libffi/src/sparc/v9.S

   1 /* -----------------------------------------------------------------------
   2    v9.S - Copyright (c) 2000, 2003, 2004, 2008 Red Hat, Inc.
   3
   4    SPARC 64-bit Foreign Function Interface
   5
   6    Permission is hereby granted, free of charge, to any person obtaining
   7    a copy of this software and associated documentation files (the
   8    ``Software''), to deal in the Software without restriction, including
   9    without limitation the rights to use, copy, modify, merge, publish,
  10    distribute, sublicense, and/or sell copies of the Software, and to
  11    permit persons to whom the Software is furnished to do so, subject to
  12    the following conditions:
  13
  14    The above copyright notice and this permission notice shall be included
  15    in all copies or substantial portions of the Software.
  16
  17    THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND,
  18    EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19    MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  20    NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
  21    HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
  22    WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  23    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  24    DEALINGS IN THE SOFTWARE.
  25    ----------------------------------------------------------------------- */
  26
  27 #define LIBFFI_ASM
  28 #include <fficonfig.h>
  29 #include <ffi.h>
  30 #include "internal.h"
  31
  32 #ifdef SPARC64
  33
  34 #define C2(X, Y)  X ## Y
  35 #define C1(X, Y)  C2(X, Y)
  36
  37 #ifdef __USER_LABEL_PREFIX__
  38 # define C(Y)   C1(__USER_LABEL_PREFIX__, Y)
  39 #else
  40 # define C(Y)   Y
  41 #endif
  42 #define L(Y)    C1(.L, Y)
  43
  44 #if defined(__sun__) && defined(__svr4__)
  45 # define E(INDEX)       .align 16
  46 #else
  47 # define E(INDEX)       .align 16; .org 2b + INDEX * 16
  48 #endif
  49
  50 #define STACK_BIAS 2047
  51
  52         .text
  53         .align 8
  54         .globl  C(ffi_call_v9)
  55         .type   C(ffi_call_v9),#function
  56         FFI_HIDDEN(C(ffi_call_v9))
  57
  58 C(ffi_call_v9):
  59 .LUW0:
  60         save    %sp, %o4, %sp
  61 .LUW1:
  62         mov     %i0, %o0                        ! copy cif
  63         add     %sp, STACK_BIAS+128+48, %o1     ! load args area
  64         mov     %i2, %o2                        ! copy rvalue
  65         call    C(ffi_prep_args_v9)
  66          mov    %i3, %o3                        ! copy avalue
  67
  68         andcc   %o0, SPARC_FLAG_FP_ARGS, %g0    ! need fp regs?
  69         add     %sp, 48, %sp                    ! deallocate prep frame
  70         be,pt   %xcc, 1f
  71          mov    %o0, %l0                        ! save flags
  72
  73         ldd     [%sp+STACK_BIAS+128], %f0       ! load all fp arg regs
  74         ldd     [%sp+STACK_BIAS+128+8], %f2
  75         ldd     [%sp+STACK_BIAS+128+16], %f4
  76         ldd     [%sp+STACK_BIAS+128+24], %f6
  77         ldd     [%sp+STACK_BIAS+128+32], %f8
  78         ldd     [%sp+STACK_BIAS+128+40], %f10
  79         ldd     [%sp+STACK_BIAS+128+48], %f12
  80         ldd     [%sp+STACK_BIAS+128+56], %f14
  81         ldd     [%sp+STACK_BIAS+128+64], %f16
  82         ldd     [%sp+STACK_BIAS+128+72], %f18
  83         ldd     [%sp+STACK_BIAS+128+80], %f20
  84         ldd     [%sp+STACK_BIAS+128+88], %f22
  85         ldd     [%sp+STACK_BIAS+128+96], %f24
  86         ldd     [%sp+STACK_BIAS+128+104], %f26
  87         ldd     [%sp+STACK_BIAS+128+112], %f28
  88         ldd     [%sp+STACK_BIAS+128+120], %f30
  89
  90 1:      ldx     [%sp+STACK_BIAS+128], %o0       ! load all int arg regs
