crypto/async_tx/async_raid6_recov.c

   1 /*
   2  * Asynchronous RAID-6 recovery calculations ASYNC_TX API.
   3  * Copyright(c) 2009 Intel Corporation
   4  *
   5  * based on raid6recov.c:
   6  *   Copyright 2002 H. Peter Anvin
   7  *
   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)
  11  * any later version.
  12  *
  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
  16  * more details.
  17  *
  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.
  21  *
  22  */
  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>
  28
  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)
  32 {
  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;
  37         u8 ax, bx;
  38         u8 *a, *b, *c;
  39
  40         if (dma) {
  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;
  46
  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,
  53                                              len, dma_flags);
  54                 if (tx) {
  55                         async_tx_submit(chan, tx, submit);
  56                         return tx;
  57                 }
  58
  59                 /* could not get a descriptor, unmap and fall through to
  60                  * the synchronous path
  61                  */
  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);
  65         }
  66
  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);
  74
  75         while (len--) {
  76                 ax    = amul[*a++];
  77                 bx    = bmul[*b++];
  78                 *c++ = ax ^ bx;
  79         }
  80
  81         return NULL;
  82 }
  83
  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)
  87 {
  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 */
  92         u8 *d, *s;
  93
  94         if (dma) {
  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;
 100
 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,
 106                                              len, dma_flags);
 107                 if (tx) {
 108                         async_tx_submit(chan, tx, submit);
 109                         return tx;
 110                 }
 111
 112                 /* could not get a descriptor, unmap and fall through to
 113                  * the synchronous path
 114                  */
 115                 dma_unmap_page(dev, dma_dest[1], len, DMA_BIDIRECTIONAL);
 116                 dma_unmap_page(dev, dma_src[0], len, DMA_TO_DEVICE);
 117         }
 118
 119         /* no channel available, or failed to allocate a descriptor, so
 120          * perform the operation synchronously
 121          */
 122         async_tx_quiesce(&submit->depend_tx);
 123         qmul  = raid6_gfmul[coef];
 124         d = page_address(dest);
 125         s = page_address(src);
 126
 127         while (len--)
 128                 *d++ = qmul[*s++];
 129
 130         return NULL;
 131 }
 132
 133 static struct dma_async_tx_descriptor *
 134 __2data_recov_4(size_t bytes, int faila, int failb, struct page **blocks,
 135               struct async_submit_ctl *submit)
 136 {
 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;
 145
 146         p = blocks[4-2];
 147         q = blocks[4-1];
 148
 149         a = blocks[faila];
 150         b = blocks[failb];
 151
 152         /* in the 4 disk case P + Pxy == P and Q + Qxy == Q */
 153         /* Dx = A*(P+Pxy) + B*(Q+Qxy) */
 154         srcs[0] = p;
 155         srcs[1] = q;
 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);
 160
 161         /* Dy = P+Pxy+Dx */
 162         srcs[0] = p;
 163         srcs[1] = b;
 164         init_async_submit(submit, flags | ASYNC_TX_XOR_ZERO_DST, tx, cb_fn,
 165                           cb_param, scribble);
 166         tx = async_xor(a, srcs, 0, 2, bytes, submit);
 167
 168         return tx;
 169
 170 }
 171
 172 static struct dma_async_tx_descriptor *
 173 __2data_recov_5(size_t bytes, int faila, int failb, struct page **blocks,
 174               struct async_submit_ctl *submit)
 175 {
 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 uninitialized_var(good);
 185         int i;
 186
 187         for (i = 0; i < 3; i++) {
 188                 if (i == faila || i == failb)
 189                         continue;
 190                 else {
 191                         good = i;
 192                         break;
 193                 }
 194         }
 195         BUG_ON(i >= 3);
 196
 197         p = blocks[5-2];
 198         q = blocks[5-1];
 199         g = blocks[good];
 200
 201         /* Compute syndrome with zero for the missing data pages
 202          * Use the dead data pages as temporary storage for delta p and
 203          * delta q
 204          */
 205         dp = blocks[faila];
 206         dq = blocks[failb];
 207
 208         init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
 209         tx = async_memcpy(dp, g, 0, 0, bytes, submit);
 210         init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
 211         tx = async_mult(dq, g, raid6_gfexp[good], bytes, submit);
 212
 213         /* compute P + Pxy */
 214         srcs[0] = dp;
 215         srcs[1] = p;
