Commit | Line | Data |
---|---|---|
225c7b1f | 1 | /* |
51a379d0 | 2 | * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved. |
225c7b1f RD |
3 | * Copyright (c) 2006, 2007 Cisco Systems, Inc. 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 | ||
225c7b1f | 34 | #include <linux/errno.h> |
9cbe05c7 | 35 | #include <linux/mm.h> |
5b0bf5e2 | 36 | #include <linux/scatterlist.h> |
5a0e3ad6 | 37 | #include <linux/slab.h> |
225c7b1f RD |
38 | |
39 | #include <linux/mlx4/cmd.h> | |
40 | ||
41 | #include "mlx4.h" | |
42 | #include "icm.h" | |
43 | #include "fw.h" | |
44 | ||
45 | /* | |
46 | * We allocate in as big chunks as we can, up to a maximum of 256 KB | |
47 | * per chunk. | |
48 | */ | |
49 | enum { | |
50 | MLX4_ICM_ALLOC_SIZE = 1 << 18, | |
51 | MLX4_TABLE_CHUNK_SIZE = 1 << 18 | |
52 | }; | |
53 | ||
5b0bf5e2 | 54 | static void mlx4_free_icm_pages(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk) |
225c7b1f | 55 | { |
225c7b1f RD |
56 | int i; |
57 | ||
5b0bf5e2 JM |
58 | if (chunk->nsg > 0) |
59 | pci_unmap_sg(dev->pdev, chunk->mem, chunk->npages, | |
60 | PCI_DMA_BIDIRECTIONAL); | |
61 | ||
62 | for (i = 0; i < chunk->npages; ++i) | |
45711f1a | 63 | __free_pages(sg_page(&chunk->mem[i]), |
5b0bf5e2 JM |
64 | get_order(chunk->mem[i].length)); |
65 | } | |
225c7b1f | 66 | |
5b0bf5e2 JM |
67 | static void mlx4_free_icm_coherent(struct mlx4_dev *dev, struct mlx4_icm_chunk *chunk) |
68 | { | |
69 | int i; | |
70 | ||
71 | for (i = 0; i < chunk->npages; ++i) | |
72 | dma_free_coherent(&dev->pdev->dev, chunk->mem[i].length, | |
45711f1a | 73 | lowmem_page_address(sg_page(&chunk->mem[i])), |
5b0bf5e2 JM |
74 | sg_dma_address(&chunk->mem[i])); |
75 | } | |
76 | ||
77 | void mlx4_free_icm(struct mlx4_dev *dev, struct mlx4_icm *icm, int coherent) | |
78 | { | |
79 | struct mlx4_icm_chunk *chunk, *tmp; | |
80 | ||
81 | if (!icm) | |
82 | return; | |
83 | ||
84 | list_for_each_entry_safe(chunk, tmp, &icm->chunk_list, list) { | |
85 | if (coherent) | |
86 | mlx4_free_icm_coherent(dev, chunk); | |
87 | else | |
88 | mlx4_free_icm_pages(dev, chunk); | |
225c7b1f RD |
89 | |
90 | kfree(chunk); | |
91 | } | |
92 | ||
93 | kfree(icm); | |
94 | } | |
95 | ||
5b0bf5e2 JM |
96 | static int mlx4_alloc_icm_pages(struct scatterlist *mem, int order, gfp_t gfp_mask) |
97 | { | |
45711f1a JA |
98 | struct page *page; |
99 | ||
100 | page = alloc_pages(gfp_mask, order); | |
101 | if (!page) | |
5b0bf5e2 JM |
102 | return -ENOMEM; |
103 | ||
642f1490 | 104 | sg_set_page(mem, page, PAGE_SIZE << order, 0); |
5b0bf5e2 JM |
105 | return 0; |
106 | } | |
107 | ||
108 | static int mlx4_alloc_icm_coherent(struct device *dev, struct scatterlist *mem, | |
109 | int order, gfp_t gfp_mask) | |
