/* Copyright © 2009, 2010, 2011, 2012 ARM Ltd.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */

if defined(aarch64) || defined(arm64) define LIBFFI_ASM include <fficonfig.h> include <ffi.h> include <ffi_cfi.h> include “internal.h”

ifdef HAVE_MACHINE_ASM_H include <machine/asm.h> else ifdef USER_LABEL_PREFIX define CONCAT1(a, b) CONCAT2(a, b) define CONCAT2(a, b) a ## b

/* Use the right prefix for global labels. */ define CNAME(x) CONCAT1 (USER_LABEL_PREFIX, x) else define CNAME(x) x endif endif

ifdef __APPLE__ # define L(X) CONCAT1(L, X) else # define L(X) CONCAT1(.L, X) endif

ifdef AARCH64EB # define BE(X) X else # define BE(X) 0 endif

ifdef ILP32 define PTR_REG(n) w##n else define PTR_REG(n) x##n endif

ifdef ILP32 define PTR_SIZE 4 else define PTR_SIZE 8 endif

define BTI_C hint #34 define BTI_J hint #36 /*

* The ELF Notes section needs to indicate if BTI is supported, as the first ELF loaded that doesn't
* declare this support disables it for memory region containing the loaded library.
*/

# define GNU_PROPERTY_AARCH64_BTI (1 << 0) /* Has Branch Target Identification */

.text
.align 4

/* ffi_call_SYSV

extern void ffi_call_SYSV (void *stack, void *frame,
                           void (*fn)(void), void *rvalue,
                           int flags, void *closure);

Therefore on entry we have:

x0 stack
x1 frame
x2 fn
x3 rvalue
x4 flags
x5 closure

*/

CNAME(ffi_call_SYSV):

cfi_startproc
BTI_C
PAC_CFI_WINDOW_SAVE
/* Sign the lr with x1 since that is the CFA which is the modifer used in auth instructions */
SIGN_LR_WITH_REG(x1)

if defined(HAVE_ARM64E_PTRAUTH) && defined(__APPLE__)

/* darwin's libunwind assumes that the cfa is the sp and that's the data
 * used to sign the lr.  In order to allow unwinding through this
 * function it is necessary to point the cfa at the signing register.
 */
cfi_def_cfa(x1, 0);

endif

/* Use a stack frame allocated by our caller.  */
stp     x29, x30, [x1]
cfi_def_cfa_register(x1)
cfi_rel_offset (x29, 0)
cfi_rel_offset (x30, 8)
mov     x9, sp
str     x9, [x1, #32]
mov     x29, x1
cfi_def_cfa_register(x29)
mov     sp, x0

mov     x9, x2                  /* save fn */
mov     x8, x3                  /* install structure return */

ifdef FFI_GO_CLOSURES

mov     x18, x5                 /* install static chain */

endif

stp     x3, x4, [x29, #16]      /* save rvalue and flags */

/* Load the vector argument passing registers, if necessary.  */
tbz     w4, #AARCH64_FLAG_ARG_V_BIT, 1f
ldp     q0, q1, [sp, #0]
ldp     q2, q3, [sp, #32]
ldp     q4, q5, [sp, #64]
ldp     q6, q7, [sp, #96]

1:

/* Load the core argument passing registers, including
   the structure return pointer.  */
ldp     x0, x1, [sp, #16*N_V_ARG_REG + 0]
ldp     x2, x3, [sp, #16*N_V_ARG_REG + 16]
ldp     x4, x5, [sp, #16*N_V_ARG_REG + 32]
ldp     x6, x7, [sp, #16*N_V_ARG_REG + 48]

/* Deallocate the context, leaving the stacked arguments.  */
add     sp, sp, #CALL_CONTEXT_SIZE

BRANCH_AND_LINK_TO_REG     x9                   /* call fn */

ldp     x3, x4, [x29, #16]      /* reload rvalue and flags */

/* Partially deconstruct the stack frame.  */
ldr     x9, [x29, #32]
mov     sp, x9
cfi_def_cfa_register (sp)
mov     x2, x29                 /* Preserve for auth */
ldp     x29, x30, [x29]

/* Save the return value as directed.  */
adr     x5, 0f
and     w4, w4, #AARCH64_RET_MASK
add     x5, x5, x4, lsl #4
br      x5

