arch/powerpc/lib/copyuser_power7.S

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License as published by
   4  * the Free Software Foundation; either version 2 of the License, or
   5  * (at your option) any later version.
   6  *
   7  * This program is distributed in the hope that it will be useful,
   8  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  10  * GNU General Public License for more details.
  11  *
  12  * You should have received a copy of the GNU General Public License
  13  * along with this program; if not, write to the Free Software
  14  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  15  *
  16  * Copyright (C) IBM Corporation, 2011
  17  *
  18  * Author: Anton Blanchard <anton@au.ibm.com>
  19  */
  20 #include <asm/ppc_asm.h>
  21
  22 #ifdef __BIG_ENDIAN__
  23 #define LVS(VRT,RA,RB)          lvsl    VRT,RA,RB
  24 #define VPERM(VRT,VRA,VRB,VRC)  vperm   VRT,VRA,VRB,VRC
  25 #else
  26 #define LVS(VRT,RA,RB)          lvsr    VRT,RA,RB
  27 #define VPERM(VRT,VRA,VRB,VRC)  vperm   VRT,VRB,VRA,VRC
  28 #endif
  29
  30         .macro err1
  31 100:
  32         .section __ex_table,"a"
  33         .align 3
  34         .llong 100b,.Ldo_err1
  35         .previous
  36         .endm
  37
  38         .macro err2
  39 200:
  40         .section __ex_table,"a"
  41         .align 3
  42         .llong 200b,.Ldo_err2
  43         .previous
  44         .endm
  45
  46 #ifdef CONFIG_ALTIVEC
  47         .macro err3
  48 300:
  49         .section __ex_table,"a"
  50         .align 3
  51         .llong 300b,.Ldo_err3
  52         .previous
  53         .endm
  54
  55         .macro err4
  56 400:
  57         .section __ex_table,"a"
  58         .align 3
  59         .llong 400b,.Ldo_err4
  60         .previous
  61         .endm
  62
  63
  64 .Ldo_err4:
  65         ld      r16,STK_REG(R16)(r1)
  66         ld      r15,STK_REG(R15)(r1)
  67         ld      r14,STK_REG(R14)(r1)
  68 .Ldo_err3:
  69         bl      exit_vmx_usercopy
  70         ld      r0,STACKFRAMESIZE+16(r1)
  71         mtlr    r0
  72         b       .Lexit
  73 #endif /* CONFIG_ALTIVEC */
  74
  75 .Ldo_err2:
  76         ld      r22,STK_REG(R22)(r1)
  77         ld      r21,STK_REG(R21)(r1)
  78         ld      r20,STK_REG(R20)(r1)
  79         ld      r19,STK_REG(R19)(r1)
  80         ld      r18,STK_REG(R18)(r1)
  81         ld      r17,STK_REG(R17)(r1)
  82         ld      r16,STK_REG(R16)(r1)
  83         ld      r15,STK_REG(R15)(r1)
  84         ld      r14,STK_REG(R14)(r1)
  85 .Lexit:
  86         addi    r1,r1,STACKFRAMESIZE
  87 .Ldo_err1:
  88         ld      r3,-STACKFRAMESIZE+STK_REG(R31)(r1)
  89         ld      r4,-STACKFRAMESIZE+STK_REG(R30)(r1)
  90         ld      r5,-STACKFRAMESIZE+STK_REG(R29)(r1)
  91         b       __copy_tofrom_user_base
  92
  93
  94 _GLOBAL(__copy_tofrom_user_power7)
  95 #ifdef CONFIG_ALTIVEC
  96         cmpldi  r5,16
  97         cmpldi  cr1,r5,4096
  98
  99         std     r3,-STACKFRAMESIZE+STK_REG(R31)(r1)
 100         std     r4,-STACKFRAMESIZE+STK_REG(R30)(r1)
