xref: /linux/drivers/net/ethernet/mellanox/mlx4/mr.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
1 /*
2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005, 2006, 2007, 2008 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 #include <linux/export.h>
38 #include <linux/slab.h>
39 #include <linux/kernel.h>
40 
41 #include <linux/mlx4/cmd.h>
42 
43 #include "mlx4.h"
44 #include "icm.h"
45 
46 #define MLX4_MPT_FLAG_SW_OWNS	    (0xfUL << 28)
47 #define MLX4_MPT_FLAG_FREE	    (0x3UL << 28)
48 #define MLX4_MPT_FLAG_MIO	    (1 << 17)
49 #define MLX4_MPT_FLAG_BIND_ENABLE   (1 << 15)
50 #define MLX4_MPT_FLAG_PHYSICAL	    (1 <<  9)
51 #define MLX4_MPT_FLAG_REGION	    (1 <<  8)
52 
53 #define MLX4_MPT_PD_FLAG_FAST_REG   (1 << 27)
54 #define MLX4_MPT_PD_FLAG_RAE	    (1 << 28)
55 #define MLX4_MPT_PD_FLAG_EN_INV	    (3 << 24)
56 
57 #define MLX4_MPT_STATUS_SW		0xF0
58 #define MLX4_MPT_STATUS_HW		0x00
59 
60 static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order)
61 {
62 	int o;
63 	int m;
64 	u32 seg;
65 
66 	spin_lock(&buddy->lock);
67 
68 	for (o = order; o <= buddy->max_order; ++o)
69 		if (buddy->num_free[o]) {
70 			m = 1 << (buddy->max_order - o);
71 			seg = find_first_bit(buddy->bits[o], m);
72 			if (seg < m)
73 				goto found;
74 		}
75 
76 	spin_unlock(&buddy->lock);
77 	return -1;
78 
79  found:
80 	clear_bit(seg, buddy->bits[o]);
81 	--buddy->num_free[o];
82 
83 	while (o > order) {
84 		--o;
85 		seg <<= 1;
86 		set_bit(seg ^ 1, buddy->bits[o]);
87 		++buddy->num_free[o];
88 	}
89 
90 	spin_unlock(&buddy->lock);
91 
92 	seg <<= order;
93 
94 	return seg;
95 }
96 
97 static void mlx4_buddy_free(struct mlx4_buddy *buddy, u32 seg, int order)
98 {
99 	seg >>= order;
100 
101 	spin_lock(&buddy->lock);
102 
103 	while (test_bit(seg ^ 1, buddy->bits[order])) {
104 		clear_bit(seg ^ 1, buddy->bits[order]);
105 		--buddy->num_free[order];
106 		seg >>= 1;
107 		++order;
108 	}
109 
110 	set_bit(seg, buddy->bits[order]);
111 	++buddy->num_free[order];
112 
113 	spin_unlock(&buddy->lock);
114 }
115 
116 static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order)
117 {
118 	int i, s;
119 
120 	buddy->max_order = max_order;
121 	spin_lock_init(&buddy->lock);
122 
123 	buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *),
124 			      GFP_KERNEL);
125 	buddy->num_free = kcalloc((buddy->max_order + 1), sizeof *buddy->num_free,
126 				  GFP_KERNEL);
127 	if (!buddy->bits || !buddy->num_free)
128 		goto err_out;
129 
130 	for (i = 0; i <= buddy->max_order; ++i) {
131 		s = BITS_TO_LONGS(1 << (buddy->max_order - i));
132 		buddy->bits[i] = kmalloc(s * sizeof (long), GFP_KERNEL);
133 		if (!buddy->bits[i])
134 			goto err_out_free;
135 		bitmap_zero(buddy->bits[i], 1 << (buddy->max_order - i));
136 	}
137 
138 	set_bit(0, buddy->bits[buddy->max_order]);
139 	buddy->num_free[buddy->max_order] = 1;
140 
141 	return 0;
142 
143 err_out_free:
144 	for (i = 0; i <= buddy->max_order; ++i)
145 		kfree(buddy->bits[i]);
146 
147 err_out:
148 	kfree(buddy->bits);
149 	kfree(buddy->num_free);
150 
151 	return -ENOMEM;
152 }
153 
154 static void mlx4_buddy_cleanup(struct mlx4_buddy *buddy)
155 {
156 	int i;
157 
158 	for (i = 0; i <= buddy->max_order; ++i)
159 		kfree(buddy->bits[i]);
160 
