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/export.h>
38 #include <linux/bitmap.h>
39 #include <linux/dma-mapping.h>
40 #include <linux/vmalloc.h>
44 u32
mlx4_bitmap_alloc(struct mlx4_bitmap
*bitmap
)
48 spin_lock(&bitmap
->lock
);
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
)
54 obj
= find_first_zero_bit(bitmap
->table
, bitmap
->max
);
57 if (obj
< bitmap
->max
) {
58 set_bit(obj
, bitmap
->table
);
59 bitmap
->last
= (obj
+ 1);
60 if (bitmap
->last
== bitmap
->max
)
69 spin_unlock(&bitmap
->lock
);
74 void mlx4_bitmap_free(struct mlx4_bitmap
*bitmap
, u32 obj
)
76 mlx4_bitmap_free_range(bitmap
, obj
, 1);
79 u32
mlx4_bitmap_alloc_range(struct mlx4_bitmap
*bitmap
, int cnt
, int align
)
83 if (likely(cnt
== 1 && align
== 1))
84 return mlx4_bitmap_alloc(bitmap
);
86 spin_lock(&bitmap
->lock
);
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
)
93 obj
= bitmap_find_next_zero_area(bitmap
->table
, bitmap
->max
,
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
)
109 bitmap
->avail
-= cnt
;
111 spin_unlock(&bitmap
->lock
);
116 u32
mlx4_bitmap_avail(struct mlx4_bitmap
*bitmap
)
118 return bitmap
->avail
;
121 void mlx4_bitmap_free_range(struct mlx4_bitmap
*bitmap
, u32 obj
, int cnt
)
123 obj
&= bitmap
->max
+ bitmap
->reserved_top
- 1;
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
)
130 bitmap
->avail
+= cnt
;
131 spin_unlock(&bitmap
->lock
);
134 int mlx4_bitmap_init(struct mlx4_bitmap
*bitmap
, u32 num
, u32 mask
,
135 u32 reserved_bot
, u32 reserved_top
)
137 /* num must be a power of 2 */
138 if (num
!= roundup_pow_of_two(num
))
143 bitmap
->max
= num
- reserved_top
;
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
);
153 bitmap_set(bitmap
->table
, 0, reserved_bot
);
158 void mlx4_bitmap_cleanup(struct mlx4_bitmap
*bitmap
)
160 kfree(bitmap
->table
);
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.
170 int mlx4_buf_alloc(struct mlx4_dev
*dev
, int size
, int max_direct
,
171 struct mlx4_buf
*buf
)
175 if (size
<= max_direct
) {
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
)
186 while (t
& ((1 << buf
->page_shift
) - 1)) {
191 memset(buf
->direct
.buf
, 0, size
);
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
),
204 for (i
= 0; i
< buf
->nbufs
; ++i
) {
205 buf
->page_list
[i
].buf
=
206 dma_alloc_coherent(&dev
->pdev
->dev
, PAGE_SIZE
,
208 if (!buf
->page_list
[i
].buf
)
211 buf
->page_list
[i
].map
= t
;
213 memset(buf
->page_list
[i
].buf
, 0, PAGE_SIZE
);
216 if (BITS_PER_LONG
== 64) {
218 pages
= kmalloc(sizeof *pages
* buf
->nbufs
, GFP_KERNEL
);
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
);
225 if (!buf
->direct
.buf
)
233 mlx4_buf_free(dev
, size
, buf
);
237 EXPORT_SYMBOL_GPL(mlx4_buf_alloc
);
239 void mlx4_buf_free(struct mlx4_dev
*dev
, int size
, struct mlx4_buf
*buf
)
244 dma_free_coherent(&dev
->pdev
->dev
, size
, buf
->direct
.buf
,
247 if (BITS_PER_LONG
== 64 && buf
->direct
.buf
)
248 vunmap(buf
->direct
.buf
);
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
);
258 EXPORT_SYMBOL_GPL(mlx4_buf_free
);
260 static struct mlx4_db_pgdir
*mlx4_alloc_db_pgdir(struct device
*dma_device
)
262 struct mlx4_db_pgdir
*pgdir
;
264 pgdir
= kzalloc(sizeof *pgdir
, GFP_KERNEL
);
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
) {
281 static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir
*pgdir
,
282 struct mlx4_db
*db
, int order
)
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
)
296 clear_bit(i
, pgdir
->bits
[o
]);
301 set_bit(i
^ 1, pgdir
->bits
[order
]);
305 db
->db
= pgdir
->db_page
+ db
->index
;
306 db
->dma
= pgdir
->db_dma
+ db
->index
* 4;
312 int mlx4_db_alloc(struct mlx4_dev
*dev
, struct mlx4_db
*db
, int order
)
314 struct mlx4_priv
*priv
= mlx4_priv(dev
);
315 struct mlx4_db_pgdir
*pgdir
;
318 mutex_lock(&priv
->pgdir_mutex
);
320 list_for_each_entry(pgdir
, &priv
->pgdir_list
, list
)
321 if (!mlx4_alloc_db_from_pgdir(pgdir
, db
, order
))
324 pgdir
= mlx4_alloc_db_pgdir(&(dev
->pdev
->dev
));
330 list_add(&pgdir
->list
, &priv
->pgdir_list
);
332 /* This should never fail -- we just allocated an empty page: */
333 WARN_ON(mlx4_alloc_db_from_pgdir(pgdir
, db
, order
));
336 mutex_unlock(&priv
->pgdir_mutex
);
340 EXPORT_SYMBOL_GPL(mlx4_db_alloc
);
342 void mlx4_db_free(struct mlx4_dev
*dev
, struct mlx4_db
*db
)
344 struct mlx4_priv
*priv
= mlx4_priv(dev
);
348 mutex_lock(&priv
->pgdir_mutex
);
353 if (db
->order
== 0 && test_bit(i
^ 1, db
->u
.pgdir
->order0
)) {
354 clear_bit(i
^ 1, db
->u
.pgdir
->order0
);
358 set_bit(i
, db
->u
.pgdir
->bits
[o
]);
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
);
367 mutex_unlock(&priv
->pgdir_mutex
);
369 EXPORT_SYMBOL_GPL(mlx4_db_free
);
371 int mlx4_alloc_hwq_res(struct mlx4_dev
*dev
, struct mlx4_hwq_resources
*wqres
,
372 int size
, int max_direct
)
376 err
= mlx4_db_alloc(dev
, &wqres
->db
, 1);
382 err
= mlx4_buf_alloc(dev
, size
, max_direct
, &wqres
->buf
);
386 err
= mlx4_mtt_init(dev
, wqres
->buf
.npages
, wqres
->buf
.page_shift
,
391 err
= mlx4_buf_write_mtt(dev
, &wqres
->mtt
, &wqres
->buf
);
398 mlx4_mtt_cleanup(dev
, &wqres
->mtt
);
400 mlx4_buf_free(dev
, size
, &wqres
->buf
);
402 mlx4_db_free(dev
, &wqres
->db
);
406 EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res
);
408 void mlx4_free_hwq_res(struct mlx4_dev
*dev
, struct mlx4_hwq_resources
*wqres
,
411 mlx4_mtt_cleanup(dev
, &wqres
->mtt
);
412 mlx4_buf_free(dev
, size
, &wqres
->buf
);
413 mlx4_db_free(dev
, &wqres
->db
);
415 EXPORT_SYMBOL_GPL(mlx4_free_hwq_res
);