1 /* -----------------------------------------------------------------------
2 unix64.S - Copyright (c) 2002 Bo Thorsen <bo@suse.de>
3 Copyright (c) 2008 Red Hat, Inc
5 x86-64 Foreign Function Interface
7 Permission is hereby granted, free of charge, to any person obtaining
8 a copy of this software and associated documentation files (the
9 ``Software''), to deal in the Software without restriction, including
10 without limitation the rights to use, copy, modify, merge, publish,
11 distribute, sublicense, and/or sell copies of the Software, and to
12 permit persons to whom the Software is furnished to do so, subject to
13 the following conditions:
15 The above copyright notice and this permission notice shall be included
16 in all copies or substantial portions of the Software.
18 THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
21 NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
22 HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
23 WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
24 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
25 DEALINGS IN THE SOFTWARE.
26 ----------------------------------------------------------------------- */
30 #include <fficonfig.h>
35 /* ffi_call_unix64 (void *args, unsigned long bytes, unsigned flags,
36 void *raddr, void (*fnaddr)(void));
38 Bit o trickiness here -- ARGS+BYTES is the base of the stack frame
39 for this function. This has been allocated by ffi_call. We also
40 deallocate some of the stack that has been alloca'd. */
43 .globl ffi_call_unix64
44 .type ffi_call_unix64,@function
48 movq (%rsp), %r10 /* Load return address. */
49 leaq (%rdi, %rsi), %rax /* Find local stack base. */
50 movq %rdx, (%rax) /* Save flags. */
51 movq %rcx, 8(%rax) /* Save raddr. */
52 movq %rbp, 16(%rax) /* Save old frame pointer. */
53 movq %r10, 24(%rax) /* Relocate return address. */
54 movq %rax, %rbp /* Finalize local stack frame. */
56 movq %rdi, %r10 /* Save a copy of the register area. */
57 movq %r8, %r11 /* Save a copy of the target fn. */
58 movl %r9d, %eax /* Set number of SSE registers. */
60 /* Load up all argument registers. */
71 /* Deallocate the reg arg area. */
74 /* Call the user function. */
77 /* Deallocate stack arg area; local stack frame in redzone. */
80 movq 0(%rbp), %rcx /* Reload flags. */
81 movq 8(%rbp), %rdi /* Reload raddr. */
82 movq 16(%rbp), %rbp /* Reload old frame pointer. */
85 /* The first byte of the flags contains the FFI_TYPE. */
87 leaq .Lstore_table(%rip), %r11
88 movslq (%r11, %r10, 4), %r10
94 .long .Lst_void-.Lstore_table /* FFI_TYPE_VOID */
95 .long .Lst_sint32-.Lstore_table /* FFI_TYPE_INT */
96 .long .Lst_float-.Lstore_table /* FFI_TYPE_FLOAT */
97 .long .Lst_double-.Lstore_table /* FFI_TYPE_DOUBLE */
98 .long .Lst_ldouble-.Lstore_table /* FFI_TYPE_LONGDOUBLE */
99 .long .Lst_uint8-.Lstore_table /* FFI_TYPE_UINT8 */
100 .long .Lst_sint8-.Lstore_table /* FFI_TYPE_SINT8 */
101 .long .Lst_uint16-.Lstore_table /* FFI_TYPE_UINT16 */
102 .long .Lst_sint16-.Lstore_table /* FFI_TYPE_SINT16 */
103 .long .Lst_uint32-.Lstore_table /* FFI_TYPE_UINT32 */
104 .long .Lst_sint32-.Lstore_table /* FFI_TYPE_SINT32 */
105 .long .Lst_int64-.Lstore_table /* FFI_TYPE_UINT64 */
106 .long .Lst_int64-.Lstore_table /* FFI_TYPE_SINT64 */
107 .long .Lst_struct-.Lstore_table /* FFI_TYPE_STRUCT */
108 .long .Lst_int64-.Lstore_table /* FFI_TYPE_POINTER */
162 leaq -20(%rsp), %rsi /* Scratch area in redzone. */
164 /* We have to locate the values now, and since we don't want to
165 write too much data into the user's return value, we spill the
166 value to a 16 byte scratch area first. Bits 8, 9, and 10
167 control where the values are located. Only one of the three
168 bits will be set; see ffi_prep_cif_machdep for the pattern. */
182 /* Bits 12-31 contain the true size of the structure. Copy from
183 the scratch area to the true destination. */
188 /* Many times we can avoid loading any SSE registers at all.
