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 #include <linux/vmalloc.h>
41 
42 #include <linux/mlx4/cmd.h>
43 
44 #include "mlx4.h"
45 #include "icm.h"
46 
47 static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order)
48 {
49 	int o;
50 	int m;
51 	u32 seg;
52 
53 	spin_lock(&buddy->lock);
54 
55 	for (o = order; o <= buddy->max_order; ++o)
56 		if (buddy->num_free[o]) {
57 			m = 1 << (buddy->max_order - o);
58 			seg = find_first_bit(buddy->bits[o], m);
59 			if (seg < m)
60 				goto found;
61 		}
62 
63 	spin_unlock(&buddy->lock);
64 	return -1;
65 
66  found:
67 	clear_bit(seg, buddy->bits[o]);
68 	--buddy->num_free[o];
69 
70 	while (o > order) {
71 		--o;
72 		seg <<= 1;
73 		set_bit(seg ^ 1, buddy->bits[o]);
74 		++buddy->num_free[o];
75 	}
76 
77 	spin_unlock(&buddy->lock);
78 
79 	seg <<= order;
80 
81 	return seg;
82 }
83 
84 static void mlx4_buddy_free(struct mlx4_buddy *buddy, u32 seg, int order)
85 {
86 	seg >>= order;
87 
88 	spin_lock(&buddy->lock);
89 
90 	while (test_bit(seg ^ 1, buddy->bits[order])) {
91 		clear_bit(seg ^ 1, buddy->bits[order]);
92 		--buddy->num_free[order];
93 		seg >>= 1;
94 		++order;
95 	}
96 
97 	set_bit(seg, buddy->bits[order]);
98 	++buddy->num_free[order];
99 
100 	spin_unlock(&buddy->lock);
101 }
102 
103 static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order)
104 {
105 	int i, s;
106 
107 	buddy->max_order = max_order;
108 	spin_lock_init(&buddy->lock);
109 
110 	buddy->bits = kcalloc(buddy->max_order + 1, sizeof (long *),
111 			      GFP_KERNEL);
112 	buddy->num_free = kcalloc((buddy->max_order + 1), sizeof *buddy->num_free,
113 				  GFP_KERNEL);
114 	if (!buddy->bits || !buddy->num_free)
115 		goto err_out;
116 
117 	for (i = 0; i <= buddy->max_order; ++i) {
118 		s = BITS_TO_LONGS(1 << (buddy->max_order - i));
119 		buddy->bits[i] = kcalloc(s, sizeof (long), GFP_KERNEL | __GFP_NOWARN);
120 		if (!buddy->bits[i]) {
121 			buddy->bits[i] = vzalloc(s * sizeof(long));
122 			if (!buddy->bits[i])
123 				goto err_out_free;
124 		}
125 	}
126 
127 	set_bit(0, buddy->bits[buddy->max_order]);
128 	buddy->num_free[buddy->max_order] = 1;
129 
130 	return 0;
131 
132 err_out_free:
133 	for (i = 0; i <= buddy->max_order; ++i)
134 		if (buddy->bits[i] && is_vmalloc_addr(buddy->bits[i]))
135 			vfree(buddy->bits[i]);
136 		else
137 			kfree(buddy->bits[i]);
138 
139 err_out:
140 	kfree(buddy->bits);
141 	kfree(buddy->num_free);
142 
143 	return -ENOMEM;
144 }
145 
146 static void mlx4_buddy_cleanup(struct mlx4_buddy *buddy)
147 {
148 	int i;
149 
150 	for (i = 0; i <= buddy->max_order; ++i)
151 		if (is_vmalloc_addr(buddy->bits[i]))
152 			vfree(buddy->bits[i]);
153 		else
154 			kfree(buddy->bits[i]);
155 
156 	kfree(buddy->bits);
157 	kfree(buddy->num_free);
158 }
159 
160 u32 __mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order)
161 {
162 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
163 	u32 seg;
164 	int seg_order;
165 	u32 offset;
166 
167 	seg_order = max_t(int, order - log_mtts_per_seg, 0);
168 
