlx4_cmd_imm(dev, param, &param, RES_MPT, RES_OP_MAP_ICM,
381 MLX4_CMD_ALLOC_RES,
382 MLX4_CMD_TIME_CLASS_A,
383 MLX4_CMD_WRAPPED);
384 }
40f2287b 385 return __mlx4_mpt_alloc_icm(dev, index, gfp);
ea51b377
JM		 386}
387
b20e519a 388void __mlx4_mpt_free_icm(struct mlx4_dev *dev, u32 index)
ea51b377
JM		 389{
390 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
391
392 mlx4_table_put(dev, &mr_table->dmpt_table, index);
393}
394
b20e519a 395static void mlx4_mpt_free_icm(struct mlx4_dev *dev, u32 index)
ea51b377 396{
e7dbeba8 397 u64 in_param = 0;
ea51b377
JM		 398
399 if (mlx4_is_mfunc(dev)) {
400 set_param_l(&in_param, index);
401 if (mlx4_cmd(dev, in_param, RES_MPT, RES_OP_MAP_ICM,
402 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
403 MLX4_CMD_WRAPPED))
404 mlx4_warn(dev, "Failed to free icm of mr index:%d\n",
405 index);
406 return;
407 }
b20e519a 408 return __mlx4_mpt_free_icm(dev, index);
ea51b377
JM		 409}
410
411int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access,
412 int npages, int page_shift, struct mlx4_mr *mr)
413{
414 u32 index;
415 int err;
416
b20e519a 417 index = mlx4_mpt_reserve(dev);
ea51b377
JM		 418 if (index == -1)
419 return -ENOMEM;
420
421 err = mlx4_mr_alloc_reserved(dev, index, pd, iova, size,
422 access, npages, page_shift, mr);
225c7b1f 423 if (err)
b20e519a 424 mlx4_mpt_release(dev, index);
225c7b1f 425
225c7b1f
RD		 426 return err;
427}
428EXPORT_SYMBOL_GPL(mlx4_mr_alloc);
429
61083720 430static int mlx4_mr_free_reserved(struct mlx4_dev *dev, struct mlx4_mr *mr)
225c7b1f 431{
225c7b1f
RD		 432 int err;
433
b20e519a 434 if (mr->enabled == MLX4_MPT_EN_HW) {
225c7b1f
RD		 435 err = mlx4_HW2SW_MPT(dev, NULL,
436 key_to_hw_index(mr->key) &
437 (dev->caps.num_mpts - 1));
61083720
SM		 438 if (err) {
439 mlx4_warn(dev, "HW2SW_MPT failed (%d),", err);
440 mlx4_warn(dev, "MR has MWs bound to it.\n");
441 return err;
442 }
225c7b1f 443
b20e519a 444 mr->enabled = MLX4_MPT_EN_SW;
ea51b377 445 }
225c7b1f 446 mlx4_mtt_cleanup(dev, &mr->mtt);
61083720
SM		 447
448 return 0;
ea51b377 449}
ea51b377 450
61083720 451int mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr)
ea51b377 452{
61083720
SM		 453 int ret;
454
455 ret = mlx4_mr_free_reserved(dev, mr);
456 if (ret)
457 return ret;
ea51b377 458 if (mr->enabled)
b20e519a
SM		 459 mlx4_mpt_free_icm(dev, key_to_hw_index(mr->key));
460 mlx4_mpt_release(dev, key_to_hw_index(mr->key));
61083720
SM		 461
462 return 0;
225c7b1f
RD		 463}
464EXPORT_SYMBOL_GPL(mlx4_mr_free);
465
466int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr)
467{
225c7b1f
RD		 468 struct mlx4_cmd_mailbox *mailbox;
469 struct mlx4_mpt_entry *mpt_entry;
470 int err;
471
40f2287b 472 err = mlx4_mpt_alloc_icm(dev, key_to_hw_index(mr->key), GFP_KERNEL);
225c7b1f
RD		 473 if (err)
474 return err;
475
476 mailbox = mlx4_alloc_cmd_mailbox(dev);
477 if (IS_ERR(mailbox)) {
478 err = PTR_ERR(mailbox);
479 goto err_table;
480 }
481 mpt_entry = mailbox->buf;
95d04f07 482 mpt_entry->flags = cpu_to_be32(MLX4_MPT_FLAG_MIO |
225c7b1f
RD		 483 MLX4_MPT_FLAG_REGION |
484 mr->access);
225c7b1f
RD		 485
486 mpt_entry->key = cpu_to_be32(key_to_hw_index(mr->key));
