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
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>
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, dma_addr_t *dma_src,
37 struct async_submit_ctl *submit)
39 struct dma_device *dma = chan->device;
40 struct dma_async_tx_descriptor *tx = NULL;
43 dma_async_tx_callback cb_fn_orig = submit->cb_fn;
44 void *cb_param_orig = submit->cb_param;
45 enum async_tx_flags flags_orig = submit->flags;
46 enum dma_ctrl_flags dma_flags;
50 /* map the dest bidrectional in case it is re-used as a source */
51 dma_dest = dma_map_page(dma->dev, dest, offset, len, DMA_BIDIRECTIONAL);
52 for (i = 0; i < src_cnt; i++) {
53 /* only map the dest once */
54 if (unlikely(src_list[i] == dest)) {
55 dma_src[i] = dma_dest;
58 dma_src[i] = dma_map_page(dma->dev, src_list[i], offset,
63 submit->flags = flags_orig;
65 xor_src_cnt = min(src_cnt, (int)dma->max_xor);
66 /* if we are submitting additional xors, leave the chain open,
67 * clear the callback parameters, and leave the destination
70 if (src_cnt > xor_src_cnt) {
71 submit->flags &= ~ASYNC_TX_ACK;
72 dma_flags = DMA_COMPL_SKIP_DEST_UNMAP;
74 submit->cb_param = NULL;
76 submit->cb_fn = cb_fn_orig;
77 submit->cb_param = cb_param_orig;
80 dma_flags |= DMA_PREP_INTERRUPT;
82 /* Since we have clobbered the src_list we are committed
83 * to doing this asynchronously. Drivers force forward progress
84 * in case they can not provide a descriptor
86 tx = dma->device_prep_dma_xor(chan, dma_dest, &dma_src[src_off],
87 xor_src_cnt, len, dma_flags);
90 async_tx_quiesce(&submit->depend_tx);
92 /* spin wait for the preceeding transactions to complete */
93 while (unlikely(!tx)) {
94 dma_async_issue_pending(chan);
95 tx = dma->device_prep_dma_xor(chan, dma_dest,
101 async_tx_submit(chan, tx, submit);
102 submit->depend_tx = tx;
104 if (src_cnt > xor_src_cnt) {
105 /* drop completed sources */
106 src_cnt -= xor_src_cnt;
107 src_off += xor_src_cnt;
109 /* use the intermediate result a source */
110 dma_src[--src_off] = dma_dest;
120 do_sync_xor(struct page *dest, struct page **src_list, unsigned int offset,
121 int src_cnt, size_t len, struct async_submit_ctl *submit)
129 if (submit->scribble)
130 srcs = submit->scribble;
132 srcs = (void **) src_list;
134 /* convert to buffer pointers */
135 for (i = 0; i < src_cnt; i++)
136 srcs[i] = page_address(src_list[i]) + offset;
138 /* set destination address */
139 dest_buf = page_address(dest) + offset;
141 if (submit->flags & ASYNC_TX_XOR_ZERO_DST)
142 memset(dest_buf, 0, len);
144 while (src_cnt > 0) {
145 /* process up to 'MAX_XOR_BLOCKS' sources */
146 xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
147 xor_blocks(xor_src_cnt, len, dest_buf, &srcs[src_off]);
149 /* drop completed sources */
150 src_cnt -= xor_src_cnt;
151 src_off += xor_src_cnt;
154 async_tx_sync_epilog(submit);
158 * async_xor - attempt to xor a set of blocks with a dma engine.
159 * @dest: destination page
160 * @src_list: array of source pages
161 * @offset: common src/dst offset to start transaction
162 * @src_cnt: number of source pages
163 * @len: length in bytes
164 * @submit: submission / completion modifiers
166 * honored flags: ASYNC_TX_ACK, ASYNC_TX_XOR_ZERO_DST, ASYNC_TX_XOR_DROP_DST
168 * xor_blocks always uses the dest as a source so the
169 * ASYNC_TX_XOR_ZERO_DST flag must be set to not include dest data in
170 * the calculation. The assumption with dma eninges is that they only
171 * use the destination buffer as a source when it is explicity specified
172 * in the source list.
174 * src_list note: if the dest is also a source it must be at index zero.
