2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/errno.h>
35 #include <linux/slab.h>
37 #include <linux/bitmap.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/vmalloc.h>
43 u32 mlx4_bitmap_alloc(struct mlx4_bitmap *bitmap)
47 spin_lock(&bitmap->lock);
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)
53 obj = find_first_zero_bit(bitmap->table, bitmap->max);
56 if (obj < bitmap->max) {
57 set_bit(obj, bitmap->table);
58 bitmap->last = (obj + 1);
59 if (bitmap->last == bitmap->max)
68 spin_unlock(&bitmap->lock);
73 void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj)
75 mlx4_bitmap_free_range(bitmap, obj, 1);
78 static unsigned long find_aligned_range(unsigned long *bitmap,
85 start = ALIGN(start, align);
87 while ((start < nbits) && test_bit(start, bitmap))
97 for (i = start + 1; i < end; i++) {
98 if (test_bit(i, bitmap)) {
107 u32 mlx4_bitmap_alloc_range(struct mlx4_bitmap *bitmap, int cnt, int align)
111 if (likely(cnt == 1 && align == 1))
112 return mlx4_bitmap_alloc(bitmap);
114 spin_lock(&bitmap->lock);
116 obj = find_aligned_range(bitmap->table, bitmap->last,
117 bitmap->max, cnt, align);
118 if (obj >= bitmap->max) {
119 bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
121 obj = find_aligned_range(bitmap->table, 0, bitmap->max,
125 if (obj < bitmap->max) {
126 for (i = 0; i < cnt; i++)
127 set_bit(obj + i, bitmap->table);
128 if (obj == bitmap->last) {
129 bitmap->last = (obj + cnt);
130 if (bitmap->last >= bitmap->max)
138 bitmap->avail -= cnt;
140 spin_unlock(&bitmap->lock);
145 u32 mlx4_bitmap_avail(struct mlx4_bitmap *bitmap)
147 return bitmap->avail;
150 void mlx4_bitmap_free_range(struct mlx4_bitmap *bitmap, u32 obj, int cnt)
154 obj &= bitmap->max + bitmap->reserved_top - 1;
156 spin_lock(&bitmap->lock);
157 for (i = 0; i < cnt; i++)
158 clear_bit(obj + i, bitmap->table);
159 bitmap->last = min(bitmap->last, obj);
160 bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
162 bitmap->avail += cnt;
163 spin_unlock(&bitmap->lock);
166 int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask,
167 u32 reserved_bot, u32 reserved_top)
171 /* num must be a power of 2 */
172 if (num != roundup_pow_of_two(num))
177 bitmap->max = num - reserved_top;
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);
187 for (i = 0; i < reserved_bot; ++i)
188 set_bit(i, bitmap->table);
193 void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap)
195 kfree(bitmap->table);
199 * Handling for queue buffers -- we allocate a bunch of memory and
200 * register it in a memory region at HCA virtual address 0. If the
201 * requested size is > max_direct, we split the allocation into
202 * multiple pages, so we don't require too much contiguous memory.
