1/* 2 * Implement fast CRC32C with PCLMULQDQ instructions. (x86_64) 3 * 4 * The white paper on CRC32C calculations with PCLMULQDQ instruction can be 5 * downloaded from: 6 * http://download.intel.com/design/intarch/papers/323405.pdf 7 * 8 * Copyright (C) 2012 Intel Corporation. 9 * 10 * Authors: 11 * Wajdi Feghali <wajdi.k.feghali@intel.com> 12 * James Guilford <james.guilford@intel.com> 13 * David Cote <david.m.cote@intel.com> 14 * Tim Chen <tim.c.chen@linux.intel.com> 15 * 16 * This software is available to you under a choice of one of two 17 * licenses. You may choose to be licensed under the terms of the GNU 18 * General Public License (GPL) Version 2, available from the file 19 * COPYING in the main directory of this source tree, or the 20 * OpenIB.org BSD license below: 21 * 22 * Redistribution and use in source and binary forms, with or 23 * without modification, are permitted provided that the following 24 * conditions are met: 25 * 26 * - Redistributions of source code must retain the above 27 * copyright notice, this list of conditions and the following 28 * disclaimer. 29 * 30 * - Redistributions in binary form must reproduce the above 31 * copyright notice, this list of conditions and the following 32 * disclaimer in the documentation and/or other materials 33 * provided with the distribution. 34 * 35 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 36 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 37 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 38 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 39 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 40 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 41 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 42 * SOFTWARE. 43 */ 44 45## ISCSI CRC 32 Implementation with crc32 and pclmulqdq Instruction 46 47.macro LABEL prefix n 48\prefix\n\(): 49.endm 50 51.macro JMPTBL_ENTRY i 52.word crc_\i - crc_array 53.endm 54 55.macro JNC_LESS_THAN j 56 jnc less_than_\j 57.endm 58 59# Define threshold where buffers are considered "small" and routed to more 60# efficient "by-1" code. This "by-1" code only handles up to 255 bytes, so 61# SMALL_SIZE can be no larger than 255. 62 63#define SMALL_SIZE 200 64 65.if (SMALL_SIZE > 255) 66.error "SMALL_ SIZE must be < 256" 67.endif 68 69# unsigned int crc_pcl(u8 *buffer, int len, unsigned int crc_init); 70 71.global crc_pcl 72crc_pcl: 73#define bufp %rdi 74#define bufp_dw %edi 75#define bufp_w %di 76#define bufp_b %dil 77#define bufptmp %rcx 78#define block_0 %rcx 79#define block_1 %rdx 80#define block_2 %r11 81#define len %rsi 82#define len_dw %esi 83#define len_w %si 84#define len_b %sil 85#define crc_init_arg %rdx 86#define tmp %rbx 87#define crc_init %r8 88#define crc_init_dw %r8d 89#define crc1 %r9 90#define crc2 %r10 91 92 pushq %rbx 93 pushq %rdi 94 pushq %rsi 95 96 ## Move crc_init for Linux to a different 97 mov crc_init_arg, crc_init 98 99 ################################################################ 100 ## 1) ALIGN: 101 ################################################################ 102 103 mov bufp, bufptmp # rdi = *buf 104 neg bufp 105 and $7, bufp # calculate the unalignment amount of 106 # the address 107 je proc_block # Skip if aligned 108 109 ## If len is less than 8 and we're unaligned, we need to jump 110 ## to special code to avoid reading beyond the end of the buffer 111 cmp $8, len 112 jae do_align 113 # less_than_8 expects length in upper 3 bits of len_dw 114 # less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30] 115 shl $32-3+1, len_dw 116 jmp less_than_8_post_shl1 117 118do_align: 119 #### Calculate CRC of unaligned bytes of the buffer (if any) 120 movq (bufptmp), tmp # load a quadward from the buffer 121 add bufp, bufptmp # align buffer pointer for quadword 122 # processing 123 sub bufp, len # update buffer length 124align_loop: 125 crc32b %bl, crc_init_dw # compute crc32 of 1-byte 126 shr $8, tmp # get next byte 127 dec bufp 128 jne align_loop 129 130proc_block: 131 132 ################################################################ 133 ## 2) PROCESS BLOCKS: 134 ################################################################ 135 136 ## compute num of bytes to be processed 137 movq len, tmp # save num bytes in tmp 138 139 cmpq $128*24, len 140 jae full_block 141 142continue_block: 143 cmpq $SMALL_SIZE, len 144 jb small 145 146 ## len < 128*24 147 movq $2731, %rax # 2731 = ceil(2^16 / 24) 148 mul len_dw 149 shrq $16, %rax 150 151 ## eax contains floor(bytes / 24) = num 24-byte chunks to do 152 153 ## process rax 24-byte chunks (128 >= rax >= 0) 154 155 ## compute end address of each block 156 ## block 0 (base addr + RAX * 8) 157 ## block 1 (base addr + RAX * 16) 158 ## block 2 (base addr + RAX * 24) 159 lea (bufptmp, %rax, 8), block_0 160 lea (block_0, %rax, 8), block_1 161 lea (block_1, %rax, 8), block_2 162 163 xor crc1, crc1 164 xor crc2, crc2 165 166 ## branch into array 167 lea jump_table(%rip), bufp 168 movzxw (bufp, %rax, 2), len 169 offset=crc_array-jump_table 170 lea offset(bufp, len, 1), bufp 171 jmp *bufp 172 173 ################################################################ 174 ## 2a) PROCESS FULL BLOCKS: 175 ################################################################ 176full_block: 177 movq $128,%rax 178 lea 128*8*2(block_0), block_1 179 lea 128*8*3(block_0), block_2 180 add $128*8*1, block_0 181 182 xor crc1,crc1 183 xor crc2,crc2 184 185 # Fall thruogh into top of crc array (crc_128) 186 187 ################################################################ 188 ## 3) CRC Array: 189 ################################################################ 190 191crc_array: 192 i=128 193.rept 128-1 194.altmacro 195LABEL crc_ %i 196.noaltmacro 197 crc32q -i*8(block_0), crc_init 198 crc32q -i*8(block_1), crc1 199 crc32q -i*8(block_2), crc2 200 i=(i-1) 201.endr 202 203.altmacro 204LABEL crc_ %i 205.noaltmacro 206 crc32q -i*8(block_0), crc_init 207 crc32q -i*8(block_1), crc1 208# SKIP crc32 -i*8(block_2), crc2 ; Don't do this one yet 209 210 mov block_2, block_0 211 212 ################################################################ 213 ## 4) Combine three results: 214 ################################################################ 215 216 lea (K_table-16)(%rip), bufp # first entry is for idx 1 217 shlq $3, %rax # rax *= 8 218 subq %rax, tmp # tmp -= rax*8 219 shlq $1, %rax 220 subq %rax, tmp # tmp -= rax*16 221 # (total tmp -= rax*24) 222 addq %rax, bufp 223 224 movdqa (bufp), %xmm0 # 2 consts: K1:K2 225 226 movq crc_init, %xmm1 # CRC for block 1 227 pclmulqdq $0x00,%xmm0,%xmm1 # Multiply by K2 228 229 movq crc1, %xmm2 # CRC for block 2 230 pclmulqdq $0x10, %xmm0, %xmm2 # Multiply by K1 231 232 pxor %xmm2,%xmm1 233 movq %xmm1, %rax 234 xor -i*8(block_2), %rax 235 mov crc2, crc_init 236 crc32 %rax, crc_init 237 238################################################################ 239## 5) Check for end: 240################################################################ 241 242LABEL