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 static bool is_raid6_zero_block(struct page *p) 34 { 35 return p == (void *) raid6_empty_zero_page; 36 } 37 38 /* the struct page *blocks[] parameter passed to async_gen_syndrome() 39 * and async_syndrome_val() contains the 'P' destination address at 40 * blocks[disks-2] and the 'Q' destination address at blocks[disks-1] 41 * 42 * note: these are macros as they are used as lvalues 43 */ 44 #define P(b, d) (b[d-2]) 45 #define Q(b, d) (b[d-1]) 46 47 /** 48 * do_async_gen_syndrome - asynchronously calculate P and/or Q 49 */ 50 static __async_inline struct dma_async_tx_descriptor * 51 do_async_gen_syndrome(struct dma_chan *chan, struct page **blocks, 52 const unsigned char *scfs, unsigned int offset, int disks, 53 size_t len, dma_addr_t *dma_src, 54 struct async_submit_ctl *submit) 55 { 56 struct dma_async_tx_descriptor *tx = NULL; 57 struct dma_device *dma = chan->device; 58 enum dma_ctrl_flags dma_flags = 0; 59 enum async_tx_flags flags_orig = submit->flags; 60 dma_async_tx_callback cb_fn_orig = submit->cb_fn; 61 dma_async_tx_callback cb_param_orig = submit->cb_param; 62 int src_cnt = disks - 2; 63 unsigned char coefs[src_cnt]; 64 unsigned short pq_src_cnt; 65 dma_addr_t dma_dest[2]; 66 int src_off = 0; 67 int idx; 68 int i; 69 70 /* DMAs use destinations as sources, so use BIDIRECTIONAL mapping */ 71 if (P(blocks, disks)) 72 dma_dest[0] = dma_map_page(dma->dev, P(blocks, disks), offset, 73 len, DMA_BIDIRECTIONAL); 74 else 75 dma_flags |= DMA_PREP_PQ_DISABLE_P; 76 if (Q(blocks, disks)) 77 dma_dest[1] = dma_map_page(dma->dev, Q(blocks, disks), offset, 78 len, DMA_BIDIRECTIONAL); 79 else 80 dma_flags |= DMA_PREP_PQ_DISABLE_Q; 81 82 /* convert source addresses being careful to collapse 'empty' 83 * sources and update the coefficients accordingly 84 */ 85 for (i = 0, idx = 0; i < src_cnt; i++) { 86 if (is_raid6_zero_block(blocks[i])) 87 continue; 88 dma_src[idx] = dma_map_page(dma->dev, blocks[i], offset, len, 89 DMA_TO_DEVICE); 90 coefs[idx] = scfs[i]; 91 idx++; 92 } 93 src_cnt = idx; 94 95 while (src_cnt > 0) { 96 submit->flags = flags_orig; 97 pq_src_cnt = min(src_cnt, dma_maxpq(dma, dma_flags)); 98 /* if we are submitting additional pqs, leave the chain open, 99 * clear the callback parameters, and leave the destination 100 * buffers mapped 101 */ 102 if (src_cnt > pq_src_cnt) { 103 submit->flags &= ~ASYNC_TX_ACK; 104 submit->flags |= ASYNC_TX_FENCE; 105 dma_flags |= DMA_COMPL_SKIP_DEST_UNMAP; 106 submit->cb_fn = NULL; 107 submit->cb_param = NULL; 108 } else { 109 dma_flags &= ~DMA_COMPL_SKIP_DEST_UNMAP; 110 submit->cb_fn = cb_fn_orig; 111 submit->cb_param = cb_param_orig; 112 if (cb_fn_orig) 113 dma_flags |= DMA_PREP_INTERRUPT; 114 } 115 if (submit->flags & ASYNC_TX_FENCE) 116 dma_flags |= DMA_PREP_FENCE; 117 118 /* Since we have clobbered the src_list we are committed 119 * to doing this asynchronously. Drivers force forward 120 * progress in case they can not provide a descriptor 121 */ 122 for (;;) { 123 tx = dma->device_prep_dma_pq(chan, dma_dest, 124 &dma_src[src_off], 125 pq_src_cnt, 126 &coefs[src_off], len, 127 dma_flags); 128 if (likely(tx)) 