/* Note that each table entry is 4 insns, and thus 16 bytes.
   For integer data, note that we're storing into ffi_arg
   and therefore we want to extend to 64 bits; these types
   have two consecutive entries allocated for them.  */
.align  4

0: BTI_J /* VOID */

b 99f
nop 
nop

1: BTI_J /* INT64 */

str     x0, [x3]
b 99f
nop

2: BTI_J /* INT128 */

stp     x0, x1, [x3]
b 99f
nop

3: brk #1000 /* UNUSED */

b 99f
nop
nop

4: brk #1000 /* UNUSED */

b 99f
nop
nop

5: brk #1000 /* UNUSED */

b 99f
nop
nop

6: brk #1000 /* UNUSED */

b 99f
nop
nop

7: brk #1000 /* UNUSED */

b 99f
nop
nop

8: BTI_J /* S4 */

st4     { v0.s, v1.s, v2.s, v3.s }[0], [x3]
b 99f
nop

9: BTI_J /* S3 */

st3     { v0.s, v1.s, v2.s }[0], [x3]
b 99f
nop

10: BTI_J /* S2 */

stp     s0, s1, [x3]
b 99f
nop

11: BTI_J

str     s0, [x3]                /* S1 */
b 99f
nop

12: BTI_J /* D4 */

st4     { v0.d, v1.d, v2.d, v3.d }[0], [x3]
b 99f
nop

13: BTI_J /* D3 */

st3     { v0.d, v1.d, v2.d }[0], [x3]
b 99f
nop

14: BTI_J /* D2 */

stp     d0, d1, [x3]
b 99f
nop

15: BTI_J /* D1 */

str     d0, [x3]
b 99f
nop

16: BTI_J /* Q4 */

str     q3, [x3, #48]
nop
nop

17: BTI_J /* Q3 */

str     q2, [x3, #32]
nop
nop

18: BTI_J /* Q2 */

stp     q0, q1, [x3]
b 99f
nop

19: BTI_J /* Q1 */

str     q0, [x3]
b 99f
nop

20: BTI_J /* UINT8 */

uxtb    w0, w0
str     x0, [x3]
nop

21: b 99f /* reserved */

nop
nop
nop

22: BTI_J /* UINT16 */

uxth    w0, w0
str     x0, [x3]
nop

23: b 99f /* reserved */

nop
nop
nop

24: BTI_J /* UINT32 */

mov     w0, w0
str     x0, [x3]
nop

25: b 99f /* reserved */

nop
nop
nop

26: BTI_J /* SINT8 */

sxtb    x0, w0
str     x0, [x3]
nop

27: b 99f /* reserved */

nop
nop
nop

28: BTI_J /* SINT16 */

sxth    x0, w0
str     x0, [x3]
nop

29: b 99f /* reserved */

nop
nop
nop

30: BTI_J /* SINT32 */

sxtw    x0, w0
str     x0, [x3]
nop

31: b 99f /* reserved */

nop
nop
nop

/* Return now that result has been populated. */

99:

AUTH_LR_WITH_REG(x2)
ret

cfi_endproc

.globl  CNAME(ffi_call_SYSV)
FFI_HIDDEN(CNAME(ffi_call_SYSV))

ifdef __ELF__

.type   CNAME(ffi_call_SYSV), #function
.size CNAME(ffi_call_SYSV), .-CNAME(ffi_call_SYSV)

endif

if FFI_CLOSURES

/* ffi_closure_SYSV

Closure invocation glue. This is the low level code invoked directly by
the closure trampoline to setup and call a closure.

On entry x17 points to a struct ffi_closure, x16 has been clobbered
all other registers are preserved.

We allocate a call context and save the argument passing registers,
then invoked the generic C ffi_closure_SYSV_inner() function to do all
the real work, on return we load the result passing registers back from
the call context.