110 | { | |
111 | void *buf = dma_alloc_coherent(dev, PAGE_SIZE << order, | |
112 | &sg_dma_address(mem), gfp_mask); | |
113 | if (!buf) | |
114 | return -ENOMEM; | |
115 | ||
116 | sg_set_buf(mem, buf, PAGE_SIZE << order); | |
117 | BUG_ON(mem->offset); | |
118 | sg_dma_len(mem) = PAGE_SIZE << order; | |
119 | return 0; | |
120 | } | |
121 | ||
225c7b1f | 122 | struct mlx4_icm *mlx4_alloc_icm(struct mlx4_dev *dev, int npages, |
5b0bf5e2 | 123 | gfp_t gfp_mask, int coherent) |
225c7b1f RD |
124 | { |
125 | struct mlx4_icm *icm; | |
126 | struct mlx4_icm_chunk *chunk = NULL; | |
127 | int cur_order; | |
5b0bf5e2 JM |
128 | int ret; |
129 | ||
130 | /* We use sg_set_buf for coherent allocs, which assumes low memory */ | |
131 | BUG_ON(coherent && (gfp_mask & __GFP_HIGHMEM)); | |
225c7b1f RD |
132 | |
133 | icm = kmalloc(sizeof *icm, gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN)); | |
134 | if (!icm) | |
5b0bf5e2 | 135 | return NULL; |
225c7b1f RD |
136 | |
137 | icm->refcount = 0; | |
138 | INIT_LIST_HEAD(&icm->chunk_list); | |
139 | ||
140 | cur_order = get_order(MLX4_ICM_ALLOC_SIZE); | |
141 | ||
142 | while (npages > 0) { | |
143 | if (!chunk) { | |
144 | chunk = kmalloc(sizeof *chunk, | |
145 | gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN)); | |
146 | if (!chunk) | |
147 | goto fail; | |
148 | ||
45711f1a | 149 | sg_init_table(chunk->mem, MLX4_ICM_CHUNK_LEN); |
225c7b1f RD |
150 | chunk->npages = 0; |
151 | chunk->nsg = 0; | |
152 | list_add_tail(&chunk->list, &icm->chunk_list); | |
153 | } | |
154 | ||
155 | while (1 << cur_order > npages) | |
156 | --cur_order; | |
157 | ||
5b0bf5e2 JM |
158 | if (coherent) |
159 | ret = mlx4_alloc_icm_coherent(&dev->pdev->dev, | |
160 | &chunk->mem[chunk->npages], | |
161 | cur_order, gfp_mask); | |
162 | else | |
163 | ret = mlx4_alloc_icm_pages(&chunk->mem[chunk->npages], | |
164 | cur_order, gfp_mask); | |
165 | ||
c050def0 RD |
166 | if (ret) { |
167 | if (--cur_order < 0) | |
168 | goto fail; | |
169 | else | |
170 | continue; | |
171 | } | |
225c7b1f | 172 | |
c050def0 | 173 | ++chunk->npages; |
225c7b1f | 174 | |
c050def0 RD |
175 | if (coherent) |
176 | ++chunk->nsg; | |
177 | else if (chunk->npages == MLX4_ICM_CHUNK_LEN) { | |
178 | chunk->nsg = pci_map_sg(dev->pdev, chunk->mem, | |
179 | chunk->npages, | |
180 | PCI_DMA_BIDIRECTIONAL); | |
225c7b1f | 181 | |
c050def0 | 182 | if (chunk->nsg <= 0) |
225c7b1f RD |
183 | goto fail; |
184 | } | |
c050def0 RD |
185 | |
186 | if (chunk->npages == MLX4_ICM_CHUNK_LEN) | |
187 | chunk = NULL; | |
188 | ||
189 | npages -= 1 << cur_order; | |
225c7b1f RD |
190 | } |
191 | ||
5b0bf5e2 | 192 | if (!coherent && chunk) { |
225c7b1f RD |
193 | chunk->nsg = pci_map_sg(dev->pdev, chunk->mem, |