crc_ 0 243 mov tmp, len 244 cmp $128*24, tmp 245 jae full_block 246 cmp $24, tmp 247 jae continue_block 248 249less_than_24: 250 shl $32-4, len_dw # less_than_16 expects length 251 # in upper 4 bits of len_dw 252 jnc less_than_16 253 crc32q (bufptmp), crc_init 254 crc32q 8(bufptmp), crc_init 255 jz do_return 256 add $16, bufptmp 257 # len is less than 8 if we got here 258 # less_than_8 expects length in upper 3 bits of len_dw 259 # less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30] 260 shl $2, len_dw 261 jmp less_than_8_post_shl1 262 263 ####################################################################### 264 ## 6) LESS THAN 256-bytes REMAIN AT THIS POINT (8-bits of len are full) 265 ####################################################################### 266small: 267 shl $32-8, len_dw # Prepare len_dw for less_than_256 268 j=256 269.rept 5 # j = {256, 128, 64, 32, 16} 270.altmacro 271LABEL less_than_ %j # less_than_j: Length should be in 272 # upper lg(j) bits of len_dw 273 j=(j/2) 274 shl $1, len_dw # Get next MSB 275 JNC_LESS_THAN %j 276.noaltmacro 277 i=0 278.rept (j/8) 279 crc32q i(bufptmp), crc_init # Compute crc32 of 8-byte data 280 i=i+8 281.endr 282 jz do_return # Return if remaining length is zero 283 add $j, bufptmp # Advance buf 284.endr 285 286less_than_8: # Length should be stored in 287 # upper 3 bits of len_dw 288 shl $1, len_dw 289less_than_8_post_shl1: 290 jnc less_than_4 291 crc32l (bufptmp), crc_init_dw # CRC of 4 bytes 292 jz do_return # return if remaining data is zero 293 add $4, bufptmp 294less_than_4: # Length should be stored in 295 # upper 2 bits of len_dw 296 shl $1, len_dw 297 jnc less_than_2 298 crc32w (bufptmp), crc_init_dw # CRC of 2 bytes 299 jz do_return # return if remaining data is zero 300 add $2, bufptmp 301less_than_2: # Length should be stored in the MSB 302 # of len_dw 303 shl $1, len_dw 304 jnc less_than_1 305 crc32b (bufptmp), crc_init_dw # CRC of 1 byte 306less_than_1: # Length should be zero 307do_return: 308 movq crc_init, %rax 309 popq %rsi 310 popq %rdi 311 popq %rbx 312 ret 313 314 ################################################################ 315 ## jump table Table is 129 entries x 2 bytes each 316 ################################################################ 317.align 4 318jump_table: 319 i=0 320.rept 129 321.altmacro 322JMPTBL_ENTRY %i 323.noaltmacro 324 i=i+1 325.endr 326 ################################################################ 327 ## PCLMULQDQ tables 328 ## Table is 128 entries x 2 quad words each 329 ################################################################ 330.data 331.align 64 332K_table: 333 .quad 0x14cd00bd6,0x105ec76f0 334 .quad 0x0ba4fc28e,0x14cd00bd6 335 .quad 0x1d82c63da,0x0f20c0dfe 336 .quad 0x09e4addf8,0x0ba4fc28e 337 .quad 0x039d3b296,0x1384aa63a 338 .quad 0x102f9b8a2,0x1d82c63da 339 .quad 0x14237f5e6,0x01c291d04 340 .quad 0x00d3b6092,0x09e4addf8 341 .quad 0x0c96cfdc0,0x0740eef02 342 .quad 0x18266e456,0x039d3b296 343 .quad 0x0daece73e,0x0083a6eec 344 .quad 0x0ab7aff2a,0x102f9b8a2 345 .quad 0x1248ea574,0x1c1733996 346 .quad 0x083348832,0x14237f5e6 347 .quad 0x12c743124,0x02ad91c30 348 .quad 0x0b9e02b86,0x00d3b6092 349 .quad 0x018b33a4e,0x06992cea2 350 .quad 0x1b331e26a,0x0c96cfdc0 351 .quad 0x17d35ba46,0x07e908048 352 .quad 0x1bf2e8b8a,0x18266e456 353 .quad 0x1a3e0968a,0x11ed1f9d8 354 .quad 0x0ce7f39f4,0x0daece73e 355 .quad 0x061d82e56,0x0f1d0f55e 356 .quad 0x0d270f1a2,0x0ab7aff2a 357 .quad 0x1c3f5f66c,0x0a87ab8a8 358 .quad 