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#include <linux/gfp.h> 28 29/** 30 * pq_scribble_page - space to hold throwaway P or Q buffer for 31 * synchronous gen_syndrome 32 */ 33static struct page *pq_scribble_page; 34 35/* the struct page *blocks[] parameter passed to async_gen_syndrome() 36 * and async_syndrome_val() contains the 'P' destination address at 37 * blocks[disks-2] and the 'Q' destination address at blocks[disks-1] 38 * 39 * note: these are macros as they are used as lvalues 40 */ 41#define P(b, d) (b[d-2]) 42#define Q(b, d) (b[d-1]) 43 44/** 45 * do_async_gen_syndrome - asynchronously calculate P and/or Q 46 */ 47static __async_inline struct dma_async_tx_descriptor * 48do_async_gen_syndrome(struct dma_chan *chan, struct page **blocks, 49 const unsigned char *scfs, unsigned int offset, int disks, 50 size_t len, dma_addr_t *dma_src, 51 struct async_submit_ctl *submit) 52{ 53 struct dma_async_tx_descriptor *tx = NULL; 54 struct dma_device *dma = chan->device; 55 enum dma_ctrl_flags dma_flags = 0; 56 enum async_tx_flags flags_orig = submit->flags; 57 dma_async_tx_callback cb_fn_orig = submit->cb_fn; 58 dma_async_tx_callback cb_param_orig = submit->cb_param; 59 int src_cnt = disks - 2; 60 unsigned char coefs[src_cnt]; 61 unsigned short pq_src_cnt; 62 dma_addr_t dma_dest[2]; 63 int src_off = 0; 64 int idx; 65 int i; 66 67 /* DMAs use destinations as sources, so use BIDIRECTIONAL mapping */ 68 if (P(blocks, disks)) 69 dma_dest[0] = dma_map_page(dma->dev, P(blocks, disks), offset, 70 len, DMA_BIDIRECTIONAL); 71 else 72 dma_flags |= DMA_PREP_PQ_DISABLE_P; 73 if (Q(blocks, disks)) 74 dma_dest[1] = dma_map_page(dma->dev, Q(blocks, disks), offset, 75 len, DMA_BIDIRECTIONAL); 76 else 77 dma_flags |= DMA_PREP_PQ_DISABLE_Q; 78 79 /* convert source addresses being careful to collapse 'empty' 80 * sources and update the coefficients accordingly 81 */ 82 for (i = 0, idx = 0; i < src_cnt; i++) { 83 if (blocks[i] == NULL) 84 continue; 85 dma_src[idx] = dma_map_page(dma->dev, blocks[i], offset, len, 86 DMA_TO_DEVICE); 87 coefs[idx] = scfs[i]; 88 idx++; 89 } 90 src_cnt = idx; 91 92 while (src_cnt > 0) { 93 submit->flags = flags_orig; 94 pq_src_cnt = min(src_cnt, dma_maxpq(dma, dma_flags)); 95 /* if we are submitting additional pqs, leave the chain open, 96 * clear the callback parameters, and leave the destination 97 * buffers mapped 98 */ 99 if (src_cnt > pq_src_cnt) { 100 submit->flags &= ~ASYNC_TX_ACK; 101 submit->flags |= ASYNC_TX_FENCE; 102 dma_flags |= DMA_COMPL_SKIP_DEST_UNMAP; 103 submit->cb_fn = NULL; 104 submit->cb_param = NULL; 105 } else { 106 dma_flags &= ~DMA_COMPL_SKIP_DEST_UNMAP; 107 submit->cb_fn = cb_fn_orig; 108 submit->cb_param = cb_param_orig; 109 if (cb_fn_orig) 110 dma_flags |= DMA_PREP_INTERRUPT; 111 } 112 if (submit->flags & ASYNC_TX_FENCE) 113 dma_flags |= DMA_PREP_FENCE; 114 115 /* Since we have clobbered the src_list we are committed 116 * to doing this asynchronously. Drivers force forward 117 * progress in case they can not provide a descriptor 118 */ 119 for (;;) { 120 tx = dma->device_prep_dma_pq(chan, dma_dest, 121 &dma_src[src_off], 122 pq_src_cnt, 123 &coefs[src_off], len, 124 dma_flags); 125 if (likely(tx)) 126 break; 127 async_tx_quiesce(&submit->depend_tx); 128 dma_async_issue_pending(chan); 129 } 130 131 async_tx_submit(chan, tx, submit); 132 submit->depend_tx = tx; 133 134 /* drop completed sources */ 135 src_cnt -= pq_src_cnt; 136 src_off += pq_src_cnt; 137 138 dma_flags |= DMA_PREP_CONTINUE; 139 } 140 141 return tx; 142} 143 144/** 145 * do_sync_gen_syndrome - synchronously calculate a raid6 syndrome 146 */ 147static void 148do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int disks, 149 size_t len, struct async_submit_ctl *submit) 150{ 151 void **srcs; 152 int i; 153 154 if (submit->scribble) 155 srcs = submit->scribble; 156 else 157 srcs = (void **) blocks; 158 159 for (i = 0; i < disks; i++) { 160 if (blocks[i] == NULL) { 161 BUG_ON(i > disks - 3); /* P or Q can't be zero */ 162 srcs[i] = (void*)raid6_empty_zero_page; 163 } else 164 srcs[i] = page_address(blocks[i]) + offset; 165 } 166 raid6_call.gen_syndrome(disks, len, srcs); 167 async_tx_sync_epilog(submit); 168} 169 170/** 171 * async_gen_syndrome - asynchronously calculate a raid6 syndrome 172 * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1 173 * @offset: common offset into each block (src and dest) to start transaction 174 * @disks: number of blocks (including missing P or Q, see below) 175 * @len: length of operation in bytes 176 * @submit: submission/completion modifiers 177 * 178 * General note: This routine assumes a field of GF(2^8) with a 179 * primitive polynomial of 0x11d and a generator of {02}. 