Commit | Line | Data |
---|---|---|
225c7b1f RD |
1 | /* |
2 | * Copyright (c) 2004 Topspin Communications. All rights reserved. | |
3 | * Copyright (c) 2005 Mellanox Technologies. All rights reserved. | |
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 | ||
35 | #include <linux/init.h> | |
36 | #include <linux/errno.h> | |
37 | ||
38 | #include <linux/mlx4/cmd.h> | |
39 | ||
40 | #include "mlx4.h" | |
41 | #include "icm.h" | |
42 | ||
43 | /* | |
44 | * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits. | |
45 | */ | |
46 | struct mlx4_mpt_entry { | |
47 | __be32 flags; | |
48 | __be32 qpn; | |
49 | __be32 key; | |
50 | __be32 pd; | |
51 | __be64 start; | |
52 | __be64 length; | |
53 | __be32 lkey; | |
54 | __be32 win_cnt; | |
55 | u8 reserved1[3]; | |
56 | u8 mtt_rep; | |
57 | __be64 mtt_seg; | |
58 | __be32 mtt_sz; | |
59 | __be32 entity_size; | |
60 | __be32 first_byte_offset; | |
61 | } __attribute__((packed)); | |
62 | ||
63 | #define MLX4_MPT_FLAG_SW_OWNS (0xfUL << 28) | |
64 | #define MLX4_MPT_FLAG_MIO (1 << 17) | |
65 | #define MLX4_MPT_FLAG_BIND_ENABLE (1 << 15) | |
66 | #define MLX4_MPT_FLAG_PHYSICAL (1 << 9) | |
67 | #define MLX4_MPT_FLAG_REGION (1 << 8) | |
68 | ||
69 | #define MLX4_MTT_FLAG_PRESENT 1 | |
70 | ||
71 | static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order) | |
72 | { | |
73 | int o; | |
74 | int m; | |
75 | u32 seg; | |
76 | ||
77 | spin_lock(&buddy->lock); | |
78 | ||
79 | for (o = order; o <= buddy->max_order; ++o) { | |
80 | m = 1 << (buddy->max_order - o); | |
81 | seg = find_first_bit(buddy->bits[o], m); | |
82 | if (seg < m) | |
83 | goto found; | |
84 | } | |
85 | ||
86 | spin_unlock(&buddy->lock); | |
87 | return -1; | |
88 | ||
89 | found: | |
90 | clear_bit(seg, buddy->bits[o]); | |
91 | ||
92 | while (o > order) { | |
93 | --o; | |
94 | seg <<= 1; | |
95 | set_bit(seg ^ 1, buddy->bits[o]); | |
96 | } | |
97 | ||
98 | spin_unlock(&buddy->lock); | |
99 | ||
100 | seg <<= order; | |
101 | ||
102 | return seg; | |
103 | } | |
104 | ||
105 | static void mlx4_buddy_free(struct mlx4_buddy *buddy, u32 seg, int order) | |
106 | { | |
107 | seg >>= order; | |
108 | ||
109 | spin_lock(&buddy->lock); | |
110 | ||
111 | while (test_bit(seg ^ 1, buddy->bits[order])) { | |
112 | clear_bit(seg ^ 1, buddy->bits[order]); | |
113 | seg >>= 1; | |
114 | ++order; | |
115 | } | |
116 | ||
117 | set_bit(seg, buddy->bits[order]); | |
118 | ||
119 | spin_unlock(&buddy->lock); | |
120 | } | |
121 | ||
122 | static int __devinit mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order) | |
123 | { | |
124 | int i, s; | |
125 | ||
126 | buddy->max_order = max_order; | |
127 | spin_lock_init(&buddy->lock); | |
128 | ||
129 | buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *), | |
130 | GFP_KERNEL); | |
131 | if (!buddy->bits) | |
132 | goto err_out; | |
133 | ||
134 | for (i = 0; i <= buddy->max_order; ++i) { | |
135 | s = BITS_TO_LONGS(1 << (buddy->max_order - i)); | |
136 | buddy->bits[i] = kmalloc(s * sizeof (long), GFP_KERNEL); | |
137 | if (!buddy->bits[i]) | |
138 | goto err_out_free; | |
139 | bitmap_zero(buddy->bits[i], 1 << (buddy->max_order - i)); | |
140 | } | |
141 | ||
142 | set_bit(0, buddy->bits[buddy->max_order]); | |
143 | ||
144 | return 0; | |
145 | ||
146 | err_out_free: | |
147 | for (i = 0; i <= buddy->max_order; ++i) | |
148 | kfree(buddy->bits[i]); | |
149 | ||
150 | kfree(buddy->bits); | |
151 | ||
