1 /*
2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32
33 #include <linux/errno.h>
34 #include <linux/slab.h>
35 #include <linux/mm.h>
36 #include <linux/export.h>
37 #include <linux/bitmap.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mlx5/driver.h>
41
42 #include "mlx5_core.h"
43
44 struct mlx5_db_pgdir {
45 struct list_head list;
46 unsigned long *bitmap;
47 __be32 *db_page;
48 dma_addr_t db_dma;
49 };
50
51 /* Handling for queue buffers -- we allocate a bunch of memory and
52 * register it in a memory region at HCA virtual address 0.
53 */
54
mlx5_dma_zalloc_coherent_node(struct mlx5_core_dev * dev,size_t size,dma_addr_t * dma_handle,int node)55 static void *mlx5_dma_zalloc_coherent_node(struct mlx5_core_dev *dev,
56 size_t size, dma_addr_t *dma_handle,
57 int node)
58 {
59 struct mlx5_priv *priv = &dev->priv;
60 int original_node;
61 void *cpu_handle;
62
63 mutex_lock(&priv->alloc_mutex);
64 original_node = dev_to_node(&dev->pdev->dev);
65 set_dev_node(&dev->pdev->dev, node);
66 cpu_handle = dma_zalloc_coherent(&dev->pdev->dev, size,
67 dma_handle, GFP_KERNEL);
68 set_dev_node(&dev->pdev->dev, original_node);
69 mutex_unlock(&priv->alloc_mutex);
70 return cpu_handle;
71 }
72
mlx5_buf_alloc_node(struct mlx5_core_dev * dev,int size,struct mlx5_buf * buf,int node)73 int mlx5_buf_alloc_node(struct mlx5_core_dev *dev, int size,
74 struct mlx5_buf *buf, int node)
75 {
76 dma_addr_t t;
77
78 buf->size = size;
79 buf->npages = 1;
80 buf->page_shift = (u8)get_order(size) + PAGE_SHIFT;
81 buf->direct.buf = mlx5_dma_zalloc_coherent_node(dev, size,
82 &t, node);
83 if (!buf->direct.buf)
84 return -ENOMEM;
85
86 buf->direct.map = t;
87
88 while (t & ((1 << buf->page_shift) - 1)) {
89 --buf->page_shift;
90 buf->npages *= 2;
91 }
92
93 return 0;
94 }
95
mlx5_buf_alloc(struct mlx5_core_dev * dev,int size,struct mlx5_buf * buf)96 int mlx5_buf_alloc(struct mlx5_core_dev *dev, int size, struct mlx5_buf *buf)
97 {
98 return mlx5_buf_alloc_node(dev, size, buf, dev->priv.numa_node);
99 }
100 EXPORT_SYMBOL_GPL(mlx5_buf_alloc);
101
mlx5_buf_free(struct mlx5_core_dev * dev,struct mlx5_buf * buf)102 void mlx5_buf_free(struct mlx5_core_dev *dev, struct mlx5_buf *buf)
103 {
104 dma_free_coherent(&dev->pdev->dev, buf->size, buf->direct.buf,
105 buf->direct.map);
106 }
107 EXPORT_SYMBOL_GPL(mlx5_buf_free);
108
mlx5_frag_buf_alloc_node(struct mlx5_core_dev * dev,int size,struct mlx5_frag_buf * buf,int node)109 int mlx5_frag_buf_alloc_node(struct mlx5_core_dev *dev, int size,
110 struct mlx5_frag_buf *buf, int node)
111 {
112 int i;
113
114 buf->size = size;
115 buf->npages = 1 << get_order(size);
116 buf->page_shift = PAGE_SHIFT;
117 buf->frags = kcalloc(buf->npages, sizeof(struct mlx5_buf_list),
118 GFP_KERNEL);
119 if (!buf->frags)
120 goto err_out;
121
122 for (i = 0; i < buf->npages; i++) {
123 struct mlx5_buf_list *frag = &buf->frags[i];
124 int frag_sz = min_t(int, size, PAGE_SIZE);
125
126 frag->buf = mlx5_dma_zalloc_coherent_node(dev, frag_sz,
127 &frag->map, node);
128 if (!frag->buf)
129 goto err_free_buf;
130 if (frag->map & ((1 << buf->page_shift) - 1)) {
131 dma_free_coherent(&dev->pdev->dev, frag_sz,
132 buf->frags[i].buf, buf->frags[i].map);
133 mlx5_core_warn(dev, "unexpected map alignment: %pad, page_shift=%d\n",
134 &frag->map, buf->page_shift);
135 goto err_free_buf;
136 }
137 size -= frag_sz;
138 }
139
140 return 0;
141
142 err_free_buf:
143 while (i--)
144 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE, buf->frags[i].buf,
145 buf->frags[i].map);
146 kfree(buf->frags);
147 err_out:
148 return -ENOMEM;
149 }
150
mlx5_frag_buf_free(struct mlx5_core_dev * dev,struct mlx5_frag_buf * buf)151 void mlx5_frag_buf_free(struct mlx5_core_dev *dev, struct mlx5_frag_buf *buf)
152 {
153 int size = buf->size;
154 int i;
155
156 for (i = 0; i < buf->npages; i++) {
157 int frag_sz = min_t(int, size, PAGE_SIZE);
158
