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 /* the struct page *blocks[] parameter passed to async_gen_syndrome()
  34  * and async_syndrome_val() contains the 'P' destination address at
  35  * blocks[disks-2] and the 'Q' destination address at blocks[disks-1]
  36  *
  37  * note: these are macros as they are used as lvalues
  38  */
  39 #define P(b, d) (b[d-2])
  40 #define Q(b, d) (b[d-1])
  41
  42 /**
  43  * do_async_gen_syndrome - asynchronously calculate P and/or Q
  44  */
  45 static __async_inline struct dma_async_tx_descriptor *
  46 do_async_gen_syndrome(struct dma_chan *chan, struct page **blocks,
  47                       const unsigned char *scfs, unsigned int offset, int disks,
  48                       size_t len, dma_addr_t *dma_src,
  49                       struct async_submit_ctl *submit)
  50 {
  51         struct dma_async_tx_descriptor *tx = NULL;
  52         struct dma_device *dma = chan->device;
  53         enum dma_ctrl_flags dma_flags = 0;
  54         enum async_tx_flags flags_orig = submit->flags;
  55         dma_async_tx_callback cb_fn_orig = submit->cb_fn;
  56         dma_async_tx_callback cb_param_orig = submit->cb_param;
  57         int src_cnt = disks - 2;
  58         unsigned char coefs[src_cnt];
  59         unsigned short pq_src_cnt;
  60         dma_addr_t dma_dest[2];
  61         int src_off = 0;
  62         int idx;
  63         int i;
  64
  65         /* DMAs use destinations as sources, so use BIDIRECTIONAL mapping */
  66         if (P(blocks, disks))
  67                 dma_dest[0] = dma_map_page(dma->dev, P(blocks, disks), offset,
  68                                            len, DMA_BIDIRECTIONAL);
  69         else
  70                 dma_flags |= DMA_PREP_PQ_DISABLE_P;
  71         if (Q(blocks, disks))
  72                 dma_dest[1] = dma_map_page(dma->dev, Q(blocks, disks), offset,
  73                                            len, DMA_BIDIRECTIONAL);
  74         else
  75                 dma_flags |= DMA_PREP_PQ_DISABLE_Q;
  76
  77         /* convert source addresses being careful to collapse 'empty'
  78          * sources and update the coefficients accordingly
  79          */
  80         for (i = 0, idx = 0; i < src_cnt; i++) {
  81                 if (blocks[i] == NULL)
  82                         continue;
  83                 dma_src[idx] = dma_map_page(dma->dev, blocks[i], offset, len,
  84                                             DMA_TO_DEVICE);
  85                 coefs[idx] = scfs[i];
  86                 idx++;
  87         }
  88         src_cnt = idx;
  89
  90         while (src_cnt > 0) {
  91                 submit->flags = flags_orig;
  92                 pq_src_cnt = min(src_cnt, dma_maxpq(dma, dma_flags));
  93                 /* if we are submitting additional pqs, leave the chain open,
  94                  * clear the callback parameters, and leave the destination
  95                  * buffers mapped
  96                  */
  97                 if (src_cnt > pq_src_cnt) {
  98                         submit->flags &= ~ASYNC_TX_ACK;
  99                         submit->flags |= ASYNC_TX_FENCE;
 100                         dma_flags |= DMA_COMPL_SKIP_DEST_UNMAP;
 101                         submit->cb_fn = NULL;
 102                         submit->cb_param = NULL;
 103                 } else {
 104                         dma_flags &= ~DMA_COMPL_SKIP_DEST_UNMAP;
 105                         submit->cb_fn = cb_fn_orig;
 106                         submit->cb_param = cb_param_orig;
 107                         if (cb_fn_orig)
 108                                 dma_flags |= DMA_PREP_INTERRUPT;
