xref: /linux/drivers/infiniband/hw/mthca/mthca_memfree.c (revision 20d0021394c1b070bf04b22c5bc8fdb437edd4c5) 
1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Cisco Systems.  All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  *
33  * $Id$
34  */
35 
36 #include <linux/mm.h>
37 
38 #include "mthca_memfree.h"
39 #include "mthca_dev.h"
40 #include "mthca_cmd.h"
41 
42 /*
43  * We allocate in as big chunks as we can, up to a maximum of 256 KB
44  * per chunk.
45  */
46 enum {
47 	MTHCA_ICM_ALLOC_SIZE   = 1 << 18,
48 	MTHCA_TABLE_CHUNK_SIZE = 1 << 18
49 };
50 
51 struct mthca_user_db_table {
52 	struct semaphore mutex;
53 	struct {
54 		u64                uvirt;
55 		struct scatterlist mem;
56 		int                refcount;
57 	}                page[0];
58 };
59 
60 void mthca_free_icm(struct mthca_dev *dev, struct mthca_icm *icm)
61 {
62 	struct mthca_icm_chunk *chunk, *tmp;
63 	int i;
64 
65 	if (!icm)
66 		return;
67 
68 	list_for_each_entry_safe(chunk, tmp, &icm->chunk_list, list) {
69 		if (chunk->nsg > 0)
70 			pci_unmap_sg(dev->pdev, chunk->mem, chunk->npages,
71 				     PCI_DMA_BIDIRECTIONAL);
72 
73 		for (i = 0; i < chunk->npages; ++i)
74 			__free_pages(chunk->mem[i].page,
75 				     get_order(chunk->mem[i].length));
76 
77 		kfree(chunk);
78 	}
79 
80 	kfree(icm);
81 }
82 
83 struct mthca_icm *mthca_alloc_icm(struct mthca_dev *dev, int npages,
84 				  unsigned int gfp_mask)
85 {
86 	struct mthca_icm *icm;
87 	struct mthca_icm_chunk *chunk = NULL;
88 	int cur_order;
89 
90 	icm = kmalloc(sizeof *icm, gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN));
91 	if (!icm)
92 		return icm;
93 
94 	icm->refcount = 0;
95 	INIT_LIST_HEAD(&icm->chunk_list);
96 
97 	cur_order = get_order(MTHCA_ICM_ALLOC_SIZE);
98 
99 	while (npages > 0) {
100 		if (!chunk) {
101 			chunk = kmalloc(sizeof *chunk,
102 					gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN));
103 			if (!chunk)
104 				goto fail;
105 
106 			chunk->npages = 0;
107 			chunk->nsg    = 0;
108 			list_add_tail(&chunk->list, &icm->chunk_list);
109 		}
110 
111 		while (1 << cur_order > npages)
112 			--cur_order;
113 
114 		chunk->mem[chunk->npages].page = alloc_pages(gfp_mask, cur_order);
115 		if (chunk->mem[chunk->npages].page) {
116 			chunk->mem[chunk->npages].length = PAGE_SIZE << cur_order;
117 			chunk->mem[chunk->npages].offset = 0;
118 
119 			if (++chunk->npages == MTHCA_ICM_CHUNK_LEN) {
120 				chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
121 							chunk->npages,
122 							PCI_DMA_BIDIRECTIONAL);
123 
124 				if (chunk->nsg <= 0)
125 					goto fail;
126 
127 				chunk = NULL;
128 			}
129 
130 			npages -= 1 << cur_order;
131 		} else {
132 			--cur_order;
133 			if (cur_order < 0)
134 				goto fail;
135 		}
136 	}
137 
138 	if (chunk) {
139 		chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
140 					chunk->npages,
141 					PCI_DMA_BIDIRECTIONAL);
142 
143 		if (chunk->nsg <= 0)
144 			goto fail;
145 	}
146 
147 	return icm;
148 
149 fail:
150 	mthca_free_icm(dev, icm);
151 	return NULL;
152 }
153 
154 int mthca_table_get(struct mthca_dev *dev, struct mthca_icm_table *table, int obj)
155 {
156 	int i = (obj & (table->num_obj - 1)) * table->obj_size / MTHCA_TABLE_CHUNK_SIZE;
157 	int ret = 0;
158 	u8 status;
159 
160 	down(&table->mutex);
161 
162 	if (table->icm[i]) {
163 		++table->icm[i]->refcount;
164 		goto out;
165 	}
166 
167 	table->icm[i] = mthca_alloc_icm(dev, MTHCA_TABLE_CHUNK_SIZE >> PAGE_SHIFT,
168 					(table->lowmem ? GFP_KERNEL : GFP_HIGHUSER) |
169 					__GFP_NOWARN);
170 	if (!table->icm[i]) {
171 		ret = -ENOMEM;
172 		goto out;
173 	}
174 
175 	if (mthca_MAP_ICM(dev, table->icm[i], table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
176 			  &status) || status) {
