alloc.c revision 271127 
1/*
2 * Copyright (c) 2006, 2007 Cisco Systems, Inc.  All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. 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
34#include <linux/errno.h>
35#include <linux/slab.h>
36#include <linux/mm.h>
37#include <linux/bitops.h>
38#include <linux/dma-mapping.h>
39#include <linux/vmalloc.h>
40
41#include "mlx4.h"
42
43u32 mlx4_bitmap_alloc(struct mlx4_bitmap *bitmap)
44{
45	u32 obj;
46
47	spin_lock(&bitmap->lock);
48
49	obj = find_next_zero_bit(bitmap->table, bitmap->max, bitmap->last);
50	if (obj >= bitmap->max) {
51		bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
52				& bitmap->mask;
53		obj = find_first_zero_bit(bitmap->table, bitmap->max);
54	}
55
56	if (obj < bitmap->max) {
57		set_bit(obj, bitmap->table);
58		bitmap->last = (obj + 1);
59		if (bitmap->last == bitmap->max)
60			bitmap->last = 0;
61		obj |= bitmap->top;
62	} else
63		obj = -1;
64
65	if (obj != -1)
66		--bitmap->avail;
67
68	spin_unlock(&bitmap->lock);
69
70	return obj;
71}
72
73void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj)
74{
75	mlx4_bitmap_free_range(bitmap, obj, 1);
76}
77
78static unsigned long find_aligned_range(unsigned long *bitmap,
79					u32 start, u32 nbits,
80					int len, int align, u32 skip_mask)
81{
82	unsigned long end, i;
83
84again:
85	start = ALIGN(start, align);
86
87	while ((start < nbits) && (test_bit(start, bitmap) ||
88				   (start & skip_mask)))
89		start += align;
90
91	if (start >= nbits)
92		return -1;
93
94	end = start+len;
95	if (end > nbits)
96		return -1;
97
98	for (i = start + 1; i < end; i++) {
99		if (test_bit(i, bitmap) || ((u32)i & skip_mask)) {
100			start = i + 1;
101			goto again;
102		}
103	}
104
105	return start;
106}
107
108u32 mlx4_bitmap_alloc_range(struct mlx4_bitmap *bitmap, int cnt,
109			    int align, u32 skip_mask)
110{
111	u32 obj;
112
113	if (likely(cnt == 1 && align == 1 && !skip_mask))
114		return mlx4_bitmap_alloc(bitmap);
115
116	spin_lock(&bitmap->lock);
117
118	obj = find_aligned_range(bitmap->table, bitmap->last,
119				bitmap->max, cnt, align, skip_mask);
120	if (obj >= bitmap->max) {
121		bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
122				& bitmap->mask;
123		obj = find_aligned_range(bitmap->table, 0, bitmap->max,
124						cnt, align, skip_mask);
125	}
126
127	if (obj < bitmap->max) {
128		bitmap_set(bitmap->table, obj, cnt);
129		if (obj == bitmap->last) {
130			bitmap->last = (obj + cnt);
131			if (bitmap->last >= bitmap->max)
132				bitmap->last = 0;
133		}
134		obj |= bitmap->top;
135	} else
136		obj = -1;
137
138	if (obj != -1)
139		bitmap->avail -= cnt;
140
141	spin_unlock(&bitmap->lock);
142
143	return obj;
144}
145
146u32 mlx4_bitmap_avail(struct mlx4_bitmap *bitmap)
147{
148	return bitmap->avail;
149}
150
151void mlx4_bitmap_free_range(struct mlx4_bitmap *bitmap, u32 obj, int cnt)
152{
153	obj &= bitmap->max + bitmap->reserved_top - 1;
154
155	spin_lock(&bitmap->lock);
156	bitmap_clear(bitmap->table, obj, cnt);
157	bitmap->avail += cnt;
158	spin_unlock(&bitmap->lock);
159}
160
161int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask,
162		     u32 reserved_bot, u32 reserved_top)
163{
164	/* sanity check */
165	if (num <= (u64)reserved_top + reserved_bot)
166		return -EINVAL;
167
168	/* num must be a power of 2 */
169	if (num != roundup_pow_of_two(num))
170		return -EINVAL;
171
172	if (reserved_bot + reserved_top >= num)
173		return -EINVAL;
174
175	bitmap->last = 0;
176	bitmap->top  = 0;
177	bitmap->max  = num - reserved_top;
178	bitmap->mask = mask;
179	bitmap->reserved_top = reserved_top;
180	bitmap->avail = num - reserved_top - reserved_bot;
181	spin_lock_init(&bitmap->lock);
182	bitmap->table = kzalloc(BITS_TO_LONGS(bitmap->max) *
183				sizeof (long), GFP_KERNEL);
184	if (!bitmap->table)
185		return -ENOMEM;
186
187	bitmap_set(bitmap->table, 0, reserved_bot);
188
189	return 0;
190}
191
192void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap)
193{
194	kfree(bitmap->table);
195}
196
197/*
198 * Handling for queue buffers -- we allocate a bunch of memory and
199 * register it in a memory region at HCA virtual address 0.  If the
200 * requested size is > max_direct, we split the allocation into
201 * multiple pages, so we don't require too much contiguous memory.
