Commit | Line | Data |
---|---|---|
9bc89cd8 DW |
1 | /* |
2 | * xor offload engine api | |
3 | * | |
4 | * Copyright © 2006, Intel Corporation. | |
5 | * | |
6 | * Dan Williams <dan.j.williams@intel.com> | |
7 | * | |
8 | * with architecture considerations by: | |
9 | * Neil Brown <neilb@suse.de> | |
10 | * Jeff Garzik <jeff@garzik.org> | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or modify it | |
13 | * under the terms and conditions of the GNU General Public License, | |
14 | * version 2, as published by the Free Software Foundation. | |
15 | * | |
16 | * This program is distributed in the hope it will be useful, but WITHOUT | |
17 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
18 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
19 | * more details. | |
20 | * | |
21 | * You should have received a copy of the GNU General Public License along with | |
22 | * this program; if not, write to the Free Software Foundation, Inc., | |
23 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
24 | * | |
25 | */ | |
26 | #include <linux/kernel.h> | |
27 | #include <linux/interrupt.h> | |
28 | #include <linux/mm.h> | |
29 | #include <linux/dma-mapping.h> | |
30 | #include <linux/raid/xor.h> | |
31 | #include <linux/async_tx.h> | |
32 | ||
33 | static void | |
34 | do_async_xor(struct dma_async_tx_descriptor *tx, struct dma_device *device, | |
35 | struct dma_chan *chan, struct page *dest, struct page **src_list, | |
36 | unsigned int offset, unsigned int src_cnt, size_t len, | |
37 | enum async_tx_flags flags, struct dma_async_tx_descriptor *depend_tx, | |
38 | dma_async_tx_callback cb_fn, void *cb_param) | |
39 | { | |
40 | dma_addr_t dma_addr; | |
41 | enum dma_data_direction dir; | |
42 | int i; | |
43 | ||
44 | pr_debug("%s: len: %zu\n", __FUNCTION__, len); | |
45 | ||
46 | dir = (flags & ASYNC_TX_ASSUME_COHERENT) ? | |
47 | DMA_NONE : DMA_FROM_DEVICE; | |
48 | ||
49 | dma_addr = dma_map_page(device->dev, dest, offset, len, dir); | |
50 | tx->tx_set_dest(dma_addr, tx, 0); | |
51 | ||
52 | dir = (flags & ASYNC_TX_ASSUME_COHERENT) ? | |
53 | DMA_NONE : DMA_TO_DEVICE; | |
54 | ||
55 | for (i = 0; i < src_cnt; i++) { | |
56 | dma_addr = dma_map_page(device->dev, src_list[i], | |
57 | offset, len, dir); | |
58 | tx->tx_set_src(dma_addr, tx, i); | |
59 | } | |
60 | ||
61 | async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param); | |
62 | } | |
63 | ||
64 | static void | |
65 | do_sync_xor(struct page *dest, struct page **src_list, unsigned int offset, | |
66 | unsigned int src_cnt, size_t len, enum async_tx_flags flags, | |
67 | struct dma_async_tx_descriptor *depend_tx, | |
68 | dma_async_tx_callback cb_fn, void *cb_param) | |
69 | { | |
70 | void *_dest; | |
71 | int i; | |
72 | ||
73 | pr_debug("%s: len: %zu\n", __FUNCTION__, len); | |
74 | ||
75 | /* reuse the 'src_list' array to convert to buffer pointers */ | |
76 | for (i = 0; i < src_cnt; i++) | |
77 | src_list[i] = (struct page *) | |
78 | (page_address(src_list[i]) + offset); | |
79 | ||
80 | /* set destination address */ | |
81 | _dest = page_address(dest) + offset; | |
82 | ||
83 | if (flags & ASYNC_TX_XOR_ZERO_DST) | |
84 | memset(_dest, 0, len); | |
85 | ||
86 | xor_blocks(src_cnt, len, _dest, | |
87 | (void **) src_list); | |
88 | ||
89 | async_tx_sync_epilog(flags, depend_tx, cb_fn, cb_param); | |
90 | } | |
91 | ||
92 | /** | |
93 | * async_xor - attempt to xor a set of blocks with a dma engine. | |
94 | * xor_blocks always uses the dest as a source so the ASYNC_TX_XOR_ZERO_DST | |
95 | * flag must be set to not include dest data in the calculation. The | |
96 | * assumption with dma eninges is that they only use the destination | |
97 | * buffer as a source when it is explicity specified in the source list. | |
98 | * @dest: destination page | |
99 | * @src_list: array of source pages (if the dest is also a source it must be | |
100 | * at index zero). The contents of this array may be overwritten. | |
