1/* $NetBSD: bcopy.S,v 1.14 2011/01/31 12:10:58 skrll Exp $ */ 2 3/* 4 * Copyright (c) 2002 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Matthew Fredette. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32/* 33 * Copy routines for NetBSD/hppa. 34 */ 35 36#undef _LOCORE 37#define _LOCORE /* XXX fredette - unfortunate */ 38 39#if defined(SPCOPY) && !defined(_STANDALONE) 40 41#include "opt_multiprocessor.h" 42 43#include <machine/cpu.h> 44 45#endif 46 47#include <machine/asm.h> 48#include <machine/frame.h> 49#include <machine/reg.h> 50 51#if defined(LIBC_SCCS) && !defined(lint) 52RCSID("$NetBSD: bcopy.S,v 1.14 2011/01/31 12:10:58 skrll Exp $") 53#endif /* LIBC_SCCS and not lint */ 54 55/* 56 * The stbys instruction is a little asymmetric. When (%r2 & 3) 57 * is zero, stbys,b,m %r1, 4(%r2) works like stws,ma. You 58 * might then wish that when (%r2 & 3) == 0, stbys,e,m %r1, -4(%r2) 59 * worked like stws,mb. But it doesn't. 60 * 61 * This macro works around this problem. It requires that %t2 62 * hold the number of bytes that will be written by this store 63 * (meaning that it ranges from one to four). 64 * 65 * Watch the delay-slot trickery here. The comib is used to set 66 * up which instruction, either the stws or the stbys, is run 67 * in the delay slot of the b instruction. 68 */ 69#define _STBYS_E_M(r, dst_spc, dst_off) \ 70 comib,<> 4, %t2, 4 ! \ 71 b 4 ! \ 72 stws,mb r, -4(dst_spc, dst_off) ! \ 73 stbys,e,m r, 0(dst_spc, dst_off) 74 75/* 76 * This macro does a bulk copy with no shifting. cmplt and m are 77 * the completer and displacement multiplier, respectively, for 78 * the load and store instructions. 79 */ 80#define _COPY(src_spc, src_off, dst_spc, dst_off, count, cmplt, m) \ 81 ! \ 82 /* ! \ 83 * Loop storing 16 bytes at a time. Since count ! \ 84 * may be > INT_MAX, we have to be careful and ! \ 85 * avoid comparisons that treat it as a signed ! \ 86 * quantity, until after this loop, when count ! \ 87 * is guaranteed to be less than 16. ! \ 88 */ ! \ 89 comib,>>=,n 15, count, _LABEL(_skip16) ! \ 90.label _LABEL(_loop16) ! \ 91 addi -16, count, count ! \ 92 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 93 ldws,cmplt m*4(src_spc, src_off), %t2 ! \ 94 ldws,cmplt m*4(src_spc, src_off), %t3 ! \ 95 ldws,cmplt m*4(src_spc, src_off), %t4 ! \ 96 stws,cmplt %t1, m*4(dst_spc, dst_off) ! \ 97 stws,cmplt %t2, m*4(dst_spc, dst_off) ! \ 98 stws,cmplt %t3, m*4(dst_spc, dst_off) ! \ 99 comib,<< 15, count, _LABEL(_loop16) ! \ 100 stws,cmplt %t4, m*4(dst_spc, dst_off) ! \ 101.label _LABEL(_skip16) ! \ 102 ! \ 103 /* Loop storing 4 bytes at a time. */ ! \ 104 addib,<,n -4, count, _LABEL(_skip4) ! \ 105.label _LABEL(_loop4) ! \ 106 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 107 addib,>= -4, count, _LABEL(_loop4) ! \ 108 stws,cmplt %t1, m*4(dst_spc, dst_off) ! \ 109.label _LABEL(_skip4) ! \ 110 /* Restore the correct count. */ ! \ 111 addi 4, count, count ! \ 112 ! \ 113.label _LABEL(_do1) ! \ 114 ! \ 115 /* Loop storing 1 byte at a time. */ ! \ 116 addib,<,n -1, count, _LABEL(_skip1) ! \ 117.label _LABEL(_loop1) ! \ 118 ldbs,cmplt m*1(src_spc, src_off), %t1 ! \ 119 addib,>= -1, count, _LABEL(_loop1) ! \ 120 stbs,cmplt %t1, m*1(dst_spc, dst_off) ! \ 121.label _LABEL(_skip1) ! \ 122 /* Restore the correct count. */ ! \ 123 b _LABEL(_done) ! \ 124 addi 1, count, count 125 126/* 127 * This macro is definitely strange. It exists purely to 128 * allow the _COPYS macro to be reused, but because it 129 * requires this long attempt to explain it, I'm starting 130 * to doubt the value of that. 131 * 132 * Part of the expansion of the _COPYS macro below are loops 133 * that copy four words or one word at a time, performing shifts 134 * to get data to line up correctly in the destination buffer. 