;; @file
; IPRT - Unwind hacks for Windows x64.
;

;
; Copyright (C) 2006-2007 Sun Microsystems, Inc.
;
; This file is part of VirtualBox Open Source Edition (OSE), as
; available from http://www.virtualbox.org. This file is free software;
; you can redistribute it and/or modify it under the terms of the GNU
; General Public License (GPL) as published by the Free Software
; Foundation, in version 2 as it comes in the "COPYING" file of the
; VirtualBox OSE distribution. VirtualBox OSE is distributed in the
; hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
;
; The contents of this file may alternatively be used under the terms
; of the Common Development and Distribution License Version 1.0
; (CDDL) only, as it comes in the "COPYING.CDDL" file of the
; VirtualBox OSE distribution, in which case the provisions of the
; CDDL are applicable instead of those of the GPL.
;
; You may elect to license modified versions of this file under the
; terms and conditions of either the GPL or the CDDL or both.
;
; Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
; Clara, CA 95054 USA or visit http://www.sun.com if you need
; additional information or have any questions.
;

%ifndef ___iprt_ntwrap_mac
%define ___iprt_ntwrap_mac

%include "iprt/asmdefs.mac"

%ifdef RT_WITH_W64_UNWIND_HACK
%ifdef RT_ARCH_AMD64


;;
; Common prolog, take the proc name as argument.
; This creates a 0x80 byte stack frame.
;
%macro NtWrapProlog 1
[proc_frame %1]
        push    rbp
        [pushreg rbp]
        mov     rbp, rsp
        [setframe rbp, 0]
        sub     rsp, 0x80
        [allocstack 0x80]

        ; save rdi and load rbp into it
        mov     [rbp - 8h], rdi
        [savereg rdi, 0x78]
        mov     rdi, rbp
[endprolog]
%endmacro

;;
; Common epilog, take the proc name as argument.
%macro NtWrapEpilog 1
        ; restore rbp and rdi then return.
        mov     rbp, rdi
        mov     rdi, [rdi - 8h]
        leave
        ret
[endproc_frame %1]
%endmacro

;;
; Create a stack marker with the rbp. The marker is 32 byte big.
; This is 32-byte aligned and 32 byte in size.
;
; Trashes r10
%macro NtWrapCreateMarker 0
        lea     r10, [rbp - 30h]
        and     r10, ~1fh               ; 32-byte align it.
        mov     dword [r10      ], 0x20080901
        mov     dword [r10 + 04h], 0x20080902
        mov     qword [r10 + 08h], rbp
        mov     dword [r10 + 10h], 0x20080903
        mov     dword [r10 + 14h], 0x20080904
        mov     qword [r10 + 18h], rbp
%endmacro

;;
; Destroys the stack marker.
;
; Trashes r10
%macro NtWrapDestroyMarker 0
        lea     r10, [rbp - 30h]
        and     r10, ~1fh               ; 32-byte align it.
        mov     [r10      ], rbp
        mov     [r10 + 08h], rbp
        mov     [r10 + 10h], rbp
        mov     [r10 + 18h], rbp
%endmacro

;;
; Wraps a function with 4 or less argument that will go into registers.
%macro NtWrapDrv2DynFunctionWithAllRegParams 2
extern NAME(%2)
BEGINPROC %1%2
        NtWrapProlog %1%2
        NtWrapCreateMarker

        call    NAME(%2)

        NtWrapDestroyMarker
        NtWrapEpilog %1%2
ENDPROC   %1%2
%endmacro

BEGINCODE

;
; The following is not used when we're in the dynamically loaded code
; and we wish to avoid it because it drags in a dependency on IoGetStackLimits.
;
; (Could perhaps fix this differently by instantiating NtWrapLocateMarkerHelper
;  in some way, but this works just the same...)
;
%ifndef IN_DYNAMICLOAD_CODE

;;
; Find the stack marker with the rbp of the entry frame.
;
; Search the current stack page inline, call a helper function
; which does a safe search of any further stack pages.
;
; Trashes       rax, r10 and r11.
; Modifies      rbp
;
%macro NtWrapLocateMarker 0
        mov     rax, rbp
        and     rax, ~1fh               ; 32-byte align it.