194 | chunk->npages, | |
195 | PCI_DMA_BIDIRECTIONAL); | |
196 | ||
197 | if (chunk->nsg <= 0) | |
198 | goto fail; | |
199 | } | |
200 | ||
201 | return icm; | |
202 | ||
203 | fail: | |
5b0bf5e2 | 204 | mlx4_free_icm(dev, icm, coherent); |
225c7b1f RD |
205 | return NULL; |
206 | } | |
207 | ||
208 | static int mlx4_MAP_ICM(struct mlx4_dev *dev, struct mlx4_icm *icm, u64 virt) | |
209 | { | |
210 | return mlx4_map_cmd(dev, MLX4_CMD_MAP_ICM, icm, virt); | |
211 | } | |
212 | ||
9740d786 | 213 | static int mlx4_UNMAP_ICM(struct mlx4_dev *dev, u64 virt, u32 page_count) |
225c7b1f RD |
214 | { |
215 | return mlx4_cmd(dev, virt, page_count, 0, MLX4_CMD_UNMAP_ICM, | |
f9baff50 | 216 | MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE); |
225c7b1f RD |
217 | } |
218 | ||
225c7b1f RD |
219 | int mlx4_MAP_ICM_AUX(struct mlx4_dev *dev, struct mlx4_icm *icm) |
220 | { | |
221 | return mlx4_map_cmd(dev, MLX4_CMD_MAP_ICM_AUX, icm, -1); | |
222 | } | |
223 | ||
224 | int mlx4_UNMAP_ICM_AUX(struct mlx4_dev *dev) | |
225 | { | |
f9baff50 JM |
226 | return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_UNMAP_ICM_AUX, |
227 | MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE); | |
225c7b1f RD |
228 | } |
229 | ||
230 | int mlx4_table_get(struct mlx4_dev *dev, struct mlx4_icm_table *table, int obj) | |
231 | { | |
232 | int i = (obj & (table->num_obj - 1)) / (MLX4_TABLE_CHUNK_SIZE / table->obj_size); | |
233 | int ret = 0; | |
234 | ||
235 | mutex_lock(&table->mutex); | |
236 | ||
237 | if (table->icm[i]) { | |
238 | ++table->icm[i]->refcount; | |
239 | goto out; | |
240 | } | |
241 | ||
242 | table->icm[i] = mlx4_alloc_icm(dev, MLX4_TABLE_CHUNK_SIZE >> PAGE_SHIFT, | |
243 | (table->lowmem ? GFP_KERNEL : GFP_HIGHUSER) | | |
5b0bf5e2 | 244 | __GFP_NOWARN, table->coherent); |
225c7b1f RD |
245 | if (!table->icm[i]) { |
246 | ret = -ENOMEM; | |
247 | goto out; | |
248 | } | |
249 | ||
250 | if (mlx4_MAP_ICM(dev, table->icm[i], table->virt + | |
251 | (u64) i * MLX4_TABLE_CHUNK_SIZE)) { | |
5b0bf5e2 | 252 | mlx4_free_icm(dev, table->icm[i], table->coherent); |
225c7b1f RD |
253 | table->icm[i] = NULL; |
254 | ret = -ENOMEM; | |
255 | goto out; | |
256 | } | |
257 | ||
258 | ++table->icm[i]->refcount; | |
259 | ||
260 | out: | |
261 | mutex_unlock(&table->mutex); | |
262 | return ret; | |
263 | } | |
264 | ||
265 | void mlx4_table_put(struct mlx4_dev *dev, struct mlx4_icm_table *table, int obj) | |
266 | { | |
267 | int i; | |
268 | ||
269 | i = (obj & (table->num_obj - 1)) / (MLX4_TABLE_CHUNK_SIZE / table->obj_size); | |
270 | ||
271 | mutex_lock(&table->mutex); | |
272 | ||
273 | if (--table->icm[i]->refcount == 0) { | |
274 | mlx4_UNMAP_ICM(dev, table->virt + i * MLX4_TABLE_CHUNK_SIZE, | |