169 	seg = mlx4_buddy_alloc(&mr_table->mtt_buddy, seg_order);
170 	if (seg == -1)
171 		return -1;
172 
173 	offset = seg * (1 << log_mtts_per_seg);
174 
175 	if (mlx4_table_get_range(dev, &mr_table->mtt_table, offset,
176 				 offset + (1 << order) - 1)) {
177 		mlx4_buddy_free(&mr_table->mtt_buddy, seg, seg_order);
178 		return -1;
179 	}
180 
181 	return offset;
182 }
183 
184 static u32 mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order)
185 {
186 	u64 in_param = 0;
187 	u64 out_param;
188 	int err;
189 
190 	if (mlx4_is_mfunc(dev)) {
191 		set_param_l(&in_param, order);
192 		err = mlx4_cmd_imm(dev, in_param, &out_param, RES_MTT,
193 						       RES_OP_RESERVE_AND_MAP,
194 						       MLX4_CMD_ALLOC_RES,
195 						       MLX4_CMD_TIME_CLASS_A,
196 						       MLX4_CMD_WRAPPED);
197 		if (err)
198 			return -1;
199 		return get_param_l(&out_param);
200 	}
201 	return __mlx4_alloc_mtt_range(dev, order);
202 }
203 
204 int mlx4_mtt_init(struct mlx4_dev *dev, int npages, int page_shift,
205 		  struct mlx4_mtt *mtt)
206 {
207 	int i;
208 
209 	if (!npages) {
210 		mtt->order      = -1;
211 		mtt->page_shift = MLX4_ICM_PAGE_SHIFT;
212 		return 0;
213 	} else
214 		mtt->page_shift = page_shift;
215 
216 	for (mtt->order = 0, i = 1; i < npages; i <<= 1)
217 		++mtt->order;
218 
219 	mtt->offset = mlx4_alloc_mtt_range(dev, mtt->order);
220 	if (mtt->offset == -1)
221 		return -ENOMEM;
222 
223 	return 0;
224 }
225 EXPORT_SYMBOL_GPL(mlx4_mtt_init);
226 
227 void __mlx4_free_mtt_range(struct mlx4_dev *dev, u32 offset, int order)
228 {
229 	u32 first_seg;
230 	int seg_order;
231 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
232 
233 	seg_order = max_t(int, order - log_mtts_per_seg, 0);
234 	first_seg = offset / (1 << log_mtts_per_seg);
235 
236 	mlx4_buddy_free(&mr_table->mtt_buddy, first_seg, seg_order);
237 	mlx4_table_put_range(dev, &mr_table->mtt_table, offset,
238 			     offset + (1 << order) - 1);
239 }
240 
241 static void mlx4_free_mtt_range(struct mlx4_dev *dev, u32 offset, int order)
242 {
243 	u64 in_param = 0;
244 	int err;
245 
246 	if (mlx4_is_mfunc(dev)) {
247 		set_param_l(&in_param, offset);
248 		set_param_h(&in_param, order);
249 		err = mlx4_cmd(dev, in_param, RES_MTT, RES_OP_RESERVE_AND_MAP,
250 						       MLX4_CMD_FREE_RES,
251 						       MLX4_CMD_TIME_CLASS_A,
252 						       MLX4_CMD_WRAPPED);
253 		if (err)
254 			mlx4_warn(dev, "Failed to free mtt range at:"
255 				  "%d order:%d\n", offset, order);
256 		return;
257 	}
258 	 __mlx4_free_mtt_range(dev, offset, order);
259 }
260 
261 void mlx4_mtt_cleanup(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
262 {
263 	if (mtt->order < 0)
264 		return;
265 
266 	mlx4_free_mtt_range(dev, mtt->offset, mtt->order);
267 }
268 EXPORT_SYMBOL_GPL(mlx4_mtt_cleanup);
269 
270 u64 mlx4_mtt_addr(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
271 {
272 	return (u64) mtt->offset * dev->caps.mtt_entry_sz;
273 }
274 EXPORT_SYMBOL_GPL(mlx4_mtt_addr);
275 
276 static u32 hw_index_to_key(u32 ind)
277 {
278 	return (ind >> 24) | (ind << 8);
