2 * Asynchronous RAID-6 recovery calculations ASYNC_TX API.
3 * Copyright(c) 2009 Intel Corporation
5 * based on raid6recov.c:
6 * Copyright 2002 H. Peter Anvin
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License as published by the Free
10 * Software Foundation; either version 2 of the License, or (at your option)
13 * This program is distributed in the hope that it will be useful, but WITHOUT
14 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
18 * You should have received a copy of the GNU General Public License along with
19 * this program; if not, write to the Free Software Foundation, Inc., 51
20 * Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
23 #include <linux/kernel.h>
24 #include <linux/interrupt.h>
25 #include <linux/dma-mapping.h>
26 #include <linux/raid/pq.h>
27 #include <linux/async_tx.h>
29 static struct dma_async_tx_descriptor *
30 async_sum_product(struct page *dest, struct page **srcs, unsigned char *coef,
31 size_t len, struct async_submit_ctl *submit)
33 struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
34 &dest, 1, srcs, 2, len);
35 struct dma_device *dma = chan ? chan->device : NULL;
36 const u8 *amul, *bmul;
41 dma_addr_t dma_dest[2];
42 dma_addr_t dma_src[2];
43 struct device *dev = dma->dev;
44 struct dma_async_tx_descriptor *tx;
45 enum dma_ctrl_flags dma_flags = DMA_PREP_PQ_DISABLE_P;
47 if (submit->flags & ASYNC_TX_FENCE)
48 dma_flags |= DMA_PREP_FENCE;
49 dma_dest[1] = dma_map_page(dev, dest, 0, len, DMA_BIDIRECTIONAL);
50 dma_src[0] = dma_map_page(dev, srcs[0], 0, len, DMA_TO_DEVICE);
51 dma_src[1] = dma_map_page(dev, srcs[1], 0, len, DMA_TO_DEVICE);
52 tx = dma->device_prep_dma_pq(chan, dma_dest, dma_src, 2, coef,
55 async_tx_submit(chan, tx, submit);
59 /* could not get a descriptor, unmap and fall through to
60 * the synchronous path
62 dma_unmap_page(dev, dma_dest[1], len, DMA_BIDIRECTIONAL);
63 dma_unmap_page(dev, dma_src[0], len, DMA_TO_DEVICE);
64 dma_unmap_page(dev, dma_src[1], len, DMA_TO_DEVICE);
67 /* run the operation synchronously */
68 async_tx_quiesce(&submit->depend_tx);
69 amul = raid6_gfmul[coef[0]];
70 bmul = raid6_gfmul[coef[1]];
71 a = page_address(srcs[0]);
72 b = page_address(srcs[1]);
73 c = page_address(dest);
84 static struct dma_async_tx_descriptor *
85 async_mult(struct page *dest, struct page *src, u8 coef, size_t len,
86 struct async_submit_ctl *submit)
88 struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
89 &dest, 1, &src, 1, len);
90 struct dma_device *dma = chan ? chan->device : NULL;
91 const u8 *qmul; /* Q multiplier table */
95 dma_addr_t dma_dest[2];
96 dma_addr_t dma_src[1];
97 struct device *dev = dma->dev;
98 struct dma_async_tx_descriptor *tx;
99 enum dma_ctrl_flags dma_flags = DMA_PREP_PQ_DISABLE_P;
101 if (submit->flags & ASYNC_TX_FENCE)
102 dma_flags |= DMA_PREP_FENCE;
103 dma_dest[1] = dma_map_page(dev, dest, 0, len, DMA_BIDIRECTIONAL);
104 dma_src[0] = dma_map_page(dev, src, 0, len, DMA_TO_DEVICE);
105 tx = dma->device_prep_dma_pq(chan, dma_dest, dma_src, 1, &coef,
108 async_tx_submit(chan, tx, submit);
112 /* could not get a descriptor, unmap and fall through to
113 * the synchronous path
115 dma_unmap_page(dev, dma_dest[1], len, DMA_BIDIRECTIONAL);
116 dma_unmap_page(dev, dma_src[0], len, DMA_TO_DEVICE);
119 /* no channel available, or failed to allocate a descriptor, so
120 * perform the operation synchronously
122 async_tx_quiesce(&submit->depend_tx);
123 qmul = raid6_gfmul[coef];
124 d = page_address(dest);
125 s = page_address(src);
133 static struct dma_async_tx_descriptor *
134 __2data_recov_4(int disks, size_t bytes, int faila, int failb,
135 struct page **blocks, struct async_submit_ctl *submit)
137 struct dma_async_tx_descriptor *tx = NULL;
138 struct page *p, *q, *a, *b;
139 struct page *srcs[2];
140 unsigned char coef[2];
141 enum async_tx_flags flags = submit->flags;
142 dma_async_tx_callback cb_fn = submit->cb_fn;
143 void *cb_param = submit->cb_param;
144 void *scribble = submit->scribble;
152 /* in the 4 disk case P + Pxy == P and Q + Qxy == Q */
153 /* Dx = A*(P+Pxy) + B*(Q+Qxy) */
156 coef[0] = raid6_gfexi[failb-faila];
157 coef[1] = raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]];
158 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
159 tx = async_sum_product(b, srcs, coef, bytes, submit);