135 * 136 * The _COPYS macro is used when copying backwards, as well 137 * as forwards. The 4-word loop always loads into %t1, %t2, %t3, 138 * and %t4 in that order. This means that when copying forward, 139 * %t1 will have the word from the lowest address, and %t4 will 140 * have the word from the highest address. When copying 141 * backwards, the opposite is true. 142 * 143 * The shift instructions need pairs of registers with adjacent 144 * words, with the register containing the word from the lowest 145 * address *always* coming first. It is this assymetry that 146 * gives rise to this macro - depending on which direction 147 * we're copying in, these ordered pairs are different. 148 * 149 * Fortunately, we can compute those register numbers at compile 150 * time, and assemble them manually into a shift instruction. 151 * That's what this macro does. 152 * 153 * This macro takes two arguments. n ranges from 0 to 3 and 154 * is the "shift number", i.e., n = 0 means we're doing the 155 * shift for what will be the first store. 156 * 157 * m is the displacement multiplier from the _COPYS macro call. 158 * This is 1 for a forward copy and -1 for a backwards copy. 159 * So, the ((m + 1) / 2) term yields 0 for a backwards copy and 160 * 1 for a forward copy, and the ((m - 1) / 2) term yields 161 * 0 for a forward copy, and -1 for a backwards copy. 162 * These terms are used to discriminate the register computations 163 * below. 164 * 165 * When copying forward, then, the first register used with 166 * the first vshd will be 19 + (3 - ((0 - 1) & 3)), or %t4, 167 * which matches _COPYS' requirement that the word last loaded 168 * be in %t4. The first register used for the second vshd 169 * will then "wrap" around to 19 + (3 - ((1 - 1) & 3)), or %t1. 170 * And so on to %t2 and %t3. 171 * 172 * When copying forward, the second register used with the first 173 * vshd will be (19 + (3 - ((n + 0) & 3)), or %t1. It will 174 * continue to be %t2, then %t3, and finally %t4. 175 * 176 * When copying backwards, the values for the first and second 177 * register for each vshd are reversed from the forwards case. 178 * (Symmetry reclaimed!) Proving this is "left as an exercise 179 * for the reader" (remember the different discriminating values!) 180 */ 181#define _VSHD(n, m, t) \ 182 .word (0xd0000000 | \ 183 ((19 + (3 - ((n - 1 * ((m + 1) / 2)) & 3))) << 16) | \ 184 ((19 + (3 - ((n + 1 * ((m - 1) / 2)) & 3))) << 21) | \ 185 (t)) 186 187/* 188 * This macro does a bulk copy with shifting. cmplt and m are 189 * the completer and displacement multiplier, respectively, for 190 * the load and store instructions. It is assumed that the 191 * word last loaded is already in %t4. 