279 }
280 
281 static u32 key_to_hw_index(u32 key)
282 {
283 	return (key << 24) | (key >> 8);
284 }
285 
286 static int mlx4_SW2HW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
287 			  int mpt_index)
288 {
289 	return mlx4_cmd(dev, mailbox->dma, mpt_index,
290 			0, MLX4_CMD_SW2HW_MPT, MLX4_CMD_TIME_CLASS_B,
291 			MLX4_CMD_WRAPPED);
292 }
293 
294 static int mlx4_HW2SW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
295 			  int mpt_index)
296 {
297 	return mlx4_cmd_box(dev, 0, mailbox ? mailbox->dma : 0, mpt_index,
298 			    !mailbox, MLX4_CMD_HW2SW_MPT,
299 			    MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
300 }
301 
302 static int mlx4_mr_alloc_reserved(struct mlx4_dev *dev, u32 mridx, u32 pd,
303 			   u64 iova, u64 size, u32 access, int npages,
304 			   int page_shift, struct mlx4_mr *mr)
305 {
306 	mr->iova       = iova;
307 	mr->size       = size;
308 	mr->pd	       = pd;
309 	mr->access     = access;
310 	mr->enabled    = MLX4_MPT_DISABLED;
311 	mr->key	       = hw_index_to_key(mridx);
312 
313 	return mlx4_mtt_init(dev, npages, page_shift, &mr->mtt);
314 }
315 
316 static int mlx4_WRITE_MTT(struct mlx4_dev *dev,
317 			  struct mlx4_cmd_mailbox *mailbox,
318 			  int num_entries)
319 {
320 	return mlx4_cmd(dev, mailbox->dma, num_entries, 0, MLX4_CMD_WRITE_MTT,
321 			MLX4_CMD_TIME_CLASS_A,  MLX4_CMD_WRAPPED);
322 }
323 
324 int __mlx4_mpt_reserve(struct mlx4_dev *dev)
325 {
326 	struct mlx4_priv *priv = mlx4_priv(dev);
327 
328 	return mlx4_bitmap_alloc(&priv->mr_table.mpt_bitmap);
329 }
330 
331 static int mlx4_mpt_reserve(struct mlx4_dev *dev)
332 {
333 	u64 out_param;
334 
335 	if (mlx4_is_mfunc(dev)) {
336 		if (mlx4_cmd_imm(dev, 0, &out_param, RES_MPT, RES_OP_RESERVE,
337 				   MLX4_CMD_ALLOC_RES,
338 				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED))
339 			return -1;
340 		return get_param_l(&out_param);
341 	}
342 	return  __mlx4_mpt_reserve(dev);
343 }
344 
345 void __mlx4_mpt_release(struct mlx4_dev *dev, u32 index)
346 {
347 	struct mlx4_priv *priv = mlx4_priv(dev);
348 
349 	mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, index);
350 }
351 
352 static void mlx4_mpt_release(struct mlx4_dev *dev, u32 index)
353 {
354 	u64 in_param = 0;
355 
356 	if (mlx4_is_mfunc(dev)) {
357 		set_param_l(&in_param, index);
358 		if (mlx4_cmd(dev, in_param, RES_MPT, RES_OP_RESERVE,
359 			       MLX4_CMD_FREE_RES,
360 			       MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED))
361 			mlx4_warn(dev, "Failed to release mr index:%d\n",
362 				  index);
363 		return;
364 	}
365 	__mlx4_mpt_release(dev, index);
366 }
367 
368 int __mlx4_mpt_alloc_icm(struct mlx4_dev *dev, u32 index)
369 {
370 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
371 
372 	return mlx4_table_get(dev, &mr_table->dmpt_table, index);
373 }
374 
375 static int mlx4_mpt_alloc_icm(struct mlx4_dev *dev, u32 index)
376 {
377 	u64 param = 0;
378 
379 	if (mlx4_is_mfunc(dev)) {
380 		set_param_l(&param, index);
381 		return mlx4_cmd_imm(dev, param, &param, RES_MPT, RES_OP_MAP_ICM,