177 		mthca_free_icm(dev, table->icm[i]);
178 		table->icm[i] = NULL;
179 		ret = -ENOMEM;
180 		goto out;
181 	}
182 
183 	++table->icm[i]->refcount;
184 
185 out:
186 	up(&table->mutex);
187 	return ret;
188 }
189 
190 void mthca_table_put(struct mthca_dev *dev, struct mthca_icm_table *table, int obj)
191 {
192 	int i;
193 	u8 status;
194 
195 	if (!mthca_is_memfree(dev))
196 		return;
197 
198 	i = (obj & (table->num_obj - 1)) * table->obj_size / MTHCA_TABLE_CHUNK_SIZE;
199 
200 	down(&table->mutex);
201 
202 	if (--table->icm[i]->refcount == 0) {
203 		mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
204 				MTHCA_TABLE_CHUNK_SIZE >> 12, &status);
205 		mthca_free_icm(dev, table->icm[i]);
206 		table->icm[i] = NULL;
207 	}
208 
209 	up(&table->mutex);
210 }
211 
212 void *mthca_table_find(struct mthca_icm_table *table, int obj)
213 {
214 	int idx, offset, i;
215 	struct mthca_icm_chunk *chunk;
216 	struct mthca_icm *icm;
217 	struct page *page = NULL;
218 
219 	if (!table->lowmem)
220 		return NULL;
221 
222 	down(&table->mutex);
223 
224 	idx = (obj & (table->num_obj - 1)) * table->obj_size;
225 	icm = table->icm[idx / MTHCA_TABLE_CHUNK_SIZE];
226 	offset = idx % MTHCA_TABLE_CHUNK_SIZE;
227 
228 	if (!icm)
229 		goto out;
230 
231 	list_for_each_entry(chunk, &icm->chunk_list, list) {
232 		for (i = 0; i < chunk->npages; ++i) {
233 			if (chunk->mem[i].length >= offset) {
234 				page = chunk->mem[i].page;
235 				break;
236 			}
237 			offset -= chunk->mem[i].length;
238 		}
239 	}
240 
241 out:
242 	up(&table->mutex);
243 	return page ? lowmem_page_address(page) + offset : NULL;
244 }
245 
246 int mthca_table_get_range(struct mthca_dev *dev, struct mthca_icm_table *table,
247 			  int start, int end)
248 {
249 	int inc = MTHCA_TABLE_CHUNK_SIZE / table->obj_size;
250 	int i, err;
251 
252 	for (i = start; i <= end; i += inc) {
253 		err = mthca_table_get(dev, table, i);
254 		if (err)
255 			goto fail;
256 	}
257 
258 	return 0;
259 
260 fail:
261 	while (i > start) {
262 		i -= inc;
263 		mthca_table_put(dev, table, i);
264 	}
265 
266 	return err;
267 }
268 
269 void mthca_table_put_range(struct mthca_dev *dev, struct mthca_icm_table *table,
270 			   int start, int end)
271 {
272 	int i;
273 
274 	if (!mthca_is_memfree(dev))
275 		return;
276 
277 	for (i = start; i <= end; i += MTHCA_TABLE_CHUNK_SIZE / table->obj_size)
278 		mthca_table_put(dev, table, i);
279 }
280 
281 struct mthca_icm_table *mthca_alloc_icm_table(struct mthca_dev *dev,
282 					      u64 virt, int obj_size,
283 					      int nobj, int reserved,
284 					      int use_lowmem)
285 {
286 	struct mthca_icm_table *table;
287 	int num_icm;
288 	int i;
289 	u8 status;
290 
291 	num_icm = obj_size * nobj / MTHCA_TABLE_CHUNK_SIZE;
292 
293 	table = kmalloc(sizeof *table + num_icm * sizeof *table->icm, GFP_KERNEL);
294 	if (!table)
295 		return NULL;
296 
297 	table->virt     = virt;
298 	table->num_icm  = num_icm;
299 	table->num_obj  = nobj;
300 	table->obj_size = obj_size;
301 	table->lowmem   = use_lowmem;
302 	init_MUTEX(&table->mutex);
303 
304 	for (i = 0; i < num_icm; ++i)
305 		table->icm[i] = NULL;
306 
307 	for (i = 0; i * MTHCA_TABLE_CHUNK_SIZE < reserved * obj_size; ++i) {
308 		table->icm[i] = mthca_alloc_icm(dev, MTHCA_TABLE_CHUNK_SIZE >> PAGE_SHIFT,
309 						(use_lowmem ? GFP_KERNEL : GFP_HIGHUSER) |
310 						__GFP_NOWARN);
311 		if (!table->icm[i])
312 			goto err;
313 		if (mthca_MAP_ICM(dev, table->icm[i], virt + i * MTHCA_TABLE_CHUNK_SIZE,
314 				  &status) || status) {
315 			mthca_free_icm(dev, table->icm[i]);
316 			table->icm[i] = NULL;
317 			goto err;
318 		}
319 
320 		/*
321 		 * Add a reference to this ICM chunk so that it never
322 		 * gets freed (since it contains reserved firmware objects).