 101         std     r5,-STACKFRAMESIZE+STK_REG(R29)(r1)
 102
 103         blt     .Lshort_copy
 104         bgt     cr1,.Lvmx_copy
 105 #else
 106         cmpldi  r5,16
 107
 108         std     r3,-STACKFRAMESIZE+STK_REG(R31)(r1)
 109         std     r4,-STACKFRAMESIZE+STK_REG(R30)(r1)
 110         std     r5,-STACKFRAMESIZE+STK_REG(R29)(r1)
 111
 112         blt     .Lshort_copy
 113 #endif
 114
 115 .Lnonvmx_copy:
 116         /* Get the source 8B aligned */
 117         neg     r6,r4
 118         mtocrf  0x01,r6
 119         clrldi  r6,r6,(64-3)
 120
 121         bf      cr7*4+3,1f
 122 err1;   lbz     r0,0(r4)
 123         addi    r4,r4,1
 124 err1;   stb     r0,0(r3)
 125         addi    r3,r3,1
 126
 127 1:      bf      cr7*4+2,2f
 128 err1;   lhz     r0,0(r4)
 129         addi    r4,r4,2
 130 err1;   sth     r0,0(r3)
 131         addi    r3,r3,2
 132
 133 2:      bf      cr7*4+1,3f
 134 err1;   lwz     r0,0(r4)
 135         addi    r4,r4,4
 136 err1;   stw     r0,0(r3)
 137         addi    r3,r3,4
 138
 139 3:      sub     r5,r5,r6
 140         cmpldi  r5,128
 141         blt     5f
 142
 143         mflr    r0
 144         stdu    r1,-STACKFRAMESIZE(r1)
 145         std     r14,STK_REG(R14)(r1)
 146         std     r15,STK_REG(R15)(r1)
 147         std     r16,STK_REG(R16)(r1)
 148         std     r17,STK_REG(R17)(r1)
 149         std     r18,STK_REG(R18)(r1)
 150         std     r19,STK_REG(R19)(r1)
 151         std     r20,STK_REG(R20)(r1)
 152         std     r21,STK_REG(R21)(r1)
 153         std     r22,STK_REG(R22)(r1)
 154         std     r0,STACKFRAMESIZE+16(r1)
 155
 156         srdi    r6,r5,7
 157         mtctr   r6
 158
 159         /* Now do cacheline (128B) sized loads and stores. */
 160         .align  5
 161 4:
 162 err2;   ld      r0,0(r4)
 163 err2;   ld      r6,8(r4)
 164 err2;   ld      r7,16(r4)
 165 err2;   ld      r8,24(r4)
 166 err2;   ld      r9,32(r4)
 167 err2;   ld      r10,40(r4)
 168 err2;   ld      r11,48(r4)
 169 err2;   ld      r12,56(r4)
 170 err2;   ld      r14,64(r4)
 171 err2;   ld      r15,72(r4)
 172 err2;   ld      r16,80(r4)
 173 err2;   ld      r17,88(r4)
 174 err2;   ld      r18,96(r4)
 175 err2;   ld      r19,104(r4)
 176 err2;   ld      r20,112(r4)
 177 err2;   ld      r21,120(r4)
 178         addi    r4,r4,128
 179 err2;   std     r0,0(r3)
 180 err2;   std     r6,8(r3)
 181 err2;   std     r7,16(r3)
 182 err2;   std     r8,24(r3)
 183 err2;   std     r9,32(r3)
 184 err2;   std     r10,40(r3)
 185 err2;   std     r11,48(r3)
 186 err2;   std     r12,56(r3)
 187 err2;   std     r14,64(r3)
 188 err2;   std     r15,72(r3)
 189 err2;   std     r16,80(r3)
 190 err2;   std     r17,88(r3)
 191 err2;   std     r18,96(r3)
 192 err2;   std     r19,104(r3)
 193 err2;   std     r20,112(r3)
 194 err2;   std     r21,120(r3)
 195         addi    r3,r3,128
 196         bdnz    4b
 197
 198         clrldi  r5,r5,(64-7)
 199
 200         ld      r14,STK_REG(R14)(r1)
 201         ld      r15,STK_REG(R15)(r1)
 202         ld      r16,STK_REG(R16)(r1)
 203         ld      r17,STK_REG(R17)(r1)
 204         ld      r18,STK_REG(R18)(r1)
