crypto/async_tx/async_pq.c

   1 /*
   2  * Copyright(c) 2007 Yuri Tikhonov <yur@emcraft.com>
   3  * Copyright(c) 2009 Intel Corporation
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License as published by the Free
   7  * Software Foundation; either version 2 of the License, or (at your option)
   8  * any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along with
  16  * this program; if not, write to the Free Software Foundation, Inc., 59
  17  * Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  18  *
  19  * The full GNU General Public License is included in this distribution in the
  20  * file called COPYING.
  21  */
  22 #include <linux/kernel.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/module.h>
  25 #include <linux/dma-mapping.h>
  26 #include <linux/raid/pq.h>
  27 #include <linux/async_tx.h>
  28 #include <linux/gfp.h>
  29
  30 /**
  31  * pq_scribble_page - space to hold throwaway P or Q buffer for
  32  * synchronous gen_syndrome
  33  */
  34 static struct page *pq_scribble_page;
  35
  36 /* the struct page *blocks[] parameter passed to async_gen_syndrome()
  37  * and async_syndrome_val() contains the 'P' destination address at
  38  * blocks[disks-2] and the 'Q' destination address at blocks[disks-1]
  39  *
  40  * note: these are macros as they are used as lvalues
  41  */
  42 #define P(b, d) (b[d-2])
  43 #define Q(b, d) (b[d-1])
  44
  45 /**
  46  * do_async_gen_syndrome - asynchronously calculate P and/or Q
  47  */
  48 static __async_inline struct dma_async_tx_descriptor *
  49 do_async_gen_syndrome(struct dma_chan *chan,
  50                       const unsigned char *scfs, int disks,
  51                       struct dmaengine_unmap_data *unmap,
  52                       enum dma_ctrl_flags dma_flags,
  53                       struct async_submit_ctl *submit)
  54 {
  55         struct dma_async_tx_descriptor *tx = NULL;
  56         struct dma_device *dma = chan->device;
  57         enum async_tx_flags flags_orig = submit->flags;
  58         dma_async_tx_callback cb_fn_orig = submit->cb_fn;
  59         dma_async_tx_callback cb_param_orig = submit->cb_param;
  60         int src_cnt = disks - 2;
  61         unsigned short pq_src_cnt;
  62         dma_addr_t dma_dest[2];
  63         int src_off = 0;
  64
  65         if (submit->flags & ASYNC_TX_FENCE)
  66                 dma_flags |= DMA_PREP_FENCE;
  67
  68         while (src_cnt > 0) {
  69                 submit->flags = flags_orig;
  70                 pq_src_cnt = min(src_cnt, dma_maxpq(dma, dma_flags));
  71                 /* if we are submitting additional pqs, leave the chain open,
  72                  * clear the callback parameters, and leave the destination
  73                  * buffers mapped
  74                  */
  75                 if (src_cnt > pq_src_cnt) {
  76                         submit->flags &= ~ASYNC_TX_ACK;
  77                         submit->flags |= ASYNC_TX_FENCE;
  78                         submit->cb_fn = NULL;
  79                         submit->cb_param = NULL;
  80                 } else {
  81                         submit->cb_fn = cb_fn_orig;
  82                         submit->cb_param = cb_param_orig;
  83                         if (cb_fn_orig)
  84                                 dma_flags |= DMA_PREP_INTERRUPT;
  85                 }
  86
  87                 /* Drivers force forward progress in case they can not provide
  88                  * a descriptor
  89                  */
  90                 for (;;) {
  91                         dma_dest[0] = unmap->addr[disks - 2];
  92                         dma_dest[1] = unmap->addr[disks - 1];
  93                         tx = dma->device_prep_dma_pq(chan, dma_dest,
  94                                                      &unmap->addr[src_off],
  95                                                      pq_src_cnt,
  96                                                      &scfs[src_off], unmap->len,
  97                                                      dma_flags);
  98                         if (likely(tx))
  99                                 break;
 100                         async_tx_quiesce(&submit->depend_tx);
 101                         dma_async_issue_pending(chan);
 102                 }
 103
 104                 dma_set_unmap(tx, unmap);
 105                 async_tx_submit(chan, tx, submit);
 106                 submit->depend_tx = tx;
 107
 108                 /* drop completed sources */
 109                 src_cnt -= pq_src_cnt;
 110                 src_off += pq_src_cnt;
 111
 112                 dma_flags |= DMA_PREP_CONTINUE;
 113         }
 114
 115         return tx;
 116 }
 117
 118 /**
 119  * do_sync_gen_syndrome - synchronously calculate a raid6 syndrome
 120  */
 121 static void
 122 do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 123                      size_t len, struct async_submit_ctl *submit)
 124 {
 125         void **srcs;
