1/* $NetBSD: bcopy.S,v 1.7 2008/04/28 20:24:06 martin 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#include <machine/asm.h> 39#include <machine/frame.h> 40 41#if defined(LIBC_SCCS) && !defined(lint) 42RCSID("$NetBSD: bcopy.S,v 1.7 2008/04/28 20:24:06 martin Exp $") 43#endif /* LIBC_SCCS and not lint */ 44 45/* 46 * The stbys instruction is a little asymmetric. When (%r2 & 3) 47 * is zero, stbys,b,m %r1, 4(%r2) works like stws,ma. You 48 * might then wish that when (%r2 & 3) == 0, stbys,e,m %r1, -4(%r2) 49 * worked like stws,mb. But it doesn't. 50 * 51 * This macro works around this problem. It requires that %t2 52 * hold the number of bytes that will be written by this store 53 * (meaning that it ranges from one to four). 54 * 55 * Watch the delay-slot trickery here. The comib is used to set 56 * up which instruction, either the stws or the stbys, is run 57 * in the delay slot of the b instruction. 58 */ 59#define _STBYS_E_M(r, dst_spc, dst_off) \ 60 comib,<> 4, %t2, 4 ! \ 61 b 4 ! \ 62 stws,mb r, -4(dst_spc, dst_off) ! \ 63 stbys,e,m r, 0(dst_spc, dst_off) 64 65/* 66 * This macro does a bulk copy with no shifting. cmplt and m are 67 * the completer and displacement multiplier, respectively, for 68 * the load and store instructions. 69 */ 70#define _COPY(src_spc, src_off, dst_spc, dst_off, count, cmplt, m) \ 71 ! \ 72 /* ! \ 73 * Loop storing 16 bytes at a time. Since count ! \ 74 * may be > INT_MAX, we have to be careful and ! \ 75 * avoid comparisons that treat it as a signed ! \ 76 * quantity, until after this loop, when count ! \ 77 * is guaranteed to be less than 16. ! \ 78 */ ! \ 79 comib,>>=,n 15, count, _LABEL(_skip16) ! \ 80.label _LABEL(_loop16) ! \ 81 addi -16, count, count ! \ 82 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 83 ldws,cmplt m*4(src_spc, src_off), %t2 ! \ 84 ldws,cmplt m*4(src_spc, src_off), %t3 ! \ 85 ldws,cmplt m*4(src_spc, src_off), %t4 ! \ 86 stws,cmplt %t1, m*4(dst_spc, dst_off) ! \ 87 stws,cmplt %t2, m*4(dst_spc, dst_off) ! \ 88 stws,cmplt %t3, m*4(dst_spc, dst_off) ! \ 89 comib,<< 15, count, _LABEL(_loop16) ! \ 90 stws,cmplt %t4, m*4(dst_spc, dst_off) ! \ 91.label _LABEL(_skip16) ! \ 92 ! \ 93 /* Loop storing 4 bytes at a time. */ ! \ 94 addib,<,n -4, count, _LABEL(_skip4) ! \ 95.label _LABEL(_loop4) ! \ 96 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 97 addib,>= -4, count, _LABEL(_loop4) ! \ 98 stws,cmplt %t1, m*4(dst_spc, dst_off) ! \ 99.label _LABEL(_skip4) ! \ 100 /* Restore the correct count. */ ! \ 101 addi 4, count, count ! \ 102 ! \ 103.label _LABEL(_do1) ! \ 104 ! \ 105 /* Loop storing 1 byte at a time. */ ! \ 106 addib,<,n -1, count, _LABEL(_skip1) ! \ 107.label _LABEL(_loop1) ! \ 108 ldbs,cmplt m*1(src_spc, src_off), %t1 ! \ 109 addib,>= -1, count, _LABEL(_loop1) ! \ 110 stbs,cmplt %t1, m*1(dst_spc, dst_off) ! \ 111.label _LABEL(_skip1) ! \ 112 /* Restore the correct count. */ ! \ 113 b _LABEL(_done) ! \ 114 addi 1, count, count 115 116/* 117 * This macro is definitely strange. It exists purely to 118 * allow the _COPYS macro to be reused, but because it 119 * requires this long attempt to explain it, I'm starting 120 * to doubt the value of that. 121 * 122 * Part of the expansion of the _COPYS macro below are loops 123 * that copy four words or one word at a time, performing shifts 124 * to get data to line up correctly in the destination buffer. 