  91         ldx     [%sp+STACK_BIAS+128+8], %o1
  92         ldx     [%sp+STACK_BIAS+128+16], %o2
  93         ldx     [%sp+STACK_BIAS+128+24], %o3
  94         ldx     [%sp+STACK_BIAS+128+32], %o4
  95         ldx     [%sp+STACK_BIAS+128+40], %o5
  96         call    %i1
  97          mov    %i5, %g5                        ! load static chain
  98
  99 0:      call    1f              ! load pc in %o7
 100          and    %l0, SPARC_FLAG_RET_MASK, %l1
 101 1:      sll     %l1, 4, %l1
 102         add     %o7, %l1, %o7   ! o7 = 0b + ret_type*16
 103         jmp     %o7+(2f-0b)
 104          nop
 105
 106         .align  16
 107 2:
 108 E(SPARC_RET_VOID)
 109         return  %i7+8
 110          nop
 111 E(SPARC_RET_STRUCT)
 112         add     %sp, STACK_BIAS-64+128+48, %l2
 113         sub     %sp, 64, %sp
 114         b       8f
 115          stx    %o0, [%l2]
 116 E(SPARC_RET_UINT8)
 117         and     %o0, 0xff, %i0
 118         return  %i7+8
 119           stx   %o0, [%o2]
 120 E(SPARC_RET_SINT8)
 121         sll     %o0, 24, %o0
 122         sra     %o0, 24, %i0
 123         return  %i7+8
 124          stx    %o0, [%o2]
 125 E(SPARC_RET_UINT16)
 126         sll     %o0, 16, %o0
 127         srl     %o0, 16, %i0
 128         return  %i7+8
 129          stx    %o0, [%o2]
 130 E(SPARC_RET_SINT16)
 131         sll     %o0, 16, %o0
 132         sra     %o0, 16, %i0
 133         return  %i7+8
 134          stx    %o0, [%o2]
 135 E(SPARC_RET_UINT32)
 136         srl     %o0, 0, %i0
 137         return  %i7+8
 138          stx    %o0, [%o2]
 139 E(SP_V9_RET_SINT32)
 140         sra     %o0, 0, %i0
 141         return  %i7+8
 142          stx    %o0, [%o2]
 143 E(SPARC_RET_INT64)
 144         stx     %o0, [%i2]
 145         return  %i7+8
 146          nop
 147 E(SPARC_RET_INT128)
 148         stx     %o0, [%i2]
 149         stx     %o1, [%i2+8]
 150         return  %i7+8
 151          nop
 152 E(SPARC_RET_F_8)
 153         st      %f7, [%i2+7*4]
 154         nop
 155         st      %f6, [%i2+6*4]
 156         nop
 157 E(SPARC_RET_F_6)
 158         st      %f5, [%i2+5*4]
 159         nop
 160         st      %f4, [%i2+4*4]
 161         nop
 162 E(SPARC_RET_F_4)
 163         std     %f2, [%i2+2*4]
 164         return  %i7+8
 165          std    %f0, [%o2]
 166 E(SPARC_RET_F_2)
 167         return  %i7+8
 168          std    %f0, [%o2]
 169 E(SP_V9_RET_F_3)
 170         st      %f2, [%i2+2*4]
 171         nop
 172         st      %f1, [%i2+1*4]
 173         nop
 174 E(SPARC_RET_F_1)
 175         return  %i7+8
 176          st     %f0, [%o2]
 177
 178         ! Finish the SPARC_RET_STRUCT sequence.
 179         .align  8
 180 8:      stx     %o1, [%l2+8]
 181         stx     %o2, [%l2+16]
 182         stx     %o3, [%l2+24]
 183         std     %f0, [%l2+32]
 184         std     %f2, [%l2+40]
 185         std     %f4, [%l2+48]
 186         std     %f6, [%l2+56]
 187
 188         ! Copy the structure into place.
 189         srl     %l0, SPARC_SIZEMASK_SHIFT, %o0  ! load size_mask
 190         mov     %i2, %o1                        ! load dst
 191         mov     %l2, %o2                        ! load src_gp
 192         call    C(ffi_struct_float_copy)
 193          add    %l2, 32, %o3                    ! load src_fp
 194
 195         return  %i7+8
 196          nop
 197
 198 .LUW2:
 199         .size   C(ffi_call_v9), . - C(ffi_call_v9)
 200
 201
 202 #undef STACKFRAME
 203 #define STACKFRAME       336    /* 16*8 register window +
 204                                    6*8 args backing store +
 205                                    20*8 locals */
 206 #define FP              %fp+STACK_BIAS
 207
 208 /* ffi_closure_v9(...)