 126         int i;
 127         int start = -1, stop = disks - 3;
 128
 129         if (submit->scribble)
 130                 srcs = submit->scribble;
 131         else
 132                 srcs = (void **) blocks;
 133
 134         for (i = 0; i < disks; i++) {
 135                 if (blocks[i] == NULL) {
 136                         BUG_ON(i > disks - 3); /* P or Q can't be zero */
 137                         srcs[i] = (void*)raid6_empty_zero_page;
 138                 } else {
 139                         srcs[i] = page_address(blocks[i]) + offset;
 140                         if (i < disks - 2) {
 141                                 stop = i;
 142                                 if (start == -1)
 143                                         start = i;
 144                         }
 145                 }
 146         }
 147         if (submit->flags & ASYNC_TX_PQ_XOR_DST) {
 148                 BUG_ON(!raid6_call.xor_syndrome);
 149                 if (start >= 0)
 150                         raid6_call.xor_syndrome(disks, start, stop, len, srcs);
 151         } else
 152                 raid6_call.gen_syndrome(disks, len, srcs);
 153         async_tx_sync_epilog(submit);
 154 }
 155
 156 /**
 157  * async_gen_syndrome - asynchronously calculate a raid6 syndrome
 158  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 159  * @offset: common offset into each block (src and dest) to start transaction
 160  * @disks: number of blocks (including missing P or Q, see below)
 161  * @len: length of operation in bytes
 162  * @submit: submission/completion modifiers
 163  *
 164  * General note: This routine assumes a field of GF(2^8) with a
 165  * primitive polynomial of 0x11d and a generator of {02}.
 166  *
 167  * 'disks' note: callers can optionally omit either P or Q (but not
 168  * both) from the calculation by setting blocks[disks-2] or
 169  * blocks[disks-1] to NULL.  When P or Q is omitted 'len' must be <=
 170  * PAGE_SIZE as a temporary buffer of this size is used in the
 171  * synchronous path.  'disks' always accounts for both destination
 172  * buffers.  If any source buffers (blocks[i] where i < disks - 2) are
 173  * set to NULL those buffers will be replaced with the raid6_zero_page
 174  * in the synchronous path and omitted in the hardware-asynchronous
 175  * path.
 176  */
 177 struct dma_async_tx_descriptor *
 178 async_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 179                    size_t len, struct async_submit_ctl *submit)
 180 {
 181         int src_cnt = disks - 2;
 182         struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
 183                                                       &P(blocks, disks), 2,
 184                                                       blocks, src_cnt, len);
 185         struct dma_device *device = chan ? chan->device : NULL;
 186         struct dmaengine_unmap_data *unmap = NULL;
 187
 188         BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks)));
 189
 190         if (device)
 191                 unmap = dmaengine_get_unmap_data(device->dev, disks, GFP_NOWAIT);
 192
 193         /* XORing P/Q is only implemented in software */
 194         if (unmap && !(submit->flags & ASYNC_TX_PQ_XOR_DST) &&
 195             (src_cnt <= dma_maxpq(device, 0) ||
 196              dma_maxpq(device, DMA_PREP_CONTINUE) > 0) &&
 197             is_dma_pq_aligned(device, offset, 0, len)) {
 198                 struct dma_async_tx_descriptor *tx;
 199                 enum dma_ctrl_flags dma_flags = 0;
 200                 unsigned char coefs[src_cnt];
 201                 int i, j;
 202
 203                 /* run the p+q asynchronously */
 204                 pr_debug("%s: (async) disks: %d len: %zu\n",
 205                          __func__, disks, len);
 206
 207                 /* convert source addresses being careful to collapse 'empty'
 208                  * sources and update the coefficients accordingly
 209                  */
 210                 unmap->len = len;
 211                 for (i = 0, j = 0; i < src_cnt; i++) {
 212                         if (blocks[i] == NULL)
 213                                 continue;
 214                         unmap->addr[j] = dma_map_page(device->dev, blocks[i], offset,
 215                                                       len, DMA_TO_DEVICE);
 216                         coefs[j] = raid6_gfexp[i];
 217                         unmap->to_cnt++;
 218                         j++;
 219                 }
 220
 221                 /*
 222                  * DMAs use destinations as sources,
 223                  * so use BIDIRECTIONAL mapping
 224                  */
 225                 unmap->bidi_cnt++;
 226                 if (P(blocks, disks))
 227                         unmap->addr[j++] = dma_map_page(device->dev, P(blocks, disks),
 228                                                         offset, len, DMA_BIDIRECTIONAL);
 229                 else {