275 | MLX4_TABLE_CHUNK_SIZE / MLX4_ICM_PAGE_SIZE); | |
5b0bf5e2 | 276 | mlx4_free_icm(dev, table->icm[i], table->coherent); |
225c7b1f RD |
277 | table->icm[i] = NULL; |
278 | } | |
279 | ||
280 | mutex_unlock(&table->mutex); | |
281 | } | |
282 | ||
d7bb58fb | 283 | void *mlx4_table_find(struct mlx4_icm_table *table, int obj, dma_addr_t *dma_handle) |
225c7b1f | 284 | { |
d7bb58fb | 285 | int idx, offset, dma_offset, i; |
225c7b1f RD |
286 | struct mlx4_icm_chunk *chunk; |
287 | struct mlx4_icm *icm; | |
288 | struct page *page = NULL; | |
289 | ||
290 | if (!table->lowmem) | |
291 | return NULL; | |
292 | ||
293 | mutex_lock(&table->mutex); | |
294 | ||
d7bb58fb JM |
295 | idx = (obj & (table->num_obj - 1)) * table->obj_size; |
296 | icm = table->icm[idx / MLX4_TABLE_CHUNK_SIZE]; | |
297 | dma_offset = offset = idx % MLX4_TABLE_CHUNK_SIZE; | |
225c7b1f RD |
298 | |
299 | if (!icm) | |
300 | goto out; | |
301 | ||
302 | list_for_each_entry(chunk, &icm->chunk_list, list) { | |
303 | for (i = 0; i < chunk->npages; ++i) { | |
d7bb58fb JM |
304 | if (dma_handle && dma_offset >= 0) { |
305 | if (sg_dma_len(&chunk->mem[i]) > dma_offset) | |
306 | *dma_handle = sg_dma_address(&chunk->mem[i]) + | |
307 | dma_offset; | |
308 | dma_offset -= sg_dma_len(&chunk->mem[i]); | |
309 | } | |
310 | /* | |
311 | * DMA mapping can merge pages but not split them, | |
312 | * so if we found the page, dma_handle has already | |
313 | * been assigned to. | |
314 | */ | |
225c7b1f | 315 | if (chunk->mem[i].length > offset) { |
45711f1a | 316 | page = sg_page(&chunk->mem[i]); |
225c7b1f RD |
317 | goto out; |
318 | } | |
319 | offset -= chunk->mem[i].length; | |
320 | } | |
321 | } | |
322 | ||
323 | out: | |
324 | mutex_unlock(&table->mutex); | |
325 | return page ? lowmem_page_address(page) + offset : NULL; | |
326 | } | |
327 | ||
328 | int mlx4_table_get_range(struct mlx4_dev *dev, struct mlx4_icm_table *table, | |
329 | int start, int end) | |
330 | { | |
331 | int inc = MLX4_TABLE_CHUNK_SIZE / table->obj_size; | |
332 | int i, err; | |
333 | ||
334 | for (i = start; i <= end; i += inc) { | |
335 | err = mlx4_table_get(dev, table, i); | |
336 | if (err) | |
337 | goto fail; | |
338 | } | |
339 | ||
340 | return 0; | |
341 | ||
342 | fail: | |
343 | while (i > start) { | |
344 | i -= inc; | |
345 | mlx4_table_put(dev, table, i); | |
346 | } | |
347 | ||
348 | return err; | |
349 | } | |
350 | ||
351 | void mlx4_table_put_range(struct mlx4_dev *dev, struct mlx4_icm_table *table, | |
352 | int start, int end) | |
353 | { | |
354 | int i; | |
355 | ||
356 | for (i = start; i <= end; i += MLX4_TABLE_CHUNK_SIZE / table->obj_size) | |
357 | mlx4_table_put(dev, table, i); | |