152 | err_out: | |
153 | return -ENOMEM; | |
154 | } | |
155 | ||
156 | static void mlx4_buddy_cleanup(struct mlx4_buddy *buddy) | |
157 | { | |
158 | int i; | |
159 | ||
160 | for (i = 0; i <= buddy->max_order; ++i) | |
161 | kfree(buddy->bits[i]); | |
162 | ||
163 | kfree(buddy->bits); | |
164 | } | |
165 | ||
166 | static u32 mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order) | |
167 | { | |
168 | struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table; | |
169 | u32 seg; | |
170 | ||
171 | seg = mlx4_buddy_alloc(&mr_table->mtt_buddy, order); | |
172 | if (seg == -1) | |
173 | return -1; | |
174 | ||
175 | if (mlx4_table_get_range(dev, &mr_table->mtt_table, seg, | |
176 | seg + (1 << order) - 1)) { | |
177 | mlx4_buddy_free(&mr_table->mtt_buddy, seg, order); | |
178 | return -1; | |
179 | } | |
180 | ||
181 | return seg; | |
182 | } | |
183 | ||
184 | int mlx4_mtt_init(struct mlx4_dev *dev, int npages, int page_shift, | |
185 | struct mlx4_mtt *mtt) | |
186 | { | |
187 | int i; | |
188 | ||
189 | if (!npages) { | |
190 | mtt->order = -1; | |
191 | mtt->page_shift = MLX4_ICM_PAGE_SHIFT; | |
192 | return 0; | |
193 | } else | |
194 | mtt->page_shift = page_shift; | |
195 | ||
196 | for (mtt->order = 0, i = MLX4_MTT_ENTRY_PER_SEG; i < npages; i <<= 1) | |
197 | ++mtt->order; | |
198 | ||
199 | mtt->first_seg = mlx4_alloc_mtt_range(dev, mtt->order); | |
200 | if (mtt->first_seg == -1) | |
201 | return -ENOMEM; | |
202 | ||
203 | return 0; | |
204 | } | |
205 | EXPORT_SYMBOL_GPL(mlx4_mtt_init); | |
206 | ||
207 | void mlx4_mtt_cleanup(struct mlx4_dev *dev, struct mlx4_mtt *mtt) | |
208 | { | |
209 | struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table; | |
210 | ||
211 | if (mtt->order < 0) | |
212 | return; | |
213 | ||
214 | mlx4_buddy_free(&mr_table->mtt_buddy, mtt->first_seg, mtt->order); | |
215 | mlx4_table_put_range(dev, &mr_table->mtt_table, mtt->first_seg, | |
216 | mtt->first_seg + (1 << mtt->order) - 1); | |
217 | } | |
218 | EXPORT_SYMBOL_GPL(mlx4_mtt_cleanup); | |
219 | ||
220 | u64 mlx4_mtt_addr(struct mlx4_dev *dev, struct mlx4_mtt *mtt) | |
221 | { | |
222 | return (u64) mtt->first_seg * dev->caps.mtt_entry_sz; | |
223 | } | |
224 | EXPORT_SYMBOL_GPL(mlx4_mtt_addr); | |
225 | ||
226 | static u32 hw_index_to_key(u32 ind) | |
227 | { | |
228 | return (ind >> 24) | (ind << 8); | |
229 | } | |
230 | ||
231 | static u32 key_to_hw_index(u32 key) | |
232 | { | |
233 | return (key << 24) | (key >> 8); | |
234 | } | |
235 | ||
236 | static int mlx4_SW2HW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox, | |
237 | int mpt_index) | |
238 | { | |
239 | return mlx4_cmd(dev, mailbox->dma, mpt_index, 0, MLX4_CMD_SW2HW_MPT, | |
240 | MLX4_CMD_TIME_CLASS_B); | |
241 | } | |
242 | ||
243 | static int mlx4_HW2SW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox, | |
244 | int mpt_index) | |
245 | { | |
246 | return mlx4_cmd_box(dev, 0, mailbox ? mailbox->dma : 0, mpt_index, | |
247 | !mailbox, MLX4_CMD_HW2SW_MPT, MLX4_CMD_TIME_CLASS_B); | |
248 | } | |
249 | ||
250 | int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access, | |
251 | int npages, int page_shift, struct mlx4_mr *mr) | |
252 | { | |
253 | struct mlx4_priv *priv = mlx4_priv(dev); | |
254 | u32 index; | |
255 | int err; | |
256 | ||
257 | index = mlx4_bitmap_alloc(&priv->mr_table.mpt_bitmap); | |
258 | if (index == -1) { | |
259 | err = -ENOMEM; | |
260 | goto err; | |
