1/* 2 * Implement fast CRC32C with PCLMULQDQ instructions. (x86_64) 3 * 4 * The white papers on CRC32C calculations with PCLMULQDQ instruction can be 5 * downloaded from: 6 * http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/crc-iscsi-polynomial-crc32-instruction-paper.pdf 7 * http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-crc-computation-paper.pdf 8 * 9 * Copyright (C) 2012 Intel Corporation. 10 * 11 * Authors: 12 * Wajdi Feghali <wajdi.k.feghali@intel.com> 13 * James Guilford <james.guilford@intel.com> 14 * David Cote <david.m.cote@intel.com> 15 * Tim Chen <tim.c.chen@linux.intel.com> 16 * 17 * This software is available to you under a choice of one of two 18 * licenses. You may choose to be licensed under the terms of the GNU 19 * General Public License (GPL) Version 2, available from the file 20 * COPYING in the main directory of this source tree, or the 21 * OpenIB.org BSD license below: 22 * 23 * Redistribution and use in source and binary forms, with or 24 * without modification, are permitted provided that the following 25 * conditions are met: 26 * 27 * - Redistributions of source code must retain the above 28 * copyright notice, this list of conditions and the following 29 * disclaimer. 30 * 31 * - Redistributions in binary form must reproduce the above 32 * copyright notice, this list of conditions and the following 33 * disclaimer in the documentation and/or other materials 34 * provided with the distribution. 35 * 36 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 37 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 38 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 39 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 40 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 41 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 42 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 43 * SOFTWARE. 44 */ 45 46#include <linux/linkage.h> 47 48## ISCSI CRC 32 Implementation with crc32 and pclmulqdq Instruction 49 50.macro LABEL prefix n 51\prefix\n\(): 52.endm 53 54.macro JMPTBL_ENTRY i 55.word crc_\i - crc_array 56.endm 57 58.macro JNC_LESS_THAN j 59 jnc less_than_\j 60.endm 61 62# Define threshold where buffers are considered "small" and routed to more 63# efficient "by-1" code. This "by-1" code only handles up to 255 bytes, so 64# SMALL_SIZE can be no larger than 255. 65 66#define SMALL_SIZE 200 67 68.if (SMALL_SIZE > 255) 69.error "SMALL_ SIZE must be < 256" 70.endif 71 72# unsigned int crc_pcl(u8 *buffer, int len, unsigned int crc_init); 73 74ENTRY(crc_pcl) 75#define bufp %rdi 76#define bufp_dw %edi 77#define bufp_w %di 78#define bufp_b %dil 79#define bufptmp %rcx 80#define block_0 %rcx 81#define block_1 %rdx 82#define block_2 %r11 83#define len %rsi 84#define len_dw %esi 85#define len_w %si 86#define len_b %sil 87#define crc_init_arg %rdx 88#define tmp %rbx 89#define crc_init %r8 90#define crc_init_dw %r8d 91#define crc1 %r9 92#define crc2 %r10 93 94 pushq %rbx 95 pushq %rdi 96 pushq %rsi 97 98 ## Move crc_init for Linux to a different 99 mov crc_init_arg, crc_init 100 101 ################################################################ 102 ## 1) ALIGN: 103 ################################################################ 104 105 mov bufp, bufptmp # rdi = *buf 106 neg bufp 107 and $7, bufp # calculate the unalignment amount of 108 # the address 109 je proc_block # Skip if aligned 110 111 ## If len is less than 8 and we're unaligned, we need to jump 112 ## to special code to avoid reading beyond the end of the buffer 