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