Blame view
arch/cris/arch-v10/lib/usercopy.c
15.8 KB
1da177e4c
Linux-2.6.12-rc2
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 |
/*
* User address space access functions.
* The non-inlined parts of asm-cris/uaccess.h are here.
*
* Copyright (C) 2000, Axis Communications AB.
*
* Written by Hans-Peter Nilsson.
* Pieces used from memcpy, originally by Kenny Ranerup long time ago.
*/
#include <asm/uaccess.h>
/* Asm:s have been tweaked (within the domain of correctness) to give
satisfactory results for "gcc version 2.96 20000427 (experimental)".
Check regularly...
Note that the PC saved at a bus-fault is the address *after* the
faulting instruction, which means the branch-target for instructions in
delay-slots for taken branches. Note also that the postincrement in
the instruction is performed regardless of bus-fault; the register is
seen updated in fault handlers.
Oh, and on the code formatting issue, to whomever feels like "fixing
it" to Conformity: I'm too "lazy", but why don't you go ahead and "fix"
string.c too. I just don't think too many people will hack this file
for the code format to be an issue. */
/* Copy to userspace. This is based on the memcpy used for
kernel-to-kernel copying; see "string.c". */
|
dbd3c7e1b
CRIS: Export miss...
|
32 |
unsigned long __copy_user(void __user *pdst, const void *psrc, unsigned long pn) |
|
1da177e4c
Linux-2.6.12-rc2
|
33 34 35 36 37 38 |
{
/* We want the parameters put in special registers.
Make sure the compiler is able to make something useful of this.
As it is now: r10 -> r13; r11 -> r11 (nop); r12 -> r12 (nop).
FIXME: Comment for old gcc version. Check.
|
49b4ff330
spelling fixes: a...
|
39 |
If gcc was alright, it really would need no temporaries, and no |
|
1da177e4c
Linux-2.6.12-rc2
|
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 |
stack space to save stuff on. */
register char *dst __asm__ ("r13") = pdst;
register const char *src __asm__ ("r11") = psrc;
register int n __asm__ ("r12") = pn;
register int retn __asm__ ("r10") = 0;
/* When src is aligned but not dst, this makes a few extra needless
cycles. I believe it would take as many to check that the
re-alignment was unnecessary. */
if (((unsigned long) dst & 3) != 0
/* Don't align if we wouldn't copy more than a few bytes; so we
don't have to check further for overflows. */
&& n >= 3)
{
if ((unsigned long) dst & 1)
{
__asm_copy_to_user_1 (dst, src, retn);
n--;
}
if ((unsigned long) dst & 2)
{
__asm_copy_to_user_2 (dst, src, retn);
n -= 2;
}
}
/* Decide which copying method to use. */
if (n >= 44*2) /* Break even between movem and
move16 is at 38.7*2, but modulo 44. */
{
/* For large copies we use 'movem'. */
/* It is not optimal to tell the compiler about clobbering any
registers; that will move the saving/restoring of those registers
to the function prologue/epilogue, and make non-movem sizes
suboptimal.
This method is not foolproof; it assumes that the "asm reg"
declarations at the beginning of the function really are used
here (beware: they may be moved to temporary registers).
This way, we do not have to save/move the registers around into
temporaries; we can safely use them straight away.
If you want to check that the allocation was right; then
check the equalities in the first comment. It should say
"r13=r13, r11=r11, r12=r12". */
__asm__ volatile ("\
.ifnc %0%1%2%3,$r13$r11$r12$r10
\
.err
\
.endif
\
|
2b05d2b3b
CRISv10 usercopy ...
