176 lines
5.3 KiB
Plaintext
176 lines
5.3 KiB
Plaintext
option casemap:none
|
|
|
|
.CODE
|
|
|
|
; import Coroutine_Finish with its mangled Microsoft Visual C++ name
|
|
?Coroutine_Finish@@YAXXZ PROTO
|
|
|
|
; extern "C" void SaveNonVolatileRegs( uintptr_t regs[8] );
|
|
; incoming parameter is rcs
|
|
SaveNonVolatileRegs PROC FRAME
|
|
.endprolog
|
|
mov qword ptr[rcx], rbx
|
|
mov qword ptr[rcx+8], rbp
|
|
mov qword ptr[rcx+16], rsi
|
|
mov qword ptr[rcx+24], rdi
|
|
mov qword ptr[rcx+32], r12
|
|
mov qword ptr[rcx+40], r13
|
|
mov qword ptr[rcx+48], r14
|
|
mov qword ptr[rcx+56], r15
|
|
ret
|
|
SaveNonVolatileRegs ENDP
|
|
|
|
; extern "C" void NORETURN Coroutine_Launch_ASM( byte **ppStackHigh, uintptr_t **ppLaunchParentFramePtr, void (*pfnExec)( void* ), void *pvParam )
|
|
; Per Win64 ABI, incoming params are rcx, rdx, r8, r9. initial stack pointer is half-aligned due to return address
|
|
Coroutine_Launch_ASM PROC FRAME
|
|
; x64 prolog and prolog description macros:
|
|
|
|
; save caller's nonvolatile registers (pushed in reverse order to match SaveNonVolatileRegs)
|
|
; so that we can slam new values in later to trick the x64 callstack unwind procedure
|
|
push r15
|
|
.pushreg r15
|
|
push r14
|
|
.pushreg r14
|
|
push r13
|
|
.pushreg r13
|
|
push r12
|
|
.pushreg r12
|
|
push rdi
|
|
.pushreg rdi
|
|
push rsi
|
|
.pushreg rsi
|
|
push rbp
|
|
.pushreg rbp
|
|
push rbx
|
|
.pushreg rbx
|
|
|
|
; stack-allocate Win64 function call shadow space for calls to pfnExec and Coroutine_Finish,
|
|
; plus 8 additional bytes to align the stack frame properly (comes in off by 8)
|
|
sub rsp, 28h
|
|
.allocstack 28h
|
|
|
|
.endprolog
|
|
|
|
; compute top of stack for coroutine: 40 bytes for stack, 64 for saved regs, 8 for return address
|
|
; (we do not bother including the additional unused 32 byte shadow space we own above that)
|
|
lea rax, [rsp+70h]
|
|
mov qword ptr [rcx], rax
|
|
|
|
; save off the address of our saved regs so that we can memcpy over them later and trick
|
|
; the x64 stack unwind logic into walking up to a different Internal_Coroutine_Continue
|
|
lea rax, [rsp+28h]
|
|
mov qword ptr [rdx], rax
|
|
|
|
; call pfnExec(pvParam)
|
|
mov rcx, r9
|
|
call r8
|
|
|
|
; call Coroutine_Finish - does not return
|
|
call ?Coroutine_Finish@@YAXXZ
|
|
|
|
Coroutine_Launch_ASM ENDP
|
|
|
|
|
|
|
|
|
|
; Needs to match definition found in setjmp.h
|
|
_JUMP_BUFFER STRUCT
|
|
m_Frame QWORD ?
|
|
m_Rbx QWORD ?
|
|
m_Rsp QWORD ?
|
|
m_Rbp QWORD ?
|
|
m_Rsi QWORD ?
|
|
m_Rdi QWORD ?
|
|
m_R12 QWORD ?
|
|
m_R13 QWORD ?
|
|
m_R14 QWORD ?
|
|
m_R15 QWORD ?
|
|
m_Rip QWORD ?
|
|
m_MxCsr DWORD ?
|
|
m_FpCsr WORD ?
|
|
m_Spare WORD ?
|
|
m_Xmm6 XMMWORD ?
|
|
m_Xmm7 XMMWORD ?
|
|
m_Xmm8 XMMWORD ?
|
|
m_Xmm9 XMMWORD ?
|
|
m_Xmm10 XMMWORD ?
|
|
m_Xmm11 XMMWORD ?
|
|
m_Xmm12 XMMWORD ?
|
|
m_Xmm13 XMMWORD ?
|
|
m_Xmm14 XMMWORD ?
|
|
m_Xmm15 XMMWORD ?