192 */ 193#define _COPYS(src_spc, src_off, dst_spc, dst_off, count, cmplt, m) \ 194 ! \ 195 /* ! \ 196 * Loop storing 16 bytes at a time. Since count ! \ 197 * may be > INT_MAX, we have to be careful and ! \ 198 * avoid comparisons that treat it as a signed ! \ 199 * quantity, until after this loop, when count ! \ 200 * is guaranteed to be less than 16. ! \ 201 */ ! \ 202 comib,>>=,n 15, count, _LABEL(S_skip16) ! \ 203.label _LABEL(S_loop16) ! \ 204 addi -16, count, count ! \ 205 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 206 ldws,cmplt m*4(src_spc, src_off), %t2 ! \ 207 ldws,cmplt m*4(src_spc, src_off), %t3 ! \ 208 _VSHD(0, m, 1) /* vshd %t4, %t1, %r1 */ ! \ 209 ldws,cmplt m*4(src_spc, src_off), %t4 ! \ 210 _VSHD(1, m, 22) /* vshd %t1, %t2, %t1 */ ! \ 211 _VSHD(2, m, 21) /* vshd %t2, %t3, %t2 */ ! \ 212 _VSHD(3, m, 20) /* vshd %t3, %t4, %t3 */ ! \ 213 stws,cmplt %r1, m*4(dst_spc, dst_off) ! \ 214 stws,cmplt %t1, m*4(dst_spc, dst_off) ! \ 215 stws,cmplt %t2, m*4(dst_spc, dst_off) ! \ 216 comib,<< 15, count, _LABEL(S_loop16) ! \ 217 stws,cmplt %t3, m*4(dst_spc, dst_off) ! \ 218.label _LABEL(S_skip16) ! \ 219 ! \ 220 /* Loop storing 4 bytes at a time. */ ! \ 221 addib,<,n -4, count, _LABEL(S_skip4) ! \ 222.label _LABEL(S_loop4) ! \ 223 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 224 _VSHD(0, m, 1) /* into %r1 (1) */ ! \ 225 copy %t1, %t4 ! \ 226 addib,>= -4, count, _LABEL(S_loop4) ! \ 227 stws,cmplt %r1, m*4(dst_spc, dst_off) ! \ 228.label _LABEL(S_skip4) ! \ 229 ! \ 230 /* ! \ 231 * We now need to "back up" src_off by the ! \ 232 * number of bytes remaining in the FIFO ! \ 233 * (i.e., the number of bytes remaining in %t4), ! \ 234 * because (the correct) count still includes ! \ 235 * these bytes, and we intent to keep it that ! \ 236 * way, and finish with the single-byte copier. ! \ 237 * ! \ 238 * The number of bytes remaining in the FIFO is ! \ 239 * related to the shift count, so recover it, ! \ 240 * restoring the correct count at the same time. ! \ 241 */ ! \ 242 mfctl %cr11, %t1 ! \ 243 addi 4, count, count ! \ 244 shd %r0, %t1, 3, %t1 ! \ 245 ! \ 246 /* ! \ 247 * If we're copying forward, the shift count ! \ 248 * is the number of bytes remaining in the ! \ 249 * FIFO, and we want to subtract it from src_off. ! \ 250 * If we're copying backwards, (4 - shift count) ! \ 251 * is the number of bytes remaining in the FIFO, ! \ 252 * and we want to add it to src_off. ! \ 253 * ! \ 254 * We observe that x + (4 - y) = x - (y - 4), ! \ 255 * and introduce this instruction to add -4 when ! \ 256 * m is -1, although this does mean one extra ! \ 257 * instruction in the forward case. ! \ 258 */ ! \ 259 addi 4*((m - 1) / 2), %t1, %t1 ! \ 260 ! \ 261 /* Now branch to the byte-at-a-time loop. */ ! \ 262 b _LABEL(_do1) ! \ 263 sub src_off, %t1, src_off 264 265/* 266 * This macro copies a region in the forward direction. 267 */ 268#define _COPY_FORWARD(src_spc, src_off, dst_spc, dst_off, count) \ 269 ! \ 270 /* ! \ 271 * Since in the shifting-left case we will ! \ 272 * load 8 bytes before checking count, to ! \ 273 * keep things simple, branch to the byte ! \ 274 * copier unless we're copying at least 8. ! \ 275 */ ! \ 276 comib,>>,n 8, count, _LABEL(_do1) ! \ 277 ! \ 278 /* ! \ 279 * Once we 4-byte align the source offset, ! \ 280 * figure out how many bytes from the region ! \ 281 * will be in the first 4-byte word we read. ! \ 282 * Ditto for writing the destination offset. ! \ 283 */ ! \ 284 extru src_off, 31, 2, %t1 ! \ 285 extru dst_off, 31, 2, %t2 ! \ 286 subi 4, %t1, %t1 ! \ 287 subi 4, %t2, %t2 ! \ 288 ! \ 289 /* ! \ 290 * Calculate the byte