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, struct page **blocks,
  50                       const unsigned char *scfs, unsigned int offset, int disks,
  51                       size_t len, dma_addr_t *dma_src,
  52                       struct async_submit_ctl *submit)
  53 {
  54         struct dma_async_tx_descriptor *tx = NULL;
  55         struct dma_device *dma = chan->device;
  56         enum dma_ctrl_flags dma_flags = 0;
  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 char coefs[src_cnt];
  62         unsigned short pq_src_cnt;
  63         dma_addr_t dma_dest[2];
  64         int src_off = 0;
  65         int idx;
  66         int i;
  67
  68         /* DMAs use destinations as sources, so use BIDIRECTIONAL mapping */
  69         if (P(blocks, disks))
  70                 dma_dest[0] = dma_map_page(dma->dev, P(blocks, disks), offset,
  71                                            len, DMA_BIDIRECTIONAL);
  72         else
  73                 dma_flags |= DMA_PREP_PQ_DISABLE_P;
  74         if (Q(blocks, disks))
  75                 dma_dest[1] = dma_map_page(dma->dev, Q(blocks, disks), offset,
  76                                            len, DMA_BIDIRECTIONAL);
  77         else
  78                 dma_flags |= DMA_PREP_PQ_DISABLE_Q;
  79
  80         /* convert source addresses being careful to collapse 'empty'
  81          * sources and update the coefficients accordingly
  82          */
  83         for (i = 0, idx = 0; i < src_cnt; i++) {
  84                 if (blocks[i] == NULL)
  85                         continue;
  86                 dma_src[idx] = dma_map_page(dma->dev, blocks[i], offset, len,
  87                                             DMA_TO_DEVICE);
  88                 coefs[idx] = scfs[i];
  89                 idx++;
  90         }
  91         src_cnt = idx;
  92
  93         while (src_cnt > 0) {
  94                 submit->flags = flags_orig;
  95                 pq_src_cnt = min(src_cnt, dma_maxpq(dma, dma_flags));
  96                 /* if we are submitting additional pqs, leave the chain open,
  97                  * clear the callback parameters, and leave the destination
  98                  * buffers mapped
  99                  */
 100                 if (src_cnt > pq_src_cnt) {
 101                         submit->flags &= ~ASYNC_TX_ACK;
 102                         submit->flags |= ASYNC_TX_FENCE;
 103                         dma_flags |= DMA_COMPL_SKIP_DEST_UNMAP;
 104                         submit->cb_fn = NULL;
 105                         submit->cb_param = NULL;
 106                 } else {
 107                         dma_flags &= ~DMA_COMPL_SKIP_DEST_UNMAP;
 108                         submit->cb_fn = cb_fn_orig;
 109                         submit->cb_param = cb_param_orig;
 110                         if (cb_fn_orig)
 111                                 dma_flags |= DMA_PREP_INTERRUPT;
 112                 }
 113                 if (submit->flags & ASYNC_TX_FENCE)
 114                         dma_flags |= DMA_PREP_FENCE;
 115
 116                 /* Since we have clobbered the src_list we are committed
 117                  * to doing this asynchronously.  Drivers force forward
 118                  * progress in case they can not provide a descriptor
 119                  */
 120                 for (;;) {
 121                         tx = dma->device_prep_dma_pq(chan, dma_dest,
 122                                                      &dma_src[src_off],
 123                                                      pq_src_cnt,
 124                                                      &coefs[src_off], len,
 125                                                      dma_flags);
 126                         if (likely(tx))
 127                                 break;
 128                         async_tx_quiesce(&submit->depend_tx);
 129                         dma_async_issue_pending(chan);
 130                 }
 131
 132                 async_tx_submit(chan, tx, submit);
 133                 submit->depend_tx = tx;
 134
 135                 /* drop completed sources */
 136                 src_cnt -= pq_src_cnt;
 137                 src_off += pq_src_cnt;
 138
 139                 dma_flags |= DMA_PREP_CONTINUE;
 140         }
 141
 142         return tx;
 143 }
 144
 145 /**
 146  * do_sync_gen_syndrome - synchronously calculate a raid6 syndrome
 147  */
 148 static void
 149 do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 150                      size_t len, struct async_submit_ctl *submit)
 151 {
 152         void **srcs;
 153         int i;
 154
 155         if (submit->scribble)
 156                 srcs = submit->scribble;
 157         else
 158                 srcs = (void **) blocks;
 159
 160         for (i = 0; i < disks; i++) {
 161                 if (blocks[i] == NULL) {
 162                         BUG_ON(i > disks - 3); /* P or Q can't be zero */
 163                         srcs[i] = (void*)raid6_empty_zero_page;
 164                 } else
 165                         srcs[i] = page_address(blocks[i]) + offset;
 166         }
 167         raid6_call.gen_syndrome(disks, len, srcs);
 168         async_tx_sync_epilog(submit);
 169 }
 170
 171 /**
 172  * async_gen_syndrome - asynchronously calculate a raid6 syndrome
 173  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 174  * @offset: common offset into each block (src and dest) to start transaction
 175  * @disks: number of blocks (including missing P or Q, see below)
 176  * @len: length of operation in bytes
 177  * @submit: submission/completion modifiers
 178  *
 179  * General note: This routine assumes a field of GF(2^8) with a
 180  * primitive polynomial of 0x11d and a generator of {02}.
