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/bitmap.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/vmalloc.h>
40
41 #include "mlx4.h"
42
mlx4_bitmap_alloc(struct mlx4_bitmap * bitmap)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) {
51 bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
52 & bitmap->mask;
53 obj = find_first_zero_bit(bitmap->table, bitmap->max);
54 }
55
56 if (obj < bitmap->max) {
57 set_bit(obj, bitmap->table);
58 bitmap->last = (obj + 1);
59 if (bitmap->last == bitmap->max)
60 bitmap->last = 0;
61 obj |= bitmap->top;
62 } else
63 obj = -1;
64
65 spin_unlock(&bitmap->lock);
66
67 return obj;
68 }
69
mlx4_bitmap_free(struct mlx4_bitmap * bitmap,u32 obj)70 void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj)
71 {
72 mlx4_bitmap_free_range(bitmap, obj, 1);
73 }
74
find_aligned_range(unsigned long * bitmap,u32 start,u32 nbits,int len,int align)75 static unsigned long find_aligned_range(unsigned long *bitmap,
76 u32 start, u32 nbits,
77 int len, int align)
78 {
79 unsigned long end, i;
80
81 again:
82 start = ALIGN(start, align);
83
84 while ((start < nbits) && test_bit(start, bitmap))
85 start += align;
86
87 if (start >= nbits)
88 return -1;
89
90 end = start+len;
91 if (end > nbits)
92 return -1;
93
94 for (i = start + 1; i < end; i++) {
95 if (test_bit(i, bitmap)) {
96 start = i + 1;
97 goto again;
98 }
99 }
100
101 return start;
102 }
103
mlx4_bitmap_alloc_range(struct mlx4_bitmap * bitmap,int cnt,int align)104 u32 mlx4_bitmap_alloc_range(struct mlx4_bitmap *bitmap, int cnt, int align)
105 {
106 u32 obj, i;
107
108 if (likely(cnt == 1 && align == 1))
109 return mlx4_bitmap_alloc(bitmap);
110
111 spin_lock(&bitmap->lock);
112
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)
117 & bitmap->mask;
118 obj = find_aligned_range(bitmap->table, 0, bitmap->max,
119 cnt, align);
120 }
121
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)
128 bitmap->last = 0;
129 }
130 obj |= bitmap->top;
131 } else
132 obj = -1;
133
134 spin_unlock(&bitmap->lock);
135
136 return obj;
137 }
138
mlx4_bitmap_free_range(struct mlx4_bitmap * bitmap,u32 obj,int cnt)139 void mlx4_bitmap_free_range(struct mlx4_bitmap *bitmap, u32 obj, int cnt)
140 {
141 u32 i;
142
143 obj &= bitmap->max + bitmap->reserved_top - 1;
144
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)
150 & bitmap->mask;
151 spin_unlock(&bitmap->lock);
152 }
153
mlx4_bitmap_init(struct mlx4_bitmap * bitmap,u32 num,u32 mask,u32 reserved_bot,u32 reserved_top)154 int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask,
155 u32 reserved_bot, u32 reserved_top)
156 {
157 int i;
158
159 /* num must be a power of 2 */
160 if (num != roundup_pow_of_two(num))
161 return -EINVAL;
162
163 bitmap->last = 0;
164 bitmap->top = 0;
165 bitmap->max = num - reserved_top;
166 bitmap->mask = mask;
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);
171 if (!bitmap->table)
172 return -ENOMEM;
173
174 for (i = 0; i < reserved_bot; ++i)
175 set_bit(i, bitmap->table);
176
177 return 0;
178 }
179
mlx4_bitmap_cleanup(struct mlx4_bitmap * bitmap)180 void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap)
181 {
182 kfree(bitmap->table);
183 }
184
185 /*
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.