261 | } | |
262 | ||
263 | mr->iova = iova; | |
264 | mr->size = size; | |
265 | mr->pd = pd; | |
266 | mr->access = access; | |
267 | mr->enabled = 0; | |
268 | mr->key = hw_index_to_key(index); | |
269 | ||
270 | err = mlx4_mtt_init(dev, npages, page_shift, &mr->mtt); | |
271 | if (err) | |
272 | goto err_index; | |
273 | ||
274 | return 0; | |
275 | ||
276 | err_index: | |
277 | mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, index); | |
278 | ||
279 | err: | |
280 | kfree(mr); | |
281 | return err; | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(mlx4_mr_alloc); | |
284 | ||
285 | void mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr) | |
286 | { | |
287 | struct mlx4_priv *priv = mlx4_priv(dev); | |
288 | int err; | |
289 | ||
290 | if (mr->enabled) { | |
291 | err = mlx4_HW2SW_MPT(dev, NULL, | |
292 | key_to_hw_index(mr->key) & | |
293 | (dev->caps.num_mpts - 1)); | |
294 | if (err) | |
295 | mlx4_warn(dev, "HW2SW_MPT failed (%d)\n", err); | |
296 | } | |
297 | ||
298 | mlx4_mtt_cleanup(dev, &mr->mtt); | |
299 | mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, key_to_hw_index(mr->key)); | |
300 | } | |
301 | EXPORT_SYMBOL_GPL(mlx4_mr_free); | |
302 | ||
303 | int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr) | |
304 | { | |
305 | struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table; | |
306 | struct mlx4_cmd_mailbox *mailbox; | |
307 | struct mlx4_mpt_entry *mpt_entry; | |
308 | int err; | |
309 | ||
310 | err = mlx4_table_get(dev, &mr_table->dmpt_table, key_to_hw_index(mr->key)); | |
311 | if (err) | |
312 | return err; | |
313 | ||
314 | mailbox = mlx4_alloc_cmd_mailbox(dev); | |
315 | if (IS_ERR(mailbox)) { | |
316 | err = PTR_ERR(mailbox); | |
317 | goto err_table; | |
318 | } | |
319 | mpt_entry = mailbox->buf; | |
320 | ||
321 | memset(mpt_entry, 0, sizeof *mpt_entry); | |
322 | ||
323 | mpt_entry->flags = cpu_to_be32(MLX4_MPT_FLAG_SW_OWNS | | |
324 | MLX4_MPT_FLAG_MIO | | |
325 | MLX4_MPT_FLAG_REGION | | |
326 | mr->access); | |
327 | if (mr->mtt.order < 0) | |
328 | mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL); | |
329 | ||
330 | mpt_entry->key = cpu_to_be32(key_to_hw_index(mr->key)); | |
331 | mpt_entry->pd = cpu_to_be32(mr->pd); | |
332 | mpt_entry->start = cpu_to_be64(mr->iova); | |
333 | mpt_entry->length = cpu_to_be64(mr->size); | |
334 | mpt_entry->entity_size = cpu_to_be32(mr->mtt.page_shift); | |
335 | mpt_entry->mtt_seg = cpu_to_be64(mlx4_mtt_addr(dev, &mr->mtt)); | |
336 | ||
337 | err = mlx4_SW2HW_MPT(dev, mailbox, | |
338 | key_to_hw_index(mr->key) & (dev->caps.num_mpts - 1)); | |
339 | if (err) { | |
340 | mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err); | |
341 | goto err_cmd; | |
342 | } | |
343 | ||
344 | mr->enabled = 1; | |
345 | ||
346 | mlx4_free_cmd_mailbox(dev, mailbox); | |
347 | ||
348 | return 0; | |
349 | ||
350 | err_cmd: | |
351 | mlx4_free_cmd_mailbox(dev, mailbox); | |
352 | ||
353 | err_table: | |
354 | mlx4_table_put(dev, &mr_table->dmpt_table, key_to_hw_index(mr->key)); | |
355 | return err; | |
356 | } | |
357 | EXPORT_SYMBOL_GPL(mlx4_mr_enable); | |
358 | ||
359 | static int mlx4_WRITE_MTT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox, | |
360 | int num_mtt) | |
361 | { | |
362 | return mlx4_cmd(dev, mailbox->dma, num_mtt, 0, MLX4_CMD_WRITE_MTT, | |
363 | MLX4_CMD_TIME_CLASS_B); | |
364 | } | |
365 | ||
366 | int mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt, | |