 205         ld      r19,STK_REG(R19)(r1)
 206         ld      r20,STK_REG(R20)(r1)
 207         ld      r21,STK_REG(R21)(r1)
 208         ld      r22,STK_REG(R22)(r1)
 209         addi    r1,r1,STACKFRAMESIZE
 210
 211         /* Up to 127B to go */
 212 5:      srdi    r6,r5,4
 213         mtocrf  0x01,r6
 214
 215 6:      bf      cr7*4+1,7f
 216 err1;   ld      r0,0(r4)
 217 err1;   ld      r6,8(r4)
 218 err1;   ld      r7,16(r4)
 219 err1;   ld      r8,24(r4)
 220 err1;   ld      r9,32(r4)
 221 err1;   ld      r10,40(r4)
 222 err1;   ld      r11,48(r4)
 223 err1;   ld      r12,56(r4)
 224         addi    r4,r4,64
 225 err1;   std     r0,0(r3)
 226 err1;   std     r6,8(r3)
 227 err1;   std     r7,16(r3)
 228 err1;   std     r8,24(r3)
 229 err1;   std     r9,32(r3)
 230 err1;   std     r10,40(r3)
 231 err1;   std     r11,48(r3)
 232 err1;   std     r12,56(r3)
 233         addi    r3,r3,64
 234
 235         /* Up to 63B to go */
 236 7:      bf      cr7*4+2,8f
 237 err1;   ld      r0,0(r4)
 238 err1;   ld      r6,8(r4)
 239 err1;   ld      r7,16(r4)
 240 err1;   ld      r8,24(r4)
 241         addi    r4,r4,32
 242 err1;   std     r0,0(r3)
 243 err1;   std     r6,8(r3)
 244 err1;   std     r7,16(r3)
 245 err1;   std     r8,24(r3)
 246         addi    r3,r3,32
 247
 248         /* Up to 31B to go */
 249 8:      bf      cr7*4+3,9f
 250 err1;   ld      r0,0(r4)
 251 err1;   ld      r6,8(r4)
 252         addi    r4,r4,16
 253 err1;   std     r0,0(r3)
 254 err1;   std     r6,8(r3)
 255         addi    r3,r3,16
 256
 257 9:      clrldi  r5,r5,(64-4)
 258
 259         /* Up to 15B to go */
 260 .Lshort_copy:
 261         mtocrf  0x01,r5
 262         bf      cr7*4+0,12f
 263 err1;   lwz     r0,0(r4)        /* Less chance of a reject with word ops */
 264 err1;   lwz     r6,4(r4)
 265         addi    r4,r4,8
 266 err1;   stw     r0,0(r3)
 267 err1;   stw     r6,4(r3)
 268         addi    r3,r3,8
 269
 270 12:     bf      cr7*4+1,13f
 271 err1;   lwz     r0,0(r4)
 272         addi    r4,r4,4
 273 err1;   stw     r0,0(r3)
 274         addi    r3,r3,4
 275
 276 13:     bf      cr7*4+2,14f
 277 err1;   lhz     r0,0(r4)
 278         addi    r4,r4,2
 279 err1;   sth     r0,0(r3)
 280         addi    r3,r3,2
 281
 282 14:     bf      cr7*4+3,15f
 283 err1;   lbz     r0,0(r4)
 284 err1;   stb     r0,0(r3)
 285
 286 15:     li      r3,0
 287         blr
 288
 289 .Lunwind_stack_nonvmx_copy:
 290         addi    r1,r1,STACKFRAMESIZE
 291         b       .Lnonvmx_copy
 292
 293 #ifdef CONFIG_ALTIVEC
 294 .Lvmx_copy:
 295         mflr    r0
 296         std     r0,16(r1)
 297         stdu    r1,-STACKFRAMESIZE(r1)
 298         bl      enter_vmx_usercopy
 299         cmpwi   cr1,r3,0
 300         ld      r0,STACKFRAMESIZE+16(r1)
 301         ld      r3,STK_REG(R31)(r1)
 302         ld      r4,STK_REG(R30)(r1)
 303         ld      r5,STK_REG(R29)(r1)
 304         mtlr    r0
 305
 306         /*
 307          * We prefetch both the source and destination using enhanced touch
 308          * instructions. We use a stream ID of 0 for the load side and
 309          * 1 for the store side.