125 * 126 * The _COPYS macro is used when copying backwards, as well 127 * as forwards. The 4-word loop always loads into %t1, %t2, %t3, 128 * and %t4 in that order. This means that when copying forward, 129 * %t1 will have the word from the lowest address, and %t4 will 130 * have the word from the highest address. When copying 131 * backwards, the opposite is true. 132 * 133 * The shift instructions need pairs of registers with adjacent 134 * words, with the register containing the word from the lowest 135 * address *always* coming first. It is this assymetry that 136 * gives rise to this macro - depending on which direction 137 * we're copying in, these ordered pairs are different. 138 * 139 * Fortunately, we can compute those register numbers at compile 140 * time, and assemble them manually into a shift instruction. 141 * That's what this macro does. 142 * 143 * This macro takes two arguments. n ranges from 0 to 3 and 144 * is the "shift number", i.e., n = 0 means we're doing the 145 * shift for what will be the first store. 146 * 147 * m is the displacement multiplier from the _COPYS macro call. 148 * This is 1 for a forward copy and -1 for a backwards copy. 149 * So, the ((m + 1) / 2) term yields 0 for a backwards copy and 150 * 1 for a forward copy, and the ((m - 1) / 2) term yields 151 * 0 for a forward copy, and -1 for a backwards copy. 152 * These terms are used to discriminate the register computations 153 * below. 154 * 155 * When copying forward, then, the first register used with 156 * the first vshd will be 19 + (3 - ((0 - 1) & 3)), or %t4, 157 * which matches _COPYS' requirement that the word last loaded 158 * be in %t4. The first register used for the second vshd 159 * will then "wrap" around to 19 + (3 - ((1 - 1) & 3)), or %t1. 160 * And so on to %t2 and %t3. 161 * 162 * When copying forward, the second register used with the first 163 * vshd will be (19 + (3 - ((n + 0) & 3)), or %t1. It will 164 * continue to be %t2, then %t3, and finally %t4. 165 * 166 * When copying backwards, the values for the first and second 167 * register for each vshd are reversed from the forwards case. 168 * (Symmetry reclaimed!) Proving this is "left as an exercise 169 * for the reader" (remember the different discriminating values!) 170 */ 171#define _VSHD(n, m, t) \ 172 .word (0xd0000000 | \ 173 ((19 + (3 - ((n - 1 * ((m + 1) / 2)) & 3))) << 16) | \ 174 ((19 + (3 - ((n + 1 * ((m - 1) / 2)) & 3))) << 21) | \ 175 (t)) 176 177/* 178 * This macro does a bulk copy with shifting. cmplt and m are 179 * the completer and displacement multiplier, respectively, for 180 * the load and store instructions. It is assumed that the 181 * word last loaded is already in %t4. 182 */ 183#define _COPYS(src_spc, src_off, dst_spc, dst_off, count, cmplt, m) \ 184 ! \ 185 /* ! \ 186 * Loop storing 16 bytes at a time. Since count ! \ 187 * may be > INT_MAX, we have to be careful and ! \ 188 * avoid comparisons that treat it as a signed ! \ 189 * quantity, until after this loop, when count ! \ 190 * is guaranteed to be less than 16. ! \ 191 */ ! \ 192 comib,>>=,n 15, count, _LABEL(S_skip16) ! \ 193.label _LABEL(S_loop16) ! \ 194 addi -16, count, count ! \ 195 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 196 ldws,cmplt