95d04f07 487 mpt_entry->pd_flags = cpu_to_be32(mr->pd | MLX4_MPT_PD_FLAG_EN_INV);
225c7b1f
RD		 488 mpt_entry->start = cpu_to_be64(mr->iova);
489 mpt_entry->length = cpu_to_be64(mr->size);
490 mpt_entry->entity_size = cpu_to_be32(mr->mtt.page_shift);
95d04f07 491
b2d9308a
JM		 492 if (mr->mtt.order < 0) {
493 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL);
2b8fb286 494 mpt_entry->mtt_addr = 0;
95d04f07 495 } else {
2b8fb286
MA		 496 mpt_entry->mtt_addr = cpu_to_be64(mlx4_mtt_addr(dev,
497 &mr->mtt));
95d04f07
RD		 498 }
499
500 if (mr->mtt.order >= 0 && mr->mtt.page_shift == 0) {
501 /* fast register MR in free state */
502 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_FREE);
c9257433
VS		 503 mpt_entry->pd_flags |= cpu_to_be32(MLX4_MPT_PD_FLAG_FAST_REG |
504 MLX4_MPT_PD_FLAG_RAE);
2b8fb286 505 mpt_entry->mtt_sz = cpu_to_be32(1 << mr->mtt.order);
95d04f07
RD		 506 } else {
507 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_SW_OWNS);
508 }
225c7b1f
RD		 509
510 err = mlx4_SW2HW_MPT(dev, mailbox,
511 key_to_hw_index(mr->key) & (dev->caps.num_mpts - 1));
512 if (err) {
513 mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err);
514 goto err_cmd;
515 }
b20e519a 516 mr->enabled = MLX4_MPT_EN_HW;
225c7b1f
RD		 517
518 mlx4_free_cmd_mailbox(dev, mailbox);
519
520 return 0;
521
522err_cmd:
523 mlx4_free_cmd_mailbox(dev, mailbox);
524
525err_table:
b20e519a 526 mlx4_mpt_free_icm(dev, key_to_hw_index(mr->key));
225c7b1f
RD		 527 return err;
528}
529EXPORT_SYMBOL_GPL(mlx4_mr_enable);
530
d7bb58fb
JM		 531static int mlx4_write_mtt_chunk(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
532 int start_index, int npages, u64 *page_list)
225c7b1f 533{
d7bb58fb
JM		 534 struct mlx4_priv *priv = mlx4_priv(dev);
535 __be64 *mtts;
536 dma_addr_t dma_handle;
537 int i;
d7bb58fb 538
2b8fb286
MA		 539 mtts = mlx4_table_find(&priv->mr_table.mtt_table, mtt->offset +
540 start_index, &dma_handle);
d7bb58fb 541
d7bb58fb
JM		 542 if (!mtts)
543 return -ENOMEM;
544
e727f5cd
RD		 545 dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle,
546 npages * sizeof (u64), DMA_TO_DEVICE);
547
d7bb58fb
JM		 548 for (i = 0; i < npages; ++i)
549 mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
550
e727f5cd
RD		 551 dma_sync_single_for_device(&dev->pdev->dev, dma_handle,
552 npages * sizeof (u64), DMA_TO_DEVICE);
d7bb58fb
JM		 553
554 return 0;
225c7b1f
RD		 555}
556
c82e9aa0 557int __mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
ea51b377 558 int start_index, int npages, u64 *page_list)
225c7b1f 559{
ea51b377 560 int err = 0;
d7bb58fb 561 int chunk;
2b8fb286
MA		 562 int mtts_per_page;
563 int max_mtts_first_page;
564
565 /* compute how may mtts fit in the first page */
566 mtts_per_page = PAGE_SIZE / sizeof(u64);
567 max_mtts_first_page = mtts_per_page - (mtt->offset + start_index)
568 % mtts_per_page;
569
570 chunk = min_t(int, max_mtts_first_page, npages);
225c7b1f 571
225c7b1f 572 while (npages > 0) {
d7bb58fb 573 err = mlx4_write_mtt_chunk(dev, mtt, start_index, chunk, page_list);
225c7b1f 574 if (err)
d7bb58fb 575 return err;
d7bb58fb
JM		 576 npages -= chunk;
577 start_index += chunk;
578 page_list += chunk;
2b8fb286
MA		 579