164 init_async_submit(submit, flags | ASYNC_TX_XOR_ZERO_DST, tx, cb_fn,
166 tx = async_xor(a, srcs, 0, 2, bytes, submit);
172 static struct dma_async_tx_descriptor *
173 __2data_recov_5(int disks, size_t bytes, int faila, int failb,
174 struct page **blocks, struct async_submit_ctl *submit)
176 struct dma_async_tx_descriptor *tx = NULL;
177 struct page *p, *q, *g, *dp, *dq;
178 struct page *srcs[2];
179 unsigned char coef[2];
180 enum async_tx_flags flags = submit->flags;
181 dma_async_tx_callback cb_fn = submit->cb_fn;
182 void *cb_param = submit->cb_param;
183 void *scribble = submit->scribble;
184 int good_srcs, good, i;
188 for (i = 0; i < disks-2; i++) {
189 if (blocks[i] == NULL)
191 if (i == faila || i == failb)
196 BUG_ON(good_srcs > 1);
202 /* Compute syndrome with zero for the missing data pages
203 * Use the dead data pages as temporary storage for delta p and
209 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
210 tx = async_memcpy(dp, g, 0, 0, bytes, submit);
211 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
212 tx = async_mult(dq, g, raid6_gfexp[good], bytes, submit);
214 /* compute P + Pxy */
217 init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
218 NULL, NULL, scribble);
219 tx = async_xor(dp, srcs, 0, 2, bytes, submit);
221 /* compute Q + Qxy */
224 init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
225 NULL, NULL, scribble);
226 tx = async_xor(dq, srcs, 0, 2, bytes, submit);
228 /* Dx = A*(P+Pxy) + B*(Q+Qxy) */
231 coef[0] = raid6_gfexi[failb-faila];
232 coef[1] = raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]];
233 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
234 tx = async_sum_product(dq, srcs, coef, bytes, submit);
239 init_async_submit(submit, flags | ASYNC_TX_XOR_DROP_DST, tx, cb_fn,
241 tx = async_xor(dp, srcs, 0, 2, bytes, submit);
246 static struct dma_async_tx_descriptor *
247 __2data_recov_n(int disks, size_t bytes, int faila, int failb,
248 struct page **blocks, struct async_submit_ctl *submit)
250 struct dma_async_tx_descriptor *tx = NULL;
251 struct page *p, *q, *dp, *dq;
252 struct page *srcs[2];
253 unsigned char coef[2];
254 enum async_tx_flags flags = submit->flags;
255 dma_async_tx_callback cb_fn = submit->cb_fn;
256 void *cb_param = submit->cb_param;
257 void *scribble = submit->scribble;
262 /* Compute syndrome with zero for the missing data pages
263 * Use the dead data pages as temporary storage for
264 * delta p and delta q
267 blocks[faila] = NULL;
268 blocks[disks-2] = dp;
270 blocks[failb] = NULL;
271 blocks[disks-1] = dq;
273 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
274 tx = async_gen_syndrome(blocks, 0, disks, bytes, submit);
276 /* Restore pointer table */
282 /* compute P + Pxy */
285 init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
286 NULL, NULL, scribble);
287 tx = async_xor(dp, srcs, 0, 2, bytes, submit);
289 /* compute Q + Qxy */
292 init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
293 NULL, NULL, scribble);
294 tx = async_xor(dq, srcs, 0, 2, bytes, submit);
296 /* Dx = A*(P+Pxy) + B*(Q+Qxy) */
299 coef[0] = raid6_gfexi[failb-faila];
300 coef[1] = raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]];
301 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
302 tx = async_sum_product(dq, srcs, coef, bytes, submit);
307 init_async_submit(submit, flags | ASYNC_TX_XOR_DROP_DST, tx, cb_fn,
309 tx = async_xor(dp, srcs, 0, 2, bytes, submit);
315 * async_raid6_2data_recov - asynchronously calculate two missing data blocks
316 * @disks: number of disks in the RAID-6 array
318 * @faila: first failed drive index
319 * @failb: second failed drive index
320 * @blocks: array of source pointers where the last two entries are p and q
321 * @submit: submission/completion modifiers
323 struct dma_async_tx_descriptor *
324 async_raid6_2data_recov(int disks, size_t bytes, int faila, int failb,
325 struct page **blocks, struct async_submit_ctl *submit)
327 void *scribble = submit->scribble;
328 int non_zero_srcs, i;
330 BUG_ON(faila == failb);
334 pr_debug("%s: disks: %d len: %zu\n", __func__, disks, bytes);
336 /* if a dma resource is not available or a scribble buffer is not
337 * available punt to the synchronous path. In the 'dma not
338 * available' case be sure to use the scribble buffer to
339 * preserve the content of 'blocks' as the caller intended.