m*4(src_spc, src_off), %t2 ! \ 197 ldws,cmplt m*4(src_spc, src_off), %t3 ! \ 198 _VSHD(0, m, 1) /* vshd %t4, %t1, %r1 */ ! \ 199 ldws,cmplt m*4(src_spc, src_off), %t4 ! \ 200 _VSHD(1, m, 22) /* vshd %t1, %t2, %t1 */ ! \ 201 _VSHD(2, m, 21) /* vshd %t2, %t3, %t2 */ ! \ 202 _VSHD(3, m, 20) /* vshd %t3, %t4, %t3 */ ! \ 203 stws,cmplt %r1, m*4(dst_spc, dst_off) ! \ 204 stws,cmplt %t1, m*4(dst_spc, dst_off) ! \ 205 stws,cmplt %t2, m*4(dst_spc, dst_off) ! \ 206 comib,<< 15, count, _LABEL(S_loop16) ! \ 207 stws,cmplt %t3, m*4(dst_spc, dst_off) ! \ 208.label _LABEL(S_skip16) ! \ 209 ! \ 210 /* Loop storing 4 bytes at a time. */ ! \ 211 addib,<,n -4, count, _LABEL(S_skip4) ! \ 212.label _LABEL(S_loop4) ! \ 213 ldws,cmplt m*4(src_spc, src_off), %t1 ! \ 214 _VSHD(0, m, 1) /* into %r1 (1) */ ! \ 215 copy %t1, %t4 ! \ 216 addib,>= -4, count, _LABEL(S_loop4) ! \ 217 stws,cmplt %r1, m*4(dst_spc, dst_off) ! \ 218.label _LABEL(S_skip4) ! \ 219 ! \ 220 /* ! \ 221 * We now need to "back up" src_off by the ! \ 222 * number of bytes remaining in the FIFO ! \ 223 * (i.e., the number of bytes remaining in %t4), ! \ 224 * because (the correct) count still includes ! \ 225 * these bytes, and we intent to keep it that ! \ 226 * way, and finish with the single-byte copier. ! \ 227 * ! \ 228 * The number of bytes remaining in the FIFO is ! \ 229 * related to the shift count, so recover it, ! \ 230 * restoring the correct count at the same time. ! \ 231 */ ! \ 232 mfctl %cr11, %t1 ! \ 233 addi 4, count, count ! \ 234 shd %r0, %t1, 3, %t1 ! \ 235 ! \ 236 /* ! \ 237 * If we're copying forward, the shift count ! \ 238 * is the number of bytes remaining in the ! \ 239 * FIFO, and we want to subtract it from src_off. ! \ 240 * If we're copying backwards, (4 - shift count) ! \ 241 * is the number of bytes remaining in the FIFO, ! \ 242 * and we want to add it to src_off. ! \ 243 * ! \ 244 * We observe that x + (4 - y) = x - (y - 4), ! \ 245 * and introduce this instruction to add -4 when ! \ 246 * m is -1, although this does mean one extra ! \ 247 * instruction in the forward case. ! \ 248 */ ! \ 249 addi 4*((m - 1) / 2), %t1, %t1 ! \ 250 ! \ 251 /* Now branch to the byte-at-a-time loop. */ ! \ 252 b _LABEL(_do1) ! \ 253 sub src_off, %t1, src_off 254 255/* 256 * This macro copies a region in the forward direction. 257 */ 258#define _COPY_FORWARD(src_spc, src_off, dst_spc, dst_off, count) \ 259 ! \ 260 /* ! \ 261 * Since in the shifting-left case we will ! \ 262 * load 8 bytes before checking count, to ! \ 263 * keep things simple, branch to the byte ! \ 264 * copier unless we're copying at least 8. ! \ 265 */ ! \ 266 comib,>>,n 8, count, _LABEL(_do1) ! \ 267 ! \ 268 /* ! \ 269 * Once we 4-byte align the source offset, ! \ 270 * figure out how many bytes from the region ! \ 271 * will be in the first 4-byte word we read. ! \ 272 * Ditto for writing the destination offset. ! \ 273 */ ! \ 274 extru src_off, 31, 2, %t1 ! \ 275 extru dst_off, 31, 2, %t2 ! \ 276 subi 4, %t1, %t1 ! \ 277 subi 4, %t2, %t2 ! \ 278 ! \ 279 /* ! \ 280 * Calculate the byte shift required. A ! \ 281 * positive value means a source 4-byte word ! \ 282 * has to be shifted to the right to line up ! \ 283 * as a destination 4-byte word. ! \ 284 */ ! \ 285 sub %t1, %t2, %t1 ! \ 286 ! \ 287 /* 4-byte align src_off. */ ! \ 288 depi 0, 31, 