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
)
65 spin_unlock(&bitmap
->lock
);
70 void mlx4_bitmap_free(struct mlx4_bitmap
*bitmap
, u32 obj
)
72 mlx4_bitmap_free_range(bitmap
, obj
, 1);
75 static unsigned long find_aligned_range(unsigned long *bitmap
,
82 start
= ALIGN(start
, align
);
84 while ((start
< nbits
) && test_bit(start
, bitmap
))
94 for (i
= start
+ 1; i
< end
; i
++) {
95 if (test_bit(i
, bitmap
)) {
104 u32
mlx4_bitmap_alloc_range(struct mlx4_bitmap
*bitmap
, int cnt
, int align
)
108 if (likely(cnt
== 1 && align
== 1))
109 return mlx4_bitmap_alloc(bitmap
);
111 spin_lock(&bitmap
->lock
);
113 obj
= find_aligned_range(bitmap
->table
, bitmap
->last
,
114 bitmap
->max
, cnt
, align
);
115 if (obj
>= bitmap
->max
) {
116 bitmap
->top
= (bitmap
->top
+ bitmap
->max
+ bitmap
->reserved_top
)
118 obj
= find_aligned_range(bitmap
->table
, 0, bitmap
->max
,
122 if (obj
< bitmap
->max
) {
123 for (i
= 0; i
< cnt
; i
++)
124 set_bit(obj
+ i
, bitmap
->table
);
125 if (obj
== bitmap
->last
) {
126 bitmap
->last
= (obj
+ cnt
);
127 if (bitmap
->last
>= bitmap
->max
)
134 spin_unlock(&bitmap
->lock
);
139 void mlx4_bitmap_free_range(struct mlx4_bitmap
*bitmap
, u32 obj
, int cnt
)
143 obj
&= bitmap
->max
+ bitmap
->reserved_top
- 1;
145 spin_lock(&bitmap
->lock
);
146 for (i
= 0; i
< cnt
; i
++)
147 clear_bit(obj
+ i
, bitmap
->table
);
148 bitmap
->last
= min(bitmap
->last
, obj
);
149 bitmap
->top
= (bitmap
->top
+ bitmap
->max
+ bitmap
->reserved_top
)
151 spin_unlock(&bitmap
->lock
);
154 int mlx4_bitmap_init(struct mlx4_bitmap
*bitmap
, u32 num
, u32 mask
,
155 u32 reserved_bot
, u32 reserved_top
)
159 /* num must be a power of 2 */
160 if (num
!= roundup_pow_of_two(num
))
165 bitmap
->max
= num
- reserved_top
;
167 bitmap
->reserved_top
= reserved_top
;
168 spin_lock_init(&bitmap
->lock
);
169 bitmap
->table
= kzalloc(BITS_TO_LONGS(bitmap
->max
) *
170 sizeof (long), GFP_KERNEL
);
174 for (i
= 0; i
< reserved_bot
; ++i
)
175 set_bit(i
, bitmap
->table
);
180 void mlx4_bitmap_cleanup(struct mlx4_bitmap
*bitmap
)
182 kfree(bitmap
->table
);
186 * Handling for queue buffers -- we allocate a bunch of memory and
187 * register it in a memory region at HCA virtual address 0. If the
188 * requested size is > max_direct, we split the allocation into
189 * multiple pages, so we don't require too much contiguous memory.
192 int mlx4_buf_alloc(struct mlx4_dev
*dev
, int size
, int max_direct
,
193 struct mlx4_buf
*buf
)
197 if (size
<= max_direct
) {
200 buf
->page_shift
= get_order(size
) + PAGE_SHIFT
;
201 buf
->direct
.buf
= dma_alloc_coherent(&dev
->pdev
->dev
,
202 size
, &t
, GFP_KERNEL
);
203 if (!buf
->direct
.buf
)
208 while (t
& ((1 << buf
->page_shift
) - 1)) {
213 memset(buf
->direct
.buf
, 0, size
);
217 buf
->nbufs
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
218 buf
->npages
= buf
->nbufs
;
219 buf
->page_shift
= PAGE_SHIFT
;
220 buf
->page_list
= kzalloc(buf
->nbufs
* sizeof *buf
->page_list
,
225 for (i
= 0; i
< buf
->nbufs
; ++i
) {
226 buf
->page_list
[i
].buf
=
227 dma_alloc_coherent(&dev
->pdev
->dev
, PAGE_SIZE
,
229 if (!buf
->page_list
[i
].buf
)
232 buf
->page_list
[i
].map
= t
;
234 memset(buf
->page_list
[i
].buf
, 0, PAGE_SIZE
);
237 if (BITS_PER_LONG
== 64) {
239 pages
= kmalloc(sizeof *pages
* buf
->nbufs
, GFP_KERNEL
);
242 for (i
= 0; i
< buf
->nbufs
; ++i
)
243 pages
[i
] = virt_to_page(buf
->page_list
[i
].buf
);
244 buf
->direct
.buf
= vmap(pages
, buf
->nbufs
, VM_MAP
, PAGE_KERNEL