580 chunk = min_t(int, mtts_per_page, npages);
225c7b1f 581 }
ea51b377
JM		 582 return err;
583}
225c7b1f 584
ea51b377
JM		 585int mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
586 int start_index, int npages, u64 *page_list)
587{
588 struct mlx4_cmd_mailbox *mailbox = NULL;
589 __be64 *inbox = NULL;
590 int chunk;
591 int err = 0;
592 int i;
593
594 if (mtt->order < 0)
595 return -EINVAL;
596
597 if (mlx4_is_mfunc(dev)) {
598 mailbox = mlx4_alloc_cmd_mailbox(dev);
599 if (IS_ERR(mailbox))
600 return PTR_ERR(mailbox);
601 inbox = mailbox->buf;
602
603 while (npages > 0) {
2b8fb286
MA		 604 chunk = min_t(int, MLX4_MAILBOX_SIZE / sizeof(u64) - 2,
605 npages);
606 inbox[0] = cpu_to_be64(mtt->offset + start_index);
ea51b377
JM		 607 inbox[1] = 0;
608 for (i = 0; i < chunk; ++i)
609 inbox[i + 2] = cpu_to_be64(page_list[i] |
610 MLX4_MTT_FLAG_PRESENT);
611 err = mlx4_WRITE_MTT(dev, mailbox, chunk);
612 if (err) {
613 mlx4_free_cmd_mailbox(dev, mailbox);
614 return err;
615 }
616
617 npages -= chunk;
618 start_index += chunk;
619 page_list += chunk;
620 }
621 mlx4_free_cmd_mailbox(dev, mailbox);
622 return err;
623 }
624
625 return __mlx4_write_mtt(dev, mtt, start_index, npages, page_list);
225c7b1f
RD		 626}
627EXPORT_SYMBOL_GPL(mlx4_write_mtt);
628
629int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
40f2287b 630 struct mlx4_buf *buf, gfp_t gfp)
225c7b1f
RD		 631{
632 u64 *page_list;
633 int err;
634 int i;
635
40f2287b
JK		 636 page_list = kmalloc(buf->npages * sizeof *page_list,
637 gfp);
225c7b1f
RD		 638 if (!page_list)
639 return -ENOMEM;
640
641 for (i = 0; i < buf->npages; ++i)
642 if (buf->nbufs == 1)
b57aacfa 643 page_list[i] = buf->direct.map + (i << buf->page_shift);
225c7b1f 644 else
b57aacfa 645 page_list[i] = buf->page_list[i].map;
225c7b1f
RD		 646
647 err = mlx4_write_mtt(dev, mtt, 0, buf->npages, page_list);
648
649 kfree(page_list);
650 return err;
651}
652EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt);
653
804d6a89
SM		 654int mlx4_mw_alloc(struct mlx4_dev *dev, u32 pd, enum mlx4_mw_type type,
655 struct mlx4_mw *mw)
656{
657 u32 index;
658
659 if ((type == MLX4_MW_TYPE_1 &&
660 !(dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)) ||
661 (type == MLX4_MW_TYPE_2 &&
662 !(dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)))
663 return -ENOTSUPP;
664
665 index = mlx4_mpt_reserve(dev);
666 if (index == -1)
667 return -ENOMEM;
668
669 mw->key = hw_index_to_key(index);
670 mw->pd = pd;
671 mw->type = type;
672 mw->enabled = MLX4_MPT_DISABLED;
673
674 return 0;
675}
676EXPORT_SYMBOL_GPL(mlx4_mw_alloc);
677
678int mlx4_mw_enable(struct mlx4_dev *dev, struct mlx4_mw *mw)
679{
680 struct mlx4_cmd_mailbox *mailbox;
681 struct mlx4_mpt_entry *mpt_entry;
682 int err;
683
40f2287b 684 err = mlx4_mpt_alloc_icm(dev, key_to_hw_index(mw->key), GFP_KERNEL);
804d6a89
SM		 685 if (err)
686 return err;
687
688 mailbox = mlx4_alloc_cmd_mailbox(dev);
689 if (IS_ERR(mailbox)) {
690 err = PTR_ERR(mailbox);
691 goto err_table;
692 }
693 mpt_entry = mailbox->buf;
694
804d6a89
SM		 695 /* Note that the MLX4_MPT_FLAG_REGION bit in mpt_entry->flags is turned
696 * off, thus creating a memory window and not a memory region.