*/

define ffi_closure_SYSV_FS (8*2 + CALL_CONTEXT_SIZE + 64)

.align 4

CNAME(ffi_closure_SYSV_V):

cfi_startproc
BTI_C
SIGN_LR
PAC_CFI_WINDOW_SAVE
stp     x29, x30, [sp, #-ffi_closure_SYSV_FS]!
cfi_adjust_cfa_offset (ffi_closure_SYSV_FS)
cfi_rel_offset (x29, 0)
cfi_rel_offset (x30, 8)

/* Save the argument passing vector registers.  */
stp     q0, q1, [sp, #16 + 0]
stp     q2, q3, [sp, #16 + 32]
stp     q4, q5, [sp, #16 + 64]
stp     q6, q7, [sp, #16 + 96]
b       0f
cfi_endproc

.globl  CNAME(ffi_closure_SYSV_V)
FFI_HIDDEN(CNAME(ffi_closure_SYSV_V))

ifdef __ELF__

.type   CNAME(ffi_closure_SYSV_V), #function
.size   CNAME(ffi_closure_SYSV_V), . - CNAME(ffi_closure_SYSV_V)

endif

.align  4

CNAME(ffi_closure_SYSV):

cfi_startproc
BTI_C
SIGN_LR
PAC_CFI_WINDOW_SAVE
stp     x29, x30, [sp, #-ffi_closure_SYSV_FS]!
cfi_adjust_cfa_offset (ffi_closure_SYSV_FS)
cfi_rel_offset (x29, 0)
cfi_rel_offset (x30, 8)

0:

mov     x29, sp

/* Save the argument passing core registers.  */
stp     x0, x1, [sp, #16 + 16*N_V_ARG_REG + 0]
stp     x2, x3, [sp, #16 + 16*N_V_ARG_REG + 16]
stp     x4, x5, [sp, #16 + 16*N_V_ARG_REG + 32]
stp     x6, x7, [sp, #16 + 16*N_V_ARG_REG + 48]

/* Load ffi_closure_inner arguments.  */
ldp     PTR_REG(0), PTR_REG(1), [x17, #FFI_TRAMPOLINE_CLOSURE_OFFSET]   /* load cif, fn */
ldr     PTR_REG(2), [x17, #FFI_TRAMPOLINE_CLOSURE_OFFSET+PTR_SIZE*2]    /* load user_data */

ifdef FFI_GO_CLOSURES L(do_closure): endif

add     x3, sp, #16                             /* load context */
add     x4, sp, #ffi_closure_SYSV_FS            /* load stack */
add     x5, sp, #16+CALL_CONTEXT_SIZE           /* load rvalue */
mov     x6, x8                                  /* load struct_rval */
bl      CNAME(ffi_closure_SYSV_inner)

/* Load the return value as directed.  */
adr     x1, 0f
and     w0, w0, #AARCH64_RET_MASK
add     x1, x1, x0, lsl #4
add     x3, sp, #16+CALL_CONTEXT_SIZE
br      x1

