kgdb, x86: Avoid invoking kgdb_nmicallback twice per NMI
[linux-2.6/pdupreez.git] / crypto / async_tx / async_xor.c
blobc029d3eb9ef0a76ed248deb4a40e47511ba12afe
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 * This routine is marked __always_inline so it can be compiled away
35 * when CONFIG_DMA_ENGINE=n
37 static __always_inline struct dma_async_tx_descriptor *
38 do_async_xor(struct dma_chan *chan, struct page *dest, struct page **src_list,
39 unsigned int offset, int src_cnt, size_t len,
40 enum async_tx_flags flags,
41 struct dma_async_tx_descriptor *depend_tx,
42 dma_async_tx_callback cb_fn, void *cb_param)
44 struct dma_device *dma = chan->device;
45 dma_addr_t *dma_src = (dma_addr_t *) src_list;
46 struct dma_async_tx_descriptor *tx = NULL;
47 int src_off = 0;
48 int i;
49 dma_async_tx_callback _cb_fn;
50 void *_cb_param;
51 enum async_tx_flags async_flags;
52 enum dma_ctrl_flags dma_flags;
53 int xor_src_cnt;
54 dma_addr_t dma_dest;
56 dma_dest = dma_map_page(dma->dev, dest, offset, len, DMA_FROM_DEVICE);
57 for (i = 0; i < src_cnt; i++)
58 dma_src[i] = dma_map_page(dma->dev, src_list[i], offset,
59 len, DMA_TO_DEVICE);
61 while (src_cnt) {
62 async_flags = flags;
63 dma_flags = 0;
64 xor_src_cnt = min(src_cnt, dma->max_xor);
65 /* if we are submitting additional xors, leave the chain open,
66 * clear the callback parameters, and leave the destination
67 * buffer mapped
69 if (src_cnt > xor_src_cnt) {
70 async_flags &= ~ASYNC_TX_ACK;
71 dma_flags = DMA_COMPL_SKIP_DEST_UNMAP;
72 _cb_fn = NULL;
73 _cb_param = NULL;
74 } else {
75 _cb_fn = cb_fn;
76 _cb_param = cb_param;
78 if (_cb_fn)
79 dma_flags |= DMA_PREP_INTERRUPT;
81 /* Since we have clobbered the src_list we are committed
82 * to doing this asynchronously. Drivers force forward progress
83 * in case they can not provide a descriptor
85 tx = dma->device_prep_dma_xor(chan, dma_dest, &dma_src[src_off],
86 xor_src_cnt, len, dma_flags);
88 if (unlikely(!tx))
89 async_tx_quiesce(&depend_tx);
91 /* spin wait for the preceeding transactions to complete */
92 while (unlikely(!tx)) {
93 dma_async_issue_pending(chan);
94 tx = dma->device_prep_dma_xor(chan, dma_dest,
95 &dma_src[src_off],
96 xor_src_cnt, len,
97 dma_flags);
100 async_tx_submit(chan, tx, async_flags, depend_tx, _cb_fn,
101 _cb_param);
103 depend_tx = tx;
104 flags |= ASYNC_TX_DEP_ACK;
106 if (src_cnt > xor_src_cnt) {
107 /* drop completed sources */
108 src_cnt -= xor_src_cnt;
109 src_off += xor_src_cnt;
111 /* use the intermediate result a source */
112 dma_src[--src_off] = dma_dest;
113 src_cnt++;
114 } else
115 break;
118 return tx;
121 static void
122 do_sync_xor(struct page *dest, struct page **src_list, unsigned int offset,
123 int src_cnt, size_t len, enum async_tx_flags flags,
124 dma_async_tx_callback cb_fn, void *cb_param)
126 int i;
127 int xor_src_cnt;
128 int src_off = 0;
129 void *dest_buf;
130 void **srcs = (void **) src_list;
132 /* reuse the 'src_list' array to convert to buffer pointers */
133 for (i = 0; i < src_cnt; i++)
134 srcs[i] = page_address(src_list[i]) + offset;
136 /* set destination address */
137 dest_buf = page_address(dest) + offset;
139 if (flags & ASYNC_TX_XOR_ZERO_DST)
140 memset(dest_buf, 0, len);
142 while (src_cnt > 0) {
143 /* process up to 'MAX_XOR_BLOCKS' sources */
144 xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
145 xor_blocks(xor_src_cnt, len, dest_buf, &srcs[src_off]);
147 /* drop completed sources */
148 src_cnt -= xor_src_cnt;
149 src_off += xor_src_cnt;
152 async_tx_sync_epilog(cb_fn, cb_param);
156 * async_xor - attempt to xor a set of blocks with a dma engine.
157 * xor_blocks always uses the dest as a source so the ASYNC_TX_XOR_ZERO_DST
158 * flag must be set to not include dest data in the calculation. The
159 * assumption with dma eninges is that they only use the destination
160 * buffer as a source when it is explicity specified in the source list.
161 * @dest: destination page
162 * @src_list: array of source pages (if the dest is also a source it must be
163 * at index zero). The contents of this array may be overwritten.
164 * @offset: offset in pages to start transaction
165 * @src_cnt: number of source pages
166 * @len: length in bytes
167 * @flags: ASYNC_TX_XOR_ZERO_DST, ASYNC_TX_XOR_DROP_DEST,
168 * ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
169 * @depend_tx: xor depends on the result of this transaction.
