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. */
57 /* Load up all argument registers. */
64 movdqa 48(%r10), %xmm0
65 movdqa 64(%r10), %xmm1
66 movdqa 80(%r10), %xmm2
67 movdqa 96(%r10), %xmm3
68 movdqa 112(%r10), %xmm4
69 movdqa 128(%r10), %xmm5
70 movdqa 144(%r10), %xmm6
71 movdqa 160(%r10), %xmm7
73 /* Deallocate the reg arg area. */
76 /* Call the user function. */
79 /* Deallocate stack arg area; local stack frame in redzone. */
82 movq 0(%rbp), %rcx /* Reload flags. */
83 movq 8(%rbp), %rdi /* Reload raddr. */
84 movq 16(%rbp), %rbp /* Reload old frame pointer. */
87 /* The first byte of the flags contains the FFI_TYPE. */
89 leaq .Lstore_table(%rip), %r11
90 movslq (%r11, %r10, 4), %r10
96 .long .Lst_void-.Lstore_table /* FFI_TYPE_VOID */
97 .long .Lst_sint32-.Lstore_table /* FFI_TYPE_INT */
98 .long .Lst_float-.Lstore_table /* FFI_TYPE_FLOAT */
99 .long .Lst_double-.Lstore_table /* FFI_TYPE_DOUBLE */
100 .long .Lst_ldouble-.Lstore_table /* FFI_TYPE_LONGDOUBLE */
101 .long .Lst_uint8-.Lstore_table /* FFI_TYPE_UINT8 */
102 .long .Lst_sint8-.Lstore_table /* FFI_TYPE_SINT8 */
103 .long .Lst_uint16-.Lstore_table /* FFI_TYPE_UINT16 */
104 .long .Lst_sint16-.Lstore_table /* FFI_TYPE_SINT16 */
105 .long .Lst_uint32-.Lstore_table /* FFI_TYPE_UINT32 */
106 .long .Lst_sint32-.Lstore_table /* FFI_TYPE_SINT32 */
107 .long .Lst_int64-.Lstore_table /* FFI_TYPE_UINT64 */
108 .long .Lst_int64-.Lstore_table /* FFI_TYPE_SINT64 */
109 .long .Lst_struct-.Lstore_table /* FFI_TYPE_STRUCT */
110 .long .Lst_int64-.Lstore_table /* FFI_TYPE_POINTER */
164 leaq -20(%rsp), %rsi /* Scratch area in redzone. */
166 /* We have to locate the values now, and since we don't want to
167 write too much data into the user's return value, we spill the
168 value to a 16 byte scratch area first. Bits 8, 9, and 10
169 control where the values are located. Only one of the three
170 bits will be set; see ffi_prep_cif_machdep for the pattern. */
184 /* Bits 11-31 contain the true size of the structure. Copy from
185 the scratch area to the true destination. */
190 .size ffi_call_unix64,.-ffi_call_unix64
193 .globl ffi_closure_unix64
194 .type ffi_closure_unix64,@function
207 movdqa %xmm0, 48(%rsp)
208 movdqa %xmm1, 64(%rsp)
209 movdqa %xmm2, 80(%rsp)
210 movdqa %xmm3, 96(%rsp)
211 movdqa %xmm4, 112(%rsp)
212 movdqa %xmm5, 128(%rsp)
213 movdqa %xmm6, 144(%rsp)
214 movdqa %xmm7, 160(%rsp)
220 call ffi_closure_unix64_inner@PLT
222 /* Deallocate stack frame early; return value is now in redzone. */
226 /* The first byte of the return value contains the FFI_TYPE. */
228 leaq .Lload_table(%rip), %r11
229 movslq (%r11, %r10, 4), %r10
235 .long .Lld_void-.Lload_table /* FFI_TYPE_VOID */
236 .long .Lld_int32-.Lload_table /* FFI_TYPE_INT */
237 .long .Lld_float-.Lload_table /* FFI_TYPE_FLOAT */
238 .long .Lld_double-.Lload_table /* FFI_TYPE_DOUBLE */
239 .long .Lld_ldouble-.Lload_table /* FFI_TYPE_LONGDOUBLE */