2, src_off ! \ 289 ! \ 290 /* ! \ 291 * It's somewhat important to note that this ! \ 292 * code thinks of count as "the number of bytes ! \ 293 * that haven't been stored yet", as opposed to ! \ 294 * "the number of bytes that haven't been copied ! \ 295 * yet". The distinction is subtle, but becomes ! \ 296 * apparent at the end of the shifting code, where ! \ 297 * we "back up" src_off to correspond to count, ! \ 298 * as opposed to flushing the FIFO. ! \ 299 * ! \ 300 * We calculated above how many bytes our first ! \ 301 * store will store, so update count now. ! \ 302 * ! \ 303 * If the shift is zero, strictly as an optimization ! \ 304 * we use a copy loop that does no shifting. ! \ 305 */ ! \ 306 comb,<> %r0, %t1, _LABEL(_shifting) ! \ 307 sub count, %t2, count ! \ 308 ! \ 309 /* Load and store the first word. */ ! \ 310 ldws,ma 4(src_spc, src_off), %t4 ! \ 311 stbys,b,m %t4, 4(dst_spc, dst_off) ! \ 312 ! \ 313 /* Do the rest of the copy. */ ! \ 314 _COPY(src_spc,src_off,dst_spc,dst_off,count,ma,1) ! \ 315 ! \ 316.label _LABEL(_shifting) ! \ 317 ! \ 318 /* ! \ 319 * If shift < 0, we need to shift words to the ! \ 320 * left. Since we can't do this directly, we ! \ 321 * adjust the shift so it's a shift to the right ! \ 322 * and load the first word into the high word of ! \ 323 * the FIFO. Otherwise, we load a zero into the ! \ 324 * high word of the FIFO. ! \ 325 */ ! \ 326 comb,<= %r0, %t1, _LABEL(_shiftingrt) ! \ 327 copy %r0, %t3 ! \ 328 addi 4, %t1, %t1 ! \ 329 ldws,ma 4(src_spc, src_off), %t3 ! \ 330.label _LABEL(_shiftingrt) ! \ 331 ! \ 332 /* ! \ 333 * Turn the shift byte count into a bit count, ! \ 334 * load the next word, set the Shift Amount ! \ 335 * Register, and form and store the first word. ! \ 336 */ ! \ 337 sh3add %t1, %r0, %t1 ! \ 338 ldws,ma 4(src_spc, src_off), %t4 ! \ 339 mtctl %t1, %cr11 ! \ 340 vshd %t3, %t4, %r1 ! \ 341 stbys,b,m %r1, 4(dst_spc, dst_off) ! \ 342 ! \ 343 /* Do the rest of the copy. */ ! \ 344 _COPYS(src_spc,src_off,dst_spc,dst_off,count,ma,1) 345 346/* This macro copies a region in the reverse direction. */ 347#define _COPY_REVERSE(src_spc, src_off, dst_spc, dst_off, count) \ 348 ! \ 349 /* Immediately add count to both offsets. */ ! \ 350 add src_off, count, src_off ! \ 351 add dst_off, count, dst_off ! \ 352 ! \ 353 /* ! \ 354 * Since in the shifting-right case we ! \ 355 * will load 8 bytes before checking ! \ 356 * count, to keep things simple, branch ! \ 357 * to the byte copier unless we're ! \ 358 * copying at least 8 bytes. ! \ 359 */ ! \ 360 comib,>>,n 8, count, _LABEL(_do1) ! \ 361 ! \ 362 /* ! \ 363 * Once we 4-byte align the source offset, ! \ 364 * figure out how many bytes from the region ! \ 365 * will be in the first 4-byte word we read. ! \ 366 * Ditto for writing the destination offset. ! \ 367 */ ! \ 368 extru,<> src_off, 31, 2, %t1 ! \ 369 ldi 4, %t1 ! \ 370 extru,<> dst_off, 31, 2, %t2 ! \ 371 ldi 4, %t2 ! \ 372 ! \ 373 /* ! \ 374 * Calculate the byte shift required. A ! \ 375 * positive value means a source 4-byte ! \ 376 * word has to be shifted to the right to ! \ 377 * line up as a destination 4-byte word. ! \ 378 */ ! \ 379 sub %t2, %t1, %t1 ! \ 380 ! \ 381 /* ! \ 382 * 4-byte align src_off, leaving it pointing ! \ 383 * to the 4-byte word *after* the next word ! \ 384 * we intend to