205 int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
206 struct mlx4_buf *buf)
210 buf->direct.buf = NULL;
211 if (size <= max_direct) {
214 buf->page_shift = get_order(size) + PAGE_SHIFT;
215 buf->direct.buf = dma_alloc_coherent(&dev->pdev->dev,
216 size, &t, GFP_KERNEL);
217 if (!buf->direct.buf)
222 while (t & ((1 << buf->page_shift) - 1)) {
227 memset(buf->direct.buf, 0, size);
231 buf->direct.buf = NULL;
233 buf->nbufs = (size + PAGE_SIZE - 1) / PAGE_SIZE;
234 buf->npages = buf->nbufs;
235 buf->page_shift = PAGE_SHIFT;
236 buf->page_list = kzalloc(buf->nbufs * sizeof *buf->page_list,
241 for (i = 0; i < buf->nbufs; ++i) {
242 buf->page_list[i].buf =
243 dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
245 if (!buf->page_list[i].buf)
248 buf->page_list[i].map = t;
250 memset(buf->page_list[i].buf, 0, PAGE_SIZE);
253 if (BITS_PER_LONG == 64) {
255 pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL);
258 for (i = 0; i < buf->nbufs; ++i)
259 pages[i] = virt_to_page(buf->page_list[i].buf);
260 buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL);
262 if (!buf->direct.buf)
270 mlx4_buf_free(dev, size, buf);
274 EXPORT_SYMBOL_GPL(mlx4_buf_alloc);
276 void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
281 dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
284 if (BITS_PER_LONG == 64 && buf->direct.buf)
285 vunmap(buf->direct.buf);
287 for (i = 0; i < buf->nbufs; ++i)
288 if (buf->page_list[i].buf)
289 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
290 buf->page_list[i].buf,
291 buf->page_list[i].map);
292 kfree(buf->page_list);
294 buf->direct.buf = NULL;
296 EXPORT_SYMBOL_GPL(mlx4_buf_free);
298 static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device)
300 struct mlx4_db_pgdir *pgdir;
302 pgdir = kzalloc(sizeof *pgdir, GFP_KERNEL);
306 bitmap_fill(pgdir->order1, MLX4_DB_PER_PAGE / 2);
307 pgdir->bits[0] = pgdir->order0;
308 pgdir->bits[1] = pgdir->order1;
309 pgdir->db_page = dma_alloc_coherent(dma_device, PAGE_SIZE,
310 &pgdir->db_dma, GFP_KERNEL);
311 if (!pgdir->db_page) {
319 static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir,
320 struct mlx4_db *db, int order)
325 for (o = order; o <= 1; ++o) {
326 i = find_first_bit(pgdir->bits[o], MLX4_DB_PER_PAGE >> o);
327 if (i < MLX4_DB_PER_PAGE >> o)
334 clear_bit(i, pgdir->bits[o]);
339 set_bit(i ^ 1, pgdir->bits[order]);
343 db->db = pgdir->db_page + db->index;
344 db->dma = pgdir->db_dma + db->index * 4;
350 int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order)
352 struct mlx4_priv *priv = mlx4_priv(dev);
353 struct mlx4_db_pgdir *pgdir;
356 mutex_lock(&priv->pgdir_mutex);
358 list_for_each_entry(pgdir, &priv->pgdir_list, list)
359 if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
362 pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev));
368 list_add(&pgdir->list, &priv->pgdir_list);
370 /* This should never fail -- we just allocated an empty page: */
371 WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order));
374 mutex_unlock(&priv->pgdir_mutex);
378 EXPORT_SYMBOL_GPL(mlx4_db_alloc);
380 void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
382 struct mlx4_priv *priv = mlx4_priv(dev);
386 mutex_lock(&priv->pgdir_mutex);
391 if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) {
392 clear_bit(i ^ 1, db->u.pgdir->order0);
396 set_bit(i, db->u.pgdir->bits[o]);
398 if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) {
399 dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
400 db->u.pgdir->db_page, db->u.pgdir->db_dma);
401 list_del(&db->u.pgdir->list);
405 mutex_unlock(&priv->pgdir_mutex);
407 EXPORT_SYMBOL_GPL(mlx4_db_free);
409 int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
410 int size, int max_direct)
414 err = mlx4_db_alloc(dev, &wqres->db, 1);
420 err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf);
424 err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
429 err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf);
436 mlx4_mtt_cleanup(dev, &wqres->mtt);
438 mlx4_buf_free(dev, size, &wqres->buf);
440 mlx4_db_free(dev, &wqres->db);
444 EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res);
446 void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
449 mlx4_mtt_cleanup(dev, &wqres->mtt);
450 mlx4_buf_free(dev, size, &wqres->buf);
451 mlx4_db_free(dev, &wqres->db);
453 EXPORT_SYMBOL_GPL(mlx4_free_hwq_res);