180 * 181 * 'disks' note: callers can optionally omit either P or Q (but not 182 * both) from the calculation by setting blocks[disks-2] or 183 * blocks[disks-1] to NULL. When P or Q is omitted 'len' must be <= 184 * PAGE_SIZE as a temporary buffer of this size is used in the 185 * synchronous path. 'disks' always accounts for both destination 186 * buffers. If any source buffers (blocks[i] where i < disks - 2) are 187 * set to NULL those buffers will be replaced with the raid6_zero_page 188 * in the synchronous path and omitted in the hardware-asynchronous 189 * path. 190 * 191 * 'blocks' note: if submit->scribble is NULL then the contents of 192 * 'blocks' may be overwritten to perform address conversions 193 * (dma_map_page() or page_address()). 194 */ 195struct dma_async_tx_descriptor * 196async_gen_syndrome(struct page **blocks, unsigned int offset, int disks, 197 size_t len, struct async_submit_ctl *submit) 198{ 199 int src_cnt = disks - 2; 200 struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ, 201 &P(blocks, disks), 2, 202 blocks, src_cnt, len); 203 struct dma_device *device = chan ? chan->device : NULL; 204 dma_addr_t *dma_src = NULL; 205 206 BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks))); 207 208 if (submit->scribble) 209 dma_src = submit->scribble; 210 else if (sizeof(dma_addr_t) <= sizeof(struct page *)) 211 dma_src = (dma_addr_t *) blocks; 212 213 if (dma_src && device && 214 (src_cnt <= dma_maxpq(device, 0) || 215 dma_maxpq(device, DMA_PREP_CONTINUE) > 0) && 216 is_dma_pq_aligned(device, offset, 0, len)) { 217 /* run the p+q asynchronously */ 218 pr_debug("%s: (async) disks: %d len: %zu\n", 219 __func__, disks, len); 220 return do_async_gen_syndrome(chan, blocks, raid6_gfexp, offset, 221 disks, len, dma_src, submit); 222 } 223 224 /* run the pq synchronously */ 225 pr_debug("%s: (sync) disks: %d len: %zu\n", __func__, disks, len); 226 227 /* wait for any prerequisite operations */ 228 async_tx_quiesce(&submit->depend_tx); 229 230 if (!P(blocks, disks)) { 231 P(blocks, disks) = pq_scribble_page; 232 BUG_ON(len + offset > PAGE_SIZE); 233 } 234 if (!Q(blocks, disks)) { 235 Q(blocks, disks) = pq_scribble_page; 236 BUG_ON(len + offset > PAGE_SIZE); 237 } 238 do_sync_gen_syndrome(blocks, offset, disks, len, submit); 239 240 return NULL; 241} 242EXPORT_SYMBOL_GPL(async_gen_syndrome); 243 244static inline struct dma_chan * 245pq_val_chan(struct async_submit_ctl *submit, struct page **blocks, int disks, size_t len) 246{ 247 #ifdef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA 248 return NULL; 249 #endif 250 return async_tx_find_channel(submit, DMA_PQ_VAL, NULL, 0, blocks, 251 disks, len); 252} 253 254/** 255 * async_syndrome_val - asynchronously validate a raid6 syndrome 256 * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1 257 * @offset: common offset into each block (src and dest) to start transaction 258 * @disks: number of blocks (including missing P or Q, see below) 259 * @len: length of operation in bytes 260 * @pqres: on val failure SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set 261 * @spare: temporary result buffer for the synchronous case 262 * @submit: submission / completion modifiers 263 * 264 * The same notes from async_gen_syndrome apply to the 'blocks', 265 * and 'disks' parameters of this routine. The synchronous path 266 * requires a temporary result buffer and submit->scribble to be 267 * specified. 