 230                         unmap->addr[j++] = 0;
 231                         dma_flags |= DMA_PREP_PQ_DISABLE_P;
 232                 }
 233
 234                 unmap->bidi_cnt++;
 235                 if (Q(blocks, disks))
 236                         unmap->addr[j++] = dma_map_page(device->dev, Q(blocks, disks),
 237                                                        offset, len, DMA_BIDIRECTIONAL);
 238                 else {
 239                         unmap->addr[j++] = 0;
 240                         dma_flags |= DMA_PREP_PQ_DISABLE_Q;
 241                 }
 242
 243                 tx = do_async_gen_syndrome(chan, coefs, j, unmap, dma_flags, submit);
 244                 dmaengine_unmap_put(unmap);
 245                 return tx;
 246         }
 247
 248         dmaengine_unmap_put(unmap);
 249
 250         /* run the pq synchronously */
 251         pr_debug("%s: (sync) disks: %d len: %zu\n", __func__, disks, len);
 252
 253         /* wait for any prerequisite operations */
 254         async_tx_quiesce(&submit->depend_tx);
 255
 256         if (!P(blocks, disks)) {
 257                 P(blocks, disks) = pq_scribble_page;
 258                 BUG_ON(len + offset > PAGE_SIZE);
 259         }
 260         if (!Q(blocks, disks)) {
 261                 Q(blocks, disks) = pq_scribble_page;
 262                 BUG_ON(len + offset > PAGE_SIZE);
 263         }
 264         do_sync_gen_syndrome(blocks, offset, disks, len, submit);
 265
 266         return NULL;
 267 }
 268 EXPORT_SYMBOL_GPL(async_gen_syndrome);
 269
 270 static inline struct dma_chan *
 271 pq_val_chan(struct async_submit_ctl *submit, struct page **blocks, int disks, size_t len)
 272 {
 273         #ifdef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA
 274         return NULL;
 275         #endif
 276         return async_tx_find_channel(submit, DMA_PQ_VAL, NULL, 0,  blocks,
 277                                      disks, len);
 278 }
 279
 280 /**
 281  * async_syndrome_val - asynchronously validate a raid6 syndrome
 282  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 283  * @offset: common offset into each block (src and dest) to start transaction
 284  * @disks: number of blocks (including missing P or Q, see below)
 285  * @len: length of operation in bytes
 286  * @pqres: on val failure SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set
 287  * @spare: temporary result buffer for the synchronous case
 288  * @submit: submission / completion modifiers
 289  *
 290  * The same notes from async_gen_syndrome apply to the 'blocks',
 291  * and 'disks' parameters of this routine.  The synchronous path
 292  * requires a temporary result buffer and submit->scribble to be
 293  * specified.
 294  */
 295 struct dma_async_tx_descriptor *
 296 async_syndrome_val(struct page **blocks, unsigned int offset, int disks,
 297                    size_t len, enum sum_check_flags *pqres, struct page *spare,
 298                    struct async_submit_ctl *submit)
 299 {
 300         struct dma_chan *chan = pq_val_chan(submit, blocks, disks, len);
 301         struct dma_device *device = chan ? chan->device : NULL;
 302         struct dma_async_tx_descriptor *tx;
 303         unsigned char coefs[disks-2];
 304         enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0;
 305         struct dmaengine_unmap_data *unmap = NULL;
 306
 307         BUG_ON(disks < 4);
 308
 309         if (device)
 310                 unmap = dmaengine_get_unmap_data(device->dev, disks, GFP_NOWAIT);
 311
 312         if (unmap && disks <= dma_maxpq(device, 0) &&
 313             is_dma_pq_aligned(device, offset, 0, len)) {
 314                 struct device *dev = device->dev;
 315                 dma_addr_t pq[2];
 316                 int i, j = 0, src_cnt = 0;
 317
 318                 pr_debug("%s: (async) disks: %d len: %zu\n",
 319                          __func__, disks, len);
 320
 321                 unmap->len = len;
 322                 for (i = 0; i < disks-2; i++)
 323                         if (likely(blocks[i])) {
 324                                 unmap->addr[j] = dma_map_page(dev, blocks[i],
 325                                                               offset, len,
 326                                                               DMA_TO_DEVICE);
 327                                 coefs[j] = raid6_gfexp[i];
 328                                 unmap->to_cnt++;
 329                                 src_cnt++;
 330                                 j++;
 331                         }
 332
 333                 if (!P(blocks, disks)) {
 334                         pq[0] = 0;
 335                         dma_flags |= DMA_PREP_PQ_DISABLE_P;
 336                 } else {
 337                         pq[0] = dma_map_page(dev, P(blocks, disks),
 338                                              offset, len,
 339                                              DMA_TO_DEVICE);
 340                         unmap->addr[j++] = pq[0];
