crypto/async_tx/async_xor.c

   1 /*
   2  * xor offload engine api
   3  *
   4  * Copyright © 2006, Intel Corporation.
   5  *
   6  *      Dan Williams <dan.j.williams@intel.com>
   7  *
   8  *      with architecture considerations by:
   9  *      Neil Brown <neilb@suse.de>
  10  *      Jeff Garzik <jeff@garzik.org>
  11  *
  12  * This program is free software; you can redistribute it and/or modify it
  13  * under the terms and conditions of the GNU General Public License,
  14  * version 2, as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope it will be useful, but WITHOUT
  17  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  18  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  19  * more details.
  20  *
  21  * You should have received a copy of the GNU General Public License along with
  22  * this program; if not, write to the Free Software Foundation, Inc.,
  23  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  24  *
  25  */
  26 #include <linux/kernel.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/mm.h>
  29 #include <linux/dma-mapping.h>
  30 #include <linux/raid/xor.h>
  31 #include <linux/async_tx.h>
  32
  33 /* do_async_xor - dma map the pages and perform the xor with an engine.
  34  *      This routine is marked __always_inline so it can be compiled away
  35  *      when CONFIG_DMA_ENGINE=n
  36  */
  37 static __always_inline void
  38 do_async_xor(struct dma_async_tx_descriptor *tx, struct dma_device *device,
  39         struct dma_chan *chan, struct page *dest, struct page **src_list,
  40         unsigned int offset, unsigned int src_cnt, size_t len,
  41         enum async_tx_flags flags, struct dma_async_tx_descriptor *depend_tx,
  42         dma_async_tx_callback cb_fn, void *cb_param)
  43 {
  44         dma_addr_t dma_addr;
  45         enum dma_data_direction dir;
  46         int i;
  47
  48         pr_debug("%s: len: %zu\n", __FUNCTION__, len);
  49
  50         dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
  51                 DMA_NONE : DMA_FROM_DEVICE;
  52
  53         dma_addr = dma_map_page(device->dev, dest, offset, len, dir);
  54         tx->tx_set_dest(dma_addr, tx, 0);
  55
  56         dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
  57                 DMA_NONE : DMA_TO_DEVICE;
  58
  59         for (i = 0; i < src_cnt; i++) {
  60                 dma_addr = dma_map_page(device->dev, src_list[i],
  61                         offset, len, dir);
  62                 tx->tx_set_src(dma_addr, tx, i);
  63         }
  64
  65         async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param);
  66 }
  67
  68 static void
  69 do_sync_xor(struct page *dest, struct page **src_list, unsigned int offset,
  70         unsigned int src_cnt, size_t len, enum async_tx_flags flags,
  71         struct dma_async_tx_descriptor *depend_tx,
  72         dma_async_tx_callback cb_fn, void *cb_param)
  73 {
  74         void *_dest;
  75         int i;
  76
  77         pr_debug("%s: len: %zu\n", __FUNCTION__, len);
  78
  79         /* reuse the 'src_list' array to convert to buffer pointers */
  80         for (i = 0; i < src_cnt; i++)
  81                 src_list[i] = (struct page *)
  82                         (page_address(src_list[i]) + offset);
  83
  84         /* set destination address */
  85         _dest = page_address(dest) + offset;
  86
  87         if (flags & ASYNC_TX_XOR_ZERO_DST)
  88                 memset(_dest, 0, len);
  89
  90         xor_blocks(src_cnt, len, _dest,
  91                 (void **) src_list);
  92
  93         async_tx_sync_epilog(flags, depend_tx, cb_fn, cb_param);
  94 }
  95
  96 /**
  97  * async_xor - attempt to xor a set of blocks with a dma engine.
  98  *      xor_blocks always uses the dest as a source so the ASYNC_TX_XOR_ZERO_DST
  99  *      flag must be set to not include dest data in the calculation.  The
 100  *      assumption with dma eninges is that they only use the destination
 101  *      buffer as a source when it is explicity specified in the source list.
 102  * @dest: destination page
 103  * @src_list: array of source pages (if the dest is also a source it must be
 104  *      at index zero).  The contents of this array may be overwritten.
 105  * @offset: offset in pages to start transaction
 106  * @src_cnt: number of source pages
 107  * @len: length in bytes
 108  * @flags: ASYNC_TX_XOR_ZERO_DST, ASYNC_TX_XOR_DROP_DEST,
 109  *      ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
 110  * @depend_tx: xor depends on the result of this transaction.
