mirror of
https://github.com/partic2/libffi
synced 2026-11-10 10:07:19 +00:00
aarch64: Move return value handling into ffi_closure_SYSV
As with the change to ffi_call_SYSV, this avoids copying data into a temporary buffer.
This commit is contained in:
Richard Henderson
committed by
Richard Henderson
Richard Henderson
parent
4fe1aea121
commit
12cf89ee04
+138
-109
@@ -37,17 +37,17 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
|
||||
#else
|
||||
#define CNAME(x) x
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#ifdef __AARCH64EB__
|
||||
# define BE(X) X
|
||||
#else
|
||||
# define BE(X) 0
|
||||
#endif
|
||||
|
||||
.text
|
||||
.align 4
|
||||
|
||||
.globl CNAME(ffi_call_SYSV)
|
||||
#ifdef __ELF__
|
||||
.type CNAME(ffi_call_SYSV), #function
|
||||
.hidden CNAME(ffi_call_SYSV)
|
||||
#endif
|
||||
|
||||
/* ffi_call_SYSV
|
||||
extern void ffi_call_SYSV (void *stack, void *frame,
|
||||
void (*fn)(void), void *rvalue, int flags);
|
||||
@@ -179,131 +179,160 @@ CNAME(ffi_call_SYSV):
|
||||
nop
|
||||
|
||||
cfi_endproc
|
||||
#ifdef __ELF__
|
||||
.size CNAME(ffi_call_SYSV), .-CNAME(ffi_call_SYSV)
|
||||
#endif
|
||||
|
||||
#define ffi_closure_SYSV_FS (8 * 2 + CALL_CONTEXT_SIZE)
|
||||
.globl CNAME(ffi_call_SYSV)
|
||||
#ifdef __ELF__
|
||||
.type CNAME(ffi_call_SYSV), #function
|
||||
.hidden CNAME(ffi_call_SYSV)
|
||||
.size CNAME(ffi_call_SYSV), .-CNAME(ffi_call_SYSV)
|
||||
#endif
|
||||
|
||||
/* 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 trampoline_data, x16 has been clobbered
|
||||
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.
|
||||
*/
|
||||
|
||||
On entry
|
||||
#define ffi_closure_SYSV_FS (8*2 + CALL_CONTEXT_SIZE + 64)
|
||||
|
||||
extern void
|
||||
ffi_closure_SYSV (struct trampoline_data *);
|
||||
.align 4
|
||||
CNAME(ffi_closure_SYSV_V):
|
||||
cfi_startproc
|
||||
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)
|
||||
|
||||
struct trampoline_data
|
||||
{
|
||||
UINT64 *ffi_closure;
|
||||
UINT64 flags;
|
||||
};
|
||||
/* 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
|
||||
|
||||
This function uses the following stack frame layout:
|
||||
.globl CNAME(ffi_closure_SYSV_V)
|
||||
#ifdef __ELF__
|
||||
.type CNAME(ffi_closure_SYSV_V), #function
|
||||
.hidden CNAME(ffi_closure_SYSV_V)
|
||||
.size CNAME(ffi_closure_SYSV_V), . - CNAME(ffi_closure_SYSV_V)
|
||||
#endif
|
||||
|
||||
==
|
||||
saved x30(lr)
|
||||
x29(fp)-> saved x29(fp)
|
||||
saved x22
|
||||
saved x21
|
||||
...