/* Note that each table entry is 4 insns, and thus 16 bytes.  */
.align  4

0: BTI_J /* VOID */

b       99f
nop
nop

1: BTI_J /* INT64 */

ldr     x0, [x3]
b       99f
nop

2: BTI_J /* INT128 */

ldp     x0, x1, [x3]
b       99f
nop

3: brk #1000 /* UNUSED */

nop
nop
nop

4: brk #1000 /* UNUSED */

nop
nop
nop

5: brk #1000 /* UNUSED */

nop
nop
nop

6: brk #1000 /* UNUSED */

nop
nop
nop

7: brk #1000 /* UNUSED */

nop
nop
nop

8: BTI_J /* S4 */

ldr     s3, [x3, #12]
nop
nop

9: BTI_J /* S3 */

ldr     s2, [x3, #8]
nop
nop

10: BTI_J /* S2 */

ldp     s0, s1, [x3]
b       99f
nop

11: BTI_J /* S1 */

ldr     s0, [x3]
b       99f
nop

12: BTI_J /* D4 */

ldr     d3, [x3, #24]
nop
nop

13: BTI_J /* D3 */

ldr     d2, [x3, #16]
nop
nop

14: BTI_J /* D2 */

ldp     d0, d1, [x3]
b       99f
nop

15: BTI_J /* D1 */

ldr     d0, [x3]
b       99f
nop

16: BTI_J /* Q4 */

ldr     q3, [x3, #48]
nop
nop

17: BTI_J /* Q3 */

ldr     q2, [x3, #32]
nop
nop

18: BTI_J /* Q2 */

ldp     q0, q1, [x3]
b       99f
nop

19: BTI_J /* Q1 */

ldr     q0, [x3]
b       99f
nop

20: BTI_J /* UINT8 */

ldrb    w0, [x3, #BE(7)]
b       99f
nop

21: brk #1000 /* reserved */

nop
nop
nop

22: BTI_J /* UINT16 */

ldrh    w0, [x3, #BE(6)]
b       99f
nop

23: brk #1000 /* reserved */

nop
nop
nop

24: BTI_J /* UINT32 */

ldr     w0, [x3, #BE(4)]
b       99f
nop

25: brk #1000 /* reserved */

nop
nop
nop

26: BTI_J /* SINT8 */

ldrsb   x0, [x3, #BE(7)]
b       99f
nop

27: brk #1000 /* reserved */

nop
nop
nop

28: BTI_J /* SINT16 */

ldrsh   x0, [x3, #BE(6)]
b       99f
nop

29: brk #1000 /* reserved */

nop
nop
nop

30: BTI_J /* SINT32 */

ldrsw   x0, [x3, #BE(4)]
nop
nop

31: /* reserved */ 99: ldp x29, x30, [sp], ffi_closure_SYSV_FS

cfi_adjust_cfa_offset (-ffi_closure_SYSV_FS)
cfi_restore (x29)
cfi_restore (x30)
AUTH_LR_AND_RET
cfi_endproc

.globl  CNAME(ffi_closure_SYSV)
FFI_HIDDEN(CNAME(ffi_closure_SYSV))

ifdef __ELF__

.type   CNAME(ffi_closure_SYSV), #function
.size   CNAME(ffi_closure_SYSV), . - CNAME(ffi_closure_SYSV)

endif

if defined(FFI_EXEC_STATIC_TRAMP)

.align 4

CNAME(ffi_closure_SYSV_V_alt):

BTI_C
/* See the comments above trampoline_code_table. */
ldr     x17, [sp, #8]                   /* Load closure in x17 */
add     sp, sp, #16                     /* Restore the stack */
b       CNAME(ffi_closure_SYSV_V)

.globl  CNAME(ffi_closure_SYSV_V_alt)
FFI_HIDDEN(CNAME(ffi_closure_SYSV_V_alt))

ifdef __ELF__

.type   CNAME(ffi_closure_SYSV_V_alt), #function
.size   CNAME(ffi_closure_SYSV_V_alt), . - CNAME(ffi_closure_SYSV_V_alt)

endif

.align 4

CNAME(ffi_closure_SYSV_alt):

BTI_C
/* See the comments above trampoline_code_table. */
ldr     x17, [sp, #8]                   /* Load closure in x17 */
add     sp, sp, #16                     /* Restore the stack */
b       CNAME(ffi_closure_SYSV)

.globl  CNAME(ffi_closure_SYSV_alt)
FFI_HIDDEN(CNAME(ffi_closure_SYSV_alt))

ifdef __ELF__

.type   CNAME(ffi_closure_SYSV_alt), #function
.size   CNAME(ffi_closure_SYSV_alt), . - CNAME(ffi_closure_SYSV_alt)

endif

/*

* Below is the definition of the trampoline code table. Each element in
* the code table is a trampoline.
*/

/*

* The trampoline uses register x17. It saves the original value of x17 on
* the stack.
*
* The trampoline has two parameters - target code to jump to and data for
* the target code. The trampoline extracts the parameters from its parameter
* block (see tramp_table_map()). The trampoline saves the data address on
* the stack. Finally, it jumps to the target code.
*
* The target code can choose to:
*
* - restore the value of x17
* - load the data address in a register
* - restore the stack pointer to what it was when the trampoline was invoked.
*/
       .align  AARCH64_TRAMP_MAP_SHIFT

CNAME(trampoline_code_table):

.rept   AARCH64_TRAMP_MAP_SIZE / AARCH64_TRAMP_SIZE
sub     sp, sp, #16             /* Make space on the stack */
str     x17, [sp]               /* Save x17 on stack */
adr     x17, #16376             /* Get data address */
ldr     x17, [x17]              /* Copy data into x17 */
str     x17, [sp, #8]           /* Save data on stack */
adr     x17, #16372             /* Get code address */
ldr     x17, [x17]              /* Load code address into x17 */
br      x17                     /* Jump to code */
.endr