|
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 |
\ ;; Save the registers we'll use in the movem process \ ;; on the stack. \ subq 11*4,$sp \ movem $r10,[$sp] \ \ ;; Now we've got this: \ ;; r11 - src \ ;; r13 - dst \ ;; r12 - n \ \ ;; Update n for the first loop \ subq 44,$r12 \ \ ; Since the noted PC of a faulting instruction in a delay-slot of a taken \ ; branch, is that of the branch target, we actually point at the from-movem \ ; for this case. There is no ambiguity here; if there was a fault in that \ ; instruction (meaning a kernel oops), the faulted PC would be the address \ ; after *that* movem. \ \ 0: \ movem [$r11+],$r10 \ subq 44,$r12 \ bge 0b \ movem $r10,[$r13+] \ 1: \ addq 44,$r12 ;; compensate for last loop underflowing n \ \ ;; Restore registers from stack \ movem [$sp+],$r10 \ 2: \ .section .fixup,\"ax\" \ \ ; To provide a correct count in r10 of bytes that failed to be copied, \ ; we jump back into the loop if the loop-branch was taken. There is no \ ; performance penalty for sany use; the program will segfault soon enough. \ \ 3: \ move.d [$sp],$r10 \ addq 44,$r10 \ move.d $r10,[$sp] \ jump 0b \ 4: \ movem [$sp+],$r10 \ addq 44,$r10 \ addq 44,$r12 \ jump 2b \ \ .previous \ .section __ex_table,\"a\" \ .dword 0b,3b \ .dword 1b,4b \ |
|
1da177e4c
Linux-2.6.12-rc2
|
200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 |
.previous"
/* Outputs */ : "=r" (dst), "=r" (src), "=r" (n), "=r" (retn)
/* Inputs */ : "0" (dst), "1" (src), "2" (n), "3" (retn));
}
/* Either we directly start copying, using dword copying in a loop, or
we copy as much as possible with 'movem' and then the last block (<44
bytes) is copied here. This will work since 'movem' will have
updated SRC, DST and N. */
while (n >= 16)
{
__asm_copy_to_user_16 (dst, src, retn);
n -= 16;
}
/* Having a separate by-four loops cuts down on cache footprint.
FIXME: Test with and without; increasing switch to be 0..15. */
while (n >= 4)
{
__asm_copy_to_user_4 (dst, src, retn);
n -= 4;
}
switch (n)
{
case 0:
break;
case 1:
__asm_copy_to_user_1 (dst, src, retn);
break;
case 2:
__asm_copy_to_user_2 (dst, src, retn);
break;
case 3:
__asm_copy_to_user_3 (dst, src, retn);
break;
}
return retn;
}
|
|
dbd3c7e1b
CRIS: Export miss...
|
243 |
EXPORT_SYMBOL(__copy_user); |
|
1da177e4c
Linux-2.6.12-rc2
|
244 245 246 247 |
/* Copy from user to kernel, zeroing the bytes that were inaccessible in
userland. The return-value is the number of bytes that were
inaccessible. */
|
|
dbd3c7e1b
CRIS: Export miss...
|
248 249 |
unsigned long __copy_user_zeroing(void *pdst, const void __user *psrc, unsigned long pn) |
|
1da177e4c
Linux-2.6.12-rc2
|
250 251 252 253 254 255 |
{
/* We want the parameters put in special registers.
Make sure the compiler is able to make something useful of this.
As it is now: r10 -> r13; r11 -> r11 (nop); r12 -> r12 (nop).
FIXME: Comment for old gcc version. Check.
|
|
49b4ff330
spelling fixes: a...
|
256 |
If gcc was alright, it really would need no temporaries, and no |
|
1da177e4c
Linux-2.6.12-rc2
|
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 |
stack space to save stuff on. */
register char *dst __asm__ ("r13") = pdst;
register const char *src __asm__ ("r11") = psrc;
register int n __asm__ ("r12") = pn;
register int retn __asm__ ("r10") = 0;
/* The best reason to align src is that we then know that a read-fault
was for aligned bytes; there's no 1..3 remaining good bytes to
pickle. */
if (((unsigned long) src & 3) != 0)
{
if (((unsigned long) src & 1) && n != 0)
{
__asm_copy_from_user_1 (dst, src, retn);
n--;
}
if (((unsigned long) src & 2) && n >= 2)
{
__asm_copy_from_user_2 (dst, src, retn);
n -= 2;
}
/* We only need one check after the unalignment-adjustments, because
if both adjustments were done, either both or neither reference
had an exception. */
if (retn != 0)
goto copy_exception_bytes;
}
/* Decide which copying method to use. */
if (n >= 44*2) /* Break even between movem and
move16 is at 38.7*2, but modulo 44.
FIXME: We use move4 now. */
{
/* For large copies we use 'movem' */
/* It is not optimal to tell the compiler about clobbering any
registers; that will move the saving/restoring of those registers
to the function prologue/epilogue, and make non-movem sizes
suboptimal.
This method is not foolproof; it assumes that the "asm reg"
declarations at the beginning of the function really are used
here (beware: they may be moved to temporary registers).
This way, we do not have to save/move the registers around into
temporaries; we can safely use them straight away.