101 | * @offset: offset in pages to start transaction | |
102 | * @src_cnt: number of source pages | |
103 | * @len: length in bytes | |
104 | * @flags: ASYNC_TX_XOR_ZERO_DST, ASYNC_TX_XOR_DROP_DEST, | |
105 | * ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK | |
106 | * @depend_tx: xor depends on the result of this transaction. | |
107 | * @cb_fn: function to call when the xor completes | |
108 | * @cb_param: parameter to pass to the callback routine | |
109 | */ | |
110 | struct dma_async_tx_descriptor * | |
111 | async_xor(struct page *dest, struct page **src_list, unsigned int offset, | |
112 | int src_cnt, size_t len, enum async_tx_flags flags, | |
113 | struct dma_async_tx_descriptor *depend_tx, | |
114 | dma_async_tx_callback cb_fn, void *cb_param) | |
115 | { | |
116 | struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_XOR); | |
117 | struct dma_device *device = chan ? chan->device : NULL; | |
118 | struct dma_async_tx_descriptor *tx = NULL; | |
119 | dma_async_tx_callback _cb_fn; | |
120 | void *_cb_param; | |
121 | unsigned long local_flags; | |
122 | int xor_src_cnt; | |
123 | int i = 0, src_off = 0, int_en; | |
124 | ||
125 | BUG_ON(src_cnt <= 1); | |
126 | ||
127 | while (src_cnt) { | |
128 | local_flags = flags; | |
129 | if (device) { /* run the xor asynchronously */ | |
130 | xor_src_cnt = min(src_cnt, device->max_xor); | |
131 | /* if we are submitting additional xors | |
132 | * only set the callback on the last transaction | |
133 | */ | |
134 | if (src_cnt > xor_src_cnt) { | |
135 | local_flags &= ~ASYNC_TX_ACK; | |
136 | _cb_fn = NULL; | |
137 | _cb_param = NULL; | |
138 | } else { | |
139 | _cb_fn = cb_fn; | |
140 | _cb_param = cb_param; | |
141 | } | |
142 | ||
143 | int_en = _cb_fn ? 1 : 0; | |
144 | ||
145 | tx = device->device_prep_dma_xor( | |
146 | chan, xor_src_cnt, len, int_en); | |
147 | ||
148 | if (tx) { | |
149 | do_async_xor(tx, device, chan, dest, | |
150 | &src_list[src_off], offset, xor_src_cnt, len, | |
151 | local_flags, depend_tx, _cb_fn, | |
152 | _cb_param); | |
153 | } else /* fall through */ | |
154 | goto xor_sync; | |
155 | } else { /* run the xor synchronously */ | |
156 | xor_sync: | |
157 | /* in the sync case the dest is an implied source | |
158 | * (assumes the dest is at the src_off index) | |
159 | */ | |
160 | if (flags & ASYNC_TX_XOR_DROP_DST) { | |
161 | src_cnt--; | |
162 | src_off++; | |
163 | } | |
164 | ||
165 | /* process up to 'MAX_XOR_BLOCKS' sources */ | |
166 | xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); | |
167 | ||
168 | /* if we are submitting additional xors | |
169 | * only set the callback on the last transaction | |
170 | */ | |
171 | if (src_cnt > xor_src_cnt) { | |
172 | local_flags &= ~ASYNC_TX_ACK; | |
173 | _cb_fn = NULL; | |
174 | _cb_param = NULL; | |
175 | } else { | |
176 | _cb_fn = cb_fn; | |
177 | _cb_param = cb_param; | |
178 | } | |
179 | ||
180 | /* wait for any prerequisite operations */ | |
181 | if (depend_tx) { | |
182 | /* if ack is already set then we cannot be sure | |
183 | * we are referring to the correct operation | |
184 | */ | |
185 | BUG_ON(depend_tx->ack); | |
186 | if (dma_wait_for_async_tx(depend_tx) == | |
187 | DMA_ERROR) | |
188 | panic("%s: DMA_ERROR waiting for " | |
189 | "depend_tx\n", | |
190 | __FUNCTION__); | |
191 | } | |
192 | ||
193 | do_sync_xor(dest, &src_list[src_off], offset, | |
194 | xor_src_cnt, len, local_flags, depend_tx, | |
195 | _cb_fn, _cb_param); | |
196 | } | |
197 | ||
198 | /* the previous tx is hidden from the client, | |
199 | * so ack it | |
200 | */ | |
201 | if (i && depend_tx) | |
202 | async_tx_ack(depend_tx); | |
203 | ||
204 | depend_tx = tx; | |
205 | ||
206 | if (src_cnt > xor_src_cnt) { | |
207 | /* drop completed sources */ | |
208 | src_cnt -= xor_src_cnt; | |
209 | src_off += xor_src_cnt; | |
210 | ||
211 | /* unconditionally preserve the destination */ | |
212 | flags &= ~ASYNC_TX_XOR_ZERO_DST; | |
213 | ||
214 | /* use the intermediate result a source, but remember | |