.globl CNAME(trampoline_code_table)
FFI_HIDDEN(CNAME(trampoline_code_table))

ifdef __ELF__

.type   CNAME(trampoline_code_table), #function
.size   CNAME(trampoline_code_table), . - CNAME(trampoline_code_table)

endif

.align  AARCH64_TRAMP_MAP_SHIFT

endif /* FFI_EXEC_STATIC_TRAMP */

if FFI_EXEC_TRAMPOLINE_TABLE

ifdef __MACH__ include <mach/machine/vm_param.h>

.align PAGE_MAX_SHIFT

CNAME(ffi_closure_trampoline_table_page):

.rept PAGE_MAX_SIZE / FFI_TRAMPOLINE_SIZE
adr x16, -PAGE_MAX_SIZE
ldp x17, x16, [x16]
br x16
    nop             /* each entry in the trampoline config page is 2*sizeof(void*) so the trampoline itself cannot be smaller than 16 bytes */
.endr

.globl CNAME(ffi_closure_trampoline_table_page)
FFI_HIDDEN(CNAME(ffi_closure_trampoline_table_page))
#ifdef __ELF__
    .type   CNAME(ffi_closure_trampoline_table_page), #function
    .size   CNAME(ffi_closure_trampoline_table_page), . - CNAME(ffi_closure_trampoline_table_page)
#endif

endif

endif /* FFI_EXEC_TRAMPOLINE_TABLE */

ifdef FFI_GO_CLOSURES

.align 4

CNAME(ffi_go_closure_SYSV_V):

cfi_startproc
BTI_C
stp     x29, x30, [sp, #-ffi_closure_SYSV_FS]!
cfi_adjust_cfa_offset (ffi_closure_SYSV_FS)
cfi_rel_offset (x29, 0)
cfi_rel_offset (x30, 8)

/* Save the argument passing vector registers.  */
stp     q0, q1, [sp, #16 + 0]
stp     q2, q3, [sp, #16 + 32]
stp     q4, q5, [sp, #16 + 64]
stp     q6, q7, [sp, #16 + 96]
b       0f
cfi_endproc

.globl  CNAME(ffi_go_closure_SYSV_V)
FFI_HIDDEN(CNAME(ffi_go_closure_SYSV_V))

ifdef __ELF__

.type   CNAME(ffi_go_closure_SYSV_V), #function
.size   CNAME(ffi_go_closure_SYSV_V), . - CNAME(ffi_go_closure_SYSV_V)

endif

.align  4

CNAME(ffi_go_closure_SYSV):

cfi_startproc
BTI_C
SIGN_LR_LINUX_ONLY
PAC_CFI_WINDOW_SAVE
stp     x29, x30, [sp, #-ffi_closure_SYSV_FS]!
cfi_adjust_cfa_offset (ffi_closure_SYSV_FS)
cfi_rel_offset (x29, 0)
cfi_rel_offset (x30, 8)

0:

mov     x29, sp

/* Save the argument passing core registers.  */
stp     x0, x1, [sp, #16 + 16*N_V_ARG_REG + 0]
stp     x2, x3, [sp, #16 + 16*N_V_ARG_REG + 16]
stp     x4, x5, [sp, #16 + 16*N_V_ARG_REG + 32]
stp     x6, x7, [sp, #16 + 16*N_V_ARG_REG + 48]

/* Load ffi_closure_inner arguments.  */
ldp     PTR_REG(0), PTR_REG(1), [x18, #PTR_SIZE]/* load cif, fn */
mov     x2, x18                                 /* load user_data */
b       L(do_closure)
cfi_endproc

.globl  CNAME(ffi_go_closure_SYSV)
FFI_HIDDEN(CNAME(ffi_go_closure_SYSV))

ifdef __ELF__

.type   CNAME(ffi_go_closure_SYSV), #function
.size   CNAME(ffi_go_closure_SYSV), . - CNAME(ffi_go_closure_SYSV)

endif endif /* FFI_GO_CLOSURES */ endif /* FFI_CLOSURES */ endif /* arm64 */

if defined __ELF__ && defined __linux__

.section .note.GNU-stack,"",%progbits

.pushsection .note.gnu.property, "a";
.balign 8;
.long 4;
.long 0x10;
.long 0x5;
.asciz "GNU";
.long 0xc0000000; /* GNU_PROPERTY_AARCH64_FEATURE_1_AND */
.long 4;
.long GNU_PROPERTY_AARCH64_BTI | GNU_PROPERTY_AARCH64_POINTER_AUTH;
.long 0;
.popsection;

endif