341 if (!async_dma_find_channel(DMA_PQ) || !scribble) {
342 void **ptrs = scribble ? scribble : (void **) blocks;
344 async_tx_quiesce(&submit->depend_tx);
345 for (i = 0; i < disks; i++)
346 if (blocks[i] == NULL)
347 ptrs[i] = (void *) raid6_empty_zero_page;
349 ptrs[i] = page_address(blocks[i]);
351 raid6_2data_recov(disks, bytes, faila, failb, ptrs);
353 async_tx_sync_epilog(submit);
359 for (i = 0; i < disks-2 && non_zero_srcs < 4; i++)
362 switch (non_zero_srcs) {
365 /* There must be at least 2 sources - the failed devices. */
369 /* dma devices do not uniformly understand a zero source pq
370 * operation (in contrast to the synchronous case), so
371 * explicitly handle the special case of a 4 disk array with
372 * both data disks missing.
374 return __2data_recov_4(disks, bytes, faila, failb, blocks, submit);
376 /* dma devices do not uniformly understand a single
377 * source pq operation (in contrast to the synchronous
378 * case), so explicitly handle the special case of a 5 disk
379 * array with 2 of 3 data disks missing.
381 return __2data_recov_5(disks, bytes, faila, failb, blocks, submit);
383 return __2data_recov_n(disks, bytes, faila, failb, blocks, submit);
386 EXPORT_SYMBOL_GPL(async_raid6_2data_recov);
389 * async_raid6_datap_recov - asynchronously calculate a data and the 'p' block
390 * @disks: number of disks in the RAID-6 array
392 * @faila: failed drive index
393 * @blocks: array of source pointers where the last two entries are p and q
394 * @submit: submission/completion modifiers
396 struct dma_async_tx_descriptor *
397 async_raid6_datap_recov(int disks, size_t bytes, int faila,
398 struct page **blocks, struct async_submit_ctl *submit)
400 struct dma_async_tx_descriptor *tx = NULL;
401 struct page *p, *q, *dq;
403 enum async_tx_flags flags = submit->flags;
404 dma_async_tx_callback cb_fn = submit->cb_fn;
405 void *cb_param = submit->cb_param;
406 void *scribble = submit->scribble;
407 int good_srcs, good, i;
408 struct page *srcs[2];
410 pr_debug("%s: disks: %d len: %zu\n", __func__, disks, bytes);
412 /* if a dma resource is not available or a scribble buffer is not
413 * available punt to the synchronous path. In the 'dma not
414 * available' case be sure to use the scribble buffer to
415 * preserve the content of 'blocks' as the caller intended.
417 if (!async_dma_find_channel(DMA_PQ) || !scribble) {
418 void **ptrs = scribble ? scribble : (void **) blocks;
420 async_tx_quiesce(&submit->depend_tx);
421 for (i = 0; i < disks; i++)
422 if (blocks[i] == NULL)
423 ptrs[i] = (void*)raid6_empty_zero_page;
425 ptrs[i] = page_address(blocks[i]);
427 raid6_datap_recov(disks, bytes, faila, ptrs);
429 async_tx_sync_epilog(submit);
436 for (i = 0; i < disks-2; i++) {
446 BUG_ON(good_srcs == 0);
451 /* Compute syndrome with zero for the missing data page
452 * Use the dead data page as temporary storage for delta q
455 blocks[faila] = NULL;
456 blocks[disks-1] = dq;
458 /* in the 4-disk case we only need to perform a single source
459 * multiplication with the one good data block.
461 if (good_srcs == 1) {
462 struct page *g = blocks[good];
464 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL,
466 tx = async_memcpy(p, g, 0, 0, bytes, submit);
468 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL,
470 tx = async_mult(dq, g, raid6_gfexp[good], bytes, submit);
472 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL,
474 tx = async_gen_syndrome(blocks, 0, disks, bytes, submit);
477 /* Restore pointer table */
481 /* calculate g^{-faila} */
482 coef = raid6_gfinv[raid6_gfexp[faila]];
486 init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
487 NULL, NULL, scribble);
488 tx = async_xor(dq, srcs, 0, 2, bytes, submit);
490 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
491 tx = async_mult(dq, dq, coef, bytes, submit);
495 init_async_submit(submit, flags | ASYNC_TX_XOR_DROP_DST, tx, cb_fn,
497 tx = async_xor(p, srcs, 0, 2, bytes, submit);
501 EXPORT_SYMBOL_GPL(async_raid6_datap_recov);
504 MODULE_DESCRIPTION("asynchronous RAID-6 recovery api");
505 MODULE_LICENSE("GPL");