161 	kfree(buddy->bits);
162 	kfree(buddy->num_free);
163 }
164 
165 u32 __mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order)
166 {
167 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
168 	u32 seg;
169 	int seg_order;
170 	u32 offset;
171 
172 	seg_order = max_t(int, order - log_mtts_per_seg, 0);
173 
174 	seg = mlx4_buddy_alloc(&mr_table->mtt_buddy, seg_order);
175 	if (seg == -1)
176 		return -1;
177 
178 	offset = seg * (1 << log_mtts_per_seg);
179 
180 	if (mlx4_table_get_range(dev, &mr_table->mtt_table, offset,
181 				 offset + (1 << order) - 1)) {
182 		mlx4_buddy_free(&mr_table->mtt_buddy, seg, seg_order);
183 		return -1;
184 	}
185 
186 	return offset;
187 }
188 
189 static u32 mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order)
190 {
191 	u64 in_param;
192 	u64 out_param;
193 	int err;
194 
195 	if (mlx4_is_mfunc(dev)) {
196 		set_param_l(&in_param, order);
197 		err = mlx4_cmd_imm(dev, in_param, &out_param, RES_MTT,
198 						       RES_OP_RESERVE_AND_MAP,
199 						       MLX4_CMD_ALLOC_RES,
200 						       MLX4_CMD_TIME_CLASS_A,
201 						       MLX4_CMD_WRAPPED);
202 		if (err)
203 			return -1;
204 		return get_param_l(&out_param);
205 	}
206 	return __mlx4_alloc_mtt_range(dev, order);
207 }
208 
209 int mlx4_mtt_init(struct mlx4_dev *dev, int npages, int page_shift,
210 		  struct mlx4_mtt *mtt)
211 {
212 	int i;
213 
214 	if (!npages) {
215 		mtt->order      = -1;
216 		mtt->page_shift = MLX4_ICM_PAGE_SHIFT;
217 		return 0;
218 	} else
219 		mtt->page_shift = page_shift;
220 
221 	for (mtt->order = 0, i = 1; i < npages; i <<= 1)
222 		++mtt->order;
223 
224 	mtt->offset = mlx4_alloc_mtt_range(dev, mtt->order);
225 	if (mtt->offset == -1)
226 		return -ENOMEM;
227 
228 	return 0;
229 }
230 EXPORT_SYMBOL_GPL(mlx4_mtt_init);
231 
232 void __mlx4_free_mtt_range(struct mlx4_dev *dev, u32 offset, int order)
233 {
234 	u32 first_seg;
235 	int seg_order;
236 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
237 
238 	seg_order = max_t(int, order - log_mtts_per_seg, 0);
239 	first_seg = offset / (1 << log_mtts_per_seg);
240 
241 	mlx4_buddy_free(&mr_table->mtt_buddy, first_seg, seg_order);
242 	mlx4_table_put_range(dev, &mr_table->mtt_table, offset,
243 			     offset + (1 << order) - 1);
244 }
245 
246 static void mlx4_free_mtt_range(struct mlx4_dev *dev, u32 offset, int order)
247 {
248 	u64 in_param;
249 	int err;
250 
251 	if (mlx4_is_mfunc(dev)) {
252 		set_param_l(&in_param, offset);
253 		set_param_h(&in_param, order);
254 		err = mlx4_cmd(dev, in_param, RES_MTT, RES_OP_RESERVE_AND_MAP,
255 						       MLX4_CMD_FREE_RES,
256 						       MLX4_CMD_TIME_CLASS_A,
257 						       MLX4_CMD_WRAPPED);
258 		if (err)
259 			mlx4_warn(dev, "Failed to free mtt range at:"
260 				  "%d order:%d\n", offset, order);
261 		return;
262 	}
263 	 __mlx4_free_mtt_range(dev, offset, order);
264 }
265 
266 void mlx4_mtt_cleanup(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
267 {
268 	if (mtt->order < 0)
269 		return;
270 
271 	mlx4_free_mtt_range(dev, mtt->offset, mtt->order);
272 }
273 EXPORT_SYMBOL_GPL(mlx4_mtt_cleanup);
274 
275 u64 mlx4_mtt_addr(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
276 {
277 	return (u64) mtt->offset * dev->caps.mtt_entry_sz;
278 }