697 */
698 mpt_entry->key = cpu_to_be32(key_to_hw_index(mw->key));
699 mpt_entry->pd_flags = cpu_to_be32(mw->pd);
700 if (mw->type == MLX4_MW_TYPE_2) {
701 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_FREE);
702 mpt_entry->qpn = cpu_to_be32(MLX4_MPT_QP_FLAG_BOUND_QP);
703 mpt_entry->pd_flags |= cpu_to_be32(MLX4_MPT_PD_FLAG_EN_INV);
704 }
705
706 err = mlx4_SW2HW_MPT(dev, mailbox,
707 key_to_hw_index(mw->key) &
708 (dev->caps.num_mpts - 1));
709 if (err) {
710 mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err);
711 goto err_cmd;
712 }
713 mw->enabled = MLX4_MPT_EN_HW;
714
715 mlx4_free_cmd_mailbox(dev, mailbox);
716
717 return 0;
718
719err_cmd:
720 mlx4_free_cmd_mailbox(dev, mailbox);
721
722err_table:
723 mlx4_mpt_free_icm(dev, key_to_hw_index(mw->key));
724 return err;
725}
726EXPORT_SYMBOL_GPL(mlx4_mw_enable);
727
728void mlx4_mw_free(struct mlx4_dev *dev, struct mlx4_mw *mw)
729{
730 int err;
731
732 if (mw->enabled == MLX4_MPT_EN_HW) {
733 err = mlx4_HW2SW_MPT(dev, NULL,
734 key_to_hw_index(mw->key) &
735 (dev->caps.num_mpts - 1));
736 if (err)
737 mlx4_warn(dev, "xxx HW2SW_MPT failed (%d)\n", err);
738
739 mw->enabled = MLX4_MPT_EN_SW;
740 }
741 if (mw->enabled)
742 mlx4_mpt_free_icm(dev, key_to_hw_index(mw->key));
743 mlx4_mpt_release(dev, key_to_hw_index(mw->key));
744}
745EXPORT_SYMBOL_GPL(mlx4_mw_free);
746
3d73c288 747int mlx4_init_mr_table(struct mlx4_dev *dev)
225c7b1f 748{
ea51b377
JM		 749 struct mlx4_priv *priv = mlx4_priv(dev);
750 struct mlx4_mr_table *mr_table = &priv->mr_table;
225c7b1f
RD		 751 int err;
752
ea51b377
JM		 753 /* Nothing to do for slaves - all MR handling is forwarded
754 * to the master */
755 if (mlx4_is_slave(dev))
756 return 0;
757
a30f1bc5
JM		 758 if (!is_power_of_2(dev->caps.num_mpts))
759 return -EINVAL;
760
225c7b1f 761 err = mlx4_bitmap_init(&mr_table->mpt_bitmap, dev->caps.num_mpts,
93fc9e1b 762 ~0, dev->caps.reserved_mrws, 0);
225c7b1f
RD		 763 if (err)
764 return err;
765
766 err = mlx4_buddy_init(&mr_table->mtt_buddy,
3de819e6 767 ilog2((u32)dev->caps.num_mtts /
2b8fb286 768 (1 << log_mtts_per_seg)));
225c7b1f
RD		 769 if (err)
770 goto err_buddy;
771
772 if (dev->caps.reserved_mtts) {
ea51b377
JM		 773 priv->reserved_mtts =
774 mlx4_alloc_mtt_range(dev,
775 fls(dev->caps.reserved_mtts - 1));
776 if (priv->reserved_mtts < 0) {
3de819e6 777 mlx4_warn(dev, "MTT table of order %u is too small.\n",
225c7b1f
RD		 778 mr_table->mtt_buddy.max_order);
779 err = -ENOMEM;
780 goto err_reserve_mtts;
781 }
782 }
783
784 return 0;
785
786err_reserve_mtts:
787 mlx4_buddy_cleanup(&mr_table->mtt_buddy);
788
789err_buddy:
790 mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
791
792 return err;
793}
794
795void mlx4_cleanup_mr_table(struct mlx4_dev *dev)
796{
ea51b377