 209
 210    Receives the closure argument in %g1.   */
 211
 212         .align 8
 213         .globl  C(ffi_go_closure_v9)
 214         .type   C(ffi_go_closure_v9),#function
 215         FFI_HIDDEN(C(ffi_go_closure_v9))
 216
 217 C(ffi_go_closure_v9):
 218 .LUW3:
 219         save    %sp, -STACKFRAME, %sp
 220 .LUW4:
 221         ldx     [%g5+8], %o0
 222         ldx     [%g5+16], %o1
 223         b       0f
 224          mov    %g5, %o2
 225
 226 .LUW5:
 227         .size   C(ffi_go_closure_v9), . - C(ffi_go_closure_v9)
 228
 229         .align 8
 230         .globl  C(ffi_closure_v9)
 231         .type   C(ffi_closure_v9),#function
 232         FFI_HIDDEN(C(ffi_closure_v9))
 233
 234 C(ffi_closure_v9):
 235 .LUW6:
 236         save    %sp, -STACKFRAME, %sp
 237 .LUW7:
 238         ldx     [%g1+FFI_TRAMPOLINE_SIZE], %o0
 239         ldx     [%g1+FFI_TRAMPOLINE_SIZE+8], %o1
 240         ldx     [%g1+FFI_TRAMPOLINE_SIZE+16], %o2
 241 0:
 242         ! Store all of the potential argument registers in va_list format.
 243         stx     %i0, [FP+128+0]
 244         stx     %i1, [FP+128+8]
 245         stx     %i2, [FP+128+16]
 246         stx     %i3, [FP+128+24]
 247         stx     %i4, [FP+128+32]
 248         stx     %i5, [FP+128+40]
 249
 250         ! Store possible floating point argument registers too.
 251         std     %f0,  [FP-128]
 252         std     %f2,  [FP-120]
 253         std     %f4,  [FP-112]
 254         std     %f6,  [FP-104]
 255         std     %f8,  [FP-96]
 256         std     %f10, [FP-88]
 257         std     %f12, [FP-80]
 258         std     %f14, [FP-72]
 259         std     %f16, [FP-64]
 260         std     %f18, [FP-56]
 261         std     %f20, [FP-48]
 262         std     %f22, [FP-40]
 263         std     %f24, [FP-32]
 264         std     %f26, [FP-24]
 265         std     %f28, [FP-16]
 266         std     %f30, [FP-8]
 267
 268         ! Call ffi_closure_sparc_inner to do the bulk of the work.
 269         add     %fp, STACK_BIAS-160, %o3
 270         add     %fp, STACK_BIAS+128, %o4
 271         call    C(ffi_closure_sparc_inner_v9)
 272          add    %fp, STACK_BIAS-128, %o5
 273
 274 0:      call    1f              ! load pc in %o7
 275          and    %o0, SPARC_FLAG_RET_MASK, %o0
 276 1:      sll     %o0, 4, %o0     ! o2 = i2 * 16
 277         add     %o7, %o0, %o7   ! o7 = 0b + i2*16
 278         jmp     %o7+(2f-0b)
 279          nop
 280
 281         ! Note that we cannot load the data in the delay slot of
 282         ! the return insn because the data is in the stack frame
 283         ! that is deallocated by the return.
 284         .align  16
 285 2:
 286 E(SPARC_RET_VOID)
 287         return  %i7+8
 288          nop
 289 E(SPARC_RET_STRUCT)
 290         ldx     [FP-160], %i0
 291         ldd     [FP-160], %f0
 292         b       8f
 293          ldx    [FP-152], %i1
 294 E(SPARC_RET_UINT8)
 295         ldub    [FP-160+7], %i0
 296         return  %i7+8
 297          nop
 298 E(SPARC_RET_SINT8)
 299         ldsb    [FP-160+7], %i0
 300         return  %i7+8
 301          nop
 302 E(SPARC_RET_UINT16)
 303         lduh    [FP-160+6], %i0
 304         return  %i7+8
 305          nop
 306 E(SPARC_RET_SINT16)
 307         ldsh    [FP-160+6], %i0
 308         return  %i7+8
 309          nop
 310 E(SPARC_RET_UINT32)
 311         lduw    [FP-160+4], %i0
 312         return  %i7+8
 313          nop
 314 E(SP_V9_RET_SINT32)
 315         ldsw    [FP-160+4], %i0
 316         return  %i7+8
 317          nop
 318 E(SPARC_RET_INT64)
 319         ldx     [FP-160], %i0
 320         return  %i7+8
 321          nop
 322 E(SPARC_RET_INT128)
 323         ldx     [FP-160], %i0
 324         ldx     [FP-160+8], %i1
 325         return  %i7+8
 326          nop
 327 E(SPARC_RET_F_8)
 328         ld      [FP-160+7*4], %f7
 329         nop
 330         ld      [FP-160+6*4], %f6
 331         nop
 332 E(SPARC_RET_F_6)
 333         ld      [FP-160+5*4], %f5
 334         nop
 335         ld      [FP-160+4*4], %f4
 336         nop
 337 E(SPARC_RET_F_4)
 338         ldd     [FP-160], %f0
 339         ldd     [FP-160+8], %f2
 340         return  %i7+8
 341          nop
 342 E(SPARC_RET_F_2)
 343         ldd     [FP-160], %f0
 344         return  %i7+8
 345          nop
 346 E(SP_V9_RET_F_3)
 347         ld      [FP-160+2*4], %f2
 348         nop
 349         ld      [FP-160+1*4], %f1
 350         nop
 351 E(SPARC_RET_F_1)
 352         ld      [FP-160], %f0
 353         return  %i7+8
 354          nop
 355
 356         ! Finish the SPARC_RET_STRUCT sequence.