);
246 if (!buf
->direct
.buf
)
254 mlx4_buf_free(dev
, size
, buf
);
258 EXPORT_SYMBOL_GPL(mlx4_buf_alloc
);
260 void mlx4_buf_free(struct mlx4_dev
*dev
, int size
, struct mlx4_buf
*buf
)
265 dma_free_coherent(&dev
->pdev
->dev
, size
, buf
->direct
.buf
,
268 if (BITS_PER_LONG
== 64)
269 vunmap(buf
->direct
.buf
);
271 for (i
= 0; i
< buf
->nbufs
; ++i
)
272 if (buf
->page_list
[i
].buf
)
273 dma_free_coherent(&dev
->pdev
->dev
, PAGE_SIZE
,
274 buf
->page_list
[i
].buf
,
275 buf
->page_list
[i
].map
);
276 kfree(buf
->page_list
);
279 EXPORT_SYMBOL_GPL(mlx4_buf_free
);
281 static struct mlx4_db_pgdir
*mlx4_alloc_db_pgdir(struct device
*dma_device
)
283 struct mlx4_db_pgdir
*pgdir
;
285 pgdir
= kzalloc(sizeof *pgdir
, GFP_KERNEL
);
289 bitmap_fill(pgdir
->order1
, MLX4_DB_PER_PAGE
/ 2);
290 pgdir
->bits
[0] = pgdir
->order0
;
291 pgdir
->bits
[1] = pgdir
->order1
;
292 pgdir
->db_page
= dma_alloc_coherent(dma_device
, PAGE_SIZE
,
293 &pgdir
->db_dma
, GFP_KERNEL
);
294 if (!pgdir
->db_page
) {
302 static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir
*pgdir
,
303 struct mlx4_db
*db
, int order
)
308 for (o
= order
; o
<= 1; ++o
) {
309 i
= find_first_bit(pgdir
->bits
[o
], MLX4_DB_PER_PAGE
>> o
);
310 if (i
< MLX4_DB_PER_PAGE
>> o
)
317 clear_bit(i
, pgdir
->bits
[o
]);
322 set_bit(i
^ 1, pgdir
->bits
[order
]);
326 db
->db
= pgdir
->db_page
+ db
->index
;
327 db
->dma
= pgdir
->db_dma
+ db
->index
* 4;
333 int mlx4_db_alloc(struct mlx4_dev
*dev
, struct mlx4_db
*db
, int order
)
335 struct mlx4_priv
*priv
= mlx4_priv(dev
);
336 struct mlx4_db_pgdir
*pgdir
;
339 mutex_lock(&priv
->pgdir_mutex
);
341 list_for_each_entry(pgdir
, &priv
->pgdir_list
, list
)
342 if (!mlx4_alloc_db_from_pgdir(pgdir
, db
, order
))
345 pgdir
= mlx4_alloc_db_pgdir(&(dev
->pdev
->dev
));
351 list_add(&pgdir
->list
, &priv
->pgdir_list
);
353 /* This should never fail -- we just allocated an empty page: */
354 WARN_ON(mlx4_alloc_db_from_pgdir(pgdir
, db
, order
));
357 mutex_unlock(&priv
->pgdir_mutex
);
361 EXPORT_SYMBOL_GPL(mlx4_db_alloc
);
363 void mlx4_db_free(struct mlx4_dev
*dev
, struct mlx4_db
*db
)
365 struct mlx4_priv
*priv
= mlx4_priv(dev
);
369 mutex_lock(&priv
->pgdir_mutex
);
374 if (db
->order
== 0 && test_bit(i
^ 1, db
->u
.pgdir
->order0
)) {
375 clear_bit(i
^ 1, db
->u
.pgdir
->order0
);
379 set_bit(i
, db
->u
.pgdir
->bits
[o
]);
381 if (bitmap_full(db
->u
.pgdir
->order1
, MLX4_DB_PER_PAGE
/ 2)) {
382 dma_free_coherent(&(dev
->pdev
->dev
), PAGE_SIZE
,
383 db
->u
.pgdir
->db_page
, db
->u
.pgdir
->db_dma
);
384 list_del(&db
->u
.pgdir
->list
);
388 mutex_unlock(&priv
->pgdir_mutex
);
390 EXPORT_SYMBOL_GPL(mlx4_db_free
);
392 int mlx4_alloc_hwq_res(struct mlx4_dev
*dev
, struct mlx4_hwq_resources
*wqres
,
393 int size
, int max_direct
)
397 err
= mlx4_db_alloc(dev
, &wqres
->db
, 1);
403 err
= mlx4_buf_alloc(dev
, size
, max_direct
, &wqres
->buf
);
407 err
= mlx4_mtt_init(dev
, wqres
->buf
.npages
, wqres
->buf
.page_shift
,
412 err
= mlx4_buf_write_mtt(dev
, &wqres
->mtt
, &wqres
->buf
);
419 mlx4_mtt_cleanup(dev
, &wqres
->mtt
);
421 mlx4_buf_free(dev
, size
, &wqres
->buf
);
423 mlx4_db_free(dev
, &wqres
->db
);
427 EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res
);
429 void mlx4_free_hwq_res(struct mlx4_dev
*dev
, struct mlx4_hwq_resources
*wqres
,
432 mlx4_mtt_cleanup(dev
, &wqres
->mtt
);
433 mlx4_buf_free(dev
, size
, &wqres
->buf
);
434 mlx4_db_free(dev
, &wqres
->db
);
436 EXPORT_SYMBOL_GPL(mlx4_free_hwq_res
);