175 * The contents of this array will be overwritten if a scribble region
178 struct dma_async_tx_descriptor *
179 async_xor(struct page *dest, struct page **src_list, unsigned int offset,
180 int src_cnt, size_t len, struct async_submit_ctl *submit)
182 struct dma_chan *chan = async_tx_find_channel(submit, DMA_XOR,
185 dma_addr_t *dma_src = NULL;
187 BUG_ON(src_cnt <= 1);
189 if (submit->scribble)
190 dma_src = submit->scribble;
191 else if (sizeof(dma_addr_t) <= sizeof(struct page *))
192 dma_src = (dma_addr_t *) src_list;
194 if (dma_src && chan) {
195 /* run the xor asynchronously */
196 pr_debug("%s (async): len: %zu\n", __func__, len);
198 return do_async_xor(chan, dest, src_list, offset, src_cnt, len,
201 /* run the xor synchronously */
202 pr_debug("%s (sync): len: %zu\n", __func__, len);
203 WARN_ONCE(chan, "%s: no space for dma address conversion\n",
206 /* in the sync case the dest is an implied source
207 * (assumes the dest is the first source)
209 if (submit->flags & ASYNC_TX_XOR_DROP_DST) {
214 /* wait for any prerequisite operations */
215 async_tx_quiesce(&submit->depend_tx);
217 do_sync_xor(dest, src_list, offset, src_cnt, len, submit);
222 EXPORT_SYMBOL_GPL(async_xor);
224 static int page_is_zero(struct page *p, unsigned int offset, size_t len)
226 char *a = page_address(p) + offset;
227 return ((*(u32 *) a) == 0 &&
228 memcmp(a, a + 4, len - 4) == 0);
232 * async_xor_val - attempt a xor parity check with a dma engine.
233 * @dest: destination page used if the xor is performed synchronously
234 * @src_list: array of source pages
235 * @offset: offset in pages to start transaction
236 * @src_cnt: number of source pages
237 * @len: length in bytes
238 * @result: 0 if sum == 0 else non-zero
239 * @submit: submission / completion modifiers
241 * honored flags: ASYNC_TX_ACK
243 * src_list note: if the dest is also a source it must be at index zero.
244 * The contents of this array will be overwritten if a scribble region
247 struct dma_async_tx_descriptor *
248 async_xor_val(struct page *dest, struct page **src_list, unsigned int offset,
249 int src_cnt, size_t len, enum sum_check_flags *result,
250 struct async_submit_ctl *submit)
252 struct dma_chan *chan = async_tx_find_channel(submit, DMA_XOR_VAL,
255 struct dma_device *device = chan ? chan->device : NULL;
256 struct dma_async_tx_descriptor *tx = NULL;
257 dma_addr_t *dma_src = NULL;
259 BUG_ON(src_cnt <= 1);
261 if (submit->scribble)
262 dma_src = submit->scribble;
263 else if (sizeof(dma_addr_t) <= sizeof(struct page *))
264 dma_src = (dma_addr_t *) src_list;
266 if (dma_src && device && src_cnt <= device->max_xor) {
267 unsigned long dma_prep_flags;
270 pr_debug("%s: (async) len: %zu\n", __func__, len);
272 dma_prep_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0;
273 for (i = 0; i < src_cnt; i++)
274 dma_src[i] = dma_map_page(device->dev, src_list[i],
275 offset, len, DMA_TO_DEVICE);
277 tx = device->device_prep_dma_xor_val(chan, dma_src, src_cnt,
281 async_tx_quiesce(&submit->depend_tx);
284 dma_async_issue_pending(chan);
285 tx = device->device_prep_dma_xor_val(chan,
286 dma_src, src_cnt, len, result,
291 async_tx_submit(chan, tx, submit);
293 enum async_tx_flags flags_orig = submit->flags;
295 pr_debug("%s: (sync) len: %zu\n", __func__, len);
296 WARN_ONCE(device && src_cnt <= device->max_xor,
297 "%s: no space for dma address conversion\n",
300 submit->flags |= ASYNC_TX_XOR_DROP_DST;
301 submit->flags &= ~ASYNC_TX_ACK;
303 tx = async_xor(dest, src_list, offset, src_cnt, len, submit);
305 async_tx_quiesce(&tx);
307 *result = !page_is_zero(dest, offset, len) << SUM_CHECK_P;
309 async_tx_sync_epilog(submit);
310 submit->flags = flags_orig;
315 EXPORT_SYMBOL_GPL(async_xor_val);
317 MODULE_AUTHOR("Intel Corporation");
318 MODULE_DESCRIPTION("asynchronous xor/xor-zero-sum api");
319 MODULE_LICENSE("GPL");