 111  * @cb_fn: function to call when the xor completes
 112  * @cb_param: parameter to pass to the callback routine
 113  */
 114 struct dma_async_tx_descriptor *
 115 async_xor(struct page *dest, struct page **src_list, unsigned int offset,
 116         int src_cnt, size_t len, enum async_tx_flags flags,
 117         struct dma_async_tx_descriptor *depend_tx,
 118         dma_async_tx_callback cb_fn, void *cb_param)
 119 {
 120         struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_XOR);
 121         struct dma_device *device = chan ? chan->device : NULL;
 122         struct dma_async_tx_descriptor *tx = NULL;
 123         dma_async_tx_callback _cb_fn;
 124         void *_cb_param;
 125         unsigned long local_flags;
 126         int xor_src_cnt;
 127         int i = 0, src_off = 0, int_en;
 128
 129         BUG_ON(src_cnt <= 1);
 130
 131         while (src_cnt) {
 132                 local_flags = flags;
 133                 if (device) { /* run the xor asynchronously */
 134                         xor_src_cnt = min(src_cnt, device->max_xor);
 135                         /* if we are submitting additional xors
 136                          * only set the callback on the last transaction
 137                          */
 138                         if (src_cnt > xor_src_cnt) {
 139                                 local_flags &= ~ASYNC_TX_ACK;
 140                                 _cb_fn = NULL;
 141                                 _cb_param = NULL;
 142                         } else {
 143                                 _cb_fn = cb_fn;
 144                                 _cb_param = cb_param;
 145                         }
 146
 147                         int_en = _cb_fn ? 1 : 0;
 148
 149                         tx = device->device_prep_dma_xor(
 150                                 chan, xor_src_cnt, len, int_en);
 151
 152                         if (tx) {
 153                                 do_async_xor(tx, device, chan, dest,
 154                                 &src_list[src_off], offset, xor_src_cnt, len,
 155                                 local_flags, depend_tx, _cb_fn,
 156                                 _cb_param);
 157                         } else /* fall through */
 158                                 goto xor_sync;
 159                 } else { /* run the xor synchronously */
 160 xor_sync:
 161                         /* in the sync case the dest is an implied source
 162                          * (assumes the dest is at the src_off index)
 163                          */
 164                         if (flags & ASYNC_TX_XOR_DROP_DST) {
 165                                 src_cnt--;
 166                                 src_off++;
 167                         }
 168
 169                         /* process up to 'MAX_XOR_BLOCKS' sources */
 170                         xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
 171
 172                         /* if we are submitting additional xors
 173                          * only set the callback on the last transaction
 174                          */
 175                         if (src_cnt > xor_src_cnt) {
 176                                 local_flags &= ~ASYNC_TX_ACK;
 177                                 _cb_fn = NULL;
 178                                 _cb_param = NULL;
 179                         } else {
 180                                 _cb_fn = cb_fn;
 181                                 _cb_param = cb_param;
 182                         }
 183
 184                         /* wait for any prerequisite operations */
 185                         if (depend_tx) {
 186                                 /* if ack is already set then we cannot be sure
 187                                  * we are referring to the correct operation
 188                                  */
 189                                 BUG_ON(depend_tx->ack);
 190                                 if (dma_wait_for_async_tx(depend_tx) ==
 191                                         DMA_ERROR)
 192                                         panic("%s: DMA_ERROR waiting for "
 193                                                 "depend_tx\n",
 194                                                 __FUNCTION__);
 195                         }
 196
 197                         do_sync_xor(dest, &src_list[src_off], offset,
 198                                 xor_src_cnt, len, local_flags, depend_tx,
 199                                 _cb_fn, _cb_param);
 200                 }
 201
 202                 /* the previous tx is hidden from the client,
 203                  * so ack it
 204                  */
 205                 if (i && depend_tx)
 206                         async_tx_ack(depend_tx);
 207
 208                 depend_tx = tx;
 209
 210                 if (src_cnt > xor_src_cnt) {
 211                         /* drop completed sources */
 212                         src_cnt -= xor_src_cnt;
 213                         src_off += xor_src_cnt;
 214
 215                         /* unconditionally preserve the destination */
 216                         flags &= ~ASYNC_TX_XOR_ZERO_DST;
 217
 218                         /* use the intermediate result a source, but remember
 219                          * it's dropped, because it's implied, in the sync case
 220                          */
 221                         src_list[--src_off] = dest;
 222                         src_cnt++;
 223                         flags |= ASYNC_TX_XOR_DROP_DST;
 224                 } else
 225                         src_cnt = 0;
 226                 i++;
 227         }
 228
 229         return tx;
 230 }
 231 EXPORT_SYMBOL_GPL(async_xor);
 232
 233 static int page_is_zero(struct page *p, unsigned int offset, size_t len)
 234 {
 235         char *a = page_address(p) + offset;
 236         return ((*(u32 *) a) == 0 &&
 237                 memcmp(a, a + 4, len - 4) == 0);
 238 }
 239
 240 /**
 241  * async_xor_zero_sum - attempt a xor parity check with a dma engine.