367 | int start_index, int npages, u64 *page_list) | |
368 | { | |
369 | struct mlx4_cmd_mailbox *mailbox; | |
370 | __be64 *mtt_entry; | |
371 | int i; | |
372 | int err = 0; | |
373 | ||
374 | if (mtt->order < 0) | |
375 | return -EINVAL; | |
376 | ||
377 | mailbox = mlx4_alloc_cmd_mailbox(dev); | |
378 | if (IS_ERR(mailbox)) | |
379 | return PTR_ERR(mailbox); | |
380 | ||
381 | mtt_entry = mailbox->buf; | |
382 | ||
383 | while (npages > 0) { | |
384 | mtt_entry[0] = cpu_to_be64(mlx4_mtt_addr(dev, mtt) + start_index * 8); | |
385 | mtt_entry[1] = 0; | |
386 | ||
387 | for (i = 0; i < npages && i < MLX4_MAILBOX_SIZE / 8 - 2; ++i) | |
388 | mtt_entry[i + 2] = cpu_to_be64(page_list[i] | | |
389 | MLX4_MTT_FLAG_PRESENT); | |
390 | ||
391 | /* | |
392 | * If we have an odd number of entries to write, add | |
393 | * one more dummy entry for firmware efficiency. | |
394 | */ | |
395 | if (i & 1) | |
396 | mtt_entry[i + 2] = 0; | |
397 | ||
398 | err = mlx4_WRITE_MTT(dev, mailbox, (i + 1) & ~1); | |
399 | if (err) | |
400 | goto out; | |
401 | ||
402 | npages -= i; | |
403 | start_index += i; | |
404 | page_list += i; | |
405 | } | |
406 | ||
407 | out: | |
408 | mlx4_free_cmd_mailbox(dev, mailbox); | |
409 | ||
410 | return err; | |
411 | } | |
412 | EXPORT_SYMBOL_GPL(mlx4_write_mtt); | |
413 | ||
414 | int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt, | |
415 | struct mlx4_buf *buf) | |
416 | { | |
417 | u64 *page_list; | |
418 | int err; | |
419 | int i; | |
420 | ||
421 | page_list = kmalloc(buf->npages * sizeof *page_list, GFP_KERNEL); | |
422 | if (!page_list) | |
423 | return -ENOMEM; | |
424 | ||
425 | for (i = 0; i < buf->npages; ++i) | |
426 | if (buf->nbufs == 1) | |
427 | page_list[i] = buf->u.direct.map + (i << buf->page_shift); | |
428 | else | |
429 | page_list[i] = buf->u.page_list[i].map; | |
430 | ||
431 | err = mlx4_write_mtt(dev, mtt, 0, buf->npages, page_list); | |
432 | ||
433 | kfree(page_list); | |
434 | return err; | |
435 | } | |
436 | EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt); | |
437 | ||
438 | int __devinit mlx4_init_mr_table(struct mlx4_dev *dev) | |
439 | { | |
440 | struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table; | |
441 | int err; | |
442 | ||
443 | err = mlx4_bitmap_init(&mr_table->mpt_bitmap, dev->caps.num_mpts, | |
444 | ~0, dev->caps.reserved_mrws); | |
445 | if (err) | |
446 | return err; | |
447 | ||
448 | err = mlx4_buddy_init(&mr_table->mtt_buddy, | |
449 | ilog2(dev->caps.num_mtt_segs)); | |
450 | if (err) | |
451 | goto err_buddy; | |
452 | ||
453 | if (dev->caps.reserved_mtts) { | |
454 | if (mlx4_alloc_mtt_range(dev, ilog2(dev->caps.reserved_mtts)) == -1) { | |
455 | mlx4_warn(dev, "MTT table of order %d is too small.\n", | |
456 | mr_table->mtt_buddy.max_order); | |
457 | err = -ENOMEM; | |
458 | goto err_reserve_mtts; | |
459 | } | |
460 | } | |
461 | ||
462 | return 0; | |
463 | ||
464 | err_reserve_mtts: | |
465 | mlx4_buddy_cleanup(&mr_table->mtt_buddy); | |
466 | ||
467 | err_buddy: | |
468 | mlx4_bitmap_cleanup(&mr_table->mpt_bitmap); | |
469 | ||
470 | return err; | |
471 | } | |
472 | ||
473 | void mlx4_cleanup_mr_table(struct mlx4_dev *dev) | |
474 | { | |
475 | struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table; | |
476 | ||
477 | mlx4_buddy_cleanup(&mr_table->mtt_buddy); | |
478 | mlx4_bitmap_cleanup(&mr_table->mpt_bitmap); | |
479 | } |