 310          */
 311         clrrdi  r6,r4,7
 312         clrrdi  r9,r3,7
 313         ori     r9,r9,1         /* stream=1 */
 314
 315         srdi    r7,r5,7         /* length in cachelines, capped at 0x3FF */
 316         cmpldi  r7,0x3FF
 317         ble     1f
 318         li      r7,0x3FF
 319 1:      lis     r0,0x0E00       /* depth=7 */
 320         sldi    r7,r7,7
 321         or      r7,r7,r0
 322         ori     r10,r7,1        /* stream=1 */
 323
 324         lis     r8,0x8000       /* GO=1 */
 325         clrldi  r8,r8,32
 326
 327 .machine push
 328 .machine "power4"
 329         /* setup read stream 0 */
 330         dcbt    r0,r6,0b01000   /* addr from */
 331         dcbt    r0,r7,0b01010   /* length and depth from */
 332         /* setup write stream 1 */
 333         dcbtst  r0,r9,0b01000   /* addr to */
 334         dcbtst  r0,r10,0b01010  /* length and depth to */
 335         eieio
 336         dcbt    r0,r8,0b01010   /* all streams GO */
 337 .machine pop
 338
 339         beq     cr1,.Lunwind_stack_nonvmx_copy
 340
 341         /*
 342          * If source and destination are not relatively aligned we use a
 343          * slower permute loop.
 344          */
 345         xor     r6,r4,r3
 346         rldicl. r6,r6,0,(64-4)
 347         bne     .Lvmx_unaligned_copy
 348
 349         /* Get the destination 16B aligned */
 350         neg     r6,r3
 351         mtocrf  0x01,r6
 352         clrldi  r6,r6,(64-4)
 353
 354         bf      cr7*4+3,1f
 355 err3;   lbz     r0,0(r4)
 356         addi    r4,r4,1
 357 err3;   stb     r0,0(r3)
 358         addi    r3,r3,1
 359
 360 1:      bf      cr7*4+2,2f
 361 err3;   lhz     r0,0(r4)
 362         addi    r4,r4,2
 363 err3;   sth     r0,0(r3)
 364         addi    r3,r3,2
 365
 366 2:      bf      cr7*4+1,3f
 367 err3;   lwz     r0,0(r4)
 368         addi    r4,r4,4
 369 err3;   stw     r0,0(r3)
 370         addi    r3,r3,4
 371
 372 3:      bf      cr7*4+0,4f
 373 err3;   ld      r0,0(r4)
 374         addi    r4,r4,8
 375 err3;   std     r0,0(r3)
 376         addi    r3,r3,8
 377
 378 4:      sub     r5,r5,r6
 379
 380         /* Get the desination 128B aligned */
 381         neg     r6,r3
 382         srdi    r7,r6,4
 383         mtocrf  0x01,r7
 384         clrldi  r6,r6,(64-7)
 385
 386         li      r9,16
 387         li      r10,32
 388         li      r11,48
 389
 390         bf      cr7*4+3,5f
 391 err3;   lvx     v1,r0,r4
 392         addi    r4,r4,16
 393 err3;   stvx    v1,r0,r3
 394         addi    r3,r3,16
 395
 396 5:      bf      cr7*4+2,6f
 397 err3;   lvx     v1,r0,r4
 398 err3;   lvx     v0,r4,r9
 399         addi    r4,r4,32
 400 err3;   stvx    v1,r0,r3
 401 err3;   stvx    v0,r3,r9
 402         addi    r3,r3,32
 403
 404 6:      bf      cr7*4+1,7f
 405 err3;   lvx     v3,r0,r4
 406 err3;   lvx     v2,r4,r9
 407 err3;   lvx     v1,r4,r10
 408 err3;   lvx     v0,r4,r11
 409         addi    r4,r4,64
 410 err3;   stvx    v3,r0,r3
 411 err3;   stvx    v2,r3,r9
 412 err3;   stvx    v1,r3,r10
 413 err3;   stvx    v0,r3,r11
 414         addi    r3,r3,64
 415
 416 7:      sub     r5,r5,r6
 417         srdi    r6,r5,7
 418
 419         std     r14,STK_REG(R14)(r1)
 420         std     r15,STK_REG(R15)(r1)
 421         std     r16,STK_REG(R16)(r1)
 422
 423         li      r12,64
 424         li      r14,80
 425         li      r15,96
 426         li      r16,112
 427
 428         mtctr   r6
 429
 430         /*
 431          * Now do cacheline sized loads and stores. By this stage the
 432          * cacheline stores are also cacheline aligned.