189 It's not worth an indirect jump to load the exact set of
190 SSE registers needed; zero or all is a good compromise. */
194 movdqa 48(%r10), %xmm0
195 movdqa 64(%r10), %xmm1
196 movdqa 80(%r10), %xmm2
197 movdqa 96(%r10), %xmm3
198 movdqa 112(%r10), %xmm4
199 movdqa 128(%r10), %xmm5
200 movdqa 144(%r10), %xmm6
201 movdqa 160(%r10), %xmm7
202 jmp .Lret_from_load_sse
205 .size ffi_call_unix64,.-ffi_call_unix64
208 .globl ffi_closure_unix64
209 .type ffi_closure_unix64,@function
213 /* The carry flag is set by the trampoline iff SSE registers
214 are used. Don't clobber it before the branch instruction. */
215 leaq -200(%rsp), %rsp
230 call ffi_closure_unix64_inner@PLT
232 /* Deallocate stack frame early; return value is now in redzone. */
236 /* The first byte of the return value contains the FFI_TYPE. */
238 leaq .Lload_table(%rip), %r11
239 movslq (%r11, %r10, 4), %r10
245 .long .Lld_void-.Lload_table /* FFI_TYPE_VOID */
246 .long .Lld_int32-.Lload_table /* FFI_TYPE_INT */
247 .long .Lld_float-.Lload_table /* FFI_TYPE_FLOAT */
248 .long .Lld_double-.Lload_table /* FFI_TYPE_DOUBLE */
249 .long .Lld_ldouble-.Lload_table /* FFI_TYPE_LONGDOUBLE */
250 .long .Lld_int8-.Lload_table /* FFI_TYPE_UINT8 */
251 .long .Lld_int8-.Lload_table /* FFI_TYPE_SINT8 */
252 .long .Lld_int16-.Lload_table /* FFI_TYPE_UINT16 */
253 .long .Lld_int16-.Lload_table /* FFI_TYPE_SINT16 */
254 .long .Lld_int32-.Lload_table /* FFI_TYPE_UINT32 */
255 .long .Lld_int32-.Lload_table /* FFI_TYPE_SINT32 */
256 .long .Lld_int64-.Lload_table /* FFI_TYPE_UINT64 */
257 .long .Lld_int64-.Lload_table /* FFI_TYPE_SINT64 */
258 .long .Lld_struct-.Lload_table /* FFI_TYPE_STRUCT */
259 .long .Lld_int64-.Lload_table /* FFI_TYPE_POINTER */
268 movzbl -24(%rsp), %eax
272 movzwl -24(%rsp), %eax
285 movss -24(%rsp), %xmm0
289 movsd -24(%rsp), %xmm0
298 /* There are four possibilities here, %rax/%rdx, %xmm0/%rax,
299 %rax/%xmm0, %xmm0/%xmm1. We collapse two by always loading
300 both rdx and xmm1 with the second word. For the remaining,
301 bit 8 set means xmm0 gets the second word, and bit 9 means
302 that rax gets the second word. */
305 movq -16(%rsp), %xmm1
314 /* See the comment above .Lload_sse; the same logic applies here. */
318 movdqa %xmm0, 48(%rsp)
319 movdqa %xmm1, 64(%rsp)
320 movdqa %xmm2, 80(%rsp)
321 movdqa %xmm3, 96(%rsp)
322 movdqa %xmm4, 112(%rsp)
323 movdqa %xmm5, 128(%rsp)
324 movdqa %xmm6, 144(%rsp)
325 movdqa %xmm7, 160(%rsp)
326 jmp .Lret_from_save_sse
329 .size ffi_closure_unix64,.-ffi_closure_unix64
331 .section .eh_frame,"a",@progbits
333 .long .LECIE1-.LSCIE1 /* CIE Length */
335 .long 0 /* CIE Identifier Tag */
336 .byte 1 /* CIE Version */
337 .ascii "zR\0" /* CIE Augmentation */
338 .uleb128 1 /* CIE Code Alignment Factor */
339 .sleb128 -8 /* CIE Data Alignment Factor */
340 .byte 0x10 /* CIE RA Column */
341 .uleb128 1 /* Augmentation size */
342 .byte 0x1b /* FDE Encoding (pcrel sdata4) */
343 .byte 0xc /* DW_CFA_def_cfa, %rsp offset 8 */
346 .byte 0x80+16 /* DW_CFA_offset, %rip offset 1*-8 */
351 .long .LEFDE1-.LASFDE1 /* FDE Length */
353 .long .LASFDE1-.Lframe1 /* FDE CIE offset */
354 .long .LUW0-. /* FDE initial location */
355 .long .LUW4-.LUW0 /* FDE address range */
356 .uleb128 0x0 /* Augmentation size */
358 .byte 0x4 /* DW_CFA_advance_loc4 */
361 /* New stack frame based off rbp. This is a itty bit of unwind
362 trickery in that the CFA *has* changed. There is no easy way
363 to describe it correctly on entry to the function. Fortunately,
364 it doesn't matter too much since at all points we can correctly
365 unwind back to ffi_call. Note that the location to which we
366 moved the return address is (the new) CFA-8, so from the
367 perspective of the unwind info, it hasn't moved. */
368 .byte 0xc /* DW_CFA_def_cfa, %rbp offset 32 */
371 .byte 0x80+6 /* DW_CFA_offset, %rbp offset 2*-8 */
373 .byte 0xa /* DW_CFA_remember_state */
375 .byte 0x4 /* DW_CFA_advance_loc4 */
377 .byte 0xc /* DW_CFA_def_cfa, %rsp offset 8 */
380 .byte 0xc0+6 /* DW_CFA_restore, %rbp */
382 .byte 0x4 /* DW_CFA_advance_loc4 */
384 .byte 0xb /* DW_CFA_restore_state */
389 .long .LEFDE3-.LASFDE3 /* FDE Length */
391 .long .LASFDE3-.Lframe1 /* FDE CIE offset */
392 .long .LUW5-. /* FDE initial location */
393 .long .LUW9-.LUW5 /* FDE address range */
394 .uleb128 0x0 /* Augmentation size */
396 .byte 0x4 /* DW_CFA_advance_loc4 */
398 .byte 0xe /* DW_CFA_def_cfa_offset */
400 .byte 0xa /* DW_CFA_remember_state */
402 .byte 0x4 /* DW_CFA_advance_loc4 */
404 .byte 0xe /* DW_CFA_def_cfa_offset */
407 .byte 0x4 /* DW_CFA_advance_loc4 */
409 .byte 0xb /* DW_CFA_restore_state */
414 #endif /* __x86_64__ */
416 #if defined __ELF__ && defined __linux__
417 .section .note.GNU-stack,"",@progbits