If you want to check that the allocation was right; then
check the equalities in the first comment. It should say
"r13=r13, r11=r11, r12=r12" */
|
|
2b05d2b3b
CRISv10 usercopy ...
|
309 310 |
__asm__ volatile ("
\
|
|
1da177e4c
Linux-2.6.12-rc2
|
311 312 313 314 315 316 |
.ifnc %0%1%2%3,$r13$r11$r12$r10 \ .err \ .endif \ |
|
2b05d2b3b
CRISv10 usercopy ...
|
317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 |
\ ;; Save the registers we'll use in the movem process \ ;; on the stack. \ subq 11*4,$sp \ movem $r10,[$sp] \ \ ;; Now we've got this: \ ;; r11 - src \ ;; r13 - dst \ ;; r12 - n \ \ ;; Update n for the first loop \ subq 44,$r12 \ 0: \ movem [$r11+],$r10 \ 1: \ subq 44,$r12 \ bge 0b \ movem $r10,[$r13+] \ \ addq 44,$r12 ;; compensate for last loop underflowing n \ \ ;; Restore registers from stack \ movem [$sp+],$r10 \ 4: \ .section .fixup,\"ax\" \ \ ;; Do not jump back into the loop if we fail. For some uses, we get a \ ;; page fault somewhere on the line. Without checking for page limits, \ ;; we don't know where, but we need to copy accurately and keep an \ ;; accurate count; not just clear the whole line. To do that, we fall \ ;; down in the code below, proceeding with smaller amounts. It should \ ;; be kept in mind that we have to cater to code like what at one time \ ;; was in fs/super.c: \ ;; i = size - copy_from_user((void *)page, data, size); \ ;; which would cause repeated faults while clearing the remainder of \ ;; the SIZE bytes at PAGE after the first fault. \ ;; A caveat here is that we must not fall through from a failing page \ ;; to a valid page. \ \ 3: \ movem [$sp+],$r10 \ addq 44,$r12 ;; Get back count before faulting point. \ subq 44,$r11 ;; Get back pointer to faulting movem-line. \ jump 4b ;; Fall through, pretending the fault didn't happen. \ \ .previous \ .section __ex_table,\"a\" \ .dword 1b,3b \ |
|
1da177e4c
Linux-2.6.12-rc2
|
415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 |
.previous"
/* Outputs */ : "=r" (dst), "=r" (src), "=r" (n), "=r" (retn)
/* Inputs */ : "0" (dst), "1" (src), "2" (n), "3" (retn));
}
/* Either we directly start copying here, using dword copying in a loop,
or we copy as much as possible with 'movem' and then the last block
(<44 bytes) is copied here. This will work since 'movem' will have
updated src, dst and n. (Except with failing src.)
Since we want to keep src accurate, we can't use
__asm_copy_from_user_N with N != (1, 2, 4); it updates dst and
retn, but not src (by design; it's value is ignored elsewhere). */
while (n >= 4)
{
__asm_copy_from_user_4 (dst, src, retn);
n -= 4;
if (retn)
goto copy_exception_bytes;
}
/* If we get here, there were no memory read faults. */
switch (n)
{
/* These copies are at least "naturally aligned" (so we don't have
to check each byte), due to the src alignment code before the
movem loop. The *_3 case *will* get the correct count for retn. */
case 0:
/* This case deliberately left in (if you have doubts check the
generated assembly code). */
break;
case 1:
__asm_copy_from_user_1 (dst, src, retn);
break;
case 2:
__asm_copy_from_user_2 (dst, src, retn);
break;
case 3:
__asm_copy_from_user_3 (dst, src, retn);
break;
}
/* If we get here, retn correctly reflects the number of failing
bytes. */
return retn;
copy_exception_bytes:
/* We already have "retn" bytes cleared, and need to clear the
remaining "n" bytes. A non-optimized simple byte-for-byte in-line
memset is preferred here, since this isn't speed-critical code and
we'd rather have this a leaf-function than calling memset. */
{
char *endp;
for (endp = dst + n; dst < endp; dst++)
*dst = 0;
}
return retn + n;
}
|
|
dbd3c7e1b
CRIS: Export miss...
|
478 |
EXPORT_SYMBOL(__copy_user_zeroing); |
|
1da177e4c
Linux-2.6.12-rc2
|
479 480 |
/* Zero userspace. */ |
|
dbd3c7e1b
CRIS: Export miss...
|
481 |
unsigned long __do_clear_user(void __user *pto, unsigned long pn) |
|
1da177e4c
Linux-2.6.12-rc2
|
482 483 484 485 486 487 |
{
/* We want the parameters put in special registers.