382 							MLX4_CMD_ALLOC_RES,
383 							MLX4_CMD_TIME_CLASS_A,
384 							MLX4_CMD_WRAPPED);
385 	}
386 	return __mlx4_mpt_alloc_icm(dev, index);
387 }
388 
389 void __mlx4_mpt_free_icm(struct mlx4_dev *dev, u32 index)
390 {
391 	struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
392 
393 	mlx4_table_put(dev, &mr_table->dmpt_table, index);
394 }
395 
396 static void mlx4_mpt_free_icm(struct mlx4_dev *dev, u32 index)
397 {
398 	u64 in_param = 0;
399 
400 	if (mlx4_is_mfunc(dev)) {
401 		set_param_l(&in_param, index);
402 		if (mlx4_cmd(dev, in_param, RES_MPT, RES_OP_MAP_ICM,
403 			     MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
404 			     MLX4_CMD_WRAPPED))
405 			mlx4_warn(dev, "Failed to free icm of mr index:%d\n",
406 				  index);
407 		return;
408 	}
409 	return __mlx4_mpt_free_icm(dev, index);
410 }
411 
412 int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access,
413 		  int npages, int page_shift, struct mlx4_mr *mr)
414 {
415 	u32 index;
416 	int err;
417 
418 	index = mlx4_mpt_reserve(dev);
419 	if (index == -1)
420 		return -ENOMEM;
421 
422 	err = mlx4_mr_alloc_reserved(dev, index, pd, iova, size,
423 				     access, npages, page_shift, mr);
424 	if (err)
425 		mlx4_mpt_release(dev, index);
426 
427 	return err;
428 }
429 EXPORT_SYMBOL_GPL(mlx4_mr_alloc);
430 
431 static int mlx4_mr_free_reserved(struct mlx4_dev *dev, struct mlx4_mr *mr)
432 {
433 	int err;
434 
435 	if (mr->enabled == MLX4_MPT_EN_HW) {
436 		err = mlx4_HW2SW_MPT(dev, NULL,
437 				     key_to_hw_index(mr->key) &
438 				     (dev->caps.num_mpts - 1));
439 		if (err) {
440 			mlx4_warn(dev, "HW2SW_MPT failed (%d),", err);
441 			mlx4_warn(dev, "MR has MWs bound to it.\n");
442 			return err;
443 		}
444 
445 		mr->enabled = MLX4_MPT_EN_SW;
446 	}
447 	mlx4_mtt_cleanup(dev, &mr->mtt);
448 
449 	return 0;
450 }
451 
452 int mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr)
453 {
454 	int ret;
455 
456 	ret = mlx4_mr_free_reserved(dev, mr);
457 	if (ret)
458 		return ret;
459 	if (mr->enabled)
460 		mlx4_mpt_free_icm(dev, key_to_hw_index(mr->key));
461 	mlx4_mpt_release(dev, key_to_hw_index(mr->key));
462 
463 	return 0;
464 }
465 EXPORT_SYMBOL_GPL(mlx4_mr_free);
466 
467 int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr)
468 {
469 	struct mlx4_cmd_mailbox *mailbox;
470 	struct mlx4_mpt_entry *mpt_entry;
471 	int err;
472 
473 	err = mlx4_mpt_alloc_icm(dev, key_to_hw_index(mr->key));
474 	if (err)
475 		return err;
476 
477 	mailbox = mlx4_alloc_cmd_mailbox(dev);
478 	if (IS_ERR(mailbox)) {
479 		err = PTR_ERR(mailbox);
480 		goto err_table;
481 	}
482 	mpt_entry = mailbox->buf;
483 	mpt_entry->flags = cpu_to_be32(MLX4_MPT_FLAG_MIO	 |
484 				       MLX4_MPT_FLAG_REGION	 |
485 				       mr->access);
486 
487 	mpt_entry->key	       = cpu_to_be32(key_to_hw_index(mr->key));
488 	mpt_entry->pd_flags    = cpu_to_be32(mr->pd | MLX4_MPT_PD_FLAG_EN_INV);
489 	mpt_entry->start       = cpu_to_be64(mr->iova);
490 	mpt_entry->length      = cpu_to_be64(mr->size);