129 break; 130 async_tx_quiesce(&submit->depend_tx); 131 dma_async_issue_pending(chan); 132 } 133 134 async_tx_submit(chan, tx, submit); 135 submit->depend_tx = tx; 136 137 /* drop completed sources */ 138 src_cnt -= pq_src_cnt; 139 src_off += pq_src_cnt; 140 141 dma_flags |= DMA_PREP_CONTINUE; 142 } 143 144 return tx; 145 } 146 147 /** 148 * do_sync_gen_syndrome - synchronously calculate a raid6 syndrome 149 */ 150 static void 151 do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int disks, 152 size_t len, struct async_submit_ctl *submit) 153 { 154 void **srcs; 155 int i; 156 157 if (submit->scribble) 158 srcs = submit->scribble; 159 else 160 srcs = (void **) blocks; 161 162 for (i = 0; i < disks; i++) { 163 if (is_raid6_zero_block(blocks[i])) { 164 BUG_ON(i > disks - 3); /* P or Q can't be zero */ 165 srcs[i] = blocks[i]; 166 } else 167 srcs[i] = page_address(blocks[i]) + offset; 168 } 169 raid6_call.gen_syndrome(disks, len, srcs); 170 async_tx_sync_epilog(submit); 171 } 172 173 /** 174 * async_gen_syndrome - asynchronously calculate a raid6 syndrome 175 * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1 176 * @offset: common offset into each block (src and dest) to start transaction 177 * @disks: number of blocks (including missing P or Q, see below) 178 * @len: length of operation in bytes 179 * @submit: submission/completion modifiers 180 * 181 * General note: This routine assumes a field of GF(2^8) with a 182 * primitive polynomial of 0x11d and a generator of {02}. 183 * 184 * 'disks' note: callers can optionally omit either P or Q (but not 185 * both) from the calculation by setting blocks[disks-2] or 186 * blocks[disks-1] to NULL. When P or Q is omitted 'len' must be <= 187 * PAGE_SIZE as a temporary buffer of this size is used in the 188 * synchronous path. 'disks' always accounts for both destination 189 * buffers. 190 * 191 * 'blocks' note: if submit->scribble is NULL then the contents of 192 * 'blocks' may be overridden 193 */ 194 struct dma_async_tx_descriptor * 195 async_gen_syndrome(struct page **blocks, unsigned int offset, int disks, 196 size_t len, struct async_submit_ctl *submit) 197 { 198 int src_cnt = disks - 2; 199 struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ, 200 &P(blocks, disks), 2, 201 blocks, src_cnt, len); 202 struct dma_device *device = chan ? chan->device : NULL; 203 dma_addr_t *dma_src = NULL; 204 205 BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks))); 206 207 if (submit->scribble) 208 dma_src = submit->scribble; 209 else if (sizeof(dma_addr_t) <= sizeof(struct page *)) 210 dma_src = (dma_addr_t *) blocks; 211 212 if (dma_src && device && 213 (src_cnt <= dma_maxpq(device, 0) || 214 dma_maxpq(device, DMA_PREP_CONTINUE) > 0)) { 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 enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0; 268 dma_addr_t *dma_src = NULL; 269 270 BUG_ON(disks < 4); 271 272 if (submit->scribble) 273 dma_src = submit->scribble; 274 else if (sizeof(dma_addr_t) <= sizeof(struct page *)) 275 dma_src = (dma_addr_t *) blocks; 276 277 if (dma_src && device && disks <= dma_maxpq(device, 0)) { 278 struct device *dev = device->dev; 