323 		 */
324 		++table->icm[i]->refcount;
325 	}
326 
327 	return table;
328 
329 err:
330 	for (i = 0; i < num_icm; ++i)
331 		if (table->icm[i]) {
332 			mthca_UNMAP_ICM(dev, virt + i * MTHCA_TABLE_CHUNK_SIZE,
333 					MTHCA_TABLE_CHUNK_SIZE >> 12, &status);
334 			mthca_free_icm(dev, table->icm[i]);
335 		}
336 
337 	kfree(table);
338 
339 	return NULL;
340 }
341 
342 void mthca_free_icm_table(struct mthca_dev *dev, struct mthca_icm_table *table)
343 {
344 	int i;
345 	u8 status;
346 
347 	for (i = 0; i < table->num_icm; ++i)
348 		if (table->icm[i]) {
349 			mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
350 					MTHCA_TABLE_CHUNK_SIZE >> 12, &status);
351 			mthca_free_icm(dev, table->icm[i]);
352 		}
353 
354 	kfree(table);
355 }
356 
357 static u64 mthca_uarc_virt(struct mthca_dev *dev, struct mthca_uar *uar, int page)
358 {
359 	return dev->uar_table.uarc_base +
360 		uar->index * dev->uar_table.uarc_size +
361 		page * 4096;
362 }
363 
364 int mthca_map_user_db(struct mthca_dev *dev, struct mthca_uar *uar,
365 		      struct mthca_user_db_table *db_tab, int index, u64 uaddr)
366 {
367 	int ret = 0;
368 	u8 status;
369 	int i;
370 
371 	if (!mthca_is_memfree(dev))
372 		return 0;
373 
374 	if (index < 0 || index > dev->uar_table.uarc_size / 8)
375 		return -EINVAL;
376 
377 	down(&db_tab->mutex);
378 
379 	i = index / MTHCA_DB_REC_PER_PAGE;
380 
381 	if ((db_tab->page[i].refcount >= MTHCA_DB_REC_PER_PAGE)       ||
382 	    (db_tab->page[i].uvirt && db_tab->page[i].uvirt != uaddr) ||
383 	    (uaddr & 4095)) {
384 		ret = -EINVAL;
385 		goto out;
386 	}
387 
388 	if (db_tab->page[i].refcount) {
389 		++db_tab->page[i].refcount;
390 		goto out;
391 	}
392 
393 	ret = get_user_pages(current, current->mm, uaddr & PAGE_MASK, 1, 1, 0,
394 			     &db_tab->page[i].mem.page, NULL);
395 	if (ret < 0)
396 		goto out;
397 
398 	db_tab->page[i].mem.length = 4096;
399 	db_tab->page[i].mem.offset = uaddr & ~PAGE_MASK;
400 
401 	ret = pci_map_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE);
402 	if (ret < 0) {
403 		put_page(db_tab->page[i].mem.page);
404 		goto out;
405 	}
406 
407 	ret = mthca_MAP_ICM_page(dev, sg_dma_address(&db_tab->page[i].mem),
408 				 mthca_uarc_virt(dev, uar, i), &status);
409 	if (!ret && status)
410 		ret = -EINVAL;
411 	if (ret) {
412 		pci_unmap_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE);
413 		put_page(db_tab->page[i].mem.page);
414 		goto out;
415 	}
416 
417 	db_tab->page[i].uvirt    = uaddr;
418 	db_tab->page[i].refcount = 1;
419 
420 out:
421 	up(&db_tab->mutex);
422 	return ret;
423 }
424 
425 void mthca_unmap_user_db(struct mthca_dev *dev, struct mthca_uar *uar,
426 			 struct mthca_user_db_table *db_tab, int index)
427 {
428 	if (!mthca_is_memfree(dev))
429 		return;
430 
431 	/*
432 	 * To make our bookkeeping simpler, we don't unmap DB
433 	 * pages until we clean up the whole db table.