load. ! \ 385 * ! \ 386 * It's somewhat important to note that this ! \ 387 * code thinks of count as "the number of bytes ! \ 388 * that haven't been stored yet", as opposed to ! \ 389 * "the number of bytes that haven't been copied ! \ 390 * yet". The distinction is subtle, but becomes ! \ 391 * apparent at the end of the shifting code, where ! \ 392 * we "back up" src_off to correspond to count, ! \ 393 * as opposed to flushing the FIFO. ! \ 394 * ! \ 395 * We calculated above how many bytes our first ! \ 396 * store will store, so update count now. ! \ 397 * ! \ 398 * If the shift is zero, we use a copy loop that ! \ 399 * does no shifting. NB: unlike the forward case, ! \ 400 * this is NOT strictly an optimization. If the ! \ 401 * SAR is zero the vshds do NOT do the right thing. ! \ 402 * This is another assymetry more or less the "fault" ! \ 403 * of vshd. ! \ 404 */ ! \ 405 addi 3, src_off, src_off ! \ 406 sub count, %t2, count ! \ 407 comb,<> %r0, %t1, _LABEL(_shifting) ! \ 408 depi 0, 31, 2, src_off ! \ 409 ! \ 410 /* Load and store the first word. */ ! \ 411 ldws,mb -4(src_spc, src_off), %t4 ! \ 412 _STBYS_E_M(%t4, dst_spc, dst_off) ! \ 413 ! \ 414 /* Do the rest of the copy. */ ! \ 415 _COPY(src_spc,src_off,dst_spc,dst_off,count,mb,-1) ! \ 416 ! \ 417.label _LABEL(_shifting) ! \ 418 ! \ 419 /* ! \ 420 * If shift < 0, we need to shift words to the ! \ 421 * left. Since we can't do this directly, we ! \ 422 * adjust the shift so it's a shift to the right ! \ 423 * and load a zero in to the low word of the FIFO. ! \ 424 * Otherwise, we load the first word into the ! \ 425 * low word of the FIFO. ! \ 426 * ! \ 427 * Note the nullification trickery here. We ! \ 428 * assume that we're shifting to the left, and ! \ 429 * load zero into the low word of the FIFO. Then ! \ 430 * we nullify the addi if we're shifting to the ! \ 431 * right. If the addi is not nullified, we are ! \ 432 * shifting to the left, so we nullify the load. ! \ 433 * we branch if we're shifting to the ! \ 434 */ ! \ 435 copy %r0, %t3 ! \ 436 comb,<=,n %r0, %t1, 0 ! \ 437 addi,tr 4, %t1, %t1 ! \ 438 ldws,mb -4(src_spc, src_off), %t3 ! \ 439 ! \ 440 /* ! \ 441 * Turn the shift byte count into a bit count, ! \ 442 * load the next word, set the Shift Amount ! \ 443 * Register, and form and store the first word. ! \ 444 */ ! \ 445 sh3add %t1, %r0, %t1 ! \ 446 ldws,mb -4(src_spc, src_off), %t4 ! \ 447 mtctl %t1, %cr11 ! \ 448 vshd %t4, %t3, %r1 ! \ 449 _STBYS_E_M(%r1, dst_spc, dst_off) ! \ 450 ! \ 451 /* Do the rest of the copy. */ ! \ 452 _COPYS(src_spc,src_off,dst_spc,dst_off,count,mb,-1) 453 454/* 455 * For paranoia, when things aren't going well, enable this 456 * code to assemble byte-at-a-time-only copying. 457 */ 458#if 1 459#undef _COPY_FORWARD 460#define _COPY_FORWARD(src_spc, src_off, dst_spc, dst_off, count) \ 461 comb,=,n %r0, count, _LABEL(_done) ! \ 462 ldbs,ma 1(src_spc, src_off), %r1 ! \ 463 addib,<> -1, count, -12 ! \ 464 stbs,ma %r1, 1(dst_spc, dst_off) ! \ 465 b,n _LABEL(_done) 466#undef _COPY_REVERSE 467#define _COPY_REVERSE(src_spc, src_off, dst_spc, dst_off, count) \ 468 comb,= %r0, count, _LABEL(_done) ! \ 469 add src_off, count, src_off ! \ 470 add dst_off, count, dst_off ! \ 471 ldbs,mb -1(src_spc, src_off), %r1 ! \ 472 addib,<> -1, count, -12 ! \ 473 stbs,mb %r1, -1(dst_spc, dst_off) ! \ 474 b,n _LABEL(_done) 475#endif 476 477/* 478 * If none of the following are defined, define BCOPY. 