mirror of
https://github.com/partic2/libffi
synced 2026-11-10 18:17:19 +00:00
aarch64: Move return value handling into ffi_call_SYSV
This lets us pass return data directly to the caller of ffi_call in most cases, rather than storing it into temporary storage first.
This commit is contained in:
Richard Henderson
committed by
Richard Henderson
Richard Henderson
parent
325471ea6a
commit
4fe1aea121
+99
-26
@@ -40,9 +40,9 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
|
||||
#endif
|
||||
|
||||
.text
|
||||
.align 2
|
||||
.align 4
|
||||
|
||||
.globl CNAME(ffi_call_SYSV)
|
||||
.globl CNAME(ffi_call_SYSV)
|
||||
#ifdef __ELF__
|
||||
.type CNAME(ffi_call_SYSV), #function
|
||||
.hidden CNAME(ffi_call_SYSV)
|
||||
@@ -50,14 +50,15 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
|
||||
|
||||
/* ffi_call_SYSV
|
||||
extern void ffi_call_SYSV (void *stack, void *frame,
|
||||
void (*fn)(void), int flags);
|
||||
void (*fn)(void), void *rvalue, int flags);
|
||||
|
||||
Therefore on entry we have:
|
||||
|
||||
x0 stack
|
||||
x1 frame
|
||||
x2 fn
|
||||
x3 flags
|
||||
x3 rvalue
|
||||
x4 flags
|
||||
*/
|
||||
|
||||
cfi_startproc
|
||||
@@ -71,43 +72,111 @@ CNAME(ffi_call_SYSV):
|
||||
cfi_rel_offset (x29, 0)
|
||||
cfi_rel_offset (x30, 8)
|
||||
|
||||
str w3, [x29, #16] /* save flags */
|
||||
mov x9, x2 /* save fn */
|
||||
mov x8, x3 /* install structure return */
|
||||
stp x3, x4, [x29, #16] /* save rvalue and flags */
|
||||
|
||||
/* Load the vector argument passing registers, if necessary. */
|
||||
tbz w3, #AARCH64_FLAG_ARG_V_BIT, 1f
|
||||
ldp q0, q1, [x29, #32 + 0]
|
||||
ldp q2, q3, [x29, #32 + 32]
|
||||
ldp q4, q5, [x29, #32 + 64]
|
||||
ldp q6, q7, [x29, #32 + 96]
|
||||
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, [x29, #32 + 16*N_V_ARG_REG + 0]
|
||||
ldp x2, x3, [x29, #32 + 16*N_V_ARG_REG + 16]
|
||||
ldp x4, x5, [x29, #32 + 16*N_V_ARG_REG + 32]
|
||||
ldp x6, x7, [x29, #32 + 16*N_V_ARG_REG + 48]
|
||||
ldr x8, [x29, #32 + 16*N_V_ARG_REG + 64]
|
||||
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
|
||||
|
||||
blr x9 /* call fn */
|
||||
|
||||
ldr w3, [x29, #16] /* reload flags */
|
||||
ldp x3, x4, [x29, #16] /* reload rvalue and flags */
|
||||
|
||||
/* Partially deconstruct the stack frame. */
|
||||
mov sp, x29
|
||||
cfi_def_cfa_register (sp)
|
||||
ldp x29, x30, [x29]
|
||||
|
||||
/* Save the core return registers. */
|
||||
stp x0, x1, [sp, #32 + 16*N_V_ARG_REG]
|
||||
/* Save the return value as directed. */
|
||||
adr x5, 0f
|
||||
and w4, w4, #AARCH64_RET_MASK
|
||||
add x5, x5, x4, lsl #3
|
||||
br x5
|
||||
|
||||
/* Save the vector return registers, if necessary. */
|
||||
tbz w3, #AARCH64_FLAG_ARG_V_BIT, 1f
|
||||
stp q0, q1, [sp, #32 + 0]
|
||||
stp q2, q3, [sp, #32 + 32]
|
||||
1:
|
||||
/* All done. */
|
||||
/* Note that each table entry is 2 insns, and thus 8 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: ret /* VOID */
|
||||
nop
|
||||
1: str x0, [x3] /* INT64 */
|
||||
ret
|
||||
2: stp x0, x1, [x3] /* INT128 */
|
||||
ret
|
||||
3: brk #1000 /* UNUSED */
|
||||
ret
|
||||
4: brk #1000 /* UNUSED */
|
||||
ret
|
||||
5: brk #1000 /* UNUSED */
|
||||
ret
|
||||
6: brk #1000 /* UNUSED */
|
||||
ret
|
||||
7: brk #1000 /* UNUSED */
|
||||
ret
|
||||
8: st4 { v0.s-v3.s }[0], [x3] /* S4 */
|
||||
ret
|
||||
9: st3 { v0.s-v2.s }[0], [x3] /* S3 */
|
||||
ret
|
||||
10: stp s0, s1, [x3] /* S2 */
|
||||
ret
|
||||
11: str s0, [x3] /* S1 */
|
||||
ret
|
||||
12: st4 { v0.d-v3.d }[0], [x3] /* D4 */
|
||||
ret
|
||||
13: st3 { v0.d-v2.d }[0], [x3] /* D3 */
|
||||
ret
|
||||
14: stp d0, d1, [x3] /* D2 */
|
||||
ret
|
||||
15: str d0, [x3] /* D1 */
|
||||
ret
|
||||
16: str q3, [x3, #48] /* Q4 */
|
||||
nop
|
||||
17: str q2, [x3, #32] /* Q3 */
|
||||
nop
|
||||
18: stp q0, q1, [x3] /* Q2 */
|
||||
ret
|
||||
19: str q0, [x3] /* Q1 */
|
||||
ret
|
||||
20: uxtb w0, w0 /* UINT8 */
|
||||
str x0, [x3]
|
||||
21: ret /* reserved */
|
||||
nop
|
||||
22: uxth w0, w0 /* UINT16 */
|
||||
str x0, [x3]
|
||||
23: ret /* reserved */
|
||||
nop
|
||||
24: mov w0, w0 /* UINT32 */
|
||||
str x0, [x3]
|
||||
25: ret /* reserved */
|
||||
nop
|
||||
26: sxtb x0, w0 /* SINT8 */
|
||||
str x0, [x3]
|
||||
27: ret /* reserved */
|
||||
nop
|
||||
28: sxth x0, w0 /* SINT16 */
|
||||
str x0, [x3]
|
||||
29: ret /* reserved */
|
||||
nop
|
||||
30: sxtw x0, w0 /* SINT32 */
|
||||
str x0, [x3]
|
||||
31: ret /* reserved */
|
||||
nop
|
||||
|
||||
cfi_endproc
|
||||
#ifdef __ELF__
|
||||
@@ -154,9 +223,13 @@ CNAME(ffi_call_SYSV):
|
||||
Voila! */
|
||||
|
||||
.text
|
||||
.align 2
|
||||
.align 4
|
||||
|
||||
.globl CNAME(ffi_closure_SYSV)
|
||||
.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]!
|
||||
|
||||
Reference in New Issue
Block a user