async_tx: kill ASYNC_TX_DEP_ACK flag
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / crypto / async_tx / async_xor.c
blob3cc5dc763b545d0580262eb508f838f30c80120a
1 /*
2 * xor offload engine api
4 * Copyright © 2006, Intel Corporation.
6 * Dan Williams <dan.j.williams@intel.com>
8 * with architecture considerations by:
9 * Neil Brown <neilb@suse.de>
10 * Jeff Garzik <jeff@garzik.org>
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.
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.
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.
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>
33 /* do_async_xor - dma map the pages and perform the xor with an engine */
34 static __async_inline struct dma_async_tx_descriptor *
35 do_async_xor(struct dma_chan *chan, struct page *dest, struct page **src_list,
36 unsigned int offset, int src_cnt, size_t len,
37 enum async_tx_flags flags,
38 struct dma_async_tx_descriptor *depend_tx,
39 dma_async_tx_callback cb_fn, void *cb_param)
41 struct dma_device *dma = chan->device;
42 dma_addr_t *dma_src = (dma_addr_t *) src_list;
43 struct dma_async_tx_descriptor *tx = NULL;
44 int src_off = 0;
45 int i;
46 dma_async_tx_callback _cb_fn;
47 void *_cb_param;
48 enum async_tx_flags async_flags;
49 enum dma_ctrl_flags dma_flags;
50 int xor_src_cnt;
51 dma_addr_t dma_dest;
53 /* map the dest bidrectional in case it is re-used as a source */
54 dma_dest = dma_map_page(dma->dev, dest, offset, len, DMA_BIDIRECTIONAL);
55 for (i = 0; i < src_cnt; i++) {
56 /* only map the dest once */
57 if (unlikely(src_list[i] == dest)) {
58 dma_src[i] = dma_dest;
59 continue;
61 dma_src[i] = dma_map_page(dma->dev, src_list[i], offset,
62 len, DMA_TO_DEVICE);
65 while (src_cnt) {
66 async_flags = flags;
67 dma_flags = 0;
68 xor_src_cnt = min(src_cnt, dma->max_xor);
69 /* if we are submitting additional xors, leave the chain open,
70 * clear the callback parameters, and leave the destination
71 * buffer mapped
73 if (src_cnt > xor_src_cnt) {
74 async_flags &= ~ASYNC_TX_ACK;
75 dma_flags = DMA_COMPL_SKIP_DEST_UNMAP;
76 _cb_fn = NULL;
77 _cb_param = NULL;
78 } else {
79 _cb_fn = cb_fn;
80 _cb_param = cb_param;
82 if (_cb_fn)
83 dma_flags |= DMA_PREP_INTERRUPT;
85 /* Since we have clobbered the src_list we are committed
86 * to doing this asynchronously. Drivers force forward progress
87 * in case they can not provide a descriptor
89 tx = dma->device_prep_dma_xor(chan, dma_dest, &dma_src[src_off],
90 xor_src_cnt, len, dma_flags);
92 if (unlikely(!tx))
93 async_tx_quiesce(&depend_tx);
95 /* spin wait for the preceeding transactions to complete */
96 while (unlikely(!tx)) {
97 dma_async_issue_pending(chan);
98 tx = dma->device_prep_dma_xor(chan, dma_dest,
99 &dma_src[src_off],
100 xor_src_cnt, len,
101 dma_flags);
104 async_tx_submit(chan, tx, async_flags, depend_tx, _cb_fn,
105 _cb_param);
107 depend_tx = tx;
109 if (src_cnt > xor_src_cnt) {
110 /* drop completed sources */
111 src_cnt -= xor_src_cnt;
112 src_off += xor_src_cnt;
114 /* use the intermediate result a source */
115 dma_src[--src_off] = dma_dest;
116 src_cnt++;
117 } else
118 break;
121 return tx;
124 static void
125 do_sync_xor(struct page *dest, struct page **src_list, unsigned int offset,
126 int src_cnt, size_t len, enum async_tx_flags flags,
127 dma_async_tx_callback cb_fn, void *cb_param)
129 int i;
130 int xor_src_cnt;
131 int src_off = 0;
132 void *dest_buf;
133 void **srcs = (void **) src_list;
135 /* reuse the 'src_list' array to convert to buffer pointers */
136 for (i = 0; i < src_cnt; i++)
137 srcs[i] = page_address(src_list[i]) + offset;
139 /* set destination address */
140 dest_buf = page_address(dest) + offset;
142 if (flags & ASYNC_TX_XOR_ZERO_DST)
143 memset(dest_buf, 0, len);
145 while (src_cnt > 0) {
146 /* process up to 'MAX_XOR_BLOCKS' sources */
147 xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
148 xor_blocks(xor_src_cnt, len, dest_buf, &srcs[src_off]);
150 /* drop completed sources */
151 src_cnt -= xor_src_cnt;
152 src_off += xor_src_cnt;
155 async_tx_sync_epilog(cb_fn, cb_param);
159 * async_xor - attempt to xor a set of blocks with a dma engine.
160 * xor_blocks always uses the dest as a source so the ASYNC_TX_XOR_ZERO_DST
161 * flag must be set to not include dest data in the calculation. The
162 * assumption with dma eninges is that they only use the destination
163 * buffer as a source when it is explicity specified in the source list.
164 * @dest: destination page
165 * @src_list: array of source pages (if the dest is also a source it must be
166 * at index zero). The contents of this array may be overwritten.
