1
/* -----------------------------------------------------------------------
2
unix64.S - Copyright (c) 2002 Bo Thorsen <bo@suse.de>
4
x86-64 Foreign Function Interface
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:
14
The above copyright notice and this permission notice shall be included
15
in all copies or substantial portions of the Software.
17
THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, EXPRESS
18
OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
20
IN NO EVENT SHALL CYGNUS SOLUTIONS BE LIABLE FOR ANY CLAIM, DAMAGES OR
21
OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
22
ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
23
OTHER DEALINGS IN THE SOFTWARE.
24
----------------------------------------------------------------------- */
28
#include <fficonfig.h>
33
/* ffi_call_unix64 (void *args, unsigned long bytes, unsigned flags,
34
void *raddr, void (*fnaddr)());
36
Bit o trickiness here -- ARGS+BYTES is the base of the stack frame
37
for this function. This has been allocated by ffi_call. We also
38
deallocate some of the stack that has been alloca'd. */
41
.globl ffi_call_unix64
42
.type ffi_call_unix64,@function
46
movq (%rsp), %r10 /* Load return address. */
47
leaq (%rdi, %rsi), %rax /* Find local stack base. */
48
movq %rdx, (%rax) /* Save flags. */
49
movq %rcx, 8(%rax) /* Save raddr. */
50
movq %rbp, 16(%rax) /* Save old frame pointer. */
51
movq %r10, 24(%rax) /* Relocate return address. */
52
movq %rax, %rbp /* Finalize local stack frame. */
54
movq %rdi, %r10 /* Save a copy of the register area. */
55
movq %r8, %r11 /* Save a copy of the target fn. */
56
movl %r9d, %eax /* Set number of SSE registers. */
58
/* Load up all argument registers. */
69
/* Deallocate the reg arg area. */
72
/* Call the user function. */
75
/* Deallocate stack arg area; local stack frame in redzone. */
78
movq 0(%rbp), %rcx /* Reload flags. */
79
movq 8(%rbp), %rdi /* Reload raddr. */
80
movq 16(%rbp), %rbp /* Reload old frame pointer. */
83
/* The first byte of the flags contains the FFI_TYPE. */
85
leaq .Lstore_table(%rip), %r11
86
movslq (%r11, %r10, 4), %r10
92
.long .Lst_void-.Lstore_table /* FFI_TYPE_VOID */
93
.long .Lst_sint32-.Lstore_table /* FFI_TYPE_INT */
94
.long .Lst_float-.Lstore_table /* FFI_TYPE_FLOAT */
95
.long .Lst_double-.Lstore_table /* FFI_TYPE_DOUBLE */
96
.long .Lst_ldouble-.Lstore_table /* FFI_TYPE_LONGDOUBLE */
97
.long .Lst_uint8-.Lstore_table /* FFI_TYPE_UINT8 */
98
.long .Lst_sint8-.Lstore_table /* FFI_TYPE_SINT8 */
99
.long .Lst_uint16-.Lstore_table /* FFI_TYPE_UINT16 */
100
.long .Lst_sint16-.Lstore_table /* FFI_TYPE_SINT16 */
101
.long .Lst_uint32-.Lstore_table /* FFI_TYPE_UINT32 */
102
.long .Lst_sint32-.Lstore_table /* FFI_TYPE_SINT32 */
103
.long .Lst_int64-.Lstore_table /* FFI_TYPE_UINT64 */
104
.long .Lst_int64-.Lstore_table /* FFI_TYPE_SINT64 */
105
.long .Lst_struct-.Lstore_table /* FFI_TYPE_STRUCT */
106
.long .Lst_int64-.Lstore_table /* FFI_TYPE_POINTER */
160
leaq -20(%rsp), %rsi /* Scratch area in redzone. */
162
/* We have to locate the values now, and since we don't want to
163
write too much data into the user's return value, we spill the
164
value to a 16 byte scratch area first. Bits 8, 9, and 10
165
control where the values are located. Only one of the three
166
bits will be set; see ffi_prep_cif_machdep for the pattern. */
180
/* Bits 12-31 contain the true size of the structure. Copy from
181
the scratch area to the true destination. */
186
/* Many times we can avoid loading any SSE registers at all.