JM		 797 struct mlx4_priv *priv = mlx4_priv(dev);
798 struct mlx4_mr_table *mr_table = &priv->mr_table;
225c7b1f 799
ea51b377
JM		 800 if (mlx4_is_slave(dev))
801 return;
802 if (priv->reserved_mtts >= 0)
803 mlx4_free_mtt_range(dev, priv->reserved_mtts,
804 fls(dev->caps.reserved_mtts - 1));
225c7b1f
RD		 805 mlx4_buddy_cleanup(&mr_table->mtt_buddy);
806 mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
807}
8ad11fb6
JM		 808
809static inline int mlx4_check_fmr(struct mlx4_fmr *fmr, u64 *page_list,
810 int npages, u64 iova)
811{
812 int i, page_mask;
813
814 if (npages > fmr->max_pages)
815 return -EINVAL;
816
817 page_mask = (1 << fmr->page_shift) - 1;
818
819 /* We are getting page lists, so va must be page aligned. */
820 if (iova & page_mask)
821 return -EINVAL;
822
823 /* Trust the user not to pass misaligned data in page_list */
824 if (0)
825 for (i = 0; i < npages; ++i) {
826 if (page_list[i] & ~page_mask)
827 return -EINVAL;
828 }
829
830 if (fmr->maps >= fmr->max_maps)
831 return -EINVAL;
832
833 return 0;
834}
835
836int mlx4_map_phys_fmr(struct mlx4_dev *dev, struct mlx4_fmr *fmr, u64 *page_list,
837 int npages, u64 iova, u32 *lkey, u32 *rkey)
838{
839 u32 key;
840 int i, err;
841
842 err = mlx4_check_fmr(fmr, page_list, npages, iova);
843 if (err)
844 return err;
845
846 ++fmr->maps;
847
848 key = key_to_hw_index(fmr->mr.key);
849 key += dev->caps.num_mpts;
850 *lkey = *rkey = fmr->mr.key = hw_index_to_key(key);
851
852 *(u8 *) fmr->mpt = MLX4_MPT_STATUS_SW;
853
854 /* Make sure MPT status is visible before writing MTT entries */
855 wmb();
856
e727f5cd
RD		 857 dma_sync_single_for_cpu(&dev->pdev->dev, fmr->dma_handle,
858 npages * sizeof(u64), DMA_TO_DEVICE);
859
8ad11fb6
JM		 860 for (i = 0; i < npages; ++i)
861 fmr->mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
862
e727f5cd
RD		 863 dma_sync_single_for_device(&dev->pdev->dev, fmr->dma_handle,
864 npages * sizeof(u64), DMA_TO_DEVICE);
8ad11fb6
JM		 865
866 fmr->mpt->key = cpu_to_be32(key);
867 fmr->mpt->lkey = cpu_to_be32(key);
868 fmr->mpt->length = cpu_to_be64(npages * (1ull << fmr->page_shift));
869 fmr->mpt->start = cpu_to_be64(iova);
870
871 /* Make MTT entries are visible before setting MPT status */
872 wmb();
873
874 *(u8 *) fmr->mpt = MLX4_MPT_STATUS_HW;
875
876 /* Make sure MPT status is visible before consumer can use FMR */
877 wmb();
878
879 return 0;
880}
881EXPORT_SYMBOL_GPL(mlx4_map_phys_fmr);
882
883int mlx4_fmr_alloc(struct mlx4_dev *dev, u32 pd, u32 access, int max_pages,
884 int max_maps, u8 page_shift, struct mlx4_fmr *fmr)
885{
886 struct mlx4_priv *priv = mlx4_priv(dev);
8ad11fb6
JM		 887 int err = -ENOMEM;
888
a5bbe892
EC		 889 if (max_maps > dev->caps.max_fmr_maps)
890 return -EINVAL;
891