479 */ 480#if !(defined(SPCOPY) || defined(MEMCPY) || defined(MEMMOVE)) 481#define BCOPY 482#endif 483 484#if defined(SPCOPY) && !defined(_STANDALONE) 485#include <sys/errno.h> 486#include "assym.h" 487 488/* 489 * int spcopy(pa_space_t ssp, const void *src, pa_space_t dsp, void *dst, 490 * size_t len) 491 * 492 * We assume that the regions do not overlap. 493 */ 494LEAF_ENTRY(spcopy) 495 496 /* 497 * Setup the fault handler, and load %ret0 498 * with EFAULT, assuming the copy will fail. 499 */ 500 .import curlwp, data 501 ldil L%curlwp, %r31 502 ldw R%curlwp(%r31), %r31 503#ifdef DIAGNOSTIC 504 comb,<>,n %r0, %r31, Lspcopy_curlwp_ok 505 ldil L%panic, %r1 506 ldil L%Lspcopy_curlwp_bad, %arg0 507 ldo R%panic(%r1), %r1 508 ldo R%Lspcopy_curlwp_bad(%arg0), %arg0 509 .call 510 bv,n %r0(%r1) 511 nop 512Lspcopy_curlwp_bad: 513 .asciz "spcopy: curlwp == NULL\n" 514 .align 8 515Lspcopy_curlwp_ok: 516#endif /* DIAGNOSTIC */ 517 ldil L%spcopy_fault, %r1 518 ldw L_ADDR(%r31), %r31 519 ldo R%spcopy_fault(%r1), %r1 520 ldi EFAULT, %ret0 521 stw %r1, U_PCB+PCB_ONFAULT(%r31) 522 523 /* Setup the space registers. */ 524 mfsp %sr2, %ret1 525 mtsp %arg0, %sr1 526 mtsp %arg2, %sr2 527 528 /* Get the len argument and do the copy. */ 529 ldw HPPA_FRAME_ARG(4)(%sp), %arg0 530#define _LABEL(l) __CONCAT(spcopy,l) 531 _COPY_FORWARD(%sr1,%arg1,%sr2,%arg3,%arg0) 532_LABEL(_done): 533 534 /* Return. */ 535 copy %r0, %ret0 536ALTENTRY(spcopy_fault) 537 stw %r0, U_PCB+PCB_ONFAULT(%r31) 538 bv %r0(%rp) 539 mtsp %ret1, %sr2 540EXIT(spcopy) 541#endif /* SPCOPY && !_STANDALONE */ 542 543#ifdef MEMCPY 544/* 545 * void *memcpy(void *restrict dst, const void *restrict src, size_t len); 546 * 547 * memcpy is specifically restricted to working on 548 * non-overlapping regions, so we can just copy forward. 549 */ 550LEAF_ENTRY(memcpy) 551 copy %arg0, %ret0 552#define _LABEL(l) __CONCAT(memcpy,l) 553 _COPY_FORWARD(%sr0,%arg1,%sr0,%arg0,%arg2) 554_LABEL(_done): 555 bv,n %r0(%rp) 556 nop 557EXIT(memcpy) 558#endif /* MEMCPY */ 559 560#ifdef BCOPY 561/* 562 * void bcopy(const void *src, void *dst, size_t len); 563 */ 564LEAF_ENTRY(bcopy) 565 copy %arg0, %r1 566 copy %arg1, %arg0 567 copy %r1, %arg1 568 /* FALLTHROUGH */ 569#define _LABEL_F(l) __CONCAT(bcopy_F,l) 570#define _LABEL_R(l) __CONCAT(bcopy_R,l) 571#endif 572 573#ifdef MEMMOVE 574/* 575 * void *memmove(void *dst, const void *src, size_t len); 576 */ 577LEAF_ENTRY(memmove) 578#define _LABEL_F(l) __CONCAT(memmove_F,l) 579#define _LABEL_R(l) __CONCAT(memmove_R,l) 580 copy %arg0, %ret0 581#endif /* MEMMOVE */ 582 583#if defined(BCOPY) || defined(MEMMOVE) 584 585 /* 586 * If src >= dst or src + len <= dst, we copy 587 * forward, else we copy in reverse. 588 */ 589 add %arg1, %arg2, %r1 590 comb,>>=,n %arg1, %arg0, 0 591 comb,>>,n %r1, %arg0, _LABEL_R(_go) 592 593#define _LABEL _LABEL_F 594 _COPY_FORWARD(%sr0,%arg1,%sr0,%arg0,%arg2) 595#undef _LABEL 596 597_LABEL_R(_go): 598#define _LABEL _LABEL_R 599 _COPY_REVERSE(%sr0,%arg1,%sr0,%arg0,%arg2) 600#undef _LABEL 601 602_LABEL_F(_done): 603_LABEL_R(_done): 604 bv,n %r0(%rp) 605 nop 606#ifdef BCOPY 607EXIT(bcopy) 608#else 609EXIT(memmove) 610#endif 611#endif /* BCOPY || MEMMOVE */ 612