167 * @offset: offset in pages to start transaction
168 * @src_cnt: number of source pages
169 * @len: length in bytes
170 * @flags: ASYNC_TX_XOR_ZERO_DST, ASYNC_TX_XOR_DROP_DEST, ASYNC_TX_ACK
171 * @depend_tx: xor depends on the result of this transaction.
172 * @cb_fn: function to call when the xor completes
173 * @cb_param: parameter to pass to the callback routine
175 struct dma_async_tx_descriptor *
176 async_xor(struct page *dest, struct page **src_list, unsigned int offset,
177 int src_cnt, size_t len, enum async_tx_flags flags,
178 struct dma_async_tx_descriptor *depend_tx,
179 dma_async_tx_callback cb_fn, void *cb_param)
181 struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_XOR,
182 &dest, 1, src_list,
183 src_cnt, len);
184 BUG_ON(src_cnt <= 1);
186 if (chan) {
187 /* run the xor asynchronously */
188 pr_debug("%s (async): len: %zu\n", __func__, len);
190 return do_async_xor(chan, dest, src_list, offset, src_cnt, len,
191 flags, depend_tx, cb_fn, cb_param);
192 } else {
193 /* run the xor synchronously */
194 pr_debug("%s (sync): len: %zu\n", __func__, len);
196 /* in the sync case the dest is an implied source
197 * (assumes the dest is the first source)
199 if (flags & ASYNC_TX_XOR_DROP_DST) {
200 src_cnt--;
201 src_list++;
204 /* wait for any prerequisite operations */
205 async_tx_quiesce(&depend_tx);
207 do_sync_xor(dest, src_list, offset, src_cnt, len,
208 flags, cb_fn, cb_param);
210 return NULL;
213 EXPORT_SYMBOL_GPL(async_xor);
215 static int page_is_zero(struct page *p, unsigned int offset, size_t len)
217 char *a = page_address(p) + offset;
218 return ((*(u32 *) a) == 0 &&
219 memcmp(a, a + 4, len - 4) == 0);
223 * async_xor_val - attempt a xor parity check with a dma engine.
224 * @dest: destination page used if the xor is performed synchronously
225 * @src_list: array of source pages. The dest page must be listed as a source
226 * at index zero. The contents of this array may be overwritten.
227 * @offset: offset in pages to start transaction
228 * @src_cnt: number of source pages
229 * @len: length in bytes
230 * @result: 0 if sum == 0 else non-zero
231 * @flags: ASYNC_TX_ACK
232 * @depend_tx: xor depends on the result of this transaction.
233 * @cb_fn: function to call when the xor completes
234 * @cb_param: parameter to pass to the callback routine
236 struct dma_async_tx_descriptor *
237 async_xor_val(struct page *dest, struct page **src_list,
238 unsigned int offset, int src_cnt, size_t len,
239 u32 *result, enum async_tx_flags flags,
240 struct dma_async_tx_descriptor *depend_tx,
241 dma_async_tx_callback cb_fn, void *cb_param)
243 struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_XOR_VAL,
244 &dest, 1, src_list,
245 src_cnt, len);
246 struct dma_device *device = chan ? chan->device : NULL;
247 struct dma_async_tx_descriptor *tx = NULL;
249 BUG_ON(src_cnt <= 1);
251 if (device && src_cnt <= device->max_xor) {
252 dma_addr_t *dma_src = (dma_addr_t *) src_list;
253 unsigned long dma_prep_flags = cb_fn ? DMA_PREP_INTERRUPT : 0;
254 int i;
256 pr_debug("%s: (async) len: %zu\n", __func__, len);
258 for (i = 0; i < src_cnt; i++)
259 dma_src[i] = dma_map_page(device->dev, src_list[i],
260 offset, len, DMA_TO_DEVICE);
262 tx = device->device_prep_dma_xor_val(chan, dma_src, src_cnt,
263 len, result,
264 dma_prep_flags);
265 if (unlikely(!tx)) {
266 async_tx_quiesce(&depend_tx);
268 while (!tx) {
269 dma_async_issue_pending(chan);
270 tx = device->device_prep_dma_xor_val(chan,
271 dma_src, src_cnt, len, result,
272 dma_prep_flags);
276 async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param);
277 } else {
278 unsigned long xor_flags = flags;
280 pr_debug("%s: (sync) len: %zu\n", __func__, len);
282 xor_flags |= ASYNC_TX_XOR_DROP_DST;
283 xor_flags &= ~ASYNC_TX_ACK;
285 tx = async_xor(dest, src_list, offset, src_cnt, len, xor_flags,
286 depend_tx, NULL, NULL);
288 async_tx_quiesce(&tx);
290 *result = page_is_zero(dest, offset, len) ? 0 : 1;
292 async_tx_sync_epilog(cb_fn, cb_param);
295 return tx;
297 EXPORT_SYMBOL_GPL(async_xor_val);
299 static int __init async_xor_init(void)
301 #ifdef CONFIG_DMA_ENGINE
302 /* To conserve stack space the input src_list (array of page pointers)
303 * is reused to hold the array of dma addresses passed to the driver.
304 * This conversion is only possible when dma_addr_t is less than the
305 * the size of a pointer. HIGHMEM64G is known to violate this
306 * assumption.
308 BUILD_BUG_ON(sizeof(dma_addr_t) > sizeof(struct page *));
309 #endif
311 return 0;
314 static void __exit async_xor_exit(void)
316 do { } while (0);
319 module_init(async_xor_init);
320 module_exit(async_xor_exit);
322 MODULE_AUTHOR("Intel Corporation");
323 MODULE_DESCRIPTION("asynchronous xor/xor-zero-sum api");
324 MODULE_LICENSE("GPL");