|
||||
sp -> call_context
|
||||
==
|
||||
.align 4
|
||||
cfi_startproc
|
||||
CNAME(ffi_closure_SYSV):
|
||||
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
|
||||
|
||||
Voila! */
|
||||
/* 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]
|
||||
str x8, [sp, #16 + 16*N_V_ARG_REG + 64]
|
||||
|
||||
.text
|
||||
.align 4
|
||||
/* Load ffi_closure_inner arguments. */
|
||||
ldp x0, x1, [x17, #FFI_TRAMPOLINE_SIZE] /* load cif, fn */
|
||||
ldr x2, [x17, #FFI_TRAMPOLINE_SIZE+16] /* load user_data */
|
||||
add x3, sp, #16 /* load context */
|
||||
add x4, sp, #ffi_closure_SYSV_FS /* load stack */
|
||||
add x5, sp, #16+CALL_CONTEXT_SIZE /* load rvalue */
|
||||
bl CNAME(ffi_closure_SYSV_inner)
|
||||
|
||||
.globl CNAME(ffi_closure_SYSV)
|
||||
/* Load the return value as directed. */
|
||||
adr x1, 0f
|
||||
and w0, w0, #AARCH64_RET_MASK
|
||||
add x1, x1, x0, lsl #3
|
||||
add x3, sp, #16+CALL_CONTEXT_SIZE
|
||||
br x1
|
||||
|
||||
/* Note that each table entry is 2 insns, and thus 8 bytes. */
|
||||
.align 4
|
||||
0: b 99f /* VOID */
|
||||
nop
|
||||
1: ldr x0, [x3] /* INT64 */
|
||||
b 99f
|
||||
2: ldp x0, x1, [x3] /* INT128 */
|
||||
b 99f
|
||||
3: brk #1000 /* UNUSED */
|
||||
nop
|
||||
4: brk #1000 /* UNUSED */
|
||||
nop
|
||||
5: brk #1000 /* UNUSED */
|
||||
nop
|
||||
6: brk #1000 /* UNUSED */
|
||||
nop
|
||||
7: brk #1000 /* UNUSED */
|
||||
nop
|
||||
8: ldr s3, [x3, #12] /* S4 */
|
||||
nop
|
||||
9: ldr s2, [x2, #8] /* S3 */
|
||||
nop
|
||||
10: ldp s0, s1, [x3] /* S2 */
|
||||
b 99f
|
||||
11: ldr s0, [x3] /* S1 */
|
||||
b 99f
|
||||
12: ldr d3, [x3, #24] /* D4 */
|
||||
nop
|
||||
13: ldr d2, [x3, #16] /* D3 */
|
||||
nop
|
||||
14: ldp d0, d1, [x3] /* D2 */
|
||||
b 99f
|
||||
15: ldr d0, [x3] /* D1 */
|
||||
b 99f
|
||||
16: ldr q3, [x3, #48] /* Q4 */
|
||||
nop
|
||||
17: ldr q2, [x3, #32] /* Q3 */
|
||||
nop
|
||||
18: ldp q0, q1, [x3] /* Q2 */
|
||||
b 99f
|
||||
19: ldr q0, [x3] /* Q1 */
|
||||
b 99f
|
||||
20: ldrb w0, [x3, #BE(7)] /* UINT8 */
|
||||
b 99f
|
||||
21: brk #1000 /* reserved */
|
||||
nop
|
||||
22: ldrh w0, [x3, #BE(6)] /* UINT16 */
|
||||
b 99f
|
||||
23: brk #1000 /* reserved */
|
||||
nop
|
||||
24: ldr w0, [x3, #BE(4)] /* UINT32 */
|
||||
b 99f
|
||||
25: brk #1000 /* reserved */
|
||||
nop
|
||||
26: ldrsb x0, [x3, #BE(7)] /* SINT8 */
|
||||
b 99f
|
||||
27: brk #1000 /* reserved */
|
||||
nop
|
||||
28: ldrsh x0, [x3, #BE(6)] /* SINT16 */
|
||||
b 99f
|
||||
29: brk #1000 /* reserved */
|
||||
nop
|
||||
30: ldrsw x0, [x3, #BE(4)] /* SINT32 */
|
||||
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)
|
||||
ret
|
||||
cfi_endproc
|
||||
|
||||
.globl CNAME(ffi_closure_SYSV)
|
||||
#ifdef __ELF__
|
||||
.type CNAME(ffi_closure_SYSV), #function
|
||||
.hidden CNAME(ffi_closure_SYSV)
|
||||
#endif
|
||||
cfi_startproc
|
||||
CNAME(ffi_closure_SYSV):
|
||||
stp x29, x30, [sp, #-16]!