 216         init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
 217                           NULL, NULL, scribble);
 218         tx = async_xor(dp, srcs, 0, 2, bytes, submit);
 219
 220         /* compute Q + Qxy */
 221         srcs[0] = dq;
 222         srcs[1] = q;
 223         init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
 224                           NULL, NULL, scribble);
 225         tx = async_xor(dq, srcs, 0, 2, bytes, submit);
 226
 227         /* Dx = A*(P+Pxy) + B*(Q+Qxy) */
 228         srcs[0] = dp;
 229         srcs[1] = dq;
 230         coef[0] = raid6_gfexi[failb-faila];
 231         coef[1] = raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]];
 232         init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
 233         tx = async_sum_product(dq, srcs, coef, bytes, submit);
 234
 235         /* Dy = P+Pxy+Dx */
 236         srcs[0] = dp;
 237         srcs[1] = dq;
 238         init_async_submit(submit, flags | ASYNC_TX_XOR_DROP_DST, tx, cb_fn,
 239                           cb_param, scribble);
 240         tx = async_xor(dp, srcs, 0, 2, bytes, submit);
 241
 242         return tx;
 243 }
 244
 245 static struct dma_async_tx_descriptor *
 246 __2data_recov_n(int disks, size_t bytes, int faila, int failb,
 247               struct page **blocks, struct async_submit_ctl *submit)
 248 {
 249         struct dma_async_tx_descriptor *tx = NULL;
 250         struct page *p, *q, *dp, *dq;
 251         struct page *srcs[2];
 252         unsigned char coef[2];
 253         enum async_tx_flags flags = submit->flags;
 254         dma_async_tx_callback cb_fn = submit->cb_fn;
 255         void *cb_param = submit->cb_param;
 256         void *scribble = submit->scribble;
 257
 258         p = blocks[disks-2];
 259         q = blocks[disks-1];
 260
 261         /* Compute syndrome with zero for the missing data pages
 262          * Use the dead data pages as temporary storage for
 263          * delta p and delta q
 264          */
 265         dp = blocks[faila];
 266         blocks[faila] = (void *)raid6_empty_zero_page;
 267         blocks[disks-2] = dp;
 268         dq = blocks[failb];
 269         blocks[failb] = (void *)raid6_empty_zero_page;
 270         blocks[disks-1] = dq;
 271
 272         init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
 273         tx = async_gen_syndrome(blocks, 0, disks, bytes, submit);
 274
 275         /* Restore pointer table */
 276         blocks[faila]   = dp;
 277         blocks[failb]   = dq;
 278         blocks[disks-2] = p;
 279         blocks[disks-1] = q;
 280
 281         /* compute P + Pxy */
 282         srcs[0] = dp;
 283         srcs[1] = p;
 284         init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
 285                           NULL, NULL, scribble);
 286         tx = async_xor(dp, srcs, 0, 2, bytes, submit);
 287
 288         /* compute Q + Qxy */
 289         srcs[0] = dq;
 290         srcs[1] = q;
 291         init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
 292                           NULL, NULL, scribble);
 293         tx = async_xor(dq, srcs, 0, 2, bytes, submit);
 294
 295         /* Dx = A*(P+Pxy) + B*(Q+Qxy) */
 296         srcs[0] = dp;
 297         srcs[1] = dq;
 298         coef[0] = raid6_gfexi[failb-faila];
 299         coef[1] = raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]];
 300         init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
 301         tx = async_sum_product(dq, srcs, coef, bytes, submit);
 302
 303         /* Dy = P+Pxy+Dx */
 304         srcs[0] = dp;
 305         srcs[1] = dq;
 306         init_async_submit(submit, flags | ASYNC_TX_XOR_DROP_DST, tx, cb_fn,
 307                           cb_param, scribble);
 308         tx = async_xor(dp, srcs, 0, 2, bytes, submit);
 309
 310         return tx;
 311 }
 312
 313 /**
 314  * async_raid6_2data_recov - asynchronously calculate two missing data blocks
 315  * @disks: number of disks in the RAID-6 array
 316  * @bytes: block size
 317  * @faila: first failed drive index
 318  * @failb: second failed drive index
 319  * @blocks: array of source pointers where the last two entries are p and q
 320  * @submit: submission/completion modifiers
 321  */
 322 struct dma_async_tx_descriptor *
 323 async_raid6_2data_recov(int disks, size_t bytes, int faila, int failb,
 324                         struct page **blocks, struct async_submit_ctl *submit)
 325 {
 326         BUG_ON(faila == failb);
 327         if (failb < faila)
 328                 swap(faila, failb);
 329
 330         pr_debug("%s: disks: %d len: %zu\n", __func__, disks, bytes);
 331
 332         /* we need to preserve the contents of 'blocks' for the async
 333          * case, so punt to synchronous if a scribble buffer is not available
 334          */
 335         if (!submit->scribble) {
 336                 void **ptrs = (void **) blocks;
 337                 int i;
 338
 339                 async_tx_quiesce(&submit->depend_tx);
 340                 for (i = 0; i < disks; i++)
 341                         ptrs[i] = page_address(blocks[i]);
 342