 109                 }
 110                 if (submit->flags & ASYNC_TX_FENCE)
 111                         dma_flags |= DMA_PREP_FENCE;
 112
 113                 /* Since we have clobbered the src_list we are committed
 114                  * to doing this asynchronously.  Drivers force forward
 115                  * progress in case they can not provide a descriptor
 116                  */
 117                 for (;;) {
 118                         tx = dma->device_prep_dma_pq(chan, dma_dest,
 119                                                      &dma_src[src_off],
 120                                                      pq_src_cnt,
 121                                                      &coefs[src_off], len,
 122                                                      dma_flags);
 123                         if (likely(tx))
 124                                 break;
 125                         async_tx_quiesce(&submit->depend_tx);
 126                         dma_async_issue_pending(chan);
 127                 }
 128
 129                 async_tx_submit(chan, tx, submit);
 130                 submit->depend_tx = tx;
 131
 132                 /* drop completed sources */
 133                 src_cnt -= pq_src_cnt;
 134                 src_off += pq_src_cnt;
 135
 136                 dma_flags |= DMA_PREP_CONTINUE;
 137         }
 138
 139         return tx;
 140 }
 141
 142 /**
 143  * do_sync_gen_syndrome - synchronously calculate a raid6 syndrome
 144  */
 145 static void
 146 do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 147                      size_t len, struct async_submit_ctl *submit)
 148 {
 149         void **srcs;
 150         int i;
 151
 152         if (submit->scribble)
 153                 srcs = submit->scribble;
 154         else
 155                 srcs = (void **) blocks;
 156
 157         for (i = 0; i < disks; i++) {
 158                 if (blocks[i] == NULL) {
 159                         BUG_ON(i > disks - 3); /* P or Q can't be zero */
 160                         srcs[i] = (void*)raid6_empty_zero_page;
 161                 } else
 162                         srcs[i] = page_address(blocks[i]) + offset;
 163         }
 164         raid6_call.gen_syndrome(disks, len, srcs);
 165         async_tx_sync_epilog(submit);
 166 }
 167
 168 /**
 169  * async_gen_syndrome - asynchronously calculate a raid6 syndrome
 170  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 171  * @offset: common offset into each block (src and dest) to start transaction
 172  * @disks: number of blocks (including missing P or Q, see below)
 173  * @len: length of operation in bytes
 174  * @submit: submission/completion modifiers
 175  *
 176  * General note: This routine assumes a field of GF(2^8) with a
 177  * primitive polynomial of 0x11d and a generator of {02}.
 178  *
 179  * 'disks' note: callers can optionally omit either P or Q (but not
 180  * both) from the calculation by setting blocks[disks-2] or
 181  * blocks[disks-1] to NULL.  When P or Q is omitted 'len' must be <=
 182  * PAGE_SIZE as a temporary buffer of this size is used in the
 183  * synchronous path.  'disks' always accounts for both destination
 184  * buffers.  If any source buffers (blocks[i] where i < disks - 2) are
 185  * set to NULL those buffers will be replaced with the raid6_zero_page
 186  * in the synchronous path and omitted in the hardware-asynchronous
 187  * path.
 188  *
 189  * 'blocks' note: if submit->scribble is NULL then the contents of
 190  * 'blocks' may be overwritten to perform address conversions
 191  * (dma_map_page() or page_address()).