113 cmp $8, len 114 jae do_align 115 # less_than_8 expects length in upper 3 bits of len_dw 116 # less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30] 117 shl $32-3+1, len_dw 118 jmp less_than_8_post_shl1 119 120do_align: 121 #### Calculate CRC of unaligned bytes of the buffer (if any) 122 movq (bufptmp), tmp # load a quadward from the buffer 123 add bufp, bufptmp # align buffer pointer for quadword 124 # processing 125 sub bufp, len # update buffer length 126align_loop: 127 crc32b %bl, crc_init_dw # compute crc32 of 1-byte 128 shr $8, tmp # get next byte 129 dec bufp 130 jne align_loop 131 132proc_block: 133 134 ################################################################ 135 ## 2) PROCESS BLOCKS: 136 ################################################################ 137 138 ## compute num of bytes to be processed 139 movq len, tmp # save num bytes in tmp 140 141 cmpq $128*24, len 142 jae full_block 143 144continue_block: 145 cmpq $SMALL_SIZE, len 146 jb small 147 148 ## len < 128*24 149 movq $2731, %rax # 2731 = ceil(2^16 / 24) 150 mul len_dw 151 shrq $16, %rax 152 153 ## eax contains floor(bytes / 24) = num 24-byte chunks to do 154 155 ## process rax 24-byte chunks (128 >= rax >= 0) 156 157 ## compute end address of each block 158 ## block 0 (base addr + RAX * 8) 159 ## block 1 (base addr + RAX * 16) 160 ## block 2 (base addr + RAX * 24) 161 lea (bufptmp, %rax, 8), block_0 162 lea (block_0, %rax, 8), block_1 163 lea (block_1, %rax, 8), block_2 164 165 xor crc1, crc1 166 xor crc2, crc2 167 168 ## branch into array 169 lea jump_table(%rip), bufp 170 movzxw (bufp, %rax, 2), len 171 offset=crc_array-jump_table 172 lea offset(bufp, len, 1), bufp 173 jmp *bufp 174 175 ################################################################ 176 ## 2a) PROCESS FULL BLOCKS: 177 ################################################################ 178full_block: 179 movq $128,%rax 180 lea 128*8*2(block_0), block_1 181 lea 128*8*3(block_0), block_2 182 add $128*8*1, block_0 183 184 xor crc1,crc1 185 xor crc2,crc2 186 187 # Fall thruogh into top of crc array (crc_128) 188 189 ################################################################ 190 ## 3) CRC Array: 191 ################################################################ 192 193crc_array: 194 i=128 195.rept 128-1 196.altmacro 197LABEL crc_ %i 198.noaltmacro 199 crc32q -i*8(block_0), crc_init 200 crc32q -i*8(block_1), crc1 201 crc32q -i*8(block_2), crc2 202 i=(i-1) 203.endr 204 205.altmacro 206LABEL crc_ %i 207.noaltmacro 208 crc32q -i*8(block_0), crc_init 209 crc32q -i*8(block_1), crc1 210# SKIP crc32 -i*8(block_2), crc2 ; Don't do this one yet 211 212 mov block_2, block_0 213 214 ################################################################ 215 ## 4) Combine three results: 216 ################################################################ 217 218 lea (K_table-16)(%rip), bufp # first entry is for idx 1 219 shlq $3, %rax # rax *= 8 220 subq %rax, tmp # tmp -= rax*8 221 shlq $1, %rax 222 subq %rax, tmp # tmp -= rax*16 223 # (total tmp -= rax*24) 224 addq %rax, bufp 225 226 movdqa (bufp), %xmm0 # 2 consts: K1:K2 227 228 movq crc_init, %xmm1 # CRC for block 1 229 pclmulqdq $0x00,%xmm0,%xmm1 # Multiply by K2 230 231 movq crc1, %xmm2 # CRC for block 2 232 pclmulqdq $0x10, %xmm0, %xmm2 # Multiply by K1 233 234 pxor %xmm2,%xmm1 235 movq %xmm1, %rax 236 xor -i*8(block_2), %rax 237 mov crc2, crc_init 238 crc32 %rax, crc_init 239 240################################################################ 241## 5) Check for end: 