279 EXPORT_SYMBOL_GPL(mlx4_mtt_addr);
280 
281 static u32 hw_index_to_key(u32 ind)
282 {
283 	return (ind >> 24) | (ind << 8);
284 }
285 
286 static u32 key_to_hw_index(u32 key)
287 {
288 	return (key << 24) | (key >> 8);
289 }
290 
291 static int mlx4_SW2HW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
292 			  int mpt_index)
293 {
294 	return mlx4_cmd(dev, mailbox->dma | dev->caps.function , mpt_index,
295 			0, MLX4_CMD_SW2HW_MPT, MLX4_CMD_TIME_CLASS_B,
296 			MLX4_CMD_WRAPPED);
297 }
298 
299 static int mlx4_HW2SW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
300 			  int mpt_index)
301 {
302 	return mlx4_cmd_box(dev, 0, mailbox ? mailbox->dma : 0, mpt_index,
303 			    !mailbox, MLX4_CMD_HW2SW_MPT,
304 			    MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
305 }
306 
307 static int mlx4_mr_reserve_range(struct mlx4_dev *dev, int cnt, int align,
308 			  u32 *base_mridx)
309 {
310 	struct mlx4_priv *priv = mlx4_priv(dev);
311 	u32 mridx;
312 
313 	mridx = mlx4_bitmap_alloc_range(&priv->mr_table.mpt_bitmap, cnt, align);
314 	if (mridx == -1)
315 		return -ENOMEM;
316 
317 	*base_mridx = mridx;
318 	return 0;
319 
320 }
321 EXPORT_SYMBOL_GPL(mlx4_mr_reserve_range);
322 
323 static void mlx4_mr_release_range(struct mlx4_dev *dev, u32 base_mridx, int cnt)
324 {
325 	struct mlx4_priv *priv = mlx4_priv(dev);
326 	mlx4_bitmap_free_range(&priv->mr_table.mpt_bitmap, base_mridx, cnt);
327 }
328 EXPORT_SYMBOL_GPL(mlx4_mr_release_range);
329 
330 static int mlx4_mr_alloc_reserved(struct mlx4_dev *dev, u32 mridx, u32 pd,
331 			   u64 iova, u64 size, u32 access, int npages,
332 			   int page_shift, struct mlx4_mr *mr)
333 {
334 	mr->iova       = iova;
335 	mr->size       = size;
336 	mr->pd	       = pd;
337 	mr->access     = access;
338 	mr->enabled    = MLX4_MR_DISABLED;
339 	mr->key	       = hw_index_to_key(mridx);
340 
341 	return mlx4_mtt_init(dev, npages, page_shift, &mr->mtt);
342 }
343 EXPORT_SYMBOL_GPL(mlx4_mr_alloc_reserved);
344 
345 static int mlx4_WRITE_MTT(struct mlx4_dev *dev,
346 			  struct mlx4_cmd_mailbox *mailbox,
347 			  int num_entries)
348 {
349 	return mlx4_cmd(dev, mailbox->dma, num_entries, 0, MLX4_CMD_WRITE_MTT,
350 			MLX4_CMD_TIME_CLASS_A,  MLX4_CMD_WRAPPED);
351 }
352 
353 int __mlx4_mr_reserve(struct mlx4_dev *dev)
354 {
355 	struct mlx4_priv *priv = mlx4_priv(dev);
356 
357 	return mlx4_bitmap_alloc(&priv->mr_table.mpt_bitmap);
358 }
359 
360 static int mlx4_mr_reserve(struct mlx4_dev *dev)
361 {
362 	u64 out_param;
363 
364 	if (mlx4_is_mfunc(dev)) {
365 		if (mlx4_cmd_imm(dev, 0, &out_param, RES_MPT, RES_OP_RESERVE,
366 				   MLX4_CMD_ALLOC_RES,
367 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED))
368 			return -1;
369 		return get_param_l(&out_param);
370 	}
371 	return  __mlx4_mr_reserve(dev);
372 }
373 
374 void __mlx4_mr_release(struct mlx4_dev *dev, u32 index)
375 {
376 	struct mlx4_priv *priv = mlx4_priv(dev);
377 
378 	mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, index);
379 }
380 
381 static void mlx4_mr_release(struct mlx4_dev *dev, u32 index)
382 {
383 	u64 in_param;
384 
385 	if (mlx4_is_mfunc(dev)) {
386 		set_param_l(&in_param, index);
387 		if (mlx4_cmd(dev, in_param, RES_MPT, RES_OP_RESERVE,
388 			       MLX4_CMD_FREE_RES,