187
It's not worth an indirect jump to load the exact set of
188
SSE registers needed; zero or all is a good compromise. */
192
movdqa 48(%r10), %xmm0
193
movdqa 64(%r10), %xmm1
194
movdqa 80(%r10), %xmm2
195
movdqa 96(%r10), %xmm3
196
movdqa 112(%r10), %xmm4
197
movdqa 128(%r10), %xmm5
198
movdqa 144(%r10), %xmm6
199
movdqa 160(%r10), %xmm7
200
jmp .Lret_from_load_sse
203
.size ffi_call_unix64,.-ffi_call_unix64
206
.globl ffi_closure_unix64
207
.type ffi_closure_unix64,@function
211
/* The carry flag is set by the trampoline iff SSE registers
212
are used. Don't clobber it before the branch instruction. */
213
leaq -200(%rsp), %rsp
228
call ffi_closure_unix64_inner@PLT
230
/* Deallocate stack frame early; return value is now in redzone. */
234
/* The first byte of the return value contains the FFI_TYPE. */
236
leaq .Lload_table(%rip), %r11
237
movslq (%r11, %r10, 4), %r10
243
.long .Lld_void-.Lload_table /* FFI_TYPE_VOID */
244
.long .Lld_int32-.Lload_table /* FFI_TYPE_INT */
245
.long .Lld_float-.Lload_table /* FFI_TYPE_FLOAT */
246
.long .Lld_double-.Lload_table /* FFI_TYPE_DOUBLE */
247
.long .Lld_ldouble-.Lload_table /* FFI_TYPE_LONGDOUBLE */
248
.long .Lld_int8-.Lload_table /* FFI_TYPE_UINT8 */
249
.long .Lld_int8-.Lload_table /* FFI_TYPE_SINT8 */
250
.long .Lld_int16-.Lload_table /* FFI_TYPE_UINT16 */
251
.long .Lld_int16-.Lload_table /* FFI_TYPE_SINT16 */
252
.long .Lld_int32-.Lload_table /* FFI_TYPE_UINT32 */
253
.long .Lld_int32-.Lload_table /* FFI_TYPE_SINT32 */
254
.long .Lld_int64-.Lload_table /* FFI_TYPE_UINT64 */
255
.long .Lld_int64-.Lload_table /* FFI_TYPE_SINT64 */
256
.long .Lld_struct-.Lload_table /* FFI_TYPE_STRUCT */
257
.long .Lld_int64-.Lload_table /* FFI_TYPE_POINTER */
266
movzbl -24(%rsp), %eax
270
movzwl -24(%rsp), %eax
283
movss -24(%rsp), %xmm0
287
movsd -24(%rsp), %xmm0
296
/* There are four possibilities here, %rax/%rdx, %xmm0/%rax,
297
%rax/%xmm0, %xmm0/%xmm1. We collapse two by always loading
298
both rdx and xmm1 with the second word. For the remaining,
299
bit 8 set means xmm0 gets the second word, and bit 9 means
300
that rax gets the second word. */
303
movq -16(%rsp), %xmm1
312
/* See the comment above .Lload_sse; the same logic applies here. */
316
movdqa %xmm0, 48(%rsp)
317
movdqa %xmm1, 64(%rsp)
318
movdqa %xmm2, 80(%rsp)
319
movdqa %xmm3, 96(%rsp)
320
movdqa %xmm4, 112(%rsp)
321
movdqa %xmm5, 128(%rsp)
322
movdqa %xmm6, 144(%rsp)
323
movdqa %xmm7, 160(%rsp)
324
jmp .Lret_from_save_sse
327
.size ffi_closure_unix64,.-ffi_closure_unix64
329
.section .eh_frame,"a",@progbits
331
.long .LECIE1-.LSCIE1 /* CIE Length */
333
.long 0 /* CIE Identifier Tag */
334
.byte 1 /* CIE Version */
335
.ascii "zR\0" /* CIE Augmentation */
336
.uleb128 1 /* CIE Code Alignment Factor */
337
.sleb128 -8 /* CIE Data Alignment Factor */
338
.byte 0x10 /* CIE RA Column */
339
.uleb128 1 /* Augmentation size */
340
.byte 0x1b /* FDE Encoding (pcrel sdata4) */
341
.byte 0xc /* DW_CFA_def_cfa, %rsp offset 8 */
344
.byte 0x80+16 /* DW_CFA_offset, %rip offset 1*-8 */
349
.long .LEFDE1-.LASFDE1 /* FDE Length */
351
.long .LASFDE1-.Lframe1 /* FDE CIE offset */
352
.long .LUW0-. /* FDE initial location */
353
.long .LUW4-.LUW0 /* FDE address range */
354
.uleb128 0x0 /* Augmentation size */
356
.byte 0x4 /* DW_CFA_advance_loc4 */
359
/* New stack frame based off rbp. This is a itty bit of unwind
360
trickery in that the CFA *has* changed. There is no easy way
361
to describe it correctly on entry to the function. Fortunately,
362
it doesn't matter too much since at all points we can correctly
363
unwind back to ffi_call. Note that the location to which we
364
moved the return address is (the new) CFA-8, so from the
365
perspective of the unwind info, it hasn't moved. */
366
.byte 0xc /* DW_CFA_def_cfa, %rbp offset 32 */
369
.byte 0x80+6 /* DW_CFA_offset, %rbp offset 2*-8 */
371
.byte 0xa /* DW_CFA_remember_state */
373
.byte 0x4 /* DW_CFA_advance_loc4 */
375
.byte 0xc /* DW_CFA_def_cfa, %rsp offset 8 */
378
.byte 0xc0+6 /* DW_CFA_restore, %rbp */
380
.byte 0x4 /* DW_CFA_advance_loc4 */
382
.byte 0xb /* DW_CFA_restore_state */
387
.long .LEFDE3-.LASFDE3 /* FDE Length */
389
.long .LASFDE3-.Lframe1 /* FDE CIE offset */
390
.long .LUW5-. /* FDE initial location */
391
.long .LUW9-.LUW5 /* FDE address range */
392
.uleb128 0x0 /* Augmentation size */
394
.byte 0x4 /* DW_CFA_advance_loc4 */
396
.byte 0xe /* DW_CFA_def_cfa_offset */
398
.byte 0xa /* DW_CFA_remember_state */
400
.byte 0x4 /* DW_CFA_advance_loc4 */
402
.byte 0xe /* DW_CFA_def_cfa_offset */
405
.byte 0x4 /* DW_CFA_advance_loc4 */
407
.byte 0xb /* DW_CFA_restore_state */
412
#endif /* __x86_64__ */