|
||||
cfi_adjust_cfa_offset (16)
|
||||
cfi_rel_offset (x29, 0)
|
||||
cfi_rel_offset (x30, 8)
|
||||
|
||||
mov x29, sp
|
||||
cfi_def_cfa_register (x29)
|
||||
|
||||
sub sp, sp, #ffi_closure_SYSV_FS
|
||||
|
||||
stp x21, x22, [x29, #-16]
|
||||
cfi_rel_offset (x21, -16)
|
||||
cfi_rel_offset (x22, -8)
|
||||
|
||||
/* Load x21 with &call_context. */
|
||||
mov x21, sp
|
||||
/* Preserve our struct trampoline_data * */
|
||||
mov x22, x17
|
||||
|
||||
/* Save the rest of the argument passing registers, including
|
||||
the structure return pointer. */
|
||||
stp x0, x1, [x21, #16*N_V_ARG_REG + 0]
|
||||
stp x2, x3, [x21, #16*N_V_ARG_REG + 16]
|
||||
stp x4, x5, [x21, #16*N_V_ARG_REG + 32]
|
||||
stp x6, x7, [x21, #16*N_V_ARG_REG + 48]
|
||||
str x8, [x21, #16*N_V_ARG_REG + 64]
|
||||
|
||||
/* Figure out if we should touch the vector registers. */
|
||||
ldr x0, [x22, #8]
|
||||
tbz x0, #AARCH64_FLAG_ARG_V_BIT, 1f
|
||||
|
||||
/* Save the argument passing vector registers. */
|
||||
stp q0, q1, [x21, #0]
|
||||
stp q2, q3, [x21, #32]
|
||||
stp q4, q5, [x21, #64]
|
||||
stp q6, q7, [x21, #96]
|
||||
1:
|
||||
/* Load &ffi_closure.. */
|
||||
ldr x0, [x22, #0]
|
||||
mov x1, x21
|
||||
/* Compute the location of the stack at the point that the
|
||||
trampoline was called. */
|
||||
add x2, x29, #16
|
||||
|
||||
bl CNAME(ffi_closure_SYSV_inner)
|
||||
|
||||
/* Figure out if we should touch the vector registers. */
|
||||
ldr x0, [x22, #8]
|
||||
tbz x0, #AARCH64_FLAG_ARG_V_BIT, 1f
|
||||
|
||||
/* Load the result passing vector registers. */
|
||||
ldp q0, q1, [x21, #0]
|
||||
ldp q2, q3, [x21, #32]
|
||||
1:
|
||||
/* Load the result passing core registers. */
|
||||
ldp x0, x1, [x21, #16*N_V_ARG_REG + 0]
|
||||
|
||||
/* We are done, unwind our frame. */
|
||||
ldp x21, x22, [x29, #-16]
|
||||
cfi_restore (x21)
|
||||
cfi_restore (x22)
|
||||
|
||||
mov sp, x29
|
||||
cfi_def_cfa_register (sp)
|
||||
|
||||
ldp x29, x30, [sp], #16
|
||||
cfi_adjust_cfa_offset (-16)
|
||||
cfi_restore (x29)
|
||||
cfi_restore (x30)
|
||||
|
||||
ret
|
||||
cfi_endproc
|
||||
#ifdef __ELF__
|
||||
.size CNAME(ffi_closure_SYSV), .-CNAME(ffi_closure_SYSV)
|
||||
.size CNAME(ffi_closure_SYSV), . - CNAME(ffi_closure_SYSV)
|
||||
#endif
|
||||
|
||||
Reference in New Issue
Block a user