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; 127 128 if (submit->scribble) 129 srcs = submit->scribble; 130 else 131 srcs = (void **) blocks; 132 133 for (i = 0; i < disks; i++) { 134 if (blocks[i] == NULL) { 135 BUG_ON(i > disks - 3); /* P or Q can't be zero */ 136 srcs[i] = (void*)raid6_empty_zero_page; 137 } else 138 srcs[i] = page_address(blocks[i]) + offset; 139 } 140 raid6_call.gen_syndrome(disks, len, srcs); 141 async_tx_sync_epilog(submit); 142 } 143 144 /** 145 * async_gen_syndrome - asynchronously calculate a raid6 syndrome 146 * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1 147 * @offset: common offset into each block (src and dest) to start transaction 148 * @disks: number of blocks (including missing P or Q, see below) 149 * @len: length of operation in bytes 150 * @submit: submission/completion modifiers 151 * 152 * General note: This routine assumes a field of GF(2^8) with a 153 * primitive polynomial of 0x11d and a generator of {02}. 154 * 155 * 'disks' note: callers can optionally omit either P or Q (but not 156 * both) from the calculation by setting blocks[disks-2] or 157 * blocks[disks-1] to NULL. When P or Q is omitted 'len' must be <= 158 * PAGE_SIZE as a temporary buffer of this size is used in the 159 * synchronous path. 'disks' always accounts for both destination 160 * buffers. If any source buffers (blocks[i] where i < disks - 2) are 161 * set to NULL those buffers will be replaced with the raid6_zero_page 162 * in the synchronous path and omitted in the hardware-asynchronous 163 * path. 164 */ 165 struct dma_async_tx_descriptor * 166 async_gen_syndrome(struct page **blocks, unsigned int offset, int disks, 167 size_t len, struct async_submit_ctl *submit) 168 { 169 int src_cnt = disks - 2; 170 struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ, 171 &P(blocks, disks), 2, 172 blocks, src_cnt, len); 173 struct dma_device *device = chan ? chan->device : NULL; 174 struct dmaengine_unmap_data *unmap = NULL; 175 176 BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks))); 177 178 if (device) 179 unmap = dmaengine_get_unmap_data(device->dev, disks, GFP_NOIO); 180 181 if (unmap && 182 (src_cnt <= dma_maxpq(device, 0) || 183 dma_maxpq(device, DMA_PREP_CONTINUE) > 0) && 184 is_dma_pq_aligned(device, offset, 0, len)) { 185 struct dma_async_tx_descriptor *tx; 186 enum dma_ctrl_flags dma_flags = 0; 187 unsigned char coefs[src_cnt]; 188 int i, j; 189 190 /* run the p+q asynchronously */ 191 pr_debug("%s: (async) disks: %d len: %zu\n", 192 __func__, disks, len); 193 194 /* convert source addresses being careful to collapse 'empty' 195 * sources and update the coefficients accordingly 196 */ 197 unmap->len = len; 198 for (i = 0, j = 0; i < src_cnt; i++) { 199 if (blocks[i] == NULL) 200 continue; 201 unmap->addr[j] = dma_map_page(device->dev, blocks[i], offset, 202 len, DMA_TO_DEVICE); 203 coefs[j] = raid6_gfexp[i]; 204 unmap->to_cnt++; 205 j++; 206 } 207 208 /* 209 * DMAs use destinations as sources, 210 * so use BIDIRECTIONAL mapping 211 */ 212 unmap->bidi_cnt++; 213 if (P(blocks, disks)) 214 unmap->addr[j++] = dma_map_page(device->dev, P(blocks, disks), 215 offset, len, DMA_BIDIRECTIONAL); 216 else { 217 unmap->addr[j++] = 0; 218 dma_flags |= DMA_PREP_PQ_DISABLE_P; 219 } 220 221 unmap->bidi_cnt++; 222 if (Q(blocks, disks)) 223 unmap->addr[j++] = dma_map_page(device->dev, Q(blocks, disks), 224 offset, len, DMA_BIDIRECTIONAL); 225 else { 226 unmap->addr[j++] = 0; 227 dma_flags |= DMA_PREP_PQ_DISABLE_Q; 228 } 229 230 tx = do_async_gen_syndrome(chan, coefs, j, unmap, dma_flags, submit); 231 dmaengine_unmap_put(unmap); 232 return tx; 233 } 234 235 dmaengine_unmap_put(unmap); 236 237 /* run the pq synchronously */ 238 pr_debug("%s: (sync) disks: %d len: %zu\n", __func__, disks, len); 239 240 /* wait for any prerequisite operations */ 241 async_tx_quiesce(&submit->depend_tx); 242 243 if (!P(blocks, disks)) { 244 P(blocks, disks) = pq_scribble_page; 245 BUG_ON(len + offset > PAGE_SIZE); 246 } 247 if (!Q(blocks, disks)) { 248 Q(blocks, disks) = pq_scribble_page; 249 BUG_ON(len + offset > PAGE_SIZE); 250 } 251 do_sync_gen_syndrome(blocks, offset, disks, len, submit); 252 253 return NULL; 254 } 255 EXPORT_SYMBOL_GPL(async_gen_syndrome); 256 257 static inline struct dma_chan * 258 pq_val_chan(struct async_submit_ctl *submit, struct page **blocks, int disks, size_t len) 259 { 260 #ifdef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA 261 return NULL; 262 #endif 263 return async_tx_find_channel(submit, DMA_PQ_VAL, NULL, 0, blocks, 264 disks, len); 265 } 266 267 /** 268 * async_syndrome_val - asynchronously validate a raid6 syndrome 269 * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1 270 * @offset: common offset into each block (src and dest) to start transaction 271 * @disks: number of blocks (including missing P or Q, see below) 272 * @len: length of operation in bytes 273 * @pqres: on val failure SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set 274 * @spare: temporary result buffer for the synchronous case 275 * @submit: submission / completion modifiers 276 * 277 * The same notes from async_gen_syndrome apply to the 'blocks', 278 * and 'disks' parameters of this routine. The synchronous path 279 * requires a temporary result buffer and submit->scribble to be 280 * specified. 281 */ 282 struct dma_async_tx_descriptor * 283 async_syndrome_val(struct page **blocks, unsigned int offset, int disks, 284 size_t len, enum sum_check_flags *pqres, struct page *spare, 285 struct async_submit_ctl *submit) 286 { 287 struct dma_chan *chan = pq_val_chan(submit, blocks, disks, len); 288 struct dma_device *device = chan ? chan->device : NULL; 289 struct dma_async_tx_descriptor *tx; 290 unsigned char coefs[disks-2]; 291 enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0; 292 struct dmaengine_unmap_data *unmap = NULL; 293 294 BUG_ON(disks < 4); 295 296 if (device) 297 unmap = dmaengine_get_unmap_data(device->dev, disks, GFP_NOIO); 298 299 if (unmap && disks <= dma_maxpq(device, 0) && 300 is_dma_pq_aligned(device, offset, 0, len)) { 301 struct device *dev = device->dev; 302 dma_addr_t pq[2]; 303 int i, j = 0, src_cnt = 0; 304 305 pr_debug("%s: (async) disks: %d len: %zu\n", 306 __func__, disks, len); 307 308 unmap->len = len; 309 for (i = 0; i < disks-2; i++) 310 if (likely(blocks[i])) { 311 unmap->addr[j] = dma_map_page(dev, blocks[i], 312 offset, len, 313 DMA_TO_DEVICE); 314 coefs[j] = raid6_gfexp[i]; 315 unmap->to_cnt++; 316 src_cnt++; 