shift required. A ! \ 291 * positive value means a source 4-byte word ! \ 292 * has to be shifted to the right to line up ! \ 293 * as a destination 4-byte word. ! \ 294 */ ! \ 295 sub %t1, %t2, %t1 ! \ 296 ! \ 297 /* 4-byte align src_off. */ ! \ 298 depi 0, 31, 2, src_off ! \ 299 ! \ 300 /* ! \ 301 * It's somewhat important to note that this ! \ 302 * code thinks of count as "the number of bytes ! \ 303 * that haven't been stored yet", as opposed to ! \ 304 * "the number of bytes that haven't been copied ! \ 305 * yet". The distinction is subtle, but becomes ! \ 306 * apparent at the end of the shifting code, where ! \ 307 * we "back up" src_off to correspond to count, ! \ 308 * as opposed to flushing the FIFO. ! \ 309 * ! \ 310 * We calculated above how many bytes our first ! \ 311 * store will store, so update count now. ! \ 312 * ! \ 313 * If the shift is zero, strictly as an optimization ! \ 314 * we use a copy loop that does no shifting. ! \ 315 */ ! \ 316 comb,<> %r0, %t1, _LABEL(_shifting) ! \ 317 sub count, %t2, count ! \ 318 ! \ 319 /* Load and store the first word. */ ! \ 320 ldws,ma 4(src_spc, src_off), %t4 ! \ 321 stbys,b,m %t4, 4(dst_spc, dst_off) ! \ 322 ! \ 323 /* Do the rest of the copy. */ ! \ 324 _COPY(src_spc,src_off,dst_spc,dst_off,count,ma,1) ! \ 325 ! \ 326.label _LABEL(_shifting) ! \ 327 ! \ 328 /* ! \ 329 * If shift < 0, we need to shift words to the ! \ 330 * left. Since we can't do this directly, we ! \ 331 * adjust the shift so it's a shift to the right ! \ 332 * and load the first word into the high word of ! \ 333 * the FIFO. Otherwise, we load a zero into the ! \ 334 * high word of the FIFO. ! \ 335 */ ! \ 336 comb,<= %r0, %t1, _LABEL(_shiftingrt) ! \ 337 copy %r0, %t3 ! \ 338 addi 4, %t1, %t1 ! \ 339 ldws,ma 4(src_spc, src_off), %t3 ! \ 340.label _LABEL(_shiftingrt) ! \ 341 ! \ 342 /* ! \ 343 * Turn the shift byte count into a bit count, ! \ 344 * load the next word, set the Shift Amount ! \ 345 * Register, and form and store the first word. ! \ 346 */ ! \ 347 sh3add %t1, %r0, %t1 ! \ 348 ldws,ma 4(src_spc, src_off), %t4 ! \ 349 mtctl %t1, %cr11 ! \ 350 vshd %t3, %t4, %r1 ! \ 351 stbys,b,m %r1, 4(dst_spc, dst_off) ! \ 352 ! \ 353 /* Do the rest of the copy. */ ! \ 354 _COPYS(src_spc,src_off,dst_spc,dst_off,count,ma,1) 355 356/* This macro copies a region in the reverse direction. */ 357#define _COPY_REVERSE(src_spc, src_off, dst_spc, dst_off, count) \ 358 ! \ 359 /* Immediately add count to both offsets. */ ! \ 360 add src_off, count, src_off ! \ 361 add dst_off, count, dst_off ! \ 362 ! \ 363 /* ! \ 364 * Since in the shifting-right case we ! \ 365 * will load 8 bytes before checking ! \ 366 * count, to keep things simple, branch ! \ 367 * to the byte copier unless we're ! \ 368 * copying at least 8 bytes. ! \ 369 */ ! \ 370 comib,>>,n 8, count, _LABEL(_do1) ! \ 371 ! \ 372 /* ! \ 373 * Once we 4-byte align the source offset, ! \ 374 * figure out how many bytes from the region ! \ 375 * will be in the first 4-byte word we read. ! \ 376 * Ditto for writing the destination offset. ! \ 377 */ ! \ 378 extru,<> src_off, 31, 2, %t1 ! \ 379 ldi 4, %t1 ! \ 380 extru,<> dst_off, 31, 2, %t2 ! \ 381 ldi 4, %t2 ! \ 382 ! \ 383 /* ! \ 384 * Calculate the byte shift required. A ! \ 385 * positive value means a source 4-byte ! \ 386 * word has to be shifted to the right to ! \ 387 * line up