 192  */
 193 struct dma_async_tx_descriptor *
 194 async_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 195                    size_t len, struct async_submit_ctl *submit)
 196 {
 197         int src_cnt = disks - 2;
 198         struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
 199                                                       &P(blocks, disks), 2,
 200                                                       blocks, src_cnt, len);
 201         struct dma_device *device = chan ? chan->device : NULL;
 202         dma_addr_t *dma_src = NULL;
 203
 204         BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks)));
 205
 206         if (submit->scribble)
 207                 dma_src = submit->scribble;
 208         else if (sizeof(dma_addr_t) <= sizeof(struct page *))
 209                 dma_src = (dma_addr_t *) blocks;
 210
 211         if (dma_src && device &&
 212             (src_cnt <= dma_maxpq(device, 0) ||
 213              dma_maxpq(device, DMA_PREP_CONTINUE) > 0) &&
 214             is_dma_pq_aligned(device, offset, 0, len)) {
 215                 /* run the p+q asynchronously */
 216                 pr_debug("%s: (async) disks: %d len: %zu\n",
 217                          __func__, disks, len);
 218                 return do_async_gen_syndrome(chan, blocks, raid6_gfexp, offset,
 219                                              disks, len, dma_src, submit);
 220         }
 221
 222         /* run the pq synchronously */
 223         pr_debug("%s: (sync) disks: %d len: %zu\n", __func__, disks, len);
 224
 225         /* wait for any prerequisite operations */
 226         async_tx_quiesce(&submit->depend_tx);
 227
 228         if (!P(blocks, disks)) {
 229                 P(blocks, disks) = scribble;
 230                 BUG_ON(len + offset > PAGE_SIZE);
 231         }
 232         if (!Q(blocks, disks)) {
 233                 Q(blocks, disks) = scribble;
 234                 BUG_ON(len + offset > PAGE_SIZE);
 235         }
 236         do_sync_gen_syndrome(blocks, offset, disks, len, submit);
 237
 238         return NULL;
 239 }
 240 EXPORT_SYMBOL_GPL(async_gen_syndrome);
 241
 242 /**
 243  * async_syndrome_val - asynchronously validate a raid6 syndrome
 244  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 245  * @offset: common offset into each block (src and dest) to start transaction
 246  * @disks: number of blocks (including missing P or Q, see below)
 247  * @len: length of operation in bytes
 248  * @pqres: on val failure SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set
 249  * @spare: temporary result buffer for the synchronous case
 250  * @submit: submission / completion modifiers
 251  *
 252  * The same notes from async_gen_syndrome apply to the 'blocks',
 253  * and 'disks' parameters of this routine.  The synchronous path
 254  * requires a temporary result buffer and submit->scribble to be
 255  * specified.
 256  */
 257 struct dma_async_tx_descriptor *
 258 async_syndrome_val(struct page **blocks, unsigned int offset, int disks,
 259                    size_t len, enum sum_check_flags *pqres, struct page *spare,
 260                    struct async_submit_ctl *submit)
 261 {
 262         struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ_VAL,
 263                                                       NULL, 0,  blocks, disks,
 264                                                       len);
 265         struct dma_device *device = chan ? chan->device : NULL;
 266         struct dma_async_tx_descriptor *tx;
 267         unsigned char coefs[disks-2];
 268         enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0;
 269         dma_addr_t *dma_src = NULL;
 270         int src_cnt = 0;
 271
 272         BUG_ON(disks < 4);
 273
 274         if (submit->scribble)
 275                 dma_src = submit->scribble;
 276         else if (sizeof(dma_addr_t) <= sizeof(struct page *))
 277                 dma_src = (dma_addr_t *) blocks;
 278
 279         if (dma_src && device && disks <= dma_maxpq(device, 0) &&
 280             is_dma_pq_aligned(device, offset, 0, len)) {
 281                 struct device *dev = device->dev;
 282                 dma_addr_t *pq = &dma_src[disks-2];
 283                 int i;
 284
 285                 pr_debug("%s: (async) disks: %d len: %zu\n",
 286                          __func__, disks, len);
 287                 if (!P(blocks, disks))
 288                         dma_flags |= DMA_PREP_PQ_DISABLE_P;
 289                 else
 290                         pq[0] = dma_map_page(dev, P(blocks, disks),
 291                                              offset, len,
 292                                              DMA_TO_DEVICE);
 293                 if (!Q(blocks, disks))
 294                         dma_flags |= DMA_PREP_PQ_DISABLE_Q;
 295                 else
 296                         pq[1] = dma_map_page(dev, Q(blocks, disks),