317 j++; 318 } 319 320 if (!P(blocks, disks)) { 321 pq[0] = 0; 322 dma_flags |= DMA_PREP_PQ_DISABLE_P; 323 } else { 324 pq[0] = dma_map_page(dev, P(blocks, disks), 325 offset, len, 326 DMA_TO_DEVICE); 327 unmap->addr[j++] = pq[0]; 328 unmap->to_cnt++; 329 } 330 if (!Q(blocks, disks)) { 331 pq[1] = 0; 332 dma_flags |= DMA_PREP_PQ_DISABLE_Q; 333 } else { 334 pq[1] = dma_map_page(dev, Q(blocks, disks), 335 offset, len, 336 DMA_TO_DEVICE); 337 unmap->addr[j++] = pq[1]; 338 unmap->to_cnt++; 339 } 340 341 if (submit->flags & ASYNC_TX_FENCE) 342 dma_flags |= DMA_PREP_FENCE; 343 for (;;) { 344 tx = device->device_prep_dma_pq_val(chan, pq, 345 unmap->addr, 346 src_cnt, 347 coefs, 348 len, pqres, 349 dma_flags); 350 if (likely(tx)) 351 break; 352 async_tx_quiesce(&submit->depend_tx); 353 dma_async_issue_pending(chan); 354 } 355 356 dma_set_unmap(tx, unmap); 357 async_tx_submit(chan, tx, submit); 358 359 return tx; 360 } else { 361 struct page *p_src = P(blocks, disks); 362 struct page *q_src = Q(blocks, disks); 363 enum async_tx_flags flags_orig = submit->flags; 364 dma_async_tx_callback cb_fn_orig = submit->cb_fn; 365 void *scribble = submit->scribble; 366 void *cb_param_orig = submit->cb_param; 367 void *p, *q, *s; 368 369 pr_debug("%s: (sync) disks: %d len: %zu\n", 370 __func__, disks, len); 371 372 /* caller must provide a temporary result buffer and 373 * allow the input parameters to be preserved 374 */ 375 BUG_ON(!spare || !scribble); 376 377 /* wait for any prerequisite operations */ 378 async_tx_quiesce(&submit->depend_tx); 379 380 /* recompute p and/or q into the temporary buffer and then 381 * check to see the result matches the current value 382 */ 383 tx = NULL; 384 *pqres = 0; 385 if (p_src) { 386 init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL, 387 NULL, NULL, scribble); 388 tx = async_xor(spare, blocks, offset, disks-2, len, submit); 389 async_tx_quiesce(&tx); 390 p = page_address(p_src) + offset; 391 s = page_address(spare) + offset; 392 *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P; 393 } 394 395 if (q_src) { 396 P(blocks, disks) = NULL; 397 Q(blocks, disks) = spare; 398 init_async_submit(submit, 0, NULL, NULL, NULL, scribble); 399 tx = async_gen_syndrome(blocks, offset, disks, len, submit); 400 async_tx_quiesce(&tx); 401 q = page_address(q_src) + offset; 402 s = page_address(spare) + offset; 403 *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q; 404 } 405 406 /* restore P, Q and submit */ 407 P(blocks, disks) = p_src; 408 Q(blocks, disks) = q_src; 409 410 submit->cb_fn = cb_fn_orig; 411 submit->cb_param = cb_param_orig; 412 submit->flags = flags_orig; 413 async_tx_sync_epilog(submit); 414 415 return NULL; 416 } 417 } 418 EXPORT_SYMBOL_GPL(async_syndrome_val); 419 420 static int __init async_pq_init(void) 421 { 422 pq_scribble_page = alloc_page(GFP_KERNEL); 423 424 if (pq_scribble_page) 425 return 0; 426 427 pr_err("%s: failed to allocate required spare page\n", __func__); 428 429 return -ENOMEM; 430 } 431 432 static void __exit async_pq_exit(void) 433 { 434 put_page(pq_scribble_page); 435 } 436 437 module_init(async_pq_init); 438 module_exit(async_pq_exit); 439 440 MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation"); 441 MODULE_LICENSE("GPL"); 442