]>
Commit | Line | Data |
---|---|---|
225c7b1f RD |
1 | /* |
2 | * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved. | |
51a379d0 | 3 | * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved. |
225c7b1f RD |
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> | |
6526128c | 36 | #include <linux/mm.h> |
225c7b1f | 37 | #include <linux/bitmap.h> |
9cbe05c7 | 38 | #include <linux/dma-mapping.h> |
29c27112 | 39 | #include <linux/vmalloc.h> |
225c7b1f RD |
40 | |
41 | #include "mlx4.h" | |
42 | ||
43 | u32 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) { | |
93fc9e1b YP |
51 | bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top) |
52 | & bitmap->mask; | |
225c7b1f RD |
53 | obj = find_first_zero_bit(bitmap->table, bitmap->max); |
54 | } | |
55 | ||
56 | if (obj < bitmap->max) { | |
57 | set_bit(obj, bitmap->table); | |
93fc9e1b YP |
58 | bitmap->last = (obj + 1); |
59 | if (bitmap->last == bitmap->max) | |
60 | bitmap->last = 0; | |
225c7b1f | 61 | obj |= bitmap->top; |
225c7b1f RD |
62 | } else |
63 | obj = -1; | |
64 | ||
65 | spin_unlock(&bitmap->lock); | |
66 | ||
67 | return obj; | |
68 | } | |
69 | ||
70 | void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj) | |
71 | { | |
a3cdcbfa YP |
72 | mlx4_bitmap_free_range(bitmap, obj, 1); |
73 | } | |
74 | ||
a3cdcbfa YP |
75 | u32 mlx4_bitmap_alloc_range(struct mlx4_bitmap *bitmap, int cnt, int align) |
76 | { | |
e27cd4f8 | 77 | u32 obj; |
a3cdcbfa YP |
78 | |
79 | if (likely(cnt == 1 && align == 1)) | |
80 | return mlx4_bitmap_alloc(bitmap); | |
81 | ||
82 | spin_lock(&bitmap->lock); | |
83 | ||
43ff8b60 AM |
84 | obj = bitmap_find_next_zero_area(bitmap->table, bitmap->max, |
85 | bitmap->last, cnt, align - 1); | |
a3cdcbfa | 86 | if (obj >= bitmap->max) { |
93fc9e1b YP |
87 | bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top) |
88 | & bitmap->mask; | |
43ff8b60 AM |
89 | obj = bitmap_find_next_zero_area(bitmap->table, bitmap->max, |
90 | 0, cnt, align - 1); | |
a3cdcbfa YP |
91 | } |
92 | ||
93 | if (obj < bitmap->max) { | |
e27cd4f8 | 94 | bitmap_set(bitmap->table, obj, cnt); |
a3cdcbfa YP |
95 | if (obj == bitmap->last) { |
96 | bitmap->last = (obj + cnt); | |
97 | if (bitmap->last >= bitmap->max) | |
98 | bitmap->last = 0; | |
99 | } | |
100 | obj |= bitmap->top; | |
101 | } else | |
102 | obj = -1; | |
103 | ||
104 | spin_unlock(&bitmap->lock); | |
105 | ||
106 | return obj; | |
107 | } | |
108 | ||
109 | void mlx4_bitmap_free_range(struct mlx4_bitmap *bitmap, u32 obj, int cnt) | |
110 | { | |
93fc9e1b | 111 | obj &= bitmap->max + bitmap->reserved_top - 1; |
225c7b1f RD |
112 | |
113 | spin_lock(&bitmap->lock); | |
e27cd4f8 | 114 | bitmap_clear(bitmap->table, obj, cnt); |
225c7b1f | 115 | bitmap->last = min(bitmap->last, obj); |
93fc9e1b YP |
116 | bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top) |
117 | & bitmap->mask; | |
225c7b1f RD |
118 | spin_unlock(&bitmap->lock); |
119 | } | |
120 | ||
93fc9e1b YP |
121 | int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask, |
122 | u32 reserved_bot, u32 reserved_top) | |
225c7b1f | 123 | { |
225c7b1f RD |
124 | /* num must be a power of 2 */ |
125 | if (num != roundup_pow_of_two(num)) | |
126 | return -EINVAL; | |
127 | ||
128 | bitmap->last = 0; | |
129 | bitmap->top = 0; | |
93fc9e1b | 130 | bitmap->max = num - reserved_top; |
225c7b1f | 131 | bitmap->mask = mask; |
93fc9e1b | 132 | bitmap->reserved_top = reserved_top; |
225c7b1f | 133 | spin_lock_init(&bitmap->lock); |
93fc9e1b YP |
134 | bitmap->table = kzalloc(BITS_TO_LONGS(bitmap->max) * |
135 | sizeof (long), GFP_KERNEL); | |
225c7b1f RD |
136 | if (!bitmap->table) |
137 | return -ENOMEM; | |
138 | ||
e27cd4f8 | 139 | bitmap_set(bitmap->table, 0, reserved_bot); |
225c7b1f RD |
140 | |
141 | return 0; | |
142 | } | |
143 | ||
144 | void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap) | |
145 | { | |
146 | kfree(bitmap->table); | |
147 | } | |
148 | ||
149 | /* | |