215 | * it's dropped, because it's implied, in the sync case | |
216 | */ | |
217 | src_list[--src_off] = dest; | |
218 | src_cnt++; | |
219 | flags |= ASYNC_TX_XOR_DROP_DST; | |
220 | } else | |
221 | src_cnt = 0; | |
222 | i++; | |
223 | } | |
224 | ||
225 | return tx; | |
226 | } | |
227 | EXPORT_SYMBOL_GPL(async_xor); | |
228 | ||
229 | static int page_is_zero(struct page *p, unsigned int offset, size_t len) | |
230 | { | |
231 | char *a = page_address(p) + offset; | |
232 | return ((*(u32 *) a) == 0 && | |
233 | memcmp(a, a + 4, len - 4) == 0); | |
234 | } | |
235 | ||
236 | /** | |
237 | * async_xor_zero_sum - attempt a xor parity check with a dma engine. | |
238 | * @dest: destination page used if the xor is performed synchronously | |
239 | * @src_list: array of source pages. The dest page must be listed as a source | |
240 | * at index zero. The contents of this array may be overwritten. | |
241 | * @offset: offset in pages to start transaction | |
242 | * @src_cnt: number of source pages | |
243 | * @len: length in bytes | |
244 | * @result: 0 if sum == 0 else non-zero | |
245 | * @flags: ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK | |
246 | * @depend_tx: xor depends on the result of this transaction. | |
247 | * @cb_fn: function to call when the xor completes | |
248 | * @cb_param: parameter to pass to the callback routine | |
249 | */ | |
250 | struct dma_async_tx_descriptor * | |
251 | async_xor_zero_sum(struct page *dest, struct page **src_list, | |
252 | unsigned int offset, int src_cnt, size_t len, | |
253 | u32 *result, enum async_tx_flags flags, | |
254 | struct dma_async_tx_descriptor *depend_tx, | |
255 | dma_async_tx_callback cb_fn, void *cb_param) | |
256 | { | |
257 | struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_ZERO_SUM); | |
258 | struct dma_device *device = chan ? chan->device : NULL; | |
259 | int int_en = cb_fn ? 1 : 0; | |
260 | struct dma_async_tx_descriptor *tx = device ? | |
261 | device->device_prep_dma_zero_sum(chan, src_cnt, len, result, | |
262 | int_en) : NULL; | |
263 | int i; | |
264 | ||
265 | BUG_ON(src_cnt <= 1); | |
266 | ||
267 | if (tx) { | |
268 | dma_addr_t dma_addr; | |
269 | enum dma_data_direction dir; | |
270 | ||
271 | pr_debug("%s: (async) len: %zu\n", __FUNCTION__, len); | |
272 | ||
273 | dir = (flags & ASYNC_TX_ASSUME_COHERENT) ? | |
274 | DMA_NONE : DMA_TO_DEVICE; | |
275 | ||
276 | for (i = 0; i < src_cnt; i++) { | |
277 | dma_addr = dma_map_page(device->dev, src_list[i], | |
278 | offset, len, dir); | |
279 | tx->tx_set_src(dma_addr, tx, i); | |
280 | } | |
281 | ||
282 | async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param); | |
283 | } else { | |
284 | unsigned long xor_flags = flags; | |
285 | ||
286 | pr_debug("%s: (sync) len: %zu\n", __FUNCTION__, len); | |
287 | ||
288 | xor_flags |= ASYNC_TX_XOR_DROP_DST; | |
289 | xor_flags &= ~ASYNC_TX_ACK; | |
290 | ||
291 | tx = async_xor(dest, src_list, offset, src_cnt, len, xor_flags, | |
292 | depend_tx, NULL, NULL); | |
293 | ||
294 | if (tx) { | |
295 | if (dma_wait_for_async_tx(tx) == DMA_ERROR) | |
296 | panic("%s: DMA_ERROR waiting for tx\n", | |
297 | __FUNCTION__); | |
298 | async_tx_ack(tx); | |
299 | } | |
300 | ||
301 | *result = page_is_zero(dest, offset, len) ? 0 : 1; | |
302 | ||
303 | tx = NULL; | |
304 | ||
305 | async_tx_sync_epilog(flags, depend_tx, cb_fn, cb_param); | |
306 | } | |
307 | ||
308 | return tx; | |
309 | } | |
310 | EXPORT_SYMBOL_GPL(async_xor_zero_sum); | |
311 | ||
312 | static int __init async_xor_init(void) | |
313 | { | |
314 | return 0; | |
315 | } | |
316 | ||
317 | static void __exit async_xor_exit(void) | |
318 | { | |
319 | do { } while (0); | |
320 | } | |
321 | ||
322 | module_init(async_xor_init); | |
323 | module_exit(async_xor_exit); | |
324 | ||
325 | MODULE_AUTHOR("Intel Corporation"); | |
326 | MODULE_DESCRIPTION("asynchronous xor/xor-zero-sum api"); | |
327 | MODULE_LICENSE("GPL"); |