170 * @cb_fn: function to call when the xor completes
171 * @cb_param: parameter to pass to the callback routine
173 struct dma_async_tx_descriptor *
174 async_xor(struct page *dest, struct page **src_list, unsigned int offset,
175 int src_cnt, size_t len, enum async_tx_flags flags,
176 struct dma_async_tx_descriptor *depend_tx,
177 dma_async_tx_callback cb_fn, void *cb_param)
179 struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_XOR,
180 &dest, 1, src_list,
181 src_cnt, len);
182 BUG_ON(src_cnt <= 1);
184 if (chan) {
185 /* run the xor asynchronously */
186 pr_debug("%s (async): len: %zu\n", __func__, len);
188 return do_async_xor(chan, dest, src_list, offset, src_cnt, len,
189 flags, depend_tx, cb_fn, cb_param);
190 } else {
191 /* run the xor synchronously */
192 pr_debug("%s (sync): len: %zu\n", __func__, len);
194 /* in the sync case the dest is an implied source
195 * (assumes the dest is the first source)
197 if (flags & ASYNC_TX_XOR_DROP_DST) {
198 src_cnt--;
199 src_list++;
202 /* wait for any prerequisite operations */
203 async_tx_quiesce(&depend_tx);
205 do_sync_xor(dest, src_list, offset, src_cnt, len,
206 flags, cb_fn, cb_param);
208 return NULL;
211 EXPORT_SYMBOL_GPL(async_xor);
213 static int page_is_zero(struct page *p, unsigned int offset, size_t len)
215 char *a = page_address(p) + offset;
216 return ((*(u32 *) a) == 0 &&
217 memcmp(a, a + 4, len - 4) == 0);
221 * async_xor_zero_sum - attempt a xor parity check with a dma engine.
222 * @dest: destination page used if the xor is performed synchronously
223 * @src_list: array of source pages. The dest page must be listed as a source
224 * at index zero. The contents of this array may be overwritten.
225 * @offset: offset in pages to start transaction
226 * @src_cnt: number of source pages
227 * @len: length in bytes
228 * @result: 0 if sum == 0 else non-zero
229 * @flags: ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
230 * @depend_tx: xor depends on the result of this transaction.
231 * @cb_fn: function to call when the xor completes
232 * @cb_param: parameter to pass to the callback routine
234 struct dma_async_tx_descriptor *
235 async_xor_zero_sum(struct page *dest, struct page **src_list,
236 unsigned int offset, int src_cnt, size_t len,
237 u32 *result, enum async_tx_flags flags,
238 struct dma_async_tx_descriptor *depend_tx,
239 dma_async_tx_callback cb_fn, void *cb_param)
241 struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_ZERO_SUM,
242 &dest, 1, src_list,
243 src_cnt, len);
244 struct dma_device *device = chan ? chan->device : NULL;
245 struct dma_async_tx_descriptor *tx = NULL;
247 BUG_ON(src_cnt <= 1);
249 if (device && src_cnt <= device->max_xor) {
250 dma_addr_t *dma_src = (dma_addr_t *) src_list;
251 unsigned long dma_prep_flags = cb_fn ? DMA_PREP_INTERRUPT : 0;
252 int i;
254 pr_debug("%s: (async) len: %zu\n", __func__, len);
256 for (i = 0; i < src_cnt; i++)
257 dma_src[i] = dma_map_page(device->dev, src_list[i],
258 offset, len, DMA_TO_DEVICE);
260 tx = device->device_prep_dma_zero_sum(chan, dma_src, src_cnt,
261 len, result,
262 dma_prep_flags);
263 if (unlikely(!tx)) {
264 async_tx_quiesce(&depend_tx);
266 while (!tx) {
267 dma_async_issue_pending(chan);
268 tx = device->device_prep_dma_zero_sum(chan,
269 dma_src, src_cnt, len, result,
270 dma_prep_flags);
274 async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param);
275 } else {
276 unsigned long xor_flags = flags;
278 pr_debug("%s: (sync) len: %zu\n", __func__, len);
280 xor_flags |= ASYNC_TX_XOR_DROP_DST;
281 xor_flags &= ~ASYNC_TX_ACK;
283 tx = async_xor(dest, src_list, offset, src_cnt, len, xor_flags,
284 depend_tx, NULL, NULL);
286 async_tx_quiesce(&tx);
288 *result = page_is_zero(dest, offset, len) ? 0 : 1;
290 async_tx_sync_epilog(cb_fn, cb_param);
293 return tx;
295 EXPORT_SYMBOL_GPL(async_xor_zero_sum);
297 static int __init async_xor_init(void)
299 #ifdef CONFIG_DMA_ENGINE
300 /* To conserve stack space the input src_list (array of page pointers)
301 * is reused to hold the array of dma addresses passed to the driver.
302 * This conversion is only possible when dma_addr_t is less than the
303 * the size of a pointer. HIGHMEM64G is known to violate this
304 * assumption.
306 BUILD_BUG_ON(sizeof(dma_addr_t) > sizeof(struct page *));
307 #endif
309 return 0;
312 static void __exit async_xor_exit(void)
314 do { } while (0);
317 module_init(async_xor_init);
318 module_exit(async_xor_exit);
320 MODULE_AUTHOR("Intel Corporation");
321 MODULE_DESCRIPTION("asynchronous xor/xor-zero-sum api");
322 MODULE_LICENSE("GPL");