 433          */
 434         .align  5
 435 8:
 436 err4;   lvx     v7,r0,r4
 437 err4;   lvx     v6,r4,r9
 438 err4;   lvx     v5,r4,r10
 439 err4;   lvx     v4,r4,r11
 440 err4;   lvx     v3,r4,r12
 441 err4;   lvx     v2,r4,r14
 442 err4;   lvx     v1,r4,r15
 443 err4;   lvx     v0,r4,r16
 444         addi    r4,r4,128
 445 err4;   stvx    v7,r0,r3
 446 err4;   stvx    v6,r3,r9
 447 err4;   stvx    v5,r3,r10
 448 err4;   stvx    v4,r3,r11
 449 err4;   stvx    v3,r3,r12
 450 err4;   stvx    v2,r3,r14
 451 err4;   stvx    v1,r3,r15
 452 err4;   stvx    v0,r3,r16
 453         addi    r3,r3,128
 454         bdnz    8b
 455
 456         ld      r14,STK_REG(R14)(r1)
 457         ld      r15,STK_REG(R15)(r1)
 458         ld      r16,STK_REG(R16)(r1)
 459
 460         /* Up to 127B to go */
 461         clrldi  r5,r5,(64-7)
 462         srdi    r6,r5,4
 463         mtocrf  0x01,r6
 464
 465         bf      cr7*4+1,9f
 466 err3;   lvx     v3,r0,r4
 467 err3;   lvx     v2,r4,r9
 468 err3;   lvx     v1,r4,r10
 469 err3;   lvx     v0,r4,r11
 470         addi    r4,r4,64
 471 err3;   stvx    v3,r0,r3
 472 err3;   stvx    v2,r3,r9
 473 err3;   stvx    v1,r3,r10
 474 err3;   stvx    v0,r3,r11
 475         addi    r3,r3,64
 476
 477 9:      bf      cr7*4+2,10f
 478 err3;   lvx     v1,r0,r4
 479 err3;   lvx     v0,r4,r9
 480         addi    r4,r4,32
 481 err3;   stvx    v1,r0,r3
 482 err3;   stvx    v0,r3,r9
 483         addi    r3,r3,32
 484
 485 10:     bf      cr7*4+3,11f
 486 err3;   lvx     v1,r0,r4
 487         addi    r4,r4,16
 488 err3;   stvx    v1,r0,r3
 489         addi    r3,r3,16
 490
 491         /* Up to 15B to go */
 492 11:     clrldi  r5,r5,(64-4)
 493         mtocrf  0x01,r5
 494         bf      cr7*4+0,12f
 495 err3;   ld      r0,0(r4)
 496         addi    r4,r4,8
 497 err3;   std     r0,0(r3)
 498         addi    r3,r3,8
 499
 500 12:     bf      cr7*4+1,13f
 501 err3;   lwz     r0,0(r4)
 502         addi    r4,r4,4
 503 err3;   stw     r0,0(r3)
 504         addi    r3,r3,4
 505
 506 13:     bf      cr7*4+2,14f
 507 err3;   lhz     r0,0(r4)
 508         addi    r4,r4,2
 509 err3;   sth     r0,0(r3)
 510         addi    r3,r3,2
 511
 512 14:     bf      cr7*4+3,15f
 513 err3;   lbz     r0,0(r4)
 514 err3;   stb     r0,0(r3)
 515
 516 15:     addi    r1,r1,STACKFRAMESIZE
 517         b       exit_vmx_usercopy       /* tail call optimise */
 518
 519 .Lvmx_unaligned_copy:
 520         /* Get the destination 16B aligned */
 521         neg     r6,r3
 522         mtocrf  0x01,r6
 523         clrldi  r6,r6,(64-4)
 524
 525         bf      cr7*4+3,1f