268 */ 269struct dma_async_tx_descriptor * 270async_syndrome_val(struct page **blocks, unsigned int offset, int disks, 271 size_t len, enum sum_check_flags *pqres, struct page *spare, 272 struct async_submit_ctl *submit) 273{ 274 struct dma_chan *chan = pq_val_chan(submit, blocks, disks, len); 275 struct dma_device *device = chan ? chan->device : NULL; 276 struct dma_async_tx_descriptor *tx; 277 unsigned char coefs[disks-2]; 278 enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0; 279 dma_addr_t *dma_src = NULL; 280 int src_cnt = 0; 281 282 BUG_ON(disks < 4); 283 284 if (submit->scribble) 285 dma_src = submit->scribble; 286 else if (sizeof(dma_addr_t) <= sizeof(struct page *)) 287 dma_src = (dma_addr_t *) blocks; 288 289 if (dma_src && device && disks <= dma_maxpq(device, 0) && 290 is_dma_pq_aligned(device, offset, 0, len)) { 291 struct device *dev = device->dev; 292 dma_addr_t *pq = &dma_src[disks-2]; 293 int i; 294 295 pr_debug("%s: (async) disks: %d len: %zu\n", 296 __func__, disks, len); 297 if (!P(blocks, disks)) 298 dma_flags |= DMA_PREP_PQ_DISABLE_P; 299 else 300 pq[0] = dma_map_page(dev, P(blocks, disks), 301 offset, len, 302 DMA_TO_DEVICE); 303 if (!Q(blocks, disks)) 304 dma_flags |= DMA_PREP_PQ_DISABLE_Q; 305 else 306 pq[1] = dma_map_page(dev, Q(blocks, disks), 307 offset, len, 308 DMA_TO_DEVICE); 309 310 if (submit->flags & ASYNC_TX_FENCE) 311 dma_flags |= DMA_PREP_FENCE; 312 for (i = 0; i < disks-2; i++) 313 if (likely(blocks[i])) { 314 dma_src[src_cnt] = dma_map_page(dev, blocks[i], 315 offset, len, 316 DMA_TO_DEVICE); 317 coefs[src_cnt] = raid6_gfexp[i]; 318 src_cnt++; 319 } 320 321 for (;;) { 322 tx = device->device_prep_dma_pq_val(chan, pq, dma_src, 323 src_cnt, 324 coefs, 325 len, pqres, 326 dma_flags); 327 if (likely(tx)) 328 break; 329 async_tx_quiesce(&submit->depend_tx); 330 dma_async_issue_pending(chan); 331 } 332 async_tx_submit(chan, tx, submit); 333 334 return tx; 335 } else { 336 struct page *p_src = P(blocks, disks); 337 struct page *q_src = Q(blocks, disks); 338 enum async_tx_flags flags_orig = submit->flags; 339 dma_async_tx_callback cb_fn_orig = submit->cb_fn; 340 void *scribble = submit->scribble; 341 void *cb_param_orig = submit->cb_param; 342 void *p, *q, *s; 343 344 pr_debug("%s: (sync) disks: %d len: %zu\n", 345 __func__, disks, len); 346 347 /* caller must provide a temporary result buffer and 348 * allow the input parameters to be preserved 349 */ 350 BUG_ON(!spare || !scribble); 351 352 /* wait for any prerequisite operations */ 353 async_tx_quiesce(&submit->depend_tx); 354 355 /* recompute p and/or q into the temporary buffer and then 356 * check to see the result matches the current value 357 */ 358 tx = NULL; 359 *pqres = 0; 360 if (p_src) { 361 init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL, 362 NULL, NULL, scribble); 363 tx = async_xor(spare, blocks, offset, disks-2, len, submit); 364 async_tx_quiesce(&tx); 365 p = page_address(p_src) + offset; 366 s = page_address(spare) + offset; 367 *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P; 368 } 369 370 if (q_src) { 371 P(blocks, disks) = NULL; 372 Q(blocks, disks) = spare; 373 init_async_submit(submit, 0, NULL, NULL, NULL, scribble); 374 tx = async_gen_syndrome(blocks, offset, disks, len, submit); 375 async_tx_quiesce(&tx); 376 q = page_address(q_src) + offset; 377 s = page_address(spare) + offset; 378 *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q; 379 } 380 381 /* restore P, Q and submit */ 382 P(blocks, disks) = p_src; 383 Q(blocks, disks) = q_src; 384 385 submit->cb_fn = cb_fn_orig; 386 submit->cb_param = cb_param_orig; 387 submit->flags = flags_orig; 388 async_tx_sync_epilog(submit); 389 390 return NULL; 391 } 392} 393EXPORT_SYMBOL_GPL(async_syndrome_val); 394 395static int __init async_pq_init(void) 396{ 397 pq_scribble_page = alloc_page(GFP_KERNEL); 398 399 if (pq_scribble_page) 400 return 0; 401 402 pr_err("%s: failed to allocate required spare page\n", __func__); 403 404 return -ENOMEM; 405} 406 407static void __exit async_pq_exit(void) 408{ 409 put_page(pq_scribble_page); 410} 411 412module_init(async_pq_init); 413module_exit(async_pq_exit); 414 415MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation"); 416MODULE_LICENSE("GPL"); 417