389 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED))
390 			mlx4_warn(dev, "Failed to release mr index:%d\n",
391 				  index);
392 		return;
393 	}
394 	__mlx4_mr_release(dev, index);
395 }
396 
397 int __mlx4_mr_alloc_icm(struct mlx4_dev *dev, u32 index)
398 {
399 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
400 
401 	return mlx4_table_get(dev, &mr_table->dmpt_table, index);
402 }
403 
404 static int mlx4_mr_alloc_icm(struct mlx4_dev *dev, u32 index)
405 {
406 	u64 param;
407 
408 	if (mlx4_is_mfunc(dev)) {
409 		set_param_l(&param, index);
410 		return mlx4_cmd_imm(dev, param, &param, RES_MPT, RES_OP_MAP_ICM,
411 							MLX4_CMD_ALLOC_RES,
412 							MLX4_CMD_TIME_CLASS_A,
413 							MLX4_CMD_WRAPPED);
414 	}
415 	return __mlx4_mr_alloc_icm(dev, index);
416 }
417 
418 void __mlx4_mr_free_icm(struct mlx4_dev *dev, u32 index)
419 {
420 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
421 
422 	mlx4_table_put(dev, &mr_table->dmpt_table, index);
423 }
424 
425 static void mlx4_mr_free_icm(struct mlx4_dev *dev, u32 index)
426 {
427 	u64 in_param;
428 
429 	if (mlx4_is_mfunc(dev)) {
430 		set_param_l(&in_param, index);
431 		if (mlx4_cmd(dev, in_param, RES_MPT, RES_OP_MAP_ICM,
432 			     MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
433 			     MLX4_CMD_WRAPPED))
434 			mlx4_warn(dev, "Failed to free icm of mr index:%d\n",
435 				  index);
436 		return;
437 	}
438 	return __mlx4_mr_free_icm(dev, index);
439 }
440 
441 int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access,
442 		  int npages, int page_shift, struct mlx4_mr *mr)
443 {
444 	u32 index;
445 	int err;
446 
447 	index = mlx4_mr_reserve(dev);
448 	if (index == -1)
449 		return -ENOMEM;
450 
451 	err = mlx4_mr_alloc_reserved(dev, index, pd, iova, size,
452 				     access, npages, page_shift, mr);
453 	if (err)
454 		mlx4_mr_release(dev, index);
455 
456 	return err;
457 }
458 EXPORT_SYMBOL_GPL(mlx4_mr_alloc);
459 
460 static void mlx4_mr_free_reserved(struct mlx4_dev *dev, struct mlx4_mr *mr)
461 {
462 	int err;
463 
464 	if (mr->enabled == MLX4_MR_EN_HW) {
465 		err = mlx4_HW2SW_MPT(dev, NULL,
466 				     key_to_hw_index(mr->key) &
467 				     (dev->caps.num_mpts - 1));
468 		if (err)
469 			mlx4_warn(dev, "xxx HW2SW_MPT failed (%d)\n", err);
470 
471 		mr->enabled = MLX4_MR_EN_SW;
472 	}
473 	mlx4_mtt_cleanup(dev, &mr->mtt);
474 }
475 EXPORT_SYMBOL_GPL(mlx4_mr_free_reserved);
476 
477 void mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr)
478 {
479 	mlx4_mr_free_reserved(dev, mr);
480 	if (mr->enabled)
481 		mlx4_mr_free_icm(dev, key_to_hw_index(mr->key));
482 	mlx4_mr_release(dev, key_to_hw_index(mr->key));
483 }
484 EXPORT_SYMBOL_GPL(mlx4_mr_free);
485 
486 int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr)
487 {
488 	struct mlx4_cmd_mailbox *mailbox;
489 	struct mlx4_mpt_entry *mpt_entry;
490 	int err;
491 
492 	err = mlx4_mr_alloc_icm(dev, key_to_hw_index(mr->key));
493 	if (err)
494 		return err;
495 
496 	mailbox = mlx4_alloc_cmd_mailbox(dev);
497 	if (IS_ERR(mailbox)) {
498 		err = PTR_ERR(mailbox);
499 		goto err_table;
500 	}
501 	mpt_entry = mailbox->buf;
502 
503 	memset(mpt_entry, 0, sizeof *mpt_entry);
504 
505 	mpt_entry->flags = cpu_to_be32(MLX4_MPT_FLAG_MIO	 |
506 				       MLX4_MPT_FLAG_REGION	 |