 242  * @dest: destination page used if the xor is performed synchronously
 243  * @src_list: array of source pages.  The dest page must be listed as a source
 244  *      at index zero.  The contents of this array may be overwritten.
 245  * @offset: offset in pages to start transaction
 246  * @src_cnt: number of source pages
 247  * @len: length in bytes
 248  * @result: 0 if sum == 0 else non-zero
 249  * @flags: ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
 250  * @depend_tx: xor depends on the result of this transaction.
 251  * @cb_fn: function to call when the xor completes
 252  * @cb_param: parameter to pass to the callback routine
 253  */
 254 struct dma_async_tx_descriptor *
 255 async_xor_zero_sum(struct page *dest, struct page **src_list,
 256         unsigned int offset, int src_cnt, size_t len,
 257         u32 *result, enum async_tx_flags flags,
 258         struct dma_async_tx_descriptor *depend_tx,
 259         dma_async_tx_callback cb_fn, void *cb_param)
 260 {
 261         struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_ZERO_SUM);
 262         struct dma_device *device = chan ? chan->device : NULL;
 263         int int_en = cb_fn ? 1 : 0;
 264         struct dma_async_tx_descriptor *tx = device ?
 265                 device->device_prep_dma_zero_sum(chan, src_cnt, len, result,
 266                         int_en) : NULL;
 267         int i;
 268
 269         BUG_ON(src_cnt <= 1);
 270
 271         if (tx) {
 272                 dma_addr_t dma_addr;
 273                 enum dma_data_direction dir;
 274
 275                 pr_debug("%s: (async) len: %zu\n", __FUNCTION__, len);
 276
 277                 dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
 278                         DMA_NONE : DMA_TO_DEVICE;
 279
 280                 for (i = 0; i < src_cnt; i++) {
 281                         dma_addr = dma_map_page(device->dev, src_list[i],
 282                                 offset, len, dir);
 283                         tx->tx_set_src(dma_addr, tx, i);
 284                 }
 285
 286                 async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param);
 287         } else {
 288                 unsigned long xor_flags = flags;
 289
 290                 pr_debug("%s: (sync) len: %zu\n", __FUNCTION__, len);
 291
 292                 xor_flags |= ASYNC_TX_XOR_DROP_DST;
 293                 xor_flags &= ~ASYNC_TX_ACK;
 294
 295                 tx = async_xor(dest, src_list, offset, src_cnt, len, xor_flags,
 296                         depend_tx, NULL, NULL);
 297
 298                 if (tx) {
 299                         if (dma_wait_for_async_tx(tx) == DMA_ERROR)
 300                                 panic("%s: DMA_ERROR waiting for tx\n",
 301                                         __FUNCTION__);
 302                         async_tx_ack(tx);
 303                 }
 304
 305                 *result = page_is_zero(dest, offset, len) ? 0 : 1;
 306
 307                 tx = NULL;
 308
 309                 async_tx_sync_epilog(flags, depend_tx, cb_fn, cb_param);
 310         }
 311
 312         return tx;
 313 }
 314 EXPORT_SYMBOL_GPL(async_xor_zero_sum);
 315
 316 static int __init async_xor_init(void)
 317 {
 318         return 0;
 319 }
 320
 321 static void __exit async_xor_exit(void)
 322 {
 323         do { } while (0);
 324 }
 325
 326 module_init(async_xor_init);
 327 module_exit(async_xor_exit);
 328
 329 MODULE_AUTHOR("Intel Corporation");
 330 MODULE_DESCRIPTION("asynchronous xor/xor-zero-sum api");
 331 MODULE_LICENSE("GPL");