0x12ed0daac,0x1248ea574 359 .quad 0x065863b64,0x08462d800 360 .quad 0x11eef4f8e,0x083348832 361 .quad 0x1ee54f54c,0x071d111a8 362 .quad 0x0b3e32c28,0x12c743124 363 .quad 0x0064f7f26,0x0ffd852c6 364 .quad 0x0dd7e3b0c,0x0b9e02b86 365 .quad 0x0f285651c,0x0dcb17aa4 366 .quad 0x010746f3c,0x018b33a4e 367 .quad 0x1c24afea4,0x0f37c5aee 368 .quad 0x0271d9844,0x1b331e26a 369 .quad 0x08e766a0c,0x06051d5a2 370 .quad 0x093a5f730,0x17d35ba46 371 .quad 0x06cb08e5c,0x11d5ca20e 372 .quad 0x06b749fb2,0x1bf2e8b8a 373 .quad 0x1167f94f2,0x021f3d99c 374 .quad 0x0cec3662e,0x1a3e0968a 375 .quad 0x19329634a,0x08f158014 376 .quad 0x0e6fc4e6a,0x0ce7f39f4 377 .quad 0x08227bb8a,0x1a5e82106 378 .quad 0x0b0cd4768,0x061d82e56 379 .quad 0x13c2b89c4,0x188815ab2 380 .quad 0x0d7a4825c,0x0d270f1a2 381 .quad 0x10f5ff2ba,0x105405f3e 382 .quad 0x00167d312,0x1c3f5f66c 383 .quad 0x0f6076544,0x0e9adf796 384 .quad 0x026f6a60a,0x12ed0daac 385 .quad 0x1a2adb74e,0x096638b34 386 .quad 0x19d34af3a,0x065863b64 387 .quad 0x049c3cc9c,0x1e50585a0 388 .quad 0x068bce87a,0x11eef4f8e 389 .quad 0x1524fa6c6,0x19f1c69dc 390 .quad 0x16cba8aca,0x1ee54f54c 391 .quad 0x042d98888,0x12913343e 392 .quad 0x1329d9f7e,0x0b3e32c28 393 .quad 0x1b1c69528,0x088f25a3a 394 .quad 0x02178513a,0x0064f7f26 395 .quad 0x0e0ac139e,0x04e36f0b0 396 .quad 0x0170076fa,0x0dd7e3b0c 397 .quad 0x141a1a2e2,0x0bd6f81f8 398 .quad 0x16ad828b4,0x0f285651c 399 .quad 0x041d17b64,0x19425cbba 400 .quad 0x1fae1cc66,0x010746f3c 401 .quad 0x1a75b4b00,0x18db37e8a 402 .quad 0x0f872e54c,0x1c24afea4 403 .quad 0x01e41e9fc,0x04c144932 404 .quad 0x086d8e4d2,0x0271d9844 405 .quad 0x160f7af7a,0x052148f02 406 .quad 0x05bb8f1bc,0x08e766a0c 407 .quad 0x0a90fd27a,0x0a3c6f37a 408 .quad 0x0b3af077a,0x093a5f730 409 .quad 0x04984d782,0x1d22c238e 410 .quad 0x0ca6ef3ac,0x06cb08e5c 411 .quad 0x0234e0b26,0x063ded06a 412 .quad 0x1d88abd4a,0x06b749fb2 413 .quad 0x04597456a,0x04d56973c 414 .quad 0x0e9e28eb4,0x1167f94f2 415 .quad 0x07b3ff57a,0x19385bf2e 416 .quad 0x0c9c8b782,0x0cec3662e 417 .quad 0x13a9cba9e,0x0e417f38a 418 .quad 0x093e106a4,0x19329634a 419 .quad 0x167001a9c,0x14e727980 420 .quad 0x1ddffc5d4,0x0e6fc4e6a 421 .quad 0x00df04680,0x0d104b8fc 422 .quad 0x02342001e,0x08227bb8a 423 .quad 0x00a2a8d7e,0x05b397730 424 .quad 0x168763fa6,0x0b0cd4768 425 .quad 0x1ed5a407a,0x0e78eb416 426 .quad 0x0d2c3ed1a,0x13c2b89c4 427 .quad 0x0995a5724,0x1641378f0 428 .quad 0x19b1afbc4,0x0d7a4825c 429 .quad 0x109ffedc0,0x08d96551c 430 .quad 0x0f2271e60,0x10f5ff2ba 431 .quad 0x00b0bf8ca,0x00bf80dd2 432 .quad 0x123888b7a,0x00167d312 433 .quad 0x1e888f7dc,0x18dcddd1c 434 .quad 0x002ee03b2,0x0f6076544 435 .quad 0x183e8d8fe,0x06a45d2b2 436 .quad 0x133d7a042,0x026f6a60a 437 .quad 0x116b0f50c,0x1dd3e10e8 438 .quad 0x05fabe670,0x1a2adb74e 439 .quad 0x130004488,0x0de87806c 440 .quad 0x000bcf5f6,0x19d34af3a 441 .quad 0x18f0c7078,0x014338754 442 .quad 0x017f27698,0x049c3cc9c 443 .quad 0x058ca5f00,0x15e3e77ee 444 .quad 0x1af900c24,0x068bce87a 445 .quad 0x0b5cfca28,0x0dd07448e 446 .quad 0x0ded288f8,0x1524fa6c6 447 .quad 0x059f229bc,0x1d8048348 448 .quad 0x06d390dec,0x16cba8aca 449 .quad 0x037170390,0x0a3e3e02c 450 .quad 0x06353c1cc,0x042d98888 451 .quad 0x0c4584f5c,0x0d73c7bea 452 .quad 0x1f16a3418,0x1329d9f7e 453 .quad 0x0531377e2,0x185137662 454 .quad 0x1d8d9ca7c,0x1b1c69528 455 .quad 0x0b25b29f2,0x18a08b5bc 456 .quad 0x19fb2a8b0,0x02178513a 457 .quad 0x1a08fe6ac,0x1da758ae0 458 .quad 0x045cddf4e,0x0e0ac139e 459 .quad 0x1a91647f2,0x169cf9eb0 460 .quad 0x1a0f717c4,0x0170076fa 461