507 				       mr->access);
508 
509 	mpt_entry->key	       = cpu_to_be32(key_to_hw_index(mr->key));
510 	mpt_entry->pd_flags    = cpu_to_be32(mr->pd | MLX4_MPT_PD_FLAG_EN_INV);
511 	mpt_entry->start       = cpu_to_be64(mr->iova);
512 	mpt_entry->length      = cpu_to_be64(mr->size);
513 	mpt_entry->entity_size = cpu_to_be32(mr->mtt.page_shift);
514 
515 	if (mr->mtt.order < 0) {
516 		mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL);
517 		mpt_entry->mtt_addr = 0;
518 	} else {
519 		mpt_entry->mtt_addr = cpu_to_be64(mlx4_mtt_addr(dev,
520 						  &mr->mtt));
521 	}
522 
523 	if (mr->mtt.order >= 0 && mr->mtt.page_shift == 0) {
524 		/* fast register MR in free state */
525 		mpt_entry->flags    |= cpu_to_be32(MLX4_MPT_FLAG_FREE);
526 		mpt_entry->pd_flags |= cpu_to_be32(MLX4_MPT_PD_FLAG_FAST_REG |
527 						   MLX4_MPT_PD_FLAG_RAE);
528 		mpt_entry->mtt_sz    = cpu_to_be32(1 << mr->mtt.order);
529 	} else {
530 		mpt_entry->flags    |= cpu_to_be32(MLX4_MPT_FLAG_SW_OWNS);
531 	}
532 
533 	err = mlx4_SW2HW_MPT(dev, mailbox,
534 			     key_to_hw_index(mr->key) & (dev->caps.num_mpts - 1));
535 	if (err) {
536 		mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err);
537 		goto err_cmd;
538 	}
539 	mr->enabled = MLX4_MR_EN_HW;
540 
541 	mlx4_free_cmd_mailbox(dev, mailbox);
542 
543 	return 0;
544 
545 err_cmd:
546 	mlx4_free_cmd_mailbox(dev, mailbox);
547 
548 err_table:
549 	mlx4_mr_free_icm(dev, key_to_hw_index(mr->key));
550 	return err;
551 }
552 EXPORT_SYMBOL_GPL(mlx4_mr_enable);
553 
554 static int mlx4_write_mtt_chunk(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
555 				int start_index, int npages, u64 *page_list)
556 {
557 	struct mlx4_priv *priv = mlx4_priv(dev);
558 	__be64 *mtts;
559 	dma_addr_t dma_handle;
560 	int i;
561 
562 	mtts = mlx4_table_find(&priv->mr_table.mtt_table, mtt->offset +
563 			       start_index, &dma_handle);
564 
565 	if (!mtts)
566 		return -ENOMEM;
567 
568 	dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle,
569 				npages * sizeof (u64), DMA_TO_DEVICE);
570 
571 	for (i = 0; i < npages; ++i)
572 		mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
573 
574 	dma_sync_single_for_device(&dev->pdev->dev, dma_handle,
575 				   npages * sizeof (u64), DMA_TO_DEVICE);
576 
577 	return 0;
578 }
579 
580 int __mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
581 		     int start_index, int npages, u64 *page_list)
582 {
583 	int err = 0;
584 	int chunk;
585 	int mtts_per_page;
586 	int max_mtts_first_page;
587 
588 	/* compute how may mtts fit in the first page */
589 	mtts_per_page = PAGE_SIZE / sizeof(u64);
590 	max_mtts_first_page = mtts_per_page - (mtt->offset + start_index)
591 			      % mtts_per_page;
592 
593 	chunk = min_t(int, max_mtts_first_page, npages);
594 
595 	while (npages > 0) {
596 		err = mlx4_write_mtt_chunk(dev, mtt, start_index, chunk, page_list);
597 		if (err)
598 			return err;
599 		npages      -= chunk;
600 		start_index += chunk;
601 		page_list   += chunk;
602 
603 		chunk = min_t(int, mtts_per_page, npages);
604 	}
605 	return err;
606 }
607 
608 int mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
609 		   int start_index, int npages, u64 *page_list)
610 {
611 	struct mlx4_cmd_mailbox *mailbox = NULL;