150 | * Handling for queue buffers -- we allocate a bunch of memory and | |
151 | * register it in a memory region at HCA virtual address 0. If the | |
152 | * requested size is > max_direct, we split the allocation into | |
153 | * multiple pages, so we don't require too much contiguous memory. | |
154 | */ | |
155 | ||
156 | int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct, | |
157 | struct mlx4_buf *buf) | |
158 | { | |
159 | dma_addr_t t; | |
160 | ||
161 | if (size <= max_direct) { | |
162 | buf->nbufs = 1; | |
163 | buf->npages = 1; | |
164 | buf->page_shift = get_order(size) + PAGE_SHIFT; | |
b57aacfa | 165 | buf->direct.buf = dma_alloc_coherent(&dev->pdev->dev, |
225c7b1f | 166 | size, &t, GFP_KERNEL); |
b57aacfa | 167 | if (!buf->direct.buf) |
225c7b1f RD |
168 | return -ENOMEM; |
169 | ||
b57aacfa | 170 | buf->direct.map = t; |
225c7b1f RD |
171 | |
172 | while (t & ((1 << buf->page_shift) - 1)) { | |
173 | --buf->page_shift; | |
174 | buf->npages *= 2; | |
175 | } | |
176 | ||
b57aacfa | 177 | memset(buf->direct.buf, 0, size); |
225c7b1f RD |
178 | } else { |
179 | int i; | |
180 | ||
181 | buf->nbufs = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
182 | buf->npages = buf->nbufs; | |
183 | buf->page_shift = PAGE_SHIFT; | |
baeb2ffa | 184 | buf->page_list = kcalloc(buf->nbufs, sizeof(*buf->page_list), |
225c7b1f | 185 | GFP_KERNEL); |
b57aacfa | 186 | if (!buf->page_list) |
225c7b1f RD |
187 | return -ENOMEM; |
188 | ||
189 | for (i = 0; i < buf->nbufs; ++i) { | |
b57aacfa | 190 | buf->page_list[i].buf = |
225c7b1f RD |
191 | dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE, |
192 | &t, GFP_KERNEL); | |
b57aacfa | 193 | if (!buf->page_list[i].buf) |
225c7b1f RD |
194 | goto err_free; |
195 | ||
b57aacfa | 196 | buf->page_list[i].map = t; |
225c7b1f | 197 | |
b57aacfa | 198 | memset(buf->page_list[i].buf, 0, PAGE_SIZE); |
225c7b1f | 199 | } |
313abe55 JM |
200 | |
201 | if (BITS_PER_LONG == 64) { | |
202 | struct page **pages; | |
203 | pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL); | |
204 | if (!pages) | |
205 | goto err_free; | |
206 | for (i = 0; i < buf->nbufs; ++i) | |
b57aacfa RD |
207 | pages[i] = virt_to_page(buf->page_list[i].buf); |
208 | buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL); | |
313abe55 | 209 | kfree(pages); |
b57aacfa | 210 | if (!buf->direct.buf) |
313abe55 JM |
211 | goto err_free; |
212 | } | |
225c7b1f RD |
213 | } |
214 | ||
215 | return 0; | |
216 | ||
217 | err_free: | |
218 | mlx4_buf_free(dev, size, buf); | |
219 | ||
220 | return -ENOMEM; | |
221 | } | |
222 | EXPORT_SYMBOL_GPL(mlx4_buf_alloc); | |
223 | ||
224 | void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf) | |
225 | { | |
226 | int i; | |
227 | ||
228 | if (buf->nbufs == 1) | |
b57aacfa RD |
229 | dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf, |
230 | buf->direct.map); | |
225c7b1f | 231 | else { |
313abe55 | 232 | if (BITS_PER_LONG == 64) |
b57aacfa | 233 | vunmap(buf->direct.buf); |
313abe55 | 234 | |
225c7b1f | 235 | for (i = 0; i < buf->nbufs; ++i) |
b57aacfa | 236 | if (buf->page_list[i].buf) |
3bba11e5 | 237 | dma_free_coherent(&dev->pdev->dev, PAGE_SIZE, |
b57aacfa RD |
238 | buf->page_list[i].buf, |
239 | buf->page_list[i].map); | |
240 | kfree(buf->page_list); | |
225c7b1f RD |
241 | } |
242 | } | |
243 | EXPORT_SYMBOL_GPL(mlx4_buf_free); | |
6296883c YP |
244 | |
245 | static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device) | |
246 | { | |
247 | struct mlx4_db_pgdir *pgdir; | |
248 | ||
249 | pgdir = kzalloc(sizeof *pgdir, GFP_KERNEL); | |
250 | if (!pgdir) | |
251 | return NULL; | |
252 | ||
253 | bitmap_fill(pgdir->order1, MLX4_DB_PER_PAGE / 2); | |
254 | pgdir->bits[0] = pgdir->order0; | |
255 | pgdir->bits[1] = pgdir->order1; | |
256 | pgdir->db_page = dma_alloc_coherent(dma_device, PAGE_SIZE, | |
257 | &pgdir->db_dma, GFP_KERNEL); | |
258 | if (!pgdir->db_page) { | |
259 | kfree(pgdir); | |
260 | return NULL; | |
261 | } | |
262 | ||
263 | return pgdir; | |