202 */
203
204int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
205		   struct mlx4_buf *buf)
206{
207	dma_addr_t t;
208
209	if (size <= max_direct) {
210		buf->nbufs        = 1;
211		buf->npages       = 1;
212		buf->page_shift   = get_order(size) + PAGE_SHIFT;
213		buf->direct.buf   = dma_alloc_coherent(&dev->pdev->dev,
214						       size, &t, GFP_KERNEL);
215		if (!buf->direct.buf)
216			return -ENOMEM;
217
218		buf->direct.map = t;
219
220		while (t & ((1 << buf->page_shift) - 1)) {
221			--buf->page_shift;
222			buf->npages *= 2;
223		}
224
225		memset(buf->direct.buf, 0, size);
226	} else {
227		int i;
228
229		buf->direct.buf  = NULL;
230		buf->nbufs       = (size + PAGE_SIZE - 1) / PAGE_SIZE;
231		buf->npages      = buf->nbufs;
232		buf->page_shift  = PAGE_SHIFT;
233		buf->page_list   = kcalloc(buf->nbufs, sizeof(*buf->page_list),
234					   GFP_KERNEL);
235		if (!buf->page_list)
236			return -ENOMEM;
237
238		for (i = 0; i < buf->nbufs; ++i) {
239			buf->page_list[i].buf =
240				dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
241						   &t, GFP_KERNEL);
242			if (!buf->page_list[i].buf)
243				goto err_free;
244
245			buf->page_list[i].map = t;
246
247			memset(buf->page_list[i].buf, 0, PAGE_SIZE);
248		}
249
250		if (BITS_PER_LONG == 64) {
251			struct page **pages;
252			pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL);
253			if (!pages)
254				goto err_free;
255			for (i = 0; i < buf->nbufs; ++i)
256				pages[i] = virt_to_page(buf->page_list[i].buf);
257			buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL);
258			kfree(pages);
259			if (!buf->direct.buf)
260				goto err_free;
261		}
262	}
263
264	return 0;
265
266err_free:
267	mlx4_buf_free(dev, size, buf);
268
269	return -ENOMEM;
270}
271EXPORT_SYMBOL_GPL(mlx4_buf_alloc);
272
273void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
274{
275	int i;
276
277	if (buf->nbufs == 1)
278		dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
279				  buf->direct.map);
280	else {
281		if (BITS_PER_LONG == 64 && buf->direct.buf)
282			vunmap(buf->direct.buf);
283
284		for (i = 0; i < buf->nbufs; ++i)
285			if (buf->page_list[i].buf)
286				dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
287						  buf->page_list[i].buf,
288						  buf->page_list[i].map);
289		kfree(buf->page_list);
290	}
291}
292EXPORT_SYMBOL_GPL(mlx4_buf_free);
293
294static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device)
295{
296	struct mlx4_db_pgdir *pgdir;
297
298	pgdir = kzalloc(sizeof *pgdir, GFP_KERNEL);
299	if (!pgdir)
300		return NULL;
301
302	bitmap_fill(pgdir->order1, MLX4_DB_PER_PAGE / 2);
303	pgdir->bits[0] = pgdir->order0;
304	pgdir->bits[1] = pgdir->order1;
305	pgdir->db_page = dma_alloc_coherent(dma_device, PAGE_SIZE,
306					    &pgdir->db_dma, GFP_KERNEL);
307	if (!pgdir->db_page) {
308		kfree(pgdir);
309		return NULL;
310	}
311
312	return pgdir;
313}
314
315static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir,
316				    struct mlx4_db *db, int order)
317{
318	int o;
319	int i;
320
321	for (o = order; o <= 1; ++o) {