|
|
_JUMP_BUFFER ENDS
|
|
|
|
|
|
;This is the reference asm for __intrinsic_setjmp() in VS2015
|
|
;mov qword ptr [rcx],rdx ; intrinsic call site does "mov rdx,rbp" followed by "add rdx,0FFFFFFFFFFFFFFC0h", looks like a nonstandard abi
|
|
;mov qword ptr [rcx+8],rbx
|
|
;mov qword ptr [rcx+18h],rbp
|
|
;mov qword ptr [rcx+20h],rsi
|
|
;mov qword ptr [rcx+28h],rdi
|
|
;mov qword ptr [rcx+30h],r12
|
|
;mov qword ptr [rcx+38h],r13
|
|
;mov qword ptr [rcx+40h],r14
|
|
;mov qword ptr [rcx+48h],r15
|
|
;lea r8,[rsp+8] ; rsp set to post-return address
|
|
;mov qword ptr [rcx+10h],r8
|
|
;mov r8,qword ptr [rsp]
|
|
;mov qword ptr [rcx+50h],r8
|
|
;stmxcsr dword ptr [rcx+58h]
|
|
;fnstcw word ptr [rcx+5Ch]
|
|
;movdqa xmmword ptr [rcx+60h],xmm6
|
|
;ovdqa xmmword ptr [rcx+70h],xmm7
|
|
;movdqa xmmword ptr [rcx+80h],xmm8
|
|
;movdqa xmmword ptr [rcx+90h],xmm9
|
|
;movdqa xmmword ptr [rcx+0A0h],xmm10
|
|
;movdqa xmmword ptr [rcx+0B0h],xmm11
|
|
;movdqa xmmword ptr [rcx+0C0h],xmm12
|
|
;movdqa xmmword ptr [rcx+0D0h],xmm13
|
|
;movdqa xmmword ptr [rcx+0E0h],xmm14
|
|
;movdqa xmmword ptr [rcx+0F0h],xmm15
|
|
;xor eax,eax
|
|
;ret
|
|
|
|
|
|
; extern "C" void NORETURN Coroutine_LongJmp_UnChecked( jmp_buf buf, int nResult )
|
|
; Per Win64 ABI, incoming params are rcx, rdx, r8, r9. initial stack pointer is half-aligned due to return address
|
|
Coroutine_LongJmp_Unchecked PROC
|
|
;load nResult into result from initial setjmp()
|
|
xor rax, rax
|
|
mov eax, edx
|
|
|
|
;restore to setjmp() caller state
|
|
mov rdx, [rcx]._JUMP_BUFFER.m_Frame ; appears to be an error checking value of (_JUMP_BUFFER.m_Rbp + 0FFFFFFFFFFFFFFC0h) passed non-standardly through rdx to setjmp()
|
|
mov rbx, [rcx]._JUMP_BUFFER.m_Rbx
|
|
mov rsp, [rcx]._JUMP_BUFFER.m_Rsp
|
|
mov rbp, [rcx]._JUMP_BUFFER.m_Rbp
|
|
mov rsi, [rcx]._JUMP_BUFFER.m_Rsi
|
|
mov rdi, [rcx]._JUMP_BUFFER.m_Rdi
|
|
mov r12, [rcx]._JUMP_BUFFER.m_R12
|
|
mov r13, [rcx]._JUMP_BUFFER.m_R13
|
|
mov r14, [rcx]._JUMP_BUFFER.m_R14
|
|
mov r15, [rcx]._JUMP_BUFFER.m_R15
|
|
mov r10, [rcx]._JUMP_BUFFER.m_Rip ; store return address in r10 for return
|
|
ldmxcsr [rcx]._JUMP_BUFFER.m_MxCsr
|
|
fldcw [rcx]._JUMP_BUFFER.m_FpCsr
|
|
;[rcx]._JUMP_BUFFER.m_Spare
|
|
movaps xmm6, [rcx]._JUMP_BUFFER.m_Xmm6
|
|
movaps xmm7, [rcx]._JUMP_BUFFER.m_Xmm7
|
|
movaps xmm8, [rcx]._JUMP_BUFFER.m_Xmm8
|
|
movaps xmm9, [rcx]._JUMP_BUFFER.m_Xmm9
|
|
movaps xmm10, [rcx]._JUMP_BUFFER.m_Xmm10
|
|
movaps xmm11, [rcx]._JUMP_BUFFER.m_Xmm11
|
|
movaps xmm12, [rcx]._JUMP_BUFFER.m_Xmm12
|
|
movaps xmm13, [rcx]._JUMP_BUFFER.m_Xmm13
|
|
movaps xmm14, [rcx]._JUMP_BUFFER.m_Xmm14
|
|
movaps xmm15, [rcx]._JUMP_BUFFER.m_Xmm15
|
|
|
|
;jmp instead of ret to _JUMP_BUFFER.m_Rip because setjmp() already set the _JUMP_BUFFER.m_Rsp to the post-return state
|
|
db 048h ; emit a REX prefix on the jmp to ensure it's a full qword
|
|
jmp qword ptr r10
|
|
Coroutine_LongJmp_Unchecked ENDP
|
|
|
|
|
|
_TEXT ENDS
|
|
END
|