279 dma_addr_t *pq = &dma_src[disks-2]; 280 int i; 281 282 pr_debug("%s: (async) disks: %d len: %zu\n", 283 __func__, disks, len); 284 if (!P(blocks, disks)) 285 dma_flags |= DMA_PREP_PQ_DISABLE_P; 286 if (!Q(blocks, disks)) 287 dma_flags |= DMA_PREP_PQ_DISABLE_Q; 288 if (submit->flags & ASYNC_TX_FENCE) 289 dma_flags |= DMA_PREP_FENCE; 290 for (i = 0; i < disks; i++) 291 if (likely(blocks[i])) { 292 BUG_ON(is_raid6_zero_block(blocks[i])); 293 dma_src[i] = dma_map_page(dev, blocks[i], 294 offset, len, 295 DMA_TO_DEVICE); 296 } 297 298 for (;;) { 299 tx = device->device_prep_dma_pq_val(chan, pq, dma_src, 300 disks - 2, 301 raid6_gfexp, 302 len, pqres, 303 dma_flags); 304 if (likely(tx)) 305 break; 306 async_tx_quiesce(&submit->depend_tx); 307 dma_async_issue_pending(chan); 308 } 309 async_tx_submit(chan, tx, submit); 310 311 return tx; 312 } else { 313 struct page *p_src = P(blocks, disks); 314 struct page *q_src = Q(blocks, disks); 315 enum async_tx_flags flags_orig = submit->flags; 316 dma_async_tx_callback cb_fn_orig = submit->cb_fn; 317 void *scribble = submit->scribble; 318 void *cb_param_orig = submit->cb_param; 319 void *p, *q, *s; 320 321 pr_debug("%s: (sync) disks: %d len: %zu\n", 322 __func__, disks, len); 323 324 /* caller must provide a temporary result buffer and 325 * allow the input parameters to be preserved 326 */ 327 BUG_ON(!spare || !scribble); 328 329 /* wait for any prerequisite operations */ 330 async_tx_quiesce(&submit->depend_tx); 331 332 /* recompute p and/or q into the temporary buffer and then 333 * check to see the result matches the current value 334 */ 335 tx = NULL; 336 *pqres = 0; 337 if (p_src) { 338 init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL, 339 NULL, NULL, scribble); 340 tx = async_xor(spare, blocks, offset, disks-2, len, submit); 341 async_tx_quiesce(&tx); 342 p = page_address(p_src) + offset; 343 s = page_address(spare) + offset; 344 *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P; 345 } 346 347 if (q_src) { 348 P(blocks, disks) = NULL; 349 Q(blocks, disks) = spare; 350 init_async_submit(submit, 0, NULL, NULL, NULL, scribble); 351 tx = async_gen_syndrome(blocks, offset, disks, len, submit); 352 async_tx_quiesce(&tx); 353 q = page_address(q_src) + offset; 354 s = page_address(spare) + offset; 355 *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q; 356 } 357 358 /* restore P, Q and submit */ 359 P(blocks, disks) = p_src; 360 Q(blocks, disks) = q_src; 361 362 submit->cb_fn = cb_fn_orig; 363 submit->cb_param = cb_param_orig; 364 submit->flags = flags_orig; 365 async_tx_sync_epilog(submit); 366 367 return NULL; 368 } 369 } 370 EXPORT_SYMBOL_GPL(async_syndrome_val); 371 372 static int __init async_pq_init(void) 373 { 374 scribble = alloc_page(GFP_KERNEL); 375 376 if (scribble) 377 return 0; 378 379 pr_err("%s: failed to allocate required spare page\n", __func__); 380 381 return -ENOMEM; 382 } 383 384 static void __exit async_pq_exit(void) 385 { 386 put_page(scribble); 387 } 388 389 module_init(async_pq_init); 390 module_exit(async_pq_exit); 391 392 MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation"); 393 MODULE_LICENSE("GPL"); 394