322		i = find_first_bit(pgdir->bits[o], MLX4_DB_PER_PAGE >> o);
323		if (i < MLX4_DB_PER_PAGE >> o)
324			goto found;
325	}
326
327	return -ENOMEM;
328
329found:
330	clear_bit(i, pgdir->bits[o]);
331
332	i <<= o;
333
334	if (o > order)
335		set_bit(i ^ 1, pgdir->bits[order]);
336
337	db->u.pgdir = pgdir;
338	db->index   = i;
339	db->db      = pgdir->db_page + db->index;
340	db->dma     = pgdir->db_dma  + db->index * 4;
341	db->order   = order;
342
343	return 0;
344}
345
346int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order)
347{
348	struct mlx4_priv *priv = mlx4_priv(dev);
349	struct mlx4_db_pgdir *pgdir;
350	int ret = 0;
351
352	mutex_lock(&priv->pgdir_mutex);
353
354	list_for_each_entry(pgdir, &priv->pgdir_list, list)
355		if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
356			goto out;
357
358	pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev));
359	if (!pgdir) {
360		ret = -ENOMEM;
361		goto out;
362	}
363
364	list_add(&pgdir->list, &priv->pgdir_list);
365
366	/* This should never fail -- we just allocated an empty page: */
367	WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order));
368
369out:
370	mutex_unlock(&priv->pgdir_mutex);
371
372	return ret;
373}
374EXPORT_SYMBOL_GPL(mlx4_db_alloc);
375
376void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
377{
378	struct mlx4_priv *priv = mlx4_priv(dev);
379	int o;
380	int i;
381
382	mutex_lock(&priv->pgdir_mutex);
383
384	o = db->order;
385	i = db->index;
386
387	if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) {
388		clear_bit(i ^ 1, db->u.pgdir->order0);
389		++o;
390	}
391	i >>= o;
392	set_bit(i, db->u.pgdir->bits[o]);
393
394	if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) {
395		dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
396				  db->u.pgdir->db_page, db->u.pgdir->db_dma);
397		list_del(&db->u.pgdir->list);
398		kfree(db->u.pgdir);
399	}
400
401	mutex_unlock(&priv->pgdir_mutex);
402}
403EXPORT_SYMBOL_GPL(mlx4_db_free);
404
405int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
406		       int size, int max_direct)
407{
408	int err;
409
410	err = mlx4_db_alloc(dev, &wqres->db, 1);
411	if (err)
412		return err;
413
414	*wqres->db.db = 0;
415
416	err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf);
417	if (err)
418		goto err_db;
419
420	err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
421			    &wqres->mtt);
422	if (err)
423		goto err_buf;
424
425	err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf);
426	if (err)
427		goto err_mtt;
428
429	return 0;
430
431err_mtt:
432	mlx4_mtt_cleanup(dev, &wqres->mtt);
433err_buf:
434	mlx4_buf_free(dev, size, &wqres->buf);
435err_db:
436	mlx4_db_free(dev, &wqres->db);
437
438	return err;
439}
440EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res);
441
442void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
443		       int size)
444{
445	mlx4_mtt_cleanup(dev, &wqres->mtt);
446	mlx4_buf_free(dev, size, &wqres->buf);
447	mlx4_db_free(dev, &wqres->db);
448}
449EXPORT_SYMBOL_GPL(mlx4_free_hwq_res);
450