491 	mpt_entry->entity_size = cpu_to_be32(mr->mtt.page_shift);
492 
493 	if (mr->mtt.order < 0) {
494 		mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL);
495 		mpt_entry->mtt_addr = 0;
496 	} else {
497 		mpt_entry->mtt_addr = cpu_to_be64(mlx4_mtt_addr(dev,
498 						  &mr->mtt));
499 	}
500 
501 	if (mr->mtt.order >= 0 && mr->mtt.page_shift == 0) {
502 		/* fast register MR in free state */
503 		mpt_entry->flags    |= cpu_to_be32(MLX4_MPT_FLAG_FREE);
504 		mpt_entry->pd_flags |= cpu_to_be32(MLX4_MPT_PD_FLAG_FAST_REG |
505 						   MLX4_MPT_PD_FLAG_RAE);
506 		mpt_entry->mtt_sz    = cpu_to_be32(1 << mr->mtt.order);
507 	} else {
508 		mpt_entry->flags    |= cpu_to_be32(MLX4_MPT_FLAG_SW_OWNS);
509 	}
510 
511 	err = mlx4_SW2HW_MPT(dev, mailbox,
512 			     key_to_hw_index(mr->key) & (dev->caps.num_mpts - 1));
513 	if (err) {
514 		mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err);
515 		goto err_cmd;
516 	}
517 	mr->enabled = MLX4_MPT_EN_HW;
518 
519 	mlx4_free_cmd_mailbox(dev, mailbox);
520 
521 	return 0;
522 
523 err_cmd:
524 	mlx4_free_cmd_mailbox(dev, mailbox);
525 
526 err_table:
527 	mlx4_mpt_free_icm(dev, key_to_hw_index(mr->key));
528 	return err;
529 }
530 EXPORT_SYMBOL_GPL(mlx4_mr_enable);
531 
532 static int mlx4_write_mtt_chunk(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
533 				int start_index, int npages, u64 *page_list)
534 {
535 	struct mlx4_priv *priv = mlx4_priv(dev);
536 	__be64 *mtts;
537 	dma_addr_t dma_handle;
538 	int i;
539 
540 	mtts = mlx4_table_find(&priv->mr_table.mtt_table, mtt->offset +
541 			       start_index, &dma_handle);
542 
543 	if (!mtts)
544 		return -ENOMEM;
545 
546 	dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle,
547 				npages * sizeof (u64), DMA_TO_DEVICE);
548 
549 	for (i = 0; i < npages; ++i)
550 		mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
551 
552 	dma_sync_single_for_device(&dev->pdev->dev, dma_handle,
553 				   npages * sizeof (u64), DMA_TO_DEVICE);
554 
555 	return 0;
556 }
557 
558 int __mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
559 		     int start_index, int npages, u64 *page_list)
560 {
561 	int err = 0;
562 	int chunk;
563 	int mtts_per_page;
564 	int max_mtts_first_page;
565 
566 	/* compute how may mtts fit in the first page */
567 	mtts_per_page = PAGE_SIZE / sizeof(u64);
568 	max_mtts_first_page = mtts_per_page - (mtt->offset + start_index)
569 			      % mtts_per_page;
570 
571 	chunk = min_t(int, max_mtts_first_page, npages);
572 
573 	while (npages > 0) {
574 		err = mlx4_write_mtt_chunk(dev, mtt, start_index, chunk, page_list);
575 		if (err)
576 			return err;
577 		npages      -= chunk;
578 		start_index += chunk;
579 		page_list   += chunk;
580 
581 		chunk = min_t(int, mtts_per_page, npages);
582 	}
583 	return err;
584 }
585 
586 int mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
587 		   int start_index, int npages, u64 *page_list)
588 {
589 	struct mlx4_cmd_mailbox *mailbox = NULL;
590 	__be64 *inbox = NULL;
591 	int chunk;
592 	int err = 0;
593 	int i;