 341                         unmap->to_cnt++;
 342                 }
 343                 if (!Q(blocks, disks)) {
 344                         pq[1] = 0;
 345                         dma_flags |= DMA_PREP_PQ_DISABLE_Q;
 346                 } else {
 347                         pq[1] = dma_map_page(dev, Q(blocks, disks),
 348                                              offset, len,
 349                                              DMA_TO_DEVICE);
 350                         unmap->addr[j++] = pq[1];
 351                         unmap->to_cnt++;
 352                 }
 353
 354                 if (submit->flags & ASYNC_TX_FENCE)
 355                         dma_flags |= DMA_PREP_FENCE;
 356                 for (;;) {
 357                         tx = device->device_prep_dma_pq_val(chan, pq,
 358                                                             unmap->addr,
 359                                                             src_cnt,
 360                                                             coefs,
 361                                                             len, pqres,
 362                                                             dma_flags);
 363                         if (likely(tx))
 364                                 break;
 365                         async_tx_quiesce(&submit->depend_tx);
 366                         dma_async_issue_pending(chan);
 367                 }
 368
 369                 dma_set_unmap(tx, unmap);
 370                 async_tx_submit(chan, tx, submit);
 371         } else {
 372                 struct page *p_src = P(blocks, disks);
 373                 struct page *q_src = Q(blocks, disks);
 374                 enum async_tx_flags flags_orig = submit->flags;
 375                 dma_async_tx_callback cb_fn_orig = submit->cb_fn;
 376                 void *scribble = submit->scribble;
 377                 void *cb_param_orig = submit->cb_param;
 378                 void *p, *q, *s;
 379
 380                 pr_debug("%s: (sync) disks: %d len: %zu\n",
 381                          __func__, disks, len);
 382
 383                 /* caller must provide a temporary result buffer and
 384                  * allow the input parameters to be preserved
 385                  */
 386                 BUG_ON(!spare || !scribble);
 387
 388                 /* wait for any prerequisite operations */
 389                 async_tx_quiesce(&submit->depend_tx);
 390
 391                 /* recompute p and/or q into the temporary buffer and then
 392                  * check to see the result matches the current value
 393                  */
 394                 tx = NULL;
 395                 *pqres = 0;
 396                 if (p_src) {
 397                         init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL,
 398                                           NULL, NULL, scribble);
 399                         tx = async_xor(spare, blocks, offset, disks-2, len, submit);
 400                         async_tx_quiesce(&tx);
 401                         p = page_address(p_src) + offset;
 402                         s = page_address(spare) + offset;
 403                         *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P;
 404                 }
 405
 406                 if (q_src) {
 407                         P(blocks, disks) = NULL;
 408                         Q(blocks, disks) = spare;
 409                         init_async_submit(submit, 0, NULL, NULL, NULL, scribble);
 410                         tx = async_gen_syndrome(blocks, offset, disks, len, submit);
 411                         async_tx_quiesce(&tx);
 412                         q = page_address(q_src) + offset;
 413                         s = page_address(spare) + offset;
 414                         *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q;
 415                 }
 416
 417                 /* restore P, Q and submit */
 418                 P(blocks, disks) = p_src;
 419                 Q(blocks, disks) = q_src;
 420
 421                 submit->cb_fn = cb_fn_orig;
 422                 submit->cb_param = cb_param_orig;
 423                 submit->flags = flags_orig;
 424                 async_tx_sync_epilog(submit);
 425                 tx = NULL;
 426         }
 427         dmaengine_unmap_put(unmap);
 428
 429         return tx;
 430 }
 431 EXPORT_SYMBOL_GPL(async_syndrome_val);
 432
 433 static int __init async_pq_init(void)
 434 {
 435         pq_scribble_page = alloc_page(GFP_KERNEL);
 436
 437         if (pq_scribble_page)
 438                 return 0;
 439
 440         pr_err("%s: failed to allocate required spare page\n", __func__);
 441
 442         return -ENOMEM;
 443 }
 444
 445 static void __exit async_pq_exit(void)
 446 {
 447         __free_page(pq_scribble_page);
 448 }
 449
 450 module_init(async_pq_init);
 451 module_exit(async_pq_exit);
 452
 453 MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation");
 454 MODULE_LICENSE("GPL");