 181  *
 182  * 'disks' note: callers can optionally omit either P or Q (but not
 183  * both) from the calculation by setting blocks[disks-2] or
 184  * blocks[disks-1] to NULL.  When P or Q is omitted 'len' must be <=
 185  * PAGE_SIZE as a temporary buffer of this size is used in the
 186  * synchronous path.  'disks' always accounts for both destination
 187  * buffers.  If any source buffers (blocks[i] where i < disks - 2) are
 188  * set to NULL those buffers will be replaced with the raid6_zero_page
 189  * in the synchronous path and omitted in the hardware-asynchronous
 190  * path.
 191  *
 192  * 'blocks' note: if submit->scribble is NULL then the contents of
 193  * 'blocks' may be overwritten to perform address conversions
 194  * (dma_map_page() or page_address()).
 195  */
 196 struct dma_async_tx_descriptor *
 197 async_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
 198                    size_t len, struct async_submit_ctl *submit)
 199 {
 200         int src_cnt = disks - 2;
 201         struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
 202                                                       &P(blocks, disks), 2,
 203                                                       blocks, src_cnt, len);
 204         struct dma_device *device = chan ? chan->device : NULL;
 205         dma_addr_t *dma_src = NULL;
 206
 207         BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks)));
 208
 209         if (submit->scribble)
 210                 dma_src = submit->scribble;
 211         else if (sizeof(dma_addr_t) <= sizeof(struct page *))
 212                 dma_src = (dma_addr_t *) blocks;
 213
 214         if (dma_src && device &&
 215             (src_cnt <= dma_maxpq(device, 0) ||
 216              dma_maxpq(device, DMA_PREP_CONTINUE) > 0) &&
 217             is_dma_pq_aligned(device, offset, 0, len)) {
 218                 /* run the p+q asynchronously */
 219                 pr_debug("%s: (async) disks: %d len: %zu\n",
 220                          __func__, disks, len);
 221                 return do_async_gen_syndrome(chan, blocks, raid6_gfexp, offset,
 222                                              disks, len, dma_src, submit);
 223         }
 224
 225         /* run the pq synchronously */
 226         pr_debug("%s: (sync) disks: %d len: %zu\n", __func__, disks, len);
 227
 228         /* wait for any prerequisite operations */
 229         async_tx_quiesce(&submit->depend_tx);
 230
 231         if (!P(blocks, disks)) {
 232                 P(blocks, disks) = pq_scribble_page;
 233                 BUG_ON(len + offset > PAGE_SIZE);
 234         }
 235         if (!Q(blocks, disks)) {
 236                 Q(blocks, disks) = pq_scribble_page;
 237                 BUG_ON(len + offset > PAGE_SIZE);
 238         }
 239         do_sync_gen_syndrome(blocks, offset, disks, len, submit);
 240
 241         return NULL;
 242 }
 243 EXPORT_SYMBOL_GPL(async_gen_syndrome);
 244
 245 static inline struct dma_chan *
 246 pq_val_chan(struct async_submit_ctl *submit, struct page **blocks, int disks, size_t len)
 247 {
 248         #ifdef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA
 249         return NULL;
 250         #endif
 251         return async_tx_find_channel(submit, DMA_PQ_VAL, NULL, 0,  blocks,
 252                                      disks, len);
 253 }
 254
 255 /**
 256  * async_syndrome_val - asynchronously validate a raid6 syndrome
 257  * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1
 258  * @offset: common offset into each block (src and dest) to start transaction
 259  * @disks: number of blocks (including missing P or Q, see below)
 260  * @len: length of operation in bytes
 261  * @pqres: on val failure SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set
 262  * @spare: temporary result buffer for the synchronous case
 263  * @submit: submission / completion modifiers
 264  *
 265  * The same notes from async_gen_syndrome apply to the 'blocks',
 266  * and 'disks' parameters of this routine.  The synchronous path
 267  * requires a temporary result buffer and submit->scribble to be
 268  * specified.
