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/dma-mapping.h>
  25 #include <linux/raid/pq.h>
  26 #include <linux/async_tx.h>
  27
  28 /**
  29  * scribble - space to hold throwaway P buffer for synchronous gen_syndrome
  30  */
  31 static struct page *scribble;
  32
  33 static bool is_raid6_zero_block(struct page *p)
  34 {
  35         return p == (void *) raid6_empty_zero_page;
  36 }
  37
  38 /* the struct page *blocks[] parameter passed to async_gen_syndrome()
  39  * and async_syndrome_val() contains the 'P' destination address at
  40  * blocks[disks-2] and the 'Q' destination address at blocks[disks-1]
  41  *
  42  * note: these are macros as they are used as lvalues
  43  */
  44 #define P(b, d) (b[d-2])
  45 #define Q(b, d) (b[d-1])
  46
  47 /**
  48  * do_async_gen_syndrome - asynchronously calculate P and/or Q
  49  */
  50 static __async_inline struct dma_async_tx_descriptor *
  51 do_async_gen_syndrome(struct dma_chan *chan, struct page **blocks,
  52                       const unsigned char *scfs, unsigned int offset, int disks,
  53                       size_t len, dma_addr_t *dma_src,
  54                       struct async_submit_ctl *submit)
  55 {
  56         struct dma_async_tx_descriptor *tx = NULL;
  57         struct dma_device *dma = chan->device;
  58         enum dma_ctrl_flags dma_flags = 0;
  59         enum async_tx_flags flags_orig = submit->flags;
  60         dma_async_tx_callback cb_fn_orig = submit->cb_fn;
  61         dma_async_tx_callback cb_param_orig = submit->cb_param;
  62         int src_cnt = disks - 2;
  63         unsigned char coefs[src_cnt];
  64         unsigned short pq_src_cnt;
  65         dma_addr_t dma_dest[2];
  66         int src_off = 0;
  67         int idx;
  68         int i;
  69
  70         /* DMAs use destinations as sources, so use BIDIRECTIONAL mapping */
  71         if (P(blocks, disks))
  72                 dma_dest[0] = dma_map_page(dma->dev, P(blocks, disks), offset,
  73                                            len, DMA_BIDIRECTIONAL);
  74         else
  75                 dma_flags |= DMA_PREP_PQ_DISABLE_P;
  76         if (Q(blocks, disks))
  77                 dma_dest[1] = dma_map_page(dma->dev, Q(blocks, disks), offset,
  78                                            len, DMA_BIDIRECTIONAL);
  79         else
  80                 dma_flags |= DMA_PREP_PQ_DISABLE_Q;
  81
  82         /* convert source addresses being careful to collapse 'empty'
  83          * sources and update the coefficients accordingly
  84          */
  85         for (i = 0, idx = 0; i < src_cnt; i++) {
  86                 if (is_raid6_zero_block(blocks[i]))
  87                         continue;
  88                 dma_src[idx] = dma_map_page(dma->dev, blocks[i], offset, len,
  89                                             DMA_TO_DEVICE);
  90                 coefs[idx] = scfs[i];
  91                 idx++;
  92         }
  93         src_cnt = idx;
  94
  95         while (src_cnt > 0) {
  96                 submit->flags = flags_orig;
  97                 pq_src_cnt = min(src_cnt, dma_maxpq(dma, dma_flags));
  98                 /* if we are submitting additional pqs, leave the chain open,
  99                  * clear the callback parameters, and leave the destination
 100                  * buffers mapped
 101                  */
 102                 if (src_cnt > pq_src_cnt) {
 103                         submit->flags &= ~ASYNC_TX_ACK;
 104                         submit->flags |= ASYNC_TX_FENCE;
 105                         dma_flags |= DMA_COMPL_SKIP_DEST_UNMAP;
 106                         submit->cb_fn = NULL;
 107                         submit->cb_param = NULL;
 108                 } else {
 109                         dma_flags &= ~DMA_COMPL_SKIP_DEST_UNMAP;
 110                         submit->cb_fn = cb_fn_orig;
 111                         submit->cb_param = cb_param_orig;
 112                         if (cb_fn_orig)
 113                                 dma_flags |= DMA_PREP_INTERRUPT;
 114                 }
 115                 if (submit->flags & ASYNC_TX_FENCE)
 116                         dma_flags |= DMA_PREP_FENCE;
 117
 118                 /* Since we have clobbered the src_list we are committed
 119                  * to doing this asynchronously.  Drivers force forward
 120                  * progress in case they can not provide a descriptor
 121                  */
 122                 for (;;) {
 123                         tx = dma->device_prep_dma_pq(chan, dma_dest,
 124                                                      &dma_src[src_off],
 125                                                      pq_src_cnt,
 126                                                      &coefs[src_off], len,
 127                                                      dma_flags);
 128                         if (likely(tx))
 129                                 break;
 130                         async_tx_quiesce(&submit->depend_tx);
 131                         dma_async_issue_pending(chan);
 132                 }
 133
 134                 async_tx_submit(chan, tx, submit);