 343                 raid6_2data_recov(disks, bytes, faila, failb, ptrs);
 344
 345                 async_tx_sync_epilog(submit);
 346
 347                 return NULL;
 348         }
 349
 350         switch (disks) {
 351         case 4:
 352                 /* dma devices do not uniformly understand a zero source pq
 353                  * operation (in contrast to the synchronous case), so
 354                  * explicitly handle the 4 disk special case
 355                  */
 356                 return __2data_recov_4(bytes, faila, failb, blocks, submit);
 357         case 5:
 358                 /* dma devices do not uniformly understand a single
 359                  * source pq operation (in contrast to the synchronous
 360                  * case), so explicitly handle the 5 disk special case
 361                  */
 362                 return __2data_recov_5(bytes, faila, failb, blocks, submit);
 363         default:
 364                 return __2data_recov_n(disks, bytes, faila, failb, blocks, submit);
 365         }
 366 }
 367 EXPORT_SYMBOL_GPL(async_raid6_2data_recov);
 368
 369 /**
 370  * async_raid6_datap_recov - asynchronously calculate a data and the 'p' block
 371  * @disks: number of disks in the RAID-6 array
 372  * @bytes: block size
 373  * @faila: failed drive index
 374  * @blocks: array of source pointers where the last two entries are p and q
 375  * @submit: submission/completion modifiers
 376  */
 377 struct dma_async_tx_descriptor *
 378 async_raid6_datap_recov(int disks, size_t bytes, int faila,
 379                         struct page **blocks, struct async_submit_ctl *submit)
 380 {
 381         struct dma_async_tx_descriptor *tx = NULL;
 382         struct page *p, *q, *dq;
 383         u8 coef;
 384         enum async_tx_flags flags = submit->flags;
 385         dma_async_tx_callback cb_fn = submit->cb_fn;
 386         void *cb_param = submit->cb_param;
 387         void *scribble = submit->scribble;
 388         struct page *srcs[2];
 389
 390         pr_debug("%s: disks: %d len: %zu\n", __func__, disks, bytes);
 391
 392         /* we need to preserve the contents of 'blocks' for the async
 393          * case, so punt to synchronous if a scribble buffer is not available
 394          */
 395         if (!scribble) {
 396                 void **ptrs = (void **) blocks;
 397                 int i;
 398
 399                 async_tx_quiesce(&submit->depend_tx);
 400                 for (i = 0; i < disks; i++)
 401                         ptrs[i] = page_address(blocks[i]);
 402
 403                 raid6_datap_recov(disks, bytes, faila, ptrs);
 404
 405                 async_tx_sync_epilog(submit);
 406
 407                 return NULL;
 408         }
 409
 410         p = blocks[disks-2];
 411         q = blocks[disks-1];
 412
 413         /* Compute syndrome with zero for the missing data page
 414          * Use the dead data page as temporary storage for delta q
 415          */
 416         dq = blocks[faila];
 417         blocks[faila] = (void *)raid6_empty_zero_page;
 418         blocks[disks-1] = dq;
 419
 420         /* in the 4 disk case we only need to perform a single source
 421          * multiplication
 422          */
 423         if (disks == 4) {
 424                 int good = faila == 0 ? 1 : 0;
 425                 struct page *g = blocks[good];
 426
 427                 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL,
 428                                   scribble);
 429                 tx = async_memcpy(p, g, 0, 0, bytes, submit);
 430
 431                 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL,
 432                                   scribble);
 433                 tx = async_mult(dq, g, raid6_gfexp[good], bytes, submit);
 434         } else {
 435                 init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL,
 436                                   scribble);
 437                 tx = async_gen_syndrome(blocks, 0, disks, bytes, submit);
 438         }
 439
 440         /* Restore pointer table */
 441         blocks[faila]   = dq;
 442         blocks[disks-1] = q;
 443
 444         /* calculate g^{-faila} */
 445         coef = raid6_gfinv[raid6_gfexp[faila]];
 446
 447         srcs[0] = dq;
 448         srcs[1] = q;
 449         init_async_submit(submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
 450                           NULL, NULL, scribble);
 451         tx = async_xor(dq, srcs, 0, 2, bytes, submit);
 452
 453         init_async_submit(submit, ASYNC_TX_FENCE, tx, NULL, NULL, scribble);
 454         tx = async_mult(dq, dq, coef, bytes, submit);
 455
 456         srcs[0] = p;
 457         srcs[1] = dq;
 458         init_async_submit(submit, flags | ASYNC_TX_XOR_DROP_DST, tx, cb_fn,
 459                           cb_param, scribble);
 460         tx = async_xor(p, srcs, 0, 2, bytes, submit);
 461
 462         return tx;
 463 }
 464 EXPORT_SYMBOL_GPL(async_raid6_datap_recov);
 465
 466 MODULE_AUTHOR("Dan Williams <dan.j.williams@intel.com>");
 467 MODULE_DESCRIPTION("asynchronous RAID-6 recovery api");
 468 MODULE_LICENSE("GPL");