264 | } | |
265 | ||
266 | static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir, | |
267 | struct mlx4_db *db, int order) | |
268 | { | |
269 | int o; | |
270 | int i; | |
271 | ||
272 | for (o = order; o <= 1; ++o) { | |
273 | i = find_first_bit(pgdir->bits[o], MLX4_DB_PER_PAGE >> o); | |
274 | if (i < MLX4_DB_PER_PAGE >> o) | |
275 | goto found; | |
276 | } | |
277 | ||
278 | return -ENOMEM; | |
279 | ||
280 | found: | |
281 | clear_bit(i, pgdir->bits[o]); | |
282 | ||
283 | i <<= o; | |
284 | ||
285 | if (o > order) | |
286 | set_bit(i ^ 1, pgdir->bits[order]); | |
287 | ||
288 | db->u.pgdir = pgdir; | |
289 | db->index = i; | |
290 | db->db = pgdir->db_page + db->index; | |
291 | db->dma = pgdir->db_dma + db->index * 4; | |
292 | db->order = order; | |
293 | ||
294 | return 0; | |
295 | } | |
296 | ||
297 | int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order) | |
298 | { | |
299 | struct mlx4_priv *priv = mlx4_priv(dev); | |
300 | struct mlx4_db_pgdir *pgdir; | |
301 | int ret = 0; | |
302 | ||
303 | mutex_lock(&priv->pgdir_mutex); | |
304 | ||
305 | list_for_each_entry(pgdir, &priv->pgdir_list, list) | |
306 | if (!mlx4_alloc_db_from_pgdir(pgdir, db, order)) | |
307 | goto out; | |
308 | ||
309 | pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev)); | |
310 | if (!pgdir) { | |
311 | ret = -ENOMEM; | |
312 | goto out; | |
313 | } | |
314 | ||
315 | list_add(&pgdir->list, &priv->pgdir_list); | |
316 | ||
317 | /* This should never fail -- we just allocated an empty page: */ | |
318 | WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order)); | |
319 | ||
320 | out: | |
321 | mutex_unlock(&priv->pgdir_mutex); | |
322 | ||
323 | return ret; | |
324 | } | |
325 | EXPORT_SYMBOL_GPL(mlx4_db_alloc); | |
326 | ||
327 | void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db) | |
328 | { | |
329 | struct mlx4_priv *priv = mlx4_priv(dev); | |
330 | int o; | |
331 | int i; | |
332 | ||
333 | mutex_lock(&priv->pgdir_mutex); | |
334 | ||
335 | o = db->order; | |
336 | i = db->index; | |
337 | ||
338 | if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) { | |
339 | clear_bit(i ^ 1, db->u.pgdir->order0); | |
340 | ++o; | |
341 | } | |
342 | i >>= o; | |
343 | set_bit(i, db->u.pgdir->bits[o]); | |
344 | ||
345 | if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) { | |
346 | dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE, | |
347 | db->u.pgdir->db_page, db->u.pgdir->db_dma); | |
348 | list_del(&db->u.pgdir->list); | |
349 | kfree(db->u.pgdir); | |
350 | } | |
351 | ||
352 | mutex_unlock(&priv->pgdir_mutex); | |
353 | } | |
354 | EXPORT_SYMBOL_GPL(mlx4_db_free); | |
38ae6a53 YP |
355 | |
356 | int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres, | |
357 | int size, int max_direct) | |
358 | { | |
359 | int err; | |
360 | ||
361 | err = mlx4_db_alloc(dev, &wqres->db, 1); | |
362 | if (err) | |
363 | return err; | |
364 | ||
365 | *wqres->db.db = 0; | |
366 | ||
367 | err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf); | |
368 | if (err) | |
369 | goto err_db; | |
370 | ||
371 | err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift, | |
372 | &wqres->mtt); | |
373 | if (err) | |
374 | goto err_buf; | |
375 | ||
376 | err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf); | |
377 | if (err) | |
378 | goto err_mtt; | |
379 | ||
380 | return 0; | |
381 | ||
382 | err_mtt: | |
383 | mlx4_mtt_cleanup(dev, &wqres->mtt); | |
384 | err_buf: | |
385 | mlx4_buf_free(dev, size, &wqres->buf); | |
386 | err_db: | |
387 | mlx4_db_free(dev, &wqres->db); | |
388 | ||
389 | return err; | |
390 | } | |
391 | EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res); | |
392 | ||
393 | void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres, | |
394 | int size) | |
395 | { | |
396 | mlx4_mtt_cleanup(dev, &wqres->mtt); | |
397 | mlx4_buf_free(dev, size, &wqres->buf); | |
398 | mlx4_db_free(dev, &wqres->db); | |
399 | } | |
400 | EXPORT_SYMBOL_GPL(mlx4_free_hwq_res); |