159 dma_free_coherent(&dev->pdev->dev, frag_sz, buf->frags[i].buf,
160 buf->frags[i].map);
161 size -= frag_sz;
162 }
163 kfree(buf->frags);
164 }
165
mlx5_alloc_db_pgdir(struct mlx5_core_dev * dev,int node)166 static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev,
167 int node)
168 {
169 u32 db_per_page = PAGE_SIZE / cache_line_size();
170 struct mlx5_db_pgdir *pgdir;
171
172 pgdir = kzalloc(sizeof(*pgdir), GFP_KERNEL);
173 if (!pgdir)
174 return NULL;
175
176 pgdir->bitmap = kcalloc(BITS_TO_LONGS(db_per_page),
177 sizeof(unsigned long),
178 GFP_KERNEL);
179
180 if (!pgdir->bitmap) {
181 kfree(pgdir);
182 return NULL;
183 }
184
185 bitmap_fill(pgdir->bitmap, db_per_page);
186
187 pgdir->db_page = mlx5_dma_zalloc_coherent_node(dev, PAGE_SIZE,
188 &pgdir->db_dma, node);
189 if (!pgdir->db_page) {
190 kfree(pgdir->bitmap);
191 kfree(pgdir);
192 return NULL;
193 }
194
195 return pgdir;
196 }
197
mlx5_alloc_db_from_pgdir(struct mlx5_db_pgdir * pgdir,struct mlx5_db * db)198 static int mlx5_alloc_db_from_pgdir(struct mlx5_db_pgdir *pgdir,
199 struct mlx5_db *db)
200 {
201 u32 db_per_page = PAGE_SIZE / cache_line_size();
202 int offset;
203 int i;
204
205 i = find_first_bit(pgdir->bitmap, db_per_page);
206 if (i >= db_per_page)
207 return -ENOMEM;
208
209 __clear_bit(i, pgdir->bitmap);
210
211 db->u.pgdir = pgdir;
212 db->index = i;
213 offset = db->index * cache_line_size();
214 db->db = pgdir->db_page + offset / sizeof(*pgdir->db_page);
215 db->dma = pgdir->db_dma + offset;
216
217 db->db[0] = 0;
218 db->db[1] = 0;
219
220 return 0;
221 }
222
mlx5_db_alloc_node(struct mlx5_core_dev * dev,struct mlx5_db * db,int node)223 int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db, int node)
224 {
225 struct mlx5_db_pgdir *pgdir;
226 int ret = 0;
227
228 mutex_lock(&dev->priv.pgdir_mutex);
229
230 list_for_each_entry(pgdir, &dev->priv.pgdir_list, list)
231 if (!mlx5_alloc_db_from_pgdir(pgdir, db))
232 goto out;
233
234 pgdir = mlx5_alloc_db_pgdir(dev, node);
235 if (!pgdir) {
236 ret = -ENOMEM;
237 goto out;
238 }
239
240 list_add(&pgdir->list, &dev->priv.pgdir_list);
241
242 /* This should never fail -- we just allocated an empty page: */
243 WARN_ON(mlx5_alloc_db_from_pgdir(pgdir, db));
244
245 out:
246 mutex_unlock(&dev->priv.pgdir_mutex);
247
248 return ret;
249 }
250 EXPORT_SYMBOL_GPL(mlx5_db_alloc_node);
251
mlx5_db_alloc(struct mlx5_core_dev * dev,struct mlx5_db * db)252 int mlx5_db_alloc(struct mlx5_core_dev *dev, struct mlx5_db *db)
253 {
254 return mlx5_db_alloc_node(dev, db, dev->priv.numa_node);
255 }
256 EXPORT_SYMBOL_GPL(mlx5_db_alloc);
257
mlx5_db_free(struct mlx5_core_dev * dev,struct mlx5_db * db)258 void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db)
259 {
260 u32 db_per_page = PAGE_SIZE / cache_line_size();
261
262 mutex_lock(&dev->priv.pgdir_mutex);
263
264 __set_bit(db->index, db->u.pgdir->bitmap);
265
266 if (bitmap_full(db->u.pgdir->bitmap, db_per_page)) {
267 dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
268 db->u.pgdir->db_page, db->u.pgdir->db_dma);
269 list_del(&db->u.pgdir->list);
270 kfree(db->u.pgdir->bitmap);
271 kfree(db->u.pgdir);
272 }
273
274 mutex_unlock(&dev->priv.pgdir_mutex);
275 }
276 EXPORT_SYMBOL_GPL(mlx5_db_free);
277
mlx5_fill_page_array(struct mlx5_buf * buf,__be64 * pas)278 void mlx5_fill_page_array(struct mlx5_buf *buf, __be64 *pas)
279 {
280 u64 addr;
281 int i;
282
283 for (i = 0; i < buf->npages; i++) {
284 addr = buf->direct.map + (i << buf->page_shift);
285
286 pas[i] = cpu_to_be64(addr);
287 }
288 }
289 EXPORT_SYMBOL_GPL(mlx5_fill_page_array);
290
mlx5_fill_page_frag_array(struct mlx5_frag_buf * buf,__be64 * pas)291 void mlx5_fill_page_frag_array(struct mlx5_frag_buf *buf, __be64 *pas)
292 {
293 int i;
294
295 for (i = 0; i < buf->npages; i++)
296 pas[i] = cpu_to_be64(buf->frags[i].map);
297 }
298 EXPORT_SYMBOL_GPL(mlx5_fill_page_frag_array);
299