434 	 */
435 
436 	down(&db_tab->mutex);
437 
438 	--db_tab->page[index / MTHCA_DB_REC_PER_PAGE].refcount;
439 
440 	up(&db_tab->mutex);
441 }
442 
443 struct mthca_user_db_table *mthca_init_user_db_tab(struct mthca_dev *dev)
444 {
445 	struct mthca_user_db_table *db_tab;
446 	int npages;
447 	int i;
448 
449 	if (!mthca_is_memfree(dev))
450 		return NULL;
451 
452 	npages = dev->uar_table.uarc_size / 4096;
453 	db_tab = kmalloc(sizeof *db_tab + npages * sizeof *db_tab->page, GFP_KERNEL);
454 	if (!db_tab)
455 		return ERR_PTR(-ENOMEM);
456 
457 	init_MUTEX(&db_tab->mutex);
458 	for (i = 0; i < npages; ++i) {
459 		db_tab->page[i].refcount = 0;
460 		db_tab->page[i].uvirt    = 0;
461 	}
462 
463 	return db_tab;
464 }
465 
466 void mthca_cleanup_user_db_tab(struct mthca_dev *dev, struct mthca_uar *uar,
467 			       struct mthca_user_db_table *db_tab)
468 {
469 	int i;
470 	u8 status;
471 
472 	if (!mthca_is_memfree(dev))
473 		return;
474 
475 	for (i = 0; i < dev->uar_table.uarc_size / 4096; ++i) {
476 		if (db_tab->page[i].uvirt) {
477 			mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, uar, i), 1, &status);
478 			pci_unmap_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE);
479 			put_page(db_tab->page[i].mem.page);
480 		}
481 	}
482 }
483 
484 int mthca_alloc_db(struct mthca_dev *dev, int type, u32 qn, u32 **db)
485 {
486 	int group;
487 	int start, end, dir;
488 	int i, j;
489 	struct mthca_db_page *page;
490 	int ret = 0;
491 	u8 status;
492 
493 	down(&dev->db_tab->mutex);
494 
495 	switch (type) {
496 	case MTHCA_DB_TYPE_CQ_ARM:
497 	case MTHCA_DB_TYPE_SQ:
498 		group = 0;
499 		start = 0;
500 		end   = dev->db_tab->max_group1;
501 		dir   = 1;
502 		break;
503 
504 	case MTHCA_DB_TYPE_CQ_SET_CI:
505 	case MTHCA_DB_TYPE_RQ:
506 	case MTHCA_DB_TYPE_SRQ:
507 		group = 1;
508 		start = dev->db_tab->npages - 1;
509 		end   = dev->db_tab->min_group2;
510 		dir   = -1;
511 		break;
512 
513 	default:
514 		ret = -EINVAL;
515 		goto out;
516 	}
517 
518 	for (i = start; i != end; i += dir)
519 		if (dev->db_tab->page[i].db_rec &&
520 		    !bitmap_full(dev->db_tab->page[i].used,
521 				 MTHCA_DB_REC_PER_PAGE)) {
522 			page = dev->db_tab->page + i;
523 			goto found;
524 		}
525 
526 	if (dev->db_tab->max_group1 >= dev->db_tab->min_group2 - 1) {
527 		ret = -ENOMEM;
528 		goto out;
529 	}
530 
531 	page = dev->db_tab->page + end;
532 	page->db_rec = dma_alloc_coherent(&dev->pdev->dev, 4096,
533 					  &page->mapping, GFP_KERNEL);
534 	if (!page->db_rec) {
535 		ret = -ENOMEM;
536 		goto out;
537 	}
538 	memset(page->db_rec, 0, 4096);
539 
540 	ret = mthca_MAP_ICM_page(dev, page->mapping,
541 				 mthca_uarc_virt(dev, &dev->driver_uar, i), &status);
542 	if (!ret && status)
543 		ret = -EINVAL;
544 	if (ret) {
545 		dma_free_coherent(&dev->pdev->dev, 4096,
546 				  page->db_rec, page->mapping);
547 		goto out;
548 	}
549 
550 	bitmap_zero(page->used, MTHCA_DB_REC_PER_PAGE);
551 	if (group == 0)
552 		++dev->db_tab->max_group1;
553 	else
554 		--dev->db_tab->min_group2;
555 
556 found:
557 	j = find_first_zero_bit(page->used, MTHCA_DB_REC_PER_PAGE);
558 	set_bit(j, page->used);
559 
560 	if (group == 1)
561 		j = MTHCA_DB_REC_PER_PAGE - 1 - j;
562 
563 	ret = i * MTHCA_DB_REC_PER_PAGE + j;
564 
565 	page->db_rec[j] = cpu_to_be64((qn << 8) | (type << 5));
566 
567 	*db = (u32 *) &page->db_rec[j];
568 
569 out:
570 	up(&dev->db_tab->mutex);
571 
572 	return ret;
573 }
574 
575 void mthca_free_db(struct mthca_dev *dev, int type, int db_index)
576 {
577 	int i, j;
578 	struct mthca_db_page *page;
579 	u8 status;
580 
581 	i = db_index / MTHCA_DB_REC_PER_PAGE;
582 	j = db_index % MTHCA_DB_REC_PER_PAGE;
583 
584 	page = dev->db_tab->page + i;
585 
586 	down(&dev->db_tab->mutex);
587 
588 	page->db_rec[j] = 0;
589 	if (i >= dev->db_tab->min_group2)
590 		j = MTHCA_DB_REC_PER_PAGE - 1 - j;
591 	clear_bit(j, page->used);
592 
593 	if (bitmap_empty(page->used, MTHCA_DB_REC_PER_PAGE) &&
594 	    i >= dev->db_tab->max_group1 - 1) {
595 		mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1, &status);
596 
597 		dma_free_coherent(&dev->pdev->dev, 4096,
598 				  page->db_rec, page->mapping);
599 		page->db_rec = NULL;
600 
601 		if (i == dev->db_tab->max_group1) {
602 			--dev->db_tab->max_group1;
603 			/* XXX may be able to unmap more pages now */
604 		}
605 		if (i == dev->db_tab->min_group2)
606 			++dev->db_tab->min_group2;
607 	}
608 
609 	up(&dev->db_tab->mutex);
610 }
611 
612 int mthca_init_db_tab(struct mthca_dev *dev)
613 {
614 	int i;
615 
616 	if (!mthca_is_memfree(dev))
617 		return 0;
618 
619 	dev->db_tab = kmalloc(sizeof *dev->db_tab, GFP_KERNEL);
620 	if (!dev->db_tab)
621 		return -ENOMEM;
622 
623 	init_MUTEX(&dev->db_tab->mutex);
624 
625 	dev->db_tab->npages     = dev->uar_table.uarc_size / 4096;
626 	dev->db_tab->max_group1 = 0;
627 	dev->db_tab->min_group2 = dev->db_tab->npages - 1;
628 
629 	dev->db_tab->page = kmalloc(dev->db_tab->npages *
630 				    sizeof *dev->db_tab->page,
631 				    GFP_KERNEL);
632 	if (!dev->db_tab->page) {
633 		kfree(dev->db_tab);
634 		return -ENOMEM;
635 	}
636 
637 	for (i = 0; i < dev->db_tab->npages; ++i)
638 		dev->db_tab->page[i].db_rec = NULL;
639 
640 	return 0;
641 }
642 
643 void mthca_cleanup_db_tab(struct mthca_dev *dev)
644 {
645 	int i;
646 	u8 status;
647 
648 	if (!mthca_is_memfree(dev))
649 		return;
650 
651 	/*
652 	 * Because we don't always free our UARC pages when they
653 	 * become empty to make mthca_free_db() simpler we need to
654 	 * make a sweep through the doorbell pages and free any
655 	 * leftover pages now.
656 	 */
657 	for (i = 0; i < dev->db_tab->npages; ++i) {
658 		if (!dev->db_tab->page[i].db_rec)
659 			continue;
660 
661 		if (!bitmap_empty(dev->db_tab->page[i].used, MTHCA_DB_REC_PER_PAGE))
662 			mthca_warn(dev, "Kernel UARC page %d not empty\n", i);
663 
664 		mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1, &status);
665 
666 		dma_free_coherent(&dev->pdev->dev, 4096,
667 				  dev->db_tab->page[i].db_rec,
668 				  dev->db_tab->page[i].mapping);
669 	}
670 
671 	kfree(dev->db_tab->page);
672 	kfree(dev->db_tab);
673 }
674