612 	__be64 *inbox = NULL;
613 	int chunk;
614 	int err = 0;
615 	int i;
616 
617 	if (mtt->order < 0)
618 		return -EINVAL;
619 
620 	if (mlx4_is_mfunc(dev)) {
621 		mailbox = mlx4_alloc_cmd_mailbox(dev);
622 		if (IS_ERR(mailbox))
623 			return PTR_ERR(mailbox);
624 		inbox = mailbox->buf;
625 
626 		while (npages > 0) {
627 			chunk = min_t(int, MLX4_MAILBOX_SIZE / sizeof(u64) - 2,
628 				      npages);
629 			inbox[0] = cpu_to_be64(mtt->offset + start_index);
630 			inbox[1] = 0;
631 			for (i = 0; i < chunk; ++i)
632 				inbox[i + 2] = cpu_to_be64(page_list[i] |
633 					       MLX4_MTT_FLAG_PRESENT);
634 			err = mlx4_WRITE_MTT(dev, mailbox, chunk);
635 			if (err) {
636 				mlx4_free_cmd_mailbox(dev, mailbox);
637 				return err;
638 			}
639 
640 			npages      -= chunk;
641 			start_index += chunk;
642 			page_list   += chunk;
643 		}
644 		mlx4_free_cmd_mailbox(dev, mailbox);
645 		return err;
646 	}
647 
648 	return __mlx4_write_mtt(dev, mtt, start_index, npages, page_list);
649 }
650 EXPORT_SYMBOL_GPL(mlx4_write_mtt);
651 
652 int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
653 		       struct mlx4_buf *buf)
654 {
655 	u64 *page_list;
656 	int err;
657 	int i;
658 
659 	page_list = kmalloc(buf->npages * sizeof *page_list, GFP_KERNEL);
660 	if (!page_list)
661 		return -ENOMEM;
662 
663 	for (i = 0; i < buf->npages; ++i)
664 		if (buf->nbufs == 1)
665 			page_list[i] = buf->direct.map + (i << buf->page_shift);
666 		else
667 			page_list[i] = buf->page_list[i].map;
668 
669 	err = mlx4_write_mtt(dev, mtt, 0, buf->npages, page_list);
670 
671 	kfree(page_list);
672 	return err;
673 }
674 EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt);
675 
676 int mlx4_init_mr_table(struct mlx4_dev *dev)
677 {
678 	struct mlx4_priv *priv = mlx4_priv(dev);
679 	struct mlx4_mr_table *mr_table = &priv->mr_table;
680 	int err;
681 
682 	if (!is_power_of_2(dev->caps.num_mpts))
683 		return -EINVAL;
684 
685 	/* Nothing to do for slaves - all MR handling is forwarded
686 	* to the master */
687 	if (mlx4_is_slave(dev))
688 		return 0;
689 
690 	err = mlx4_bitmap_init(&mr_table->mpt_bitmap, dev->caps.num_mpts,
691 			       ~0, dev->caps.reserved_mrws, 0);
692 	if (err)
693 		return err;
694 
695 	err = mlx4_buddy_init(&mr_table->mtt_buddy,
696 			      ilog2(dev->caps.num_mtts /
697 			      (1 << log_mtts_per_seg)));
698 	if (err)
699 		goto err_buddy;
700 
701 	if (dev->caps.reserved_mtts) {
702 		priv->reserved_mtts =
703 			mlx4_alloc_mtt_range(dev,
704 					     fls(dev->caps.reserved_mtts - 1));
705 		if (priv->reserved_mtts < 0) {
706 			mlx4_warn(dev, "MTT table of order %d is too small.\n",
707 				  mr_table->mtt_buddy.max_order);
708 			err = -ENOMEM;
709 			goto err_reserve_mtts;
710 		}
711 	}
712 
713 	return 0;
714 
715 err_reserve_mtts:
716 	mlx4_buddy_cleanup(&mr_table->mtt_buddy);
717 
718 err_buddy:
719 	mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
720 
721 	return err;
722 }
723 
724 void mlx4_cleanup_mr_table(struct mlx4_dev *dev)
725 {
726 	struct mlx4_priv *priv = mlx4_priv(dev);
727 	struct mlx4_mr_table *mr_table = &priv->mr_table;
728 
729 	if (mlx4_is_slave(dev))
730 		return;
731 	if (priv->reserved_mtts >= 0)
732 		mlx4_free_mtt_range(dev, priv->reserved_mtts,