c5057ddc 892 if (page_shift < (ffs(dev->caps.page_size_cap) - 1) || page_shift >= 32)
8ad11fb6
JM		 893 return -EINVAL;
894
895 /* All MTTs must fit in the same page */
896 if (max_pages * sizeof *fmr->mtts > PAGE_SIZE)
897 return -EINVAL;
898
899 fmr->page_shift = page_shift;
900 fmr->max_pages = max_pages;
901 fmr->max_maps = max_maps;
902 fmr->maps = 0;
903
904 err = mlx4_mr_alloc(dev, pd, 0, 0, access, max_pages,
905 page_shift, &fmr->mr);
906 if (err)
907 return err;
908
8ad11fb6 909 fmr->mtts = mlx4_table_find(&priv->mr_table.mtt_table,
2b8fb286 910 fmr->mr.mtt.offset,
8ad11fb6 911 &fmr->dma_handle);
2b8fb286 912
8ad11fb6
JM		 913 if (!fmr->mtts) {
914 err = -ENOMEM;
915 goto err_free;
916 }
917
8ad11fb6
JM		 918 return 0;
919
920err_free:
61083720 921 (void) mlx4_mr_free(dev, &fmr->mr);
8ad11fb6
JM		 922 return err;
923}
924EXPORT_SYMBOL_GPL(mlx4_fmr_alloc);
925
926int mlx4_fmr_enable(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
927{
11e75a74
JM		 928 struct mlx4_priv *priv = mlx4_priv(dev);
929 int err;
930
931 err = mlx4_mr_enable(dev, &fmr->mr);
932 if (err)
933 return err;
934
935 fmr->mpt = mlx4_table_find(&priv->mr_table.dmpt_table,
936 key_to_hw_index(fmr->mr.key), NULL);
937 if (!fmr->mpt)
938 return -ENOMEM;
939
940 return 0;
8ad11fb6
JM		 941}
942EXPORT_SYMBOL_GPL(mlx4_fmr_enable);
943
944void mlx4_fmr_unmap(struct mlx4_dev *dev, struct mlx4_fmr *fmr,
945 u32 *lkey, u32 *rkey)
946{
ea51b377
JM		 947 struct mlx4_cmd_mailbox *mailbox;
948 int err;
949
8ad11fb6
JM		 950 if (!fmr->maps)
951 return;
952
8ad11fb6
JM		 953 fmr->maps = 0;
954
ea51b377
JM		 955 mailbox = mlx4_alloc_cmd_mailbox(dev);
956 if (IS_ERR(mailbox)) {
957 err = PTR_ERR(mailbox);
958 printk(KERN_WARNING "mlx4_ib: mlx4_alloc_cmd_mailbox"
959 " failed (%d)\n", err);
960 return;
961 }
962
963 err = mlx4_HW2SW_MPT(dev, NULL,
964 key_to_hw_index(fmr->mr.key) &
965 (dev->caps.num_mpts - 1));
966 mlx4_free_cmd_mailbox(dev, mailbox);
967 if (err) {
968 printk(KERN_WARNING "mlx4_ib: mlx4_HW2SW_MPT failed (%d)\n",
969 err);
970 return;
971 }
b20e519a 972 fmr->mr.enabled = MLX4_MPT_EN_SW;
8ad11fb6
JM		 973}
974EXPORT_SYMBOL_GPL(mlx4_fmr_unmap);
975
976int mlx4_fmr_free(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
977{
61083720
SM		 978 int ret;
979
8ad11fb6
JM		 980 if (fmr->maps)
981 return -EBUSY;
982
61083720
SM		 983 ret = mlx4_mr_free(dev, &fmr->mr);
984 if (ret)
985 return ret;
b20e519a 986 fmr->mr.enabled = MLX4_MPT_DISABLED;
8ad11fb6
JM		 987
988 return 0;
989}
990EXPORT_SYMBOL_GPL(mlx4_fmr_free);
991
992int mlx4_SYNC_TPT(struct mlx4_dev *dev)
993{
f9baff50 994 return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_SYNC_TPT, 1000,
5e92d803 995 MLX4_CMD_NATIVE);
8ad11fb6
JM		 996}
997EXPORT_SYMBOL_GPL(mlx4_SYNC_TPT);
Linux 6.x block layer and io_uring tree(s)