594 
595 	if (mtt->order < 0)
596 		return -EINVAL;
597 
598 	if (mlx4_is_mfunc(dev)) {
599 		mailbox = mlx4_alloc_cmd_mailbox(dev);
600 		if (IS_ERR(mailbox))
601 			return PTR_ERR(mailbox);
602 		inbox = mailbox->buf;
603 
604 		while (npages > 0) {
605 			chunk = min_t(int, MLX4_MAILBOX_SIZE / sizeof(u64) - 2,
606 				      npages);
607 			inbox[0] = cpu_to_be64(mtt->offset + start_index);
608 			inbox[1] = 0;
609 			for (i = 0; i < chunk; ++i)
610 				inbox[i + 2] = cpu_to_be64(page_list[i] |
611 					       MLX4_MTT_FLAG_PRESENT);
612 			err = mlx4_WRITE_MTT(dev, mailbox, chunk);
613 			if (err) {
614 				mlx4_free_cmd_mailbox(dev, mailbox);
615 				return err;
616 			}
617 
618 			npages      -= chunk;
619 			start_index += chunk;
620 			page_list   += chunk;
621 		}
622 		mlx4_free_cmd_mailbox(dev, mailbox);
623 		return err;
624 	}
625 
626 	return __mlx4_write_mtt(dev, mtt, start_index, npages, page_list);
627 }
628 EXPORT_SYMBOL_GPL(mlx4_write_mtt);
629 
630 int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
631 		       struct mlx4_buf *buf)
632 {
633 	u64 *page_list;
634 	int err;
635 	int i;
636 
637 	page_list = kmalloc(buf->npages * sizeof *page_list, GFP_KERNEL);
638 	if (!page_list)
639 		return -ENOMEM;
640 
641 	for (i = 0; i < buf->npages; ++i)
642 		if (buf->nbufs == 1)
643 			page_list[i] = buf->direct.map + (i << buf->page_shift);
644 		else
645 			page_list[i] = buf->page_list[i].map;
646 
647 	err = mlx4_write_mtt(dev, mtt, 0, buf->npages, page_list);
648 
649 	kfree(page_list);
650 	return err;
651 }
652 EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt);
653 
654 int 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 }
676 EXPORT_SYMBOL_GPL(mlx4_mw_alloc);
677 
678 int 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 
684 	err = mlx4_mpt_alloc_icm(dev, key_to_hw_index(mw->key));
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 
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 
719 err_cmd:
720 	mlx4_free_cmd_mailbox(dev, mailbox);
721 
722 err_table:
723 	mlx4_mpt_free_icm(dev, key_to_hw_index(mw->key));
724 	return err;
725 }
726 EXPORT_SYMBOL_GPL(mlx4_mw_enable);
727 
728 void 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 }
745 EXPORT_SYMBOL_GPL(mlx4_mw_free);
746 
747 int mlx4_init_mr_table(struct mlx4_dev *dev)
748 {
749 	struct mlx4_priv *priv = mlx4_priv(dev);
750 	struct mlx4_mr_table *mr_table = &priv->mr_table;
751 	int err;
752 
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 
758 	if (!is_power_of_2(dev->caps.num_mpts))
759 		return -EINVAL;
760 
761 	err = mlx4_bitmap_init(&mr_table->mpt_bitmap, dev->caps.num_mpts,
762 			       ~0, dev->caps.reserved_mrws, 0);
763 	if (err)
764 		return err;
765 
766 	err = mlx4_buddy_init(&mr_table->mtt_buddy,
767 			      ilog2((u32)dev->caps.num_mtts /
768 			      (1 << log_mtts_per_seg)));
769 	if (err)
770 		goto err_buddy;
771 
772 	if (dev->caps.reserved_mtts) {
773 		priv->reserved_mtts =
774 			mlx4_alloc_mtt_range(dev,
775 					     fls(dev->caps.reserved_mtts - 1));