242################################################################ 243 244LABEL crc_ 0 245 mov tmp, len 246 cmp $128*24, tmp 247 jae full_block 248 cmp $24, tmp 249 jae continue_block 250 251less_than_24: 252 shl $32-4, len_dw # less_than_16 expects length 253 # in upper 4 bits of len_dw 254 jnc less_than_16 255 crc32q (bufptmp), crc_init 256 crc32q 8(bufptmp), crc_init 257 jz do_return 258 add $16, bufptmp 259 # len is less than 8 if we got here 260 # less_than_8 expects length in upper 3 bits of len_dw 261 # less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30] 262 shl $2, len_dw 263 jmp less_than_8_post_shl1 264 265 ####################################################################### 266 ## 6) LESS THAN 256-bytes REMAIN AT THIS POINT (8-bits of len are full) 267 ####################################################################### 268small: 269 shl $32-8, len_dw # Prepare len_dw for less_than_256 270 j=256 271.rept 5 # j = {256, 128, 64, 32, 16} 272.altmacro 273LABEL less_than_ %j # less_than_j: Length should be in 274 # upper lg(j) bits of len_dw 275 j=(j/2) 276 shl $1, len_dw # Get next MSB 277 JNC_LESS_THAN %j 278.noaltmacro 279 i=0 280.rept (j/8) 281 crc32q i(bufptmp), crc_init # Compute crc32 of 8-byte data 282 i=i+8 283.endr 284 jz do_return # Return if remaining length is zero 285 add $j, bufptmp # Advance buf 286.endr 287 288less_than_8: # Length should be stored in 289 # upper 3 bits of len_dw 290 shl $1, len_dw 291less_than_8_post_shl1: 292 jnc less_than_4 293 crc32l (bufptmp), crc_init_dw # CRC of 4 bytes 294 jz do_return # return if remaining data is zero 295 add $4, bufptmp 296less_than_4: # Length should be stored in 297 # upper 2 bits of len_dw 298 shl $1, len_dw 299 jnc less_than_2 300 crc32w (bufptmp), crc_init_dw # CRC of 2 bytes 301 jz do_return # return if remaining data is zero 302 add $2, bufptmp 303less_than_2: # Length should be stored in the MSB 304 # of len_dw 305 shl $1, len_dw 306 jnc less_than_1 307 crc32b (bufptmp), crc_init_dw # CRC of 1 byte 308less_than_1: # Length should be zero 309do_return: 310 movq crc_init, %rax 311 popq %rsi 312 popq %rdi 313 popq %rbx 314 ret 315 316 ################################################################ 317 ## jump table Table is 129 entries x 2 bytes each 318 ################################################################ 319.align 4 320jump_table: 321 i=0 322.rept 129 323.altmacro 324JMPTBL_ENTRY %i 325.noaltmacro 326 i=i+1 327.endr 328 329ENDPROC(crc_pcl) 330 331 ################################################################ 332 ## PCLMULQDQ tables 333 ## Table is 128 entries x 2 quad words each 334 ################################################################ 335.data 336.align 64 337K_table: 338 .quad 0x14cd00bd6,0x105ec76f0 339 .quad 0x0ba4fc28e,0x14cd00bd6 340 .quad 0x1d82c63da,0x0f20c0dfe 341 .quad 0x09e4addf8,0x0ba4fc28e 342 .quad 0x039d3b296,0x1384aa63a 343 .quad 0x102f9b8a2,0x1d82c63da 344 .quad 0x14237f5e6,0x01c291d04 345 .quad 0x00d3b6092,0x09e4addf8 346 .quad 0x0c96cfdc0,0x0740eef02 347 .quad 0x18266e456,0x039d3b296 348 .quad 0x0daece73e,0x0083a6eec 349 .quad 0x0ab7aff2a,0x102f9b8a2 350 .quad 0x1248ea574,0x1c1733996 351 .quad 0x083348832,0x14237f5e6 352 .quad 0x12c743124,0x02ad91c30 353 .quad 0x0b9e02b86,0x00d3b6092 354 .quad 0x018b33a4e,0x06992cea2 355 .quad 0x1b331e26a,0x0c96cfdc0 356 .quad 0x17d35ba46,0x07e908048 357 .quad 0x1bf2e8b8a,0x18266e456 358 .quad 0x1a3e0968a,0x11ed1f9d8 359 .quad 0x0ce7f39f4,0x0daece73e 360 .quad 0x061d82e56,0x0f1d0f55e 361 .quad 0x0d270f1a2,0x0ab7aff2a 