 135                 submit->depend_tx = tx;
 136
 137                 /* drop completed sources */
 138                 src_cnt -= pq_src_cnt;
 139                 src_off += pq_src_cnt;
 140
 141                 dma_flags |= DMA_PREP_CONTINUE;
 142         }
 143
 144         return tx;
 145 }
 146
 147 /**
 148  * do_sync_gen_syndrome - synchronously calculate a raid6 syndrome
 149  */
 150 static void
 151 do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 152                      size_t len, struct async_submit_ctl *submit)
 153 {
 154         void **srcs;
 155         int i;
 156
 157         if (submit->scribble)
 158                 srcs = submit->scribble;
 159         else
 160                 srcs = (void **) blocks;
 161
 162         for (i = 0; i < disks; i++) {
 163                 if (is_raid6_zero_block(blocks[i])) {
 164                         BUG_ON(i > disks - 3); /* P or Q can't be zero */
 165                         srcs[i] = blocks[i];
 166                 } else
 167                         srcs[i] = page_address(blocks[i]) + offset;
 168         }
 169         raid6_call.gen_syndrome(disks, len, srcs);
 170         async_tx_sync_epilog(submit);
 171 }
 172
 173 /**
 174  * async_gen_syndrome - asynchronously calculate a raid6 syndrome
 175  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 176  * @offset: common offset into each block (src and dest) to start transaction
 177  * @disks: number of blocks (including missing P or Q, see below)
 178  * @len: length of operation in bytes
 179  * @submit: submission/completion modifiers
 180  *
 181  * General note: This routine assumes a field of GF(2^8) with a
 182  * primitive polynomial of 0x11d and a generator of {02}.
 183  *
 184  * 'disks' note: callers can optionally omit either P or Q (but not
 185  * both) from the calculation by setting blocks[disks-2] or
 186  * blocks[disks-1] to NULL.  When P or Q is omitted 'len' must be <=
 187  * PAGE_SIZE as a temporary buffer of this size is used in the
 188  * synchronous path.  'disks' always accounts for both destination
 189  * buffers.
 190  *
 191  * 'blocks' note: if submit->scribble is NULL then the contents of
 192  * 'blocks' may be overridden
 193  */
 194 struct dma_async_tx_descriptor *
 195 async_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 196                    size_t len, struct async_submit_ctl *submit)
 197 {
 198         int src_cnt = disks - 2;
 199         struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
 200                                                       &P(blocks, disks), 2,
 201                                                       blocks, src_cnt, len);
 202         struct dma_device *device = chan ? chan->device : NULL;
 203         dma_addr_t *dma_src = NULL;
 204
 205         BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks)));
 206
 207         if (submit->scribble)
 208                 dma_src = submit->scribble;
 209         else if (sizeof(dma_addr_t) <= sizeof(struct page *))
 210                 dma_src = (dma_addr_t *) blocks;
 211
 212         if (dma_src && device &&
 213             (src_cnt <= dma_maxpq(device, 0) ||
 214              dma_maxpq(device, DMA_PREP_CONTINUE) > 0) &&
 215             is_dma_pq_aligned(device, offset, 0, len)) {
 216                 /* run the p+q asynchronously */
 217                 pr_debug("%s: (async) disks: %d len: %zu\n",
 218                          __func__, disks, len);
 219                 return do_async_gen_syndrome(chan, blocks, raid6_gfexp, offset,
 220                                              disks, len, dma_src, submit);
 221         }
 222
 223         /* run the pq synchronously */
 224         pr_debug("%s: (sync) disks: %d len: %zu\n", __func__, disks, len);
 225
 226         /* wait for any prerequisite operations */
 227         async_tx_quiesce(&submit->depend_tx);
 228
 229         if (!P(blocks, disks)) {
 230                 P(blocks, disks) = scribble;
 231                 BUG_ON(len + offset > PAGE_SIZE);
 232         }
 233         if (!Q(blocks, disks)) {
 234                 Q(blocks, disks) = scribble;
 235                 BUG_ON(len + offset > PAGE_SIZE);
 236         }
 237         do_sync_gen_syndrome(blocks, offset, disks, len, submit);
 238
 239         return NULL;
 240 }
 241 EXPORT_SYMBOL_GPL(async_gen_syndrome);
 242
 243 /**
 244  * async_syndrome_val - asynchronously validate a raid6 syndrome
 245  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 246  * @offset: common offset into each block (src and dest) to start transaction
 247  * @disks: number of blocks (including missing P or Q, see below)
 248  * @len: length of operation in bytes
 249  * @pqres: on val failure SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set
 250  * @spare: temporary result buffer for the synchronous case
 251  * @submit: submission / completion modifiers
 252  *
 253  * The same notes from async_gen_syndrome apply to the 'blocks',
 254  * and 'disks' parameters of this routine.  The synchronous path
 255  * requires a temporary result buffer and submit->scribble to be
 256  * specified.