190 */
191
mlx4_buf_alloc(struct mlx4_dev * dev,int size,int max_direct,struct mlx4_buf * buf)192 int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
193 struct mlx4_buf *buf)
194 {
195 dma_addr_t t;
196
197 if (size <= max_direct) {
198 buf->nbufs = 1;
199 buf->npages = 1;
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)
204 return -ENOMEM;
205
206 buf->direct.map = t;
207
208 while (t & ((1 << buf->page_shift) - 1)) {
209 --buf->page_shift;
210 buf->npages *= 2;
211 }
212
213 memset(buf->direct.buf, 0, size);
214 } else {
215 int i;
216
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,
221 GFP_KERNEL);
222 if (!buf->page_list)
223 return -ENOMEM;
224
225 for (i = 0; i < buf->nbufs; ++i) {
226 buf->page_list[i].buf =
227 dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
228 &t, GFP_KERNEL);
229 if (!buf->page_list[i].buf)
230 goto err_free;
231
232 buf->page_list[i].map = t;
233
234 memset(buf->page_list[i].buf, 0, PAGE_SIZE);
235 }
236
237 if (BITS_PER_LONG == 64) {
238 struct page **pages;
239 pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL);
240 if (!pages)
241 goto err_free;
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);
245 kfree(pages);
246 if (!buf->direct.buf)
247 goto err_free;
248 }
249 }
250
251 return 0;
252
253 err_free:
254 mlx4_buf_free(dev, size, buf);
255
256 return -ENOMEM;
257 }
258 EXPORT_SYMBOL_GPL(mlx4_buf_alloc);
259
mlx4_buf_free(struct mlx4_dev * dev,int size,struct mlx4_buf * buf)260 void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
261 {
262 int i;
263
264 if (buf->nbufs == 1)
265 dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
266 buf->direct.map);
267 else {
268 if (BITS_PER_LONG == 64)
269 vunmap(buf->direct.buf);
270
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);
277 }
278 }
279 EXPORT_SYMBOL_GPL(mlx4_buf_free);
280
mlx4_alloc_db_pgdir(struct device * dma_device)281 static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device)
282 {
283 struct mlx4_db_pgdir *pgdir;
284
285 pgdir = kzalloc(sizeof *pgdir, GFP_KERNEL);
286 if (!pgdir)
287 return NULL;
288
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) {
295 kfree(pgdir);
296 return NULL;
297 }
298
299 return pgdir;
300 }
301
mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir * pgdir,struct mlx4_db * db,int order)302 static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir,
303 struct mlx4_db *db, int order)
304 {
305 int o;
306 int i;
307
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)
311 goto found;
312 }
313
314 return -ENOMEM;
315
316 found:
317 clear_bit(i, pgdir->bits[o]);
318
319 i <<= o;
320
321 if (o > order)
322 set_bit(i ^ 1, pgdir->bits[order]);
323
324 db->u.pgdir = pgdir;
325 db->index = i;
326 db->db = pgdir->db_page + db->index;
327 db->dma = pgdir->db_dma + db->index * 4;
328 db->order = order;
329
330 return 0;
331 }
332
mlx4_db_alloc(struct mlx4_dev * dev,struct mlx4_db * db,int order)333 int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order)
334 {
335 struct mlx4_priv *priv = mlx4_priv(dev);
336 struct mlx4_db_pgdir *pgdir;
337 int ret = 0;
338
339 mutex_lock(&priv->pgdir_mutex);
340
341 list_for_each_entry(pgdir, &priv->pgdir_list, list)
342 if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
343 goto out;
344
345 pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev));
346 if (!pgdir) {
347 ret = -ENOMEM;
348 goto out;
349 }
350
351 list_add(&pgdir->list, &priv->pgdir_list);
352
353 /* This should never fail -- we just allocated an empty page: */
354 WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order));
355
356 out:
357 mutex_unlock(&priv->pgdir_mutex);
358
359 return ret;
360 }
361 EXPORT_SYMBOL_GPL(mlx4_db_alloc);
362
mlx4_db_free(struct mlx4_dev * dev,struct mlx4_db * db)363 void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
364 {
365 struct mlx4_priv *priv = mlx4_priv(dev);
366 int o;
367 int i;
368
369 mutex_lock(&priv->pgdir_mutex);
370
371 o = db->order;
372 i = db->index;
373
374 if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) {
375 clear_bit(i ^ 1, db->u.pgdir->order0);
376 ++o;
377 }
378 i >>= o;
379 set_bit(i, db->u.pgdir->bits[o]);
380
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);
385 kfree(db->u.pgdir);
386 }
387
388 mutex_unlock(&priv->pgdir_mutex);
389 }
390 EXPORT_SYMBOL_GPL(mlx4_db_free);
391
mlx4_alloc_hwq_res(struct mlx4_dev * dev,struct mlx4_hwq_resources * wqres,int size,int max_direct)392 int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
393 int size, int max_direct)
394 {
395 int err;
396
397 err = mlx4_db_alloc(dev, &wqres->db, 1);
398 if (err)
399 return err;
400
401 *wqres->db.db = 0;
402
403 err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf);
404 if (err)
405 goto err_db;
406
407 err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
408 &wqres->mtt);
409 if (err)
410 goto err_buf;
411
412 err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf);
413 if (err)
414 goto err_mtt;
415
416 return 0;
417
418 err_mtt:
419 mlx4_mtt_cleanup(dev, &wqres->mtt);
420 err_buf:
421 mlx4_buf_free(dev, size, &wqres->buf);
422 err_db:
423 mlx4_db_free(dev, &wqres->db);
424
425 return err;
426 }
427 EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res);
428
mlx4_free_hwq_res(struct mlx4_dev * dev,struct mlx4_hwq_resources * wqres,int size)429 void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
430 int size)
431 {
432 mlx4_mtt_cleanup(dev, &wqres->mtt);
433 mlx4_buf_free(dev, size, &wqres->buf);
434 mlx4_db_free(dev, &wqres->db);
435 }
436 EXPORT_SYMBOL_GPL(mlx4_free_hwq_res);
437