362 .quad 0x1c3f5f66c,0x0a87ab8a8 363 .quad 0x12ed0daac,0x1248ea574 364 .quad 0x065863b64,0x08462d800 365 .quad 0x11eef4f8e,0x083348832 366 .quad 0x1ee54f54c,0x071d111a8 367 .quad 0x0b3e32c28,0x12c743124 368 .quad 0x0064f7f26,0x0ffd852c6 369 .quad 0x0dd7e3b0c,0x0b9e02b86 370 .quad 0x0f285651c,0x0dcb17aa4 371 .quad 0x010746f3c,0x018b33a4e 372 .quad 0x1c24afea4,0x0f37c5aee 373 .quad 0x0271d9844,0x1b331e26a 374 .quad 0x08e766a0c,0x06051d5a2 375 .quad 0x093a5f730,0x17d35ba46 376 .quad 0x06cb08e5c,0x11d5ca20e 377 .quad 0x06b749fb2,0x1bf2e8b8a 378 .quad 0x1167f94f2,0x021f3d99c 379 .quad 0x0cec3662e,0x1a3e0968a 380 .quad 0x19329634a,0x08f158014 381 .quad 0x0e6fc4e6a,0x0ce7f39f4 382 .quad 0x08227bb8a,0x1a5e82106 383 .quad 0x0b0cd4768,0x061d82e56 384 .quad 0x13c2b89c4,0x188815ab2 385 .quad 0x0d7a4825c,0x0d270f1a2 386 .quad 0x10f5ff2ba,0x105405f3e 387 .quad 0x00167d312,0x1c3f5f66c 388 .quad 0x0f6076544,0x0e9adf796 389 .quad 0x026f6a60a,0x12ed0daac 390 .quad 0x1a2adb74e,0x096638b34 391 .quad 0x19d34af3a,0x065863b64 392 .quad 0x049c3cc9c,0x1e50585a0 393 .quad 0x068bce87a,0x11eef4f8e 394 .quad 0x1524fa6c6,0x19f1c69dc 395 .quad 0x16cba8aca,0x1ee54f54c 396 .quad 0x042d98888,0x12913343e 397 .quad 0x1329d9f7e,0x0b3e32c28 398 .quad 0x1b1c69528,0x088f25a3a 399 .quad 0x02178513a,0x0064f7f26 400 .quad 0x0e0ac139e,0x04e36f0b0 401 .quad 0x0170076fa,0x0dd7e3b0c 402 .quad 0x141a1a2e2,0x0bd6f81f8 403 .quad 0x16ad828b4,0x0f285651c 404 .quad 0x041d17b64,0x19425cbba 405 .quad 0x1fae1cc66,0x010746f3c 406 .quad 0x1a75b4b00,0x18db37e8a 407 .quad 0x0f872e54c,0x1c24afea4 408 .quad 0x01e41e9fc,0x04c144932 409 .quad 0x086d8e4d2,0x0271d9844 410 .quad 0x160f7af7a,0x052148f02 411 .quad 0x05bb8f1bc,0x08e766a0c 412 .quad 0x0a90fd27a,0x0a3c6f37a 413 .quad 0x0b3af077a,0x093a5f730 414 .quad 0x04984d782,0x1d22c238e 415 .quad 0x0ca6ef3ac,0x06cb08e5c 416 .quad 0x0234e0b26,0x063ded06a 417 .quad 0x1d88abd4a,0x06b749fb2 418 .quad 0x04597456a,0x04d56973c 419 .quad 0x0e9e28eb4,0x1167f94f2 420 .quad 0x07b3ff57a,0x19385bf2e 421 .quad 0x0c9c8b782,0x0cec3662e 422 .quad 0x13a9cba9e,0x0e417f38a 423 .quad 0x093e106a4,0x19329634a 424 .quad 0x167001a9c,0x14e727980 425 .quad 0x1ddffc5d4,0x0e6fc4e6a 426 .quad 0x00df04680,0x0d104b8fc 427 .quad 0x02342001e,0x08227bb8a 428 .quad 0x00a2a8d7e,0x05b397730 429 .quad 0x168763fa6,0x0b0cd4768 430 .quad 0x1ed5a407a,0x0e78eb416 431 .quad 0x0d2c3ed1a,0x13c2b89c4 432 .quad 0x0995a5724,0x1641378f0 433 .quad 0x19b1afbc4,0x0d7a4825c 434 .quad 0x109ffedc0,0x08d96551c 435 .quad 0x0f2271e60,0x10f5ff2ba 436 .quad 0x00b0bf8ca,0x00bf80dd2 437 .quad 0x123888b7a,0x00167d312 438 .quad 0x1e888f7dc,0x18dcddd1c 439 .quad 0x002ee03b2,0x0f6076544 440 .quad 0x183e8d8fe,0x06a45d2b2 441 .quad 0x133d7a042,0x026f6a60a 442 .quad 0x116b0f50c,0x1dd3e10e8 443 .quad 0x05fabe670,0x1a2adb74e 444 .quad 0x130004488,0x0de87806c 445 .quad 0x000bcf5f6,0x19d34af3a 446 .quad 0x18f0c7078,0x014338754 447 .quad 0x017f27698,0x049c3cc9c 448 .quad 0x058ca5f00,0x15e3e77ee 449 .quad 0x1af900c24,0x068bce87a 450 .quad 0x0b5cfca28,0x0dd07448e 451 .quad 0x0ded288f8,0x1524fa6c6 452 .quad 0x059f229bc,0x1d8048348 453 .quad 0x06d390dec,0x16cba8aca 454 .quad 0x037170390,0x0a3e3e02c 455 .quad 0x06353c1cc,0x042d98888 456 .quad 0x0c4584f5c,0x0d73c7bea 457 .quad 0x1f16a3418,0x1329d9f7e 458 .quad 0x0531377e2,0x185137662 459 .quad 0x1d8d9ca7c,0x1b1c69528 460 .quad 0x0b25b29f2,0x18a08b5bc 461 .quad 0x19fb2a8b0,0x02178513a 462 .quad 0x1a08fe6ac,0x1da758ae0 463 .quad 0x045cddf4e,0x0e0ac139e 464 .quad 0x1a91647f2,0x169cf9eb0 465 .quad 0x1a0f717c4,0x0170076fa 466