 526 err3;   lbz     r0,0(r4)
 527         addi    r4,r4,1
 528 err3;   stb     r0,0(r3)
 529         addi    r3,r3,1
 530
 531 1:      bf      cr7*4+2,2f
 532 err3;   lhz     r0,0(r4)
 533         addi    r4,r4,2
 534 err3;   sth     r0,0(r3)
 535         addi    r3,r3,2
 536
 537 2:      bf      cr7*4+1,3f
 538 err3;   lwz     r0,0(r4)
 539         addi    r4,r4,4
 540 err3;   stw     r0,0(r3)
 541         addi    r3,r3,4
 542
 543 3:      bf      cr7*4+0,4f
 544 err3;   lwz     r0,0(r4)        /* Less chance of a reject with word ops */
 545 err3;   lwz     r7,4(r4)
 546         addi    r4,r4,8
 547 err3;   stw     r0,0(r3)
 548 err3;   stw     r7,4(r3)
 549         addi    r3,r3,8
 550
 551 4:      sub     r5,r5,r6
 552
 553         /* Get the desination 128B aligned */
 554         neg     r6,r3
 555         srdi    r7,r6,4
 556         mtocrf  0x01,r7
 557         clrldi  r6,r6,(64-7)
 558
 559         li      r9,16
 560         li      r10,32
 561         li      r11,48
 562
 563         LVS(v16,0,r4)           /* Setup permute control vector */
 564 err3;   lvx     v0,0,r4
 565         addi    r4,r4,16
 566
 567         bf      cr7*4+3,5f
 568 err3;   lvx     v1,r0,r4
 569         VPERM(v8,v0,v1,v16)
 570         addi    r4,r4,16
 571 err3;   stvx    v8,r0,r3
 572         addi    r3,r3,16
 573         vor     v0,v1,v1
 574
 575 5:      bf      cr7*4+2,6f
 576 err3;   lvx     v1,r0,r4
 577         VPERM(v8,v0,v1,v16)
 578 err3;   lvx     v0,r4,r9
 579         VPERM(v9,v1,v0,v16)
 580         addi    r4,r4,32
 581 err3;   stvx    v8,r0,r3
 582 err3;   stvx    v9,r3,r9
 583         addi    r3,r3,32
 584
 585 6:      bf      cr7*4+1,7f
 586 err3;   lvx     v3,r0,r4
 587         VPERM(v8,v0,v3,v16)
 588 err3;   lvx     v2,r4,r9
 589         VPERM(v9,v3,v2,v16)
 590 err3;   lvx     v1,r4,r10
 591         VPERM(v10,v2,v1,v16)
 592 err3;   lvx     v0,r4,r11
 593         VPERM(v11,v1,v0,v16)
 594         addi    r4,r4,64
 595 err3;   stvx    v8,r0,r3
 596 err3;   stvx    v9,r3,r9
 597 err3;   stvx    v10,r3,r10
 598 err3;   stvx    v11,r3,r11
 599         addi    r3,r3,64
 600
 601 7:      sub     r5,r5,r6
 602         srdi    r6,r5,7
 603
 604         std     r14,STK_REG(R14)(r1)
 605         std     r15,STK_REG(R15)(r1)
 606         std     r16,STK_REG(R16)(r1)
 607
 608         li      r12,64
 609         li      r14,80
 610         li      r15,96
 611         li      r16,112
 612
 613         mtctr   r6
 614
 615         /*
 616          * Now do cacheline sized loads and stores. By this stage the
 617          * cacheline stores are also cacheline aligned.