733 				    fls(dev->caps.reserved_mtts - 1));
734 	mlx4_buddy_cleanup(&mr_table->mtt_buddy);
735 	mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
736 }
737 
738 static inline int mlx4_check_fmr(struct mlx4_fmr *fmr, u64 *page_list,
739 				  int npages, u64 iova)
740 {
741 	int i, page_mask;
742 
743 	if (npages > fmr->max_pages)
744 		return -EINVAL;
745 
746 	page_mask = (1 << fmr->page_shift) - 1;
747 
748 	/* We are getting page lists, so va must be page aligned. */
749 	if (iova & page_mask)
750 		return -EINVAL;
751 
752 	/* Trust the user not to pass misaligned data in page_list */
753 	if (0)
754 		for (i = 0; i < npages; ++i) {
755 			if (page_list[i] & ~page_mask)
756 				return -EINVAL;
757 		}
758 
759 	if (fmr->maps >= fmr->max_maps)
760 		return -EINVAL;
761 
762 	return 0;
763 }
764 
765 int mlx4_map_phys_fmr(struct mlx4_dev *dev, struct mlx4_fmr *fmr, u64 *page_list,
766 		      int npages, u64 iova, u32 *lkey, u32 *rkey)
767 {
768 	u32 key;
769 	int i, err;
770 
771 	err = mlx4_check_fmr(fmr, page_list, npages, iova);
772 	if (err)
773 		return err;
774 
775 	++fmr->maps;
776 
777 	key = key_to_hw_index(fmr->mr.key);
778 	key += dev->caps.num_mpts;
779 	*lkey = *rkey = fmr->mr.key = hw_index_to_key(key);
780 
781 	*(u8 *) fmr->mpt = MLX4_MPT_STATUS_SW;
782 
783 	/* Make sure MPT status is visible before writing MTT entries */
784 	wmb();
785 
786 	dma_sync_single_for_cpu(&dev->pdev->dev, fmr->dma_handle,
787 				npages * sizeof(u64), DMA_TO_DEVICE);
788 
789 	for (i = 0; i < npages; ++i)
790 		fmr->mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
791 
792 	dma_sync_single_for_device(&dev->pdev->dev, fmr->dma_handle,
793 				   npages * sizeof(u64), DMA_TO_DEVICE);
794 
795 	fmr->mpt->key    = cpu_to_be32(key);
796 	fmr->mpt->lkey   = cpu_to_be32(key);
797 	fmr->mpt->length = cpu_to_be64(npages * (1ull << fmr->page_shift));
798 	fmr->mpt->start  = cpu_to_be64(iova);
799 
800 	/* Make MTT entries are visible before setting MPT status */
801 	wmb();
802 
803 	*(u8 *) fmr->mpt = MLX4_MPT_STATUS_HW;
804 
805 	/* Make sure MPT status is visible before consumer can use FMR */
806 	wmb();
807 
808 	return 0;
809 }
810 EXPORT_SYMBOL_GPL(mlx4_map_phys_fmr);
811 
812 int mlx4_fmr_alloc(struct mlx4_dev *dev, u32 pd, u32 access, int max_pages,
813 		   int max_maps, u8 page_shift, struct mlx4_fmr *fmr)
814 {
815 	struct mlx4_priv *priv = mlx4_priv(dev);
816 	u64 mtt_offset;
817 	int err = -ENOMEM;
818 
819 	if (page_shift < (ffs(dev->caps.page_size_cap) - 1) || page_shift >= 32)
820 		return -EINVAL;
821 
822 	/* All MTTs must fit in the same page */
823 	if (max_pages * sizeof *fmr->mtts > PAGE_SIZE)
824 		return -EINVAL;
825 
826 	fmr->page_shift = page_shift;
827 	fmr->max_pages  = max_pages;
828 	fmr->max_maps   = max_maps;
829 	fmr->maps = 0;
830 
831 	err = mlx4_mr_alloc(dev, pd, 0, 0, access, max_pages,
832 			    page_shift, &fmr->mr);
833 	if (err)
834 		return err;
835 
836 	mtt_offset = fmr->mr.mtt.offset * dev->caps.mtt_entry_sz;
837 
838 	fmr->mtts = mlx4_table_find(&priv->mr_table.mtt_table,
839 				    fmr->mr.mtt.offset,
840 				    &fmr->dma_handle);
841 
842 	if (!fmr->mtts) {
843 		err = -ENOMEM;
844 		goto err_free;
845 	}
846 
847 	return 0;
848 
849 err_free:
850 	mlx4_mr_free(dev, &fmr->mr);
851 	return err;
852 }
853 EXPORT_SYMBOL_GPL(mlx4_fmr_alloc);
854 
855 static int mlx4_fmr_alloc_reserved(struct mlx4_dev *dev, u32 mridx,
856 			    u32 pd, u32 access, int max_pages,
857 			    int max_maps, u8 page_shift, struct mlx4_fmr *fmr)
858 {
859 	struct mlx4_priv *priv = mlx4_priv(dev);
860 	int err = -ENOMEM;
861 
862 	if (page_shift < (ffs(dev->caps.page_size_cap) - 1) || page_shift >= 32)
863 		return -EINVAL;
864 
865 	/* All MTTs must fit in the same page */
866 	if (max_pages * sizeof *fmr->mtts > PAGE_SIZE)
867 		return -EINVAL;
868 
869 	fmr->page_shift = page_shift;
870 	fmr->max_pages  = max_pages;
871 	fmr->max_maps   = max_maps;
872 	fmr->maps = 0;
873 
874 	err = mlx4_mr_alloc_reserved(dev, mridx, pd, 0, 0, access, max_pages,
875 				     page_shift, &fmr->mr);
876 	if (err)
877 		return err;
878 
879 	fmr->mtts = mlx4_table_find(&priv->mr_table.mtt_table,
880 				    fmr->mr.mtt.offset,
881 				    &fmr->dma_handle);
882 	if (!fmr->mtts) {
883 		err = -ENOMEM;
884 		goto err_free;
885 	}
886 
887 	return 0;
888 
889 err_free:
890 	mlx4_mr_free_reserved(dev, &fmr->mr);
891 	return err;
892 }
893 EXPORT_SYMBOL_GPL(mlx4_fmr_alloc_reserved);
894 
895 int mlx4_fmr_enable(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
896 {
897 	struct mlx4_priv *priv = mlx4_priv(dev);
898 	int err;
899 
900 	err = mlx4_mr_enable(dev, &fmr->mr);
901 	if (err)
902 		return err;
903 
904 	fmr->mpt = mlx4_table_find(&priv->mr_table.dmpt_table,
905 				    key_to_hw_index(fmr->mr.key), NULL);
906 	if (!fmr->mpt)
907 		return -ENOMEM;
908 
909 	return 0;
910 }
911 EXPORT_SYMBOL_GPL(mlx4_fmr_enable);
912 
913 void mlx4_fmr_unmap(struct mlx4_dev *dev, struct mlx4_fmr *fmr,
914 		    u32 *lkey, u32 *rkey)
915 {
916 	struct mlx4_cmd_mailbox *mailbox;
917 	int err;
918 
919 	if (!fmr->maps)
920 		return;
921 
922 	fmr->maps = 0;
923 
924 	mailbox = mlx4_alloc_cmd_mailbox(dev);
925 	if (IS_ERR(mailbox)) {
926 		err = PTR_ERR(mailbox);
927 		printk(KERN_WARNING "mlx4_ib: mlx4_alloc_cmd_mailbox"
928 		       " failed (%d)\n", err);
929 		return;
930 	}
931 
932 	err = mlx4_HW2SW_MPT(dev, NULL,
933 			     key_to_hw_index(fmr->mr.key) &
934 			     (dev->caps.num_mpts - 1));
935 	mlx4_free_cmd_mailbox(dev, mailbox);
936 	if (err) {
937 		printk(KERN_WARNING "mlx4_ib: mlx4_HW2SW_MPT failed (%d)\n",
938 		       err);
939 		return;
940 	}
941 	fmr->mr.enabled = MLX4_MR_EN_SW;
942 }
943 EXPORT_SYMBOL_GPL(mlx4_fmr_unmap);
944 
945 int mlx4_fmr_free(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
946 {
947 	if (fmr->maps)
948 		return -EBUSY;
949 
950 	mlx4_mr_free(dev, &fmr->mr);
951 	fmr->mr.enabled = MLX4_MR_DISABLED;
952 
953 	return 0;
954 }
955 EXPORT_SYMBOL_GPL(mlx4_fmr_free);
956 
957 static int mlx4_fmr_free_reserved(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
958 {
959 	if (fmr->maps)
960 		return -EBUSY;
961 
962 	mlx4_mr_free_reserved(dev, &fmr->mr);
963 	fmr->mr.enabled = MLX4_MR_DISABLED;
964 
965 	return 0;
966 }
967 EXPORT_SYMBOL_GPL(mlx4_fmr_free_reserved);
968 
969 int mlx4_SYNC_TPT(struct mlx4_dev *dev)
970 {
971 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_SYNC_TPT, 1000,
972 			MLX4_CMD_WRAPPED);
973 }
974 EXPORT_SYMBOL_GPL(mlx4_SYNC_TPT);
975