 269  */
 270 struct dma_async_tx_descriptor *
 271 async_syndrome_val(struct page **blocks, unsigned int offset, int disks,
 272                    size_t len, enum sum_check_flags *pqres, struct page *spare,
 273                    struct async_submit_ctl *submit)
 274 {
 275         struct dma_chan *chan = pq_val_chan(submit, blocks, disks, len);
 276         struct dma_device *device = chan ? chan->device : NULL;
 277         struct dma_async_tx_descriptor *tx;
 278         unsigned char coefs[disks-2];
 279         enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0;
 280         dma_addr_t *dma_src = NULL;
 281         int src_cnt = 0;
 282
 283         BUG_ON(disks < 4);
 284
 285         if (submit->scribble)
 286                 dma_src = submit->scribble;
 287         else if (sizeof(dma_addr_t) <= sizeof(struct page *))
 288                 dma_src = (dma_addr_t *) blocks;
 289
 290         if (dma_src && device && disks <= dma_maxpq(device, 0) &&
 291             is_dma_pq_aligned(device, offset, 0, len)) {
 292                 struct device *dev = device->dev;
 293                 dma_addr_t *pq = &dma_src[disks-2];
 294                 int i;
 295
 296                 pr_debug("%s: (async) disks: %d len: %zu\n",
 297                          __func__, disks, len);
 298                 if (!P(blocks, disks))
 299                         dma_flags |= DMA_PREP_PQ_DISABLE_P;
 300                 else
 301                         pq[0] = dma_map_page(dev, P(blocks, disks),
 302                                              offset, len,
 303                                              DMA_TO_DEVICE);
 304                 if (!Q(blocks, disks))
 305                         dma_flags |= DMA_PREP_PQ_DISABLE_Q;
 306                 else
 307                         pq[1] = dma_map_page(dev, Q(blocks, disks),
 308                                              offset, len,
 309                                              DMA_TO_DEVICE);
 310
 311                 if (submit->flags & ASYNC_TX_FENCE)
 312                         dma_flags |= DMA_PREP_FENCE;
 313                 for (i = 0; i < disks-2; i++)
 314                         if (likely(blocks[i])) {
 315                                 dma_src[src_cnt] = dma_map_page(dev, blocks[i],
 316                                                                 offset, len,
 317                                                                 DMA_TO_DEVICE);
 318                                 coefs[src_cnt] = raid6_gfexp[i];
 319                                 src_cnt++;
 320                         }
 321
 322                 for (;;) {
 323                         tx = device->device_prep_dma_pq_val(chan, pq, dma_src,
 324                                                             src_cnt,
 325                                                             coefs,
 326                                                             len, pqres,
 327                                                             dma_flags);
 328                         if (likely(tx))
 329                                 break;
 330                         async_tx_quiesce(&submit->depend_tx);
 331                         dma_async_issue_pending(chan);
 332                 }
 333                 async_tx_submit(chan, tx, submit);
 334
 335                 return tx;
 336         } else {
 337                 struct page *p_src = P(blocks, disks);
 338                 struct page *q_src = Q(blocks, disks);
 339                 enum async_tx_flags flags_orig = submit->flags;
 340                 dma_async_tx_callback cb_fn_orig = submit->cb_fn;
 341                 void *scribble = submit->scribble;
 342                 void *cb_param_orig = submit->cb_param;
 343                 void *p, *q, *s;
 344
 345                 pr_debug("%s: (sync) disks: %d len: %zu\n",
 346                          __func__, disks, len);
 347
 348                 /* caller must provide a temporary result buffer and
 349                  * allow the input parameters to be preserved
 350                  */
 351                 BUG_ON(!spare || !scribble);
 352
 353                 /* wait for any prerequisite operations */
 354                 async_tx_quiesce(&submit->depend_tx);
 355
 356                 /* recompute p and/or q into the temporary buffer and then
 357                  * check to see the result matches the current value
 358                  */
 359                 tx = NULL;
 360                 *pqres = 0;
 361                 if (p_src) {
 362                         init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL,
 363                                           NULL, NULL, scribble);
 364                         tx = async_xor(spare, blocks, offset, disks-2, len, submit);
 365                         async_tx_quiesce(&tx);
 366                         p = page_address(p_src) + offset;
 367                         s = page_address(spare) + offset;
 368                         *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P;
 369                 }
 370
 371                 if (q_src) {
 372                         P(blocks, disks) = NULL;
 373                         Q(blocks, disks) = spare;
 374                         init_async_submit(submit, 0, NULL, NULL, NULL, scribble);
 375                         tx = async_gen_syndrome(blocks, offset, disks, len, submit);
 376                         async_tx_quiesce(&tx);
 377                         q = page_address(q_src) + offset;
 378                         s = page_address(spare) + offset;
 379                         *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q;
 380                 }
 381
 382                 /* restore P, Q and submit */
 383                 P(blocks, disks) = p_src;
 384                 Q(blocks, disks) = q_src;
 385
 386                 submit->cb_fn = cb_fn_orig;
 387                 submit->cb_param = cb_param_orig;
 388                 submit->flags = flags_orig;
 389                 async_tx_sync_epilog(submit);
 390
 391                 return NULL;
 392         }
 393 }
 394 EXPORT_SYMBOL_GPL(async_syndrome_val);
 395
 396 static int __init async_pq_init(void)
 397 {
 398         pq_scribble_page = alloc_page(GFP_KERNEL);
 399
 400         if (pq_scribble_page)
 401                 return 0;
 402
 403         pr_err("%s: failed to allocate required spare page\n", __func__);
 404
 405         return -ENOMEM;
 406 }
 407
 408 static void __exit async_pq_exit(void)
 409 {
 410         put_page(pq_scribble_page);
 411 }
 412
 413 module_init(async_pq_init);
 414 module_exit(async_pq_exit);
 415
 416 MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation");
 417 MODULE_LICENSE("GPL");