 257  */
 258 struct dma_async_tx_descriptor *
 259 async_syndrome_val(struct page **blocks, unsigned int offset, int disks,
 260                    size_t len, enum sum_check_flags *pqres, struct page *spare,
 261                    struct async_submit_ctl *submit)
 262 {
 263         struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ_VAL,
 264                                                       NULL, 0,  blocks, disks,
 265                                                       len);
 266         struct dma_device *device = chan ? chan->device : NULL;
 267         struct dma_async_tx_descriptor *tx;
 268         enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0;
 269         dma_addr_t *dma_src = NULL;
 270
 271         BUG_ON(disks < 4);
 272
 273         if (submit->scribble)
 274                 dma_src = submit->scribble;
 275         else if (sizeof(dma_addr_t) <= sizeof(struct page *))
 276                 dma_src = (dma_addr_t *) blocks;
 277
 278         if (dma_src && device && disks <= dma_maxpq(device, 0) &&
 279             is_dma_pq_aligned(device, offset, 0, len)) {
 280                 struct device *dev = device->dev;
 281                 dma_addr_t *pq = &dma_src[disks-2];
 282                 int i;
 283
 284                 pr_debug("%s: (async) disks: %d len: %zu\n",
 285                          __func__, disks, len);
 286                 if (!P(blocks, disks))
 287                         dma_flags |= DMA_PREP_PQ_DISABLE_P;
 288                 if (!Q(blocks, disks))
 289                         dma_flags |= DMA_PREP_PQ_DISABLE_Q;
 290                 if (submit->flags & ASYNC_TX_FENCE)
 291                         dma_flags |= DMA_PREP_FENCE;
 292                 for (i = 0; i < disks; i++)
 293                         if (likely(blocks[i])) {
 294                                 BUG_ON(is_raid6_zero_block(blocks[i]));
 295                                 dma_src[i] = dma_map_page(dev, blocks[i],
 296                                                           offset, len,
 297                                                           DMA_TO_DEVICE);
 298                         }
 299
 300                 for (;;) {
 301                         tx = device->device_prep_dma_pq_val(chan, pq, dma_src,
 302                                                             disks - 2,
 303                                                             raid6_gfexp,
 304                                                             len, pqres,
 305                                                             dma_flags);
 306                         if (likely(tx))
 307                                 break;
 308                         async_tx_quiesce(&submit->depend_tx);
 309                         dma_async_issue_pending(chan);
 310                 }
 311                 async_tx_submit(chan, tx, submit);
 312
 313                 return tx;
 314         } else {
 315                 struct page *p_src = P(blocks, disks);
 316                 struct page *q_src = Q(blocks, disks);
 317                 enum async_tx_flags flags_orig = submit->flags;
 318                 dma_async_tx_callback cb_fn_orig = submit->cb_fn;
 319                 void *scribble = submit->scribble;
 320                 void *cb_param_orig = submit->cb_param;
 321                 void *p, *q, *s;
 322
 323                 pr_debug("%s: (sync) disks: %d len: %zu\n",
 324                          __func__, disks, len);
 325
 326                 /* caller must provide a temporary result buffer and
 327                  * allow the input parameters to be preserved
 328                  */
 329                 BUG_ON(!spare || !scribble);
 330
 331                 /* wait for any prerequisite operations */
 332                 async_tx_quiesce(&submit->depend_tx);
 333
 334                 /* recompute p and/or q into the temporary buffer and then
 335                  * check to see the result matches the current value
 336                  */
 337                 tx = NULL;
 338                 *pqres = 0;
 339                 if (p_src) {
 340                         init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL,
 341                                           NULL, NULL, scribble);
 342                         tx = async_xor(spare, blocks, offset, disks-2, len, submit);
 343                         async_tx_quiesce(&tx);
 344                         p = page_address(p_src) + offset;
 345                         s = page_address(spare) + offset;
 346                         *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P;
 347                 }
 348
 349                 if (q_src) {
 350                         P(blocks, disks) = NULL;
 351                         Q(blocks, disks) = spare;
 352                         init_async_submit(submit, 0, NULL, NULL, NULL, scribble);
 353                         tx = async_gen_syndrome(blocks, offset, disks, len, submit);
 354                         async_tx_quiesce(&tx);
 355                         q = page_address(q_src) + offset;
 356                         s = page_address(spare) + offset;
 357                         *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q;
 358                 }
 359
 360                 /* restore P, Q and submit */
 361                 P(blocks, disks) = p_src;
 362                 Q(blocks, disks) = q_src;
 363
 364                 submit->cb_fn = cb_fn_orig;
 365                 submit->cb_param = cb_param_orig;
 366                 submit->flags = flags_orig;
 367                 async_tx_sync_epilog(submit);
 368
 369                 return NULL;
 370         }
 371 }
 372 EXPORT_SYMBOL_GPL(async_syndrome_val);
 373
 374 static int __init async_pq_init(void)
 375 {
 376         scribble = alloc_page(GFP_KERNEL);
 377
 378         if (scribble)
 379                 return 0;
 380
 381         pr_err("%s: failed to allocate required spare page\n", __func__);
 382
 383         return -ENOMEM;
 384 }
 385
 386 static void __exit async_pq_exit(void)
 387 {
 388         put_page(scribble);
 389 }
 390
 391 module_init(async_pq_init);
 392 module_exit(async_pq_exit);
 393
 394 MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation");
 395 MODULE_LICENSE("GPL");