 297                                              offset, len,
 298                                              DMA_TO_DEVICE);
 299
 300                 if (submit->flags & ASYNC_TX_FENCE)
 301                         dma_flags |= DMA_PREP_FENCE;
 302                 for (i = 0; i < disks-2; i++)
 303                         if (likely(blocks[i])) {
 304                                 dma_src[src_cnt] = dma_map_page(dev, blocks[i],
 305                                                                 offset, len,
 306                                                                 DMA_TO_DEVICE);
 307                                 coefs[src_cnt] = raid6_gfexp[i];
 308                                 src_cnt++;
 309                         }
 310
 311                 for (;;) {
 312                         tx = device->device_prep_dma_pq_val(chan, pq, dma_src,
 313                                                             src_cnt,
 314                                                             coefs,
 315                                                             len, pqres,
 316                                                             dma_flags);
 317                         if (likely(tx))
 318                                 break;
 319                         async_tx_quiesce(&submit->depend_tx);
 320                         dma_async_issue_pending(chan);
 321                 }
 322                 async_tx_submit(chan, tx, submit);
 323
 324                 return tx;
 325         } else {
 326                 struct page *p_src = P(blocks, disks);
 327                 struct page *q_src = Q(blocks, disks);
 328                 enum async_tx_flags flags_orig = submit->flags;
 329                 dma_async_tx_callback cb_fn_orig = submit->cb_fn;
 330                 void *scribble = submit->scribble;
 331                 void *cb_param_orig = submit->cb_param;
 332                 void *p, *q, *s;
 333
 334                 pr_debug("%s: (sync) disks: %d len: %zu\n",
 335                          __func__, disks, len);
 336
 337                 /* caller must provide a temporary result buffer and
 338                  * allow the input parameters to be preserved
 339                  */
 340                 BUG_ON(!spare || !scribble);
 341
 342                 /* wait for any prerequisite operations */
 343                 async_tx_quiesce(&submit->depend_tx);
 344
 345                 /* recompute p and/or q into the temporary buffer and then
 346                  * check to see the result matches the current value
 347                  */
 348                 tx = NULL;
 349                 *pqres = 0;
 350                 if (p_src) {
 351                         init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL,
 352                                           NULL, NULL, scribble);
 353                         tx = async_xor(spare, blocks, offset, disks-2, len, submit);
 354                         async_tx_quiesce(&tx);
 355                         p = page_address(p_src) + offset;
 356                         s = page_address(spare) + offset;
 357                         *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P;
 358                 }
 359
 360                 if (q_src) {
 361                         P(blocks, disks) = NULL;
 362                         Q(blocks, disks) = spare;
 363                         init_async_submit(submit, 0, NULL, NULL, NULL, scribble);
 364                         tx = async_gen_syndrome(blocks, offset, disks, len, submit);
 365                         async_tx_quiesce(&tx);
 366                         q = page_address(q_src) + offset;
 367                         s = page_address(spare) + offset;
 368                         *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q;
 369                 }
 370
 371                 /* restore P, Q and submit */
 372                 P(blocks, disks) = p_src;
 373                 Q(blocks, disks) = q_src;
 374
 375                 submit->cb_fn = cb_fn_orig;
 376                 submit->cb_param = cb_param_orig;
 377                 submit->flags = flags_orig;
 378                 async_tx_sync_epilog(submit);
 379
 380                 return NULL;
 381         }
 382 }
 383 EXPORT_SYMBOL_GPL(async_syndrome_val);
 384
 385 static int __init async_pq_init(void)
 386 {
 387         scribble = alloc_page(GFP_KERNEL);
 388
 389         if (scribble)
 390                 return 0;
 391
 392         pr_err("%s: failed to allocate required spare page\n", __func__);
 393
 394         return -ENOMEM;
 395 }
 396
 397 static void __exit async_pq_exit(void)
 398 {
 399         put_page(scribble);
 400 }
 401
 402 module_init(async_pq_init);
 403 module_exit(async_pq_exit);
 404
 405 MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation");
 406 MODULE_LICENSE("GPL");