 357         .align  8
 358 8:      ldd     [FP-152], %f2
 359         ldx     [FP-144], %i2
 360         ldd     [FP-144], %f4
 361         ldx     [FP-136], %i3
 362         ldd     [FP-136], %f6
 363         return  %i7+8
 364          nop
 365
 366 .LUW8:
 367         .size   C(ffi_closure_v9), . - C(ffi_closure_v9)
 368
 369 #ifdef HAVE_RO_EH_FRAME
 370         .section        ".eh_frame",#alloc
 371 #else
 372         .section        ".eh_frame",#alloc,#write
 373 #endif
 374
 375 #ifdef HAVE_AS_SPARC_UA_PCREL
 376 # define FDE_RANGE(B, E)  .long %r_disp32(B), E - B
 377 #else
 378 # define FDE_RANGE(B, E)  .align 8; .xword B, E - B
 379 #endif
 380
 381         .align 8
 382 .LCIE:
 383         .long   .LECIE - .LSCIE         ! CIE Length
 384 .LSCIE:
 385         .long   0                       ! CIE Identifier Tag
 386         .byte   1                       ! CIE Version
 387         .ascii  "zR\0"                  ! CIE Augmentation
 388         .byte   4                       ! CIE Code Alignment Factor
 389         .byte   0x78                    ! CIE Data Alignment Factor
 390         .byte   15                      ! CIE RA Column
 391         .byte   1                       ! Augmentation size
 392 #ifdef HAVE_AS_SPARC_UA_PCREL
 393         .byte   0x1b                    ! FDE Encoding (pcrel sdata4)
 394 #else
 395         .byte   0x50                    ! FDE Encoding (aligned absolute)
 396 #endif
 397         .byte   0xc, 14, 0xff, 0xf      ! DW_CFA_def_cfa, %o6, offset 0x7ff
 398         .align  8
 399 .LECIE:
 400
 401         .long   .LEFDE1 - .LSFDE1       ! FDE Length
 402 .LSFDE1:
 403         .long   .LSFDE1 - .LCIE         ! FDE CIE offset
 404         FDE_RANGE(.LUW0, .LUW2)
 405         .byte   0                       ! Augmentation size
 406         .byte   0x40+1                  ! DW_CFA_advance_loc 4
 407         .byte   0xd, 30                 ! DW_CFA_def_cfa_register, %i6
 408         .byte   0x2d                    ! DW_CFA_GNU_window_save
 409         .byte   0x9, 15, 31             ! DW_CFA_register, %o7, %i7
 410         .align  8
 411 .LEFDE1:
 412
 413         .long   .LEFDE2 - .LSFDE2       ! FDE Length
 414 .LSFDE2:
 415         .long   .LSFDE2 - .LCIE         ! FDE CIE offset
 416         FDE_RANGE(.LUW3, .LUW5)
 417         .byte   0                       ! Augmentation size
 418         .byte   0x40+1                  ! DW_CFA_advance_loc 4
 419         .byte   0xd, 30                 ! DW_CFA_def_cfa_register, %i6
 420         .byte   0x2d                    ! DW_CFA_GNU_window_save
 421         .byte   0x9, 15, 31             ! DW_CFA_register, %o7, %i7
 422         .align  8
 423 .LEFDE2:
 424
 425         .long   .LEFDE3 - .LSFDE3       ! FDE Length
 426 .LSFDE3:
 427         .long   .LSFDE3 - .LCIE         ! FDE CIE offset
 428         FDE_RANGE(.LUW6, .LUW8)
 429         .byte   0                       ! Augmentation size
 430         .byte   0x40+1                  ! DW_CFA_advance_loc 4
 431         .byte   0xd, 30                 ! DW_CFA_def_cfa_register, %i6
 432         .byte   0x2d                    ! DW_CFA_GNU_window_save
 433         .byte   0x9, 15, 31             ! DW_CFA_register, %o7, %i7
 434         .align  8
 435 .LEFDE3:
 436
 437 #endif /* SPARC64 */
 438 #ifdef __linux__
 439         .section        .note.GNU-stack,"",@progbits
 440 #endif