358 | } | |
359 | ||
360 | int mlx4_init_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table, | |
361 | u64 virt, int obj_size, int nobj, int reserved, | |
5b0bf5e2 | 362 | int use_lowmem, int use_coherent) |
225c7b1f RD |
363 | { |
364 | int obj_per_chunk; | |
365 | int num_icm; | |
366 | unsigned chunk_size; | |
367 | int i; | |
368 | ||
369 | obj_per_chunk = MLX4_TABLE_CHUNK_SIZE / obj_size; | |
370 | num_icm = (nobj + obj_per_chunk - 1) / obj_per_chunk; | |
371 | ||
372 | table->icm = kcalloc(num_icm, sizeof *table->icm, GFP_KERNEL); | |
373 | if (!table->icm) | |
374 | return -ENOMEM; | |
375 | table->virt = virt; | |
376 | table->num_icm = num_icm; | |
377 | table->num_obj = nobj; | |
378 | table->obj_size = obj_size; | |
379 | table->lowmem = use_lowmem; | |
5b0bf5e2 | 380 | table->coherent = use_coherent; |
225c7b1f RD |
381 | mutex_init(&table->mutex); |
382 | ||
383 | for (i = 0; i * MLX4_TABLE_CHUNK_SIZE < reserved * obj_size; ++i) { | |
384 | chunk_size = MLX4_TABLE_CHUNK_SIZE; | |
385 | if ((i + 1) * MLX4_TABLE_CHUNK_SIZE > nobj * obj_size) | |
386 | chunk_size = PAGE_ALIGN(nobj * obj_size - i * MLX4_TABLE_CHUNK_SIZE); | |
387 | ||
388 | table->icm[i] = mlx4_alloc_icm(dev, chunk_size >> PAGE_SHIFT, | |
389 | (use_lowmem ? GFP_KERNEL : GFP_HIGHUSER) | | |
5b0bf5e2 | 390 | __GFP_NOWARN, use_coherent); |
225c7b1f RD |
391 | if (!table->icm[i]) |
392 | goto err; | |
393 | if (mlx4_MAP_ICM(dev, table->icm[i], virt + i * MLX4_TABLE_CHUNK_SIZE)) { | |
5b0bf5e2 | 394 | mlx4_free_icm(dev, table->icm[i], use_coherent); |
225c7b1f RD |
395 | table->icm[i] = NULL; |
396 | goto err; | |
397 | } | |
398 | ||
399 | /* | |
400 | * Add a reference to this ICM chunk so that it never | |
401 | * gets freed (since it contains reserved firmware objects). | |
402 | */ | |
403 | ++table->icm[i]->refcount; | |
404 | } | |
405 | ||
406 | return 0; | |
407 | ||
408 | err: | |
409 | for (i = 0; i < num_icm; ++i) | |
410 | if (table->icm[i]) { | |
411 | mlx4_UNMAP_ICM(dev, virt + i * MLX4_TABLE_CHUNK_SIZE, | |
412 | MLX4_TABLE_CHUNK_SIZE / MLX4_ICM_PAGE_SIZE); | |
5b0bf5e2 | 413 | mlx4_free_icm(dev, table->icm[i], use_coherent); |
225c7b1f RD |
414 | } |
415 | ||
416 | return -ENOMEM; | |
417 | } | |
418 | ||
419 | void mlx4_cleanup_icm_table(struct mlx4_dev *dev, struct mlx4_icm_table *table) | |
420 | { | |
421 | int i; | |
422 | ||
423 | for (i = 0; i < table->num_icm; ++i) | |
424 | if (table->icm[i]) { | |
425 | mlx4_UNMAP_ICM(dev, table->virt + i * MLX4_TABLE_CHUNK_SIZE, | |
426 | MLX4_TABLE_CHUNK_SIZE / MLX4_ICM_PAGE_SIZE); | |
5b0bf5e2 | 427 | mlx4_free_icm(dev, table->icm[i], table->coherent); |
225c7b1f RD |
428 | } |
429 | ||
430 | kfree(table->icm); | |
431 | } |