240 .long .Lld_int8-.Lload_table /* FFI_TYPE_UINT8 */
241 .long .Lld_int8-.Lload_table /* FFI_TYPE_SINT8 */
242 .long .Lld_int16-.Lload_table /* FFI_TYPE_UINT16 */
243 .long .Lld_int16-.Lload_table /* FFI_TYPE_SINT16 */
244 .long .Lld_int32-.Lload_table /* FFI_TYPE_UINT32 */
245 .long .Lld_int32-.Lload_table /* FFI_TYPE_SINT32 */
246 .long .Lld_int64-.Lload_table /* FFI_TYPE_UINT64 */
247 .long .Lld_int64-.Lload_table /* FFI_TYPE_SINT64 */
248 .long .Lld_struct-.Lload_table /* FFI_TYPE_STRUCT */
249 .long .Lld_int64-.Lload_table /* FFI_TYPE_POINTER */
258 movzbl -24(%rsp), %eax
262 movzwl -24(%rsp), %eax
275 movss -24(%rsp), %xmm0
279 movsd -24(%rsp), %xmm0
288 /* There are four possibilities here, %rax/%rdx, %xmm0/%rax,
289 %rax/%xmm0, %xmm0/%xmm1. We collapse two by always loading
290 both rdx and xmm1 with the second word. For the remaining,
291 bit 8 set means xmm0 gets the second word, and bit 9 means
292 that rax gets the second word. */
295 movq -16(%rsp), %xmm1
304 .size ffi_closure_unix64,.-ffi_closure_unix64
306 .section .eh_frame,"a",@progbits
308 .long .LECIE1-.LSCIE1 /* CIE Length */
310 .long 0 /* CIE Identifier Tag */
311 .byte 1 /* CIE Version */
312 .ascii "zR\0" /* CIE Augmentation */
313 .uleb128 1 /* CIE Code Alignment Factor */
314 .sleb128 -8 /* CIE Data Alignment Factor */
315 .byte 0x10 /* CIE RA Column */
316 .uleb128 1 /* Augmentation size */
317 .byte 0x1b /* FDE Encoding (pcrel sdata4) */
318 .byte 0xc /* DW_CFA_def_cfa, %rsp offset 8 */
321 .byte 0x80+16 /* DW_CFA_offset, %rip offset 1*-8 */
326 .long .LEFDE1-.LASFDE1 /* FDE Length */
328 .long .LASFDE1-.Lframe1 /* FDE CIE offset */
329 .long .LUW0-. /* FDE initial location */
330 .long .LUW3-.LUW0 /* FDE address range */
331 .uleb128 0x0 /* Augmentation size */
333 .byte 0x4 /* DW_CFA_advance_loc4 */
336 /* New stack frame based off rbp. This is a itty bit of unwind
337 trickery in that the CFA *has* changed. There is no easy way
338 to describe it correctly on entry to the function. Fortunately,
339 it doesn't matter too much since at all points we can correctly
340 unwind back to ffi_call. Note that the location to which we
341 moved the return address is (the new) CFA-8, so from the
342 perspective of the unwind info, it hasn't moved. */
343 .byte 0xc /* DW_CFA_def_cfa, %rbp offset 32 */
346 .byte 0x80+6 /* DW_CFA_offset, %rbp offset 2*-8 */
349 .byte 0x4 /* DW_CFA_advance_loc4 */
351 .byte 0xc /* DW_CFA_def_cfa, %rsp offset 8 */
354 .byte 0xc0+6 /* DW_CFA_restore, %rbp */
358 .long .LEFDE3-.LASFDE3 /* FDE Length */
360 .long .LASFDE3-.Lframe1 /* FDE CIE offset */
361 .long .LUW4-. /* FDE initial location */
362 .long .LUW7-.LUW4 /* FDE address range */
363 .uleb128 0x0 /* Augmentation size */
364 .byte 0x4 /* DW_CFA_advance_loc4 */
366 .byte 0xe /* DW_CFA_def_cfa_offset */
368 .byte 0x4 /* DW_CFA_advance_loc4 */
370 .byte 0xe /* DW_CFA_def_cfa_offset */
375 #endif /* __x86_64__ */