as a destination 4-byte word. ! \ 388 */ ! \ 389 sub %t2, %t1, %t1 ! \ 390 ! \ 391 /* ! \ 392 * 4-byte align src_off, leaving it pointing ! \ 393 * to the 4-byte word *after* the next word ! \ 394 * we intend to load. ! \ 395 * ! \ 396 * It's somewhat important to note that this ! \ 397 * code thinks of count as "the number of bytes ! \ 398 * that haven't been stored yet", as opposed to ! \ 399 * "the number of bytes that haven't been copied ! \ 400 * yet". The distinction is subtle, but becomes ! \ 401 * apparent at the end of the shifting code, where ! \ 402 * we "back up" src_off to correspond to count, ! \ 403 * as opposed to flushing the FIFO. ! \ 404 * ! \ 405 * We calculated above how many bytes our first ! \ 406 * store will store, so update count now. ! \ 407 * ! \ 408 * If the shift is zero, we use a copy loop that ! \ 409 * does no shifting. NB: unlike the forward case, ! \ 410 * this is NOT strictly an optimization. If the ! \ 411 * SAR is zero the vshds do NOT do the right thing. ! \ 412 * This is another assymetry more or less the "fault" ! \ 413 * of vshd. ! \ 414 */ ! \ 415 addi 3, src_off, src_off ! \ 416 sub count, %t2, count ! \ 417 comb,<> %r0, %t1, _LABEL(_shifting) ! \ 418 depi 0, 31, 2, src_off ! \ 419 ! \ 420 /* Load and store the first word. */ ! \ 421 ldws,mb -4(src_spc, src_off), %t4 ! \ 422 _STBYS_E_M(%t4, dst_spc, dst_off) ! \ 423 ! \ 424 /* Do the rest of the copy. */ ! \ 425 _COPY(src_spc,src_off,dst_spc,dst_off,count,mb,-1) ! \ 426 ! \ 427.label _LABEL(_shifting) ! \ 428 ! \ 429 /* ! \ 430 * If shift < 0, we need to shift words to the ! \ 431 * left. Since we can't do this directly, we ! \ 432 * adjust the shift so it's a shift to the right ! \ 433 * and load a zero in to the low word of the FIFO. ! \ 434 * Otherwise, we load the first word into the ! \ 435 * low word of the FIFO. ! \ 436 * ! \ 437 * Note the nullification trickery here. We ! \ 438 * assume that we're shifting to the left, and ! \ 439 * load zero into the low word of the FIFO. Then ! \ 440 * we nullify the addi if we're shifting to the ! \ 441 * right. If the addi is not nullified, we are ! \ 442 * shifting to the left, so we nullify the load. ! \ 443 * we branch if we're shifting to the ! \ 444 */ ! \ 445 copy %r0, %t3 ! \ 446 comb,<=,n %r0, %t1, 0 ! \ 447 addi,tr 4, %t1, %t1 ! \ 448 ldws,mb -4(src_spc, src_off), %t3 ! \ 449 ! \ 450 /* ! \ 451 * Turn the shift byte count into a bit count, ! \ 452 * load the next word, set the Shift Amount ! \ 453 * Register, and form and store the first word. ! \ 454 */ ! \ 455 sh3add %t1, %r0, %t1 ! \ 456 ldws,mb -4(src_spc, src_off), %t4 ! \ 457 mtctl %t1, %cr11 ! \ 458 vshd %t4, %t3, %r1 ! \ 459 _STBYS_E_M(%r1, dst_spc, dst_off) ! \ 460 ! \ 461 /* Do the rest of the copy. */ ! \ 462 _COPYS(src_spc,src_off,dst_spc,dst_off,count,mb,-1) 463 464/* 465 * For paranoia, when things aren't going well, enable this 466 * code to assemble byte-at-a-time-only copying. 467 */ 468#if 1 469#undef _COPY_FORWARD 470#define _COPY_FORWARD(src_spc, src_off, dst_spc, dst_off, count) \ 471 comb,=,n %r0, count, _LABEL(_done) ! \ 472 ldbs,ma 1(src_spc, src_off), %r1 ! \ 473 addib,<> -1, count, -12 ! \ 474 stbs,ma %r1, 1(dst_spc, dst_off) ! \ 475 b,n _LABEL(_done) 476#undef _COPY_REVERSE 477#define _COPY_REVERSE(src_spc, src_off, dst_spc, dst_off, count) \ 478 comb,= %r0, count, _LABEL(_done) ! \ 479 add src_off, count, src_off ! \ 480 add dst_off, count, dst_off ! \ 481 ldbs,mb -1(src_spc, src_off), %r1 ! \ 482 addib,<> -1, count, -12 ! \ 483 stbs,mb %r1, -1(dst_spc, dst_off) ! \ 484 b,n _LABEL(_done) 485#endif 486 487/* 488 * If none of the following are defined, define BCOPY. 