 618          */
 619         .align  5
 620 8:
 621 err4;   lvx     v7,r0,r4
 622         VPERM(v8,v0,v7,v16)
 623 err4;   lvx     v6,r4,r9
 624         VPERM(v9,v7,v6,v16)
 625 err4;   lvx     v5,r4,r10
 626         VPERM(v10,v6,v5,v16)
 627 err4;   lvx     v4,r4,r11
 628         VPERM(v11,v5,v4,v16)
 629 err4;   lvx     v3,r4,r12
 630         VPERM(v12,v4,v3,v16)
 631 err4;   lvx     v2,r4,r14
 632         VPERM(v13,v3,v2,v16)
 633 err4;   lvx     v1,r4,r15
 634         VPERM(v14,v2,v1,v16)
 635 err4;   lvx     v0,r4,r16
 636         VPERM(v15,v1,v0,v16)
 637         addi    r4,r4,128
 638 err4;   stvx    v8,r0,r3
 639 err4;   stvx    v9,r3,r9
 640 err4;   stvx    v10,r3,r10
 641 err4;   stvx    v11,r3,r11
 642 err4;   stvx    v12,r3,r12
 643 err4;   stvx    v13,r3,r14
 644 err4;   stvx    v14,r3,r15
 645 err4;   stvx    v15,r3,r16
 646         addi    r3,r3,128
 647         bdnz    8b
 648
 649         ld      r14,STK_REG(R14)(r1)
 650         ld      r15,STK_REG(R15)(r1)
 651         ld      r16,STK_REG(R16)(r1)
 652
 653         /* Up to 127B to go */
 654         clrldi  r5,r5,(64-7)
 655         srdi    r6,r5,4
 656         mtocrf  0x01,r6
 657
 658         bf      cr7*4+1,9f
 659 err3;   lvx     v3,r0,r4
 660         VPERM(v8,v0,v3,v16)
 661 err3;   lvx     v2,r4,r9
 662         VPERM(v9,v3,v2,v16)
 663 err3;   lvx     v1,r4,r10
 664         VPERM(v10,v2,v1,v16)
 665 err3;   lvx     v0,r4,r11
 666         VPERM(v11,v1,v0,v16)
 667         addi    r4,r4,64
 668 err3;   stvx    v8,r0,r3
 669 err3;   stvx    v9,r3,r9
 670 err3;   stvx    v10,r3,r10
 671 err3;   stvx    v11,r3,r11
 672         addi    r3,r3,64
 673
 674 9:      bf      cr7*4+2,10f
 675 err3;   lvx     v1,r0,r4
 676         VPERM(v8,v0,v1,v16)
 677 err3;   lvx     v0,r4,r9
 678         VPERM(v9,v1,v0,v16)
 679         addi    r4,r4,32
 680 err3;   stvx    v8,r0,r3
 681 err3;   stvx    v9,r3,r9
 682         addi    r3,r3,32
 683
 684 10:     bf      cr7*4+3,11f
 685 err3;   lvx     v1,r0,r4
 686         VPERM(v8,v0,v1,v16)
 687         addi    r4,r4,16
 688 err3;   stvx    v8,r0,r3
 689         addi    r3,r3,16
 690
 691         /* Up to 15B to go */
 692 11:     clrldi  r5,r5,(64-4)
 693         addi    r4,r4,-16       /* Unwind the +16 load offset */
 694         mtocrf  0x01,r5
 695         bf      cr7*4+0,12f
 696 err3;   lwz     r0,0(r4)        /* Less chance of a reject with word ops */
 697 err3;   lwz     r6,4(r4)
 698         addi    r4,r4,8
 699 err3;   stw     r0,0(r3)
 700 err3;   stw     r6,4(r3)
 701         addi    r3,r3,8
 702
 703 12:     bf      cr7*4+1,13f
 704 err3;   lwz     r0,0(r4)
 705         addi    r4,r4,4
 706 err3;   stw     r0,0(r3)
 707         addi    r3,r3,4
 708
 709 13:     bf      cr7*4+2,14f
 710 err3;   lhz     r0,0(r4)
 711         addi    r4,r4,2
 712 err3;   sth     r0,0(r3)
 713         addi    r3,r3,2
 714
 715 14:     bf      cr7*4+3,15f
 716 err3;   lbz     r0,0(r4)
 717 err3;   stb     r0,0(r3)
 718
 719 15:     addi    r1,r1,STACKFRAMESIZE
 720         b       exit_vmx_usercopy       /* tail call optimise */
 721 #endif /* CONFIG_ALTIVEC */