Make sure the compiler is able to make something useful of this.
As it is now: r10 -> r13; r11 -> r11 (nop); r12 -> r12 (nop).
FIXME: Comment for old gcc version. Check.
|
|
49b4ff330
spelling fixes: a...
|
488 |
If gcc was alright, it really would need no temporaries, and no |
|
1da177e4c
Linux-2.6.12-rc2
|
489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 |
stack space to save stuff on. */
register char *dst __asm__ ("r13") = pto;
register int n __asm__ ("r12") = pn;
register int retn __asm__ ("r10") = 0;
if (((unsigned long) dst & 3) != 0
/* Don't align if we wouldn't copy more than a few bytes. */
&& n >= 3)
{
if ((unsigned long) dst & 1)
{
__asm_clear_1 (dst, retn);
n--;
}
if ((unsigned long) dst & 2)
{
__asm_clear_2 (dst, retn);
n -= 2;
}
}
/* Decide which copying method to use.
FIXME: This number is from the "ordinary" kernel memset. */
if (n >= (1*48))
{
/* For large clears we use 'movem' */
/* It is not optimal to tell the compiler about clobbering any
call-saved registers; that will move the saving/restoring of
those registers to the function prologue/epilogue, and make
non-movem sizes suboptimal.
This method is not foolproof; it assumes that the "asm reg"
declarations at the beginning of the function really are used
here (beware: they may be moved to temporary registers).
This way, we do not have to save/move the registers around into
temporaries; we can safely use them straight away.
If you want to check that the allocation was right; then
check the equalities in the first comment. It should say
something like "r13=r13, r11=r11, r12=r12". */
|
|
2b05d2b3b
CRISv10 usercopy ...
|
533 534 |
__asm__ volatile ("
\
|
|
1da177e4c
Linux-2.6.12-rc2
|
535 536 537 538 539 540 |
.ifnc %0%1%2,$r13$r12$r10 \ .err \ .endif \ |
|
2b05d2b3b
CRISv10 usercopy ...
|
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 |
\ ;; Save the registers we'll clobber in the movem process \ ;; on the stack. Don't mention them to gcc, it will only be \ ;; upset. \ subq 11*4,$sp \ movem $r10,[$sp] \ \ clear.d $r0 \ clear.d $r1 \ clear.d $r2 \ clear.d $r3 \ clear.d $r4 \ clear.d $r5 \ clear.d $r6 \ clear.d $r7 \ clear.d $r8 \ clear.d $r9 \ clear.d $r10 \ clear.d $r11 \ \ ;; Now we've got this: \ ;; r13 - dst \ ;; r12 - n \ \ ;; Update n for the first loop \ subq 12*4,$r12 \ 0: \ subq 12*4,$r12 \ bge 0b \ movem $r11,[$r13+] \ 1: \ addq 12*4,$r12 ;; compensate for last loop underflowing n \ \ ;; Restore registers from stack \ movem [$sp+],$r10 \ 2: \ .section .fixup,\"ax\" \ 3: \ move.d [$sp],$r10 \ addq 12*4,$r10 \ move.d $r10,[$sp] \ clear.d $r10 \ jump 0b \ \ 4: \ movem [$sp+],$r10 \ addq 12*4,$r10 \ addq 12*4,$r12 \ jump 2b \ \ .previous \ .section __ex_table,\"a\" \ .dword 0b,3b \ .dword 1b,4b \ |
|
1da177e4c
Linux-2.6.12-rc2
|
649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 |
.previous"
/* Outputs */ : "=r" (dst), "=r" (n), "=r" (retn)
/* Inputs */ : "0" (dst), "1" (n), "2" (retn)
/* Clobber */ : "r11");
}
while (n >= 16)
{
__asm_clear_16 (dst, retn);
n -= 16;
}
/* Having a separate by-four loops cuts down on cache footprint.
FIXME: Test with and without; increasing switch to be 0..15. */
while (n >= 4)
{
__asm_clear_4 (dst, retn);
n -= 4;
}
switch (n)
{
case 0:
break;
case 1:
__asm_clear_1 (dst, retn);
break;
case 2:
__asm_clear_2 (dst, retn);
break;
case 3:
__asm_clear_3 (dst, retn);
break;
}
return retn;
}
|
|
dbd3c7e1b
CRIS: Export miss...
|
687 |
EXPORT_SYMBOL(__do_clear_user); |