776 		if (priv->reserved_mtts < 0) {
777 			mlx4_warn(dev, "MTT table of order %u is too small.\n",
778 				  mr_table->mtt_buddy.max_order);
779 			err = -ENOMEM;
780 			goto err_reserve_mtts;
781 		}
782 	}
783 
784 	return 0;
785 
786 err_reserve_mtts:
787 	mlx4_buddy_cleanup(&mr_table->mtt_buddy);
788 
789 err_buddy:
790 	mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
791 
792 	return err;
793 }
794 
795 void mlx4_cleanup_mr_table(struct mlx4_dev *dev)
796 {
797 	struct mlx4_priv *priv = mlx4_priv(dev);
798 	struct mlx4_mr_table *mr_table = &priv->mr_table;
799 
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));
805 	mlx4_buddy_cleanup(&mr_table->mtt_buddy);
806 	mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
807 }
808 
809 static 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 
836 int 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 
857 	dma_sync_single_for_cpu(&dev->pdev->dev, fmr->dma_handle,
858 				npages * sizeof(u64), DMA_TO_DEVICE);
859 
860 	for (i = 0; i < npages; ++i)
861 		fmr->mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
862 
863 	dma_sync_single_for_device(&dev->pdev->dev, fmr->dma_handle,
864 				   npages * sizeof(u64), DMA_TO_DEVICE);
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 }
881 EXPORT_SYMBOL_GPL(mlx4_map_phys_fmr);
882 
883 int 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);
887 	int err = -ENOMEM;
888 
889 	if (max_maps > dev->caps.max_fmr_maps)
890 		return -EINVAL;
891 
892 	if (page_shift < (ffs(dev->caps.page_size_cap) - 1) || page_shift >= 32)
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 
909 	fmr->mtts = mlx4_table_find(&priv->mr_table.mtt_table,
910 				    fmr->mr.mtt.offset,
911 				    &fmr->dma_handle);
912 
913 	if (!fmr->mtts) {
914 		err = -ENOMEM;
915 		goto err_free;
916 	}
917 
918 	return 0;
919 
920 err_free:
921 	(void) mlx4_mr_free(dev, &fmr->mr);
922 	return err;
923 }
924 EXPORT_SYMBOL_GPL(mlx4_fmr_alloc);
925 
926 int mlx4_fmr_enable(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
927 {
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;
941 }
942 EXPORT_SYMBOL_GPL(mlx4_fmr_enable);
943 
944 void mlx4_fmr_unmap(struct mlx4_dev *dev, struct mlx4_fmr *fmr,
945 		    u32 *lkey, u32 *rkey)
946 {
947 	struct mlx4_cmd_mailbox *mailbox;
948 	int err;
949 
950 	if (!fmr->maps)
951 		return;
952 
953 	fmr->maps = 0;
954 
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 	}
972 	fmr->mr.enabled = MLX4_MPT_EN_SW;
973 }
974 EXPORT_SYMBOL_GPL(mlx4_fmr_unmap);
975 
976 int mlx4_fmr_free(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
977 {
978 	int ret;
979 
980 	if (fmr->maps)
981 		return -EBUSY;
982 
983 	ret = mlx4_mr_free(dev, &fmr->mr);
984 	if (ret)
985 		return ret;
986 	fmr->mr.enabled = MLX4_MPT_DISABLED;
987 
988 	return 0;
989 }
990 EXPORT_SYMBOL_GPL(mlx4_fmr_free);
991 
992 int mlx4_SYNC_TPT(struct mlx4_dev *dev)
993 {
994 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_SYNC_TPT, 1000,
995 			MLX4_CMD_NATIVE);
996 }
997 EXPORT_SYMBOL_GPL(mlx4_SYNC_TPT);
998