489 */ 490#if !(defined(SPCOPY) || defined(MEMCPY) || defined(MEMMOVE)) 491#define BCOPY 492#endif 493 494#if defined(SPCOPY) && !defined(_STANDALONE) 495 496#include <sys/errno.h> 497#include "assym.h" 498 499/* 500 * int spcopy(pa_space_t ssp, const void *src, pa_space_t dsp, void *dst, 501 * size_t len) 502 * 503 * We assume that the regions do not overlap. 504 */ 505LEAF_ENTRY(spcopy) 506 507 /* 508 * Setup the fault handler, which will fill in %ret0 if triggered. 509 */ 510 GET_CURLWP(%r31) 511#ifdef DIAGNOSTIC 512 comb,<>,n %r0, %r31, Lspcopy_curlwp_ok 513 ldil L%panic, %r1 514 ldil L%Lspcopy_curlwp_bad, %arg0 515 ldo R%panic(%r1), %r1 516 ldo R%Lspcopy_curlwp_bad(%arg0), %arg0 517 .call 518 bv,n %r0(%r1) 519 nop 520Lspcopy_curlwp_bad: 521 .asciz "spcopy: curlwp == NULL\n" 522 .align 8 523Lspcopy_curlwp_ok: 524#endif /* DIAGNOSTIC */ 525 ldil L%spcopy_fault, %r1 526 ldw L_PCB(%r31), %r31 527 ldo R%spcopy_fault(%r1), %r1 528 stw %r1, PCB_ONFAULT(%r31) 529 530 /* Setup the space registers. */ 531 mfsp %sr2, %ret1 532 mtsp %arg0, %sr1 533 mtsp %arg2, %sr2 534 535 /* Get the len argument and do the copy. */ 536 ldw HPPA_FRAME_ARG(4)(%sp), %arg0 537#define _LABEL(l) __CONCAT(spcopy,l) 538 _COPY_FORWARD(%sr1,%arg1,%sr2,%arg3,%arg0) 539_LABEL(_done): 540 541 /* Return. */ 542 copy %r0, %ret0 543ALTENTRY(spcopy_fault) 544 stw %r0, PCB_ONFAULT(%r31) 545 bv %r0(%rp) 546 mtsp %ret1, %sr2 547EXIT(spcopy) 548#endif /* SPCOPY && !_STANDALONE */ 549 550#ifdef MEMCPY 551/* 552 * void *memcpy(void *restrict dst, const void *restrict src, size_t len); 553 * 554 * memcpy is specifically restricted to working on 555 * non-overlapping regions, so we can just copy forward. 556 */ 557LEAF_ENTRY(memcpy) 558 copy %arg0, %ret0 559#define _LABEL(l) __CONCAT(memcpy,l) 560 _COPY_FORWARD(%sr0,%arg1,%sr0,%arg0,%arg2) 561_LABEL(_done): 562 bv,n %r0(%rp) 563 nop 564EXIT(memcpy) 565#endif /* MEMCPY */ 566 567#ifdef BCOPY 568/* 569 * void bcopy(const void *src, void *dst, size_t len); 570 */ 571LEAF_ENTRY(bcopy) 572 copy %arg0, %r1 573 copy %arg1, %arg0 574 copy %r1, %arg1 575 /* FALLTHROUGH */ 576#define _LABEL_F(l) __CONCAT(bcopy_F,l) 577#define _LABEL_R(l) __CONCAT(bcopy_R,l) 578#endif 579 580#ifdef MEMMOVE 581/* 582 * void *memmove(void *dst, const void *src, size_t len); 583 */ 584LEAF_ENTRY(memmove) 585#define _LABEL_F(l) __CONCAT(memmove_F,l) 586#define _LABEL_R(l) __CONCAT(memmove_R,l) 587 copy %arg0, %ret0 588#endif /* MEMMOVE */ 589 590#if defined(BCOPY) || defined(MEMMOVE) 591 592 /* 593 * If src >= dst or src + len <= dst, we copy 594 * forward, else we copy in reverse. 595 */ 596 add %arg1, %arg2, %r1 597 comb,>>=,n %arg1, %arg0, 0 598 comb,>>,n %r1, %arg0, _LABEL_R(_go) 599 600#define _LABEL _LABEL_F 601 _COPY_FORWARD(%sr0,%arg1,%sr0,%arg0,%arg2) 602#undef _LABEL 603 604_LABEL_R(_go): 605#define _LABEL _LABEL_R 606 _COPY_REVERSE(%sr0,%arg1,%sr0,%arg0,%arg2) 607#undef _LABEL 608 609_LABEL_F(_done): 610_LABEL_R(_done): 611 bv,n %r0(%rp) 612 nop 613#ifdef BCOPY 614EXIT(bcopy) 615#else 616EXIT(memmove) 617#endif 618#endif /* BCOPY || MEMMOVE */ 619