        ;
        ; Calc remainig space in the current page. If we're on a
        ; page boundrary, we'll search the entire previous page.
        ;
        mov     r10, rax
        neg     r10
        and     r10, 0fffh
        inc     r10
        shr     r10, 5                  ; /= 32 bytes
        jz      %%not_found             ; If zero, take the slow path

        ;
        ; The search loop.
        ;
%%again:
        dec     r10
        lea     rax, [rax + 20h]
        jz      %%not_found
        cmp     dword [rax      ], 0x20080901
        je      %%candidate
        jmp     %%again

%%not_found:
        call    NAME(NtWrapLocateMarkerHelper)
        jmp     %%done

%%candidate:
        cmp     dword [rax + 04h], 0x20080902
        jne     %%again
        cmp     dword [rax + 10h], 0x20080903
        jne     %%again
        cmp     dword [rax + 14h], 0x20080904
        jne     %%again
        mov     r11,  [rax + 08h]
        cmp     r11,  [rax + 18h]
        jne     %%again

        ; found it, change rbp.
        mov     rbp, r11
%%done:
%endmacro

;;
; Wraps a function with 4 or less argument that will go into registers.
%macro NtWrapDyn2DrvFunctionWithAllRegParams 2
extern NAME(%2)
BEGINPROC %1%2
        NtWrapProlog %1%2
        NtWrapLocateMarker

        call    NAME(%2)

        NtWrapEpilog %1%2
ENDPROC   %1%2
%endmacro

;;
; Wraps a function with 5 argument, where the first 4 goes into registers.
%macro NtWrapDyn2DrvFunctionWith5Params 2
extern NAME(%2)
BEGINPROC %1%2
        NtWrapProlog %1%2
        NtWrapLocateMarker

        mov     r11, [rdi + 30h]
        mov     [rsp + 20h], r11
        call    NAME(%2)

        NtWrapEpilog %1%2
ENDPROC   %1%2
%endmacro

;;
; Wraps a function with 6 argument, where the first 4 goes into registers.
%macro NtWrapDyn2DrvFunctionWith6Params 2
extern NAME(%2)
BEGINPROC %1%2
        NtWrapProlog %1%2
        NtWrapLocateMarker

        mov     r11, [rdi + 30h]
        mov     [rsp + 20h], r11
        mov     r10, [rdi + 38h]
        mov     [rsp + 28h], r10
        call    NAME(%2)

        NtWrapEpilog %1%2
ENDPROC   %1%2
%endmacro

;;
; Wraps a function with 7 argument, where the first 4 goes into registers.
%macro NtWrapDyn2DrvFunctionWith7Params 2
extern NAME(%2)
BEGINPROC %1%2
        NtWrapProlog %1%2
        NtWrapLocateMarker

        mov     r11, [rdi + 30h]
        mov     [rsp + 20h], r11
        mov     r10, [rdi + 38h]
        mov     [rsp + 28h], r10
        mov     rax, [rdi + 40h]
        mov     [rsp + 30h], rax
        call    NAME(%2)

        NtWrapEpilog %1%2
ENDPROC   %1%2
%endmacro


extern IoGetStackLimits

;;
; Helper that cautiously continues the stack marker search
; NtWrapLocateMarker started.
;
; The stack layout at the time is something like this.
;       rbp+08h         callers return address.
;       rbp-00h         saved rbp
;       rbp-08h         saved rdi
;       rbp-09h
;         thru          unused.
;       rbp-80h
;       rbp-88h         our return address.
;       rbp-89h
;         thru          callee register dump zone.
;       rbp-a0h
;
; @param    rax         Current stack location.
; @param    rdi         Parent stack frame pointer. (This should equal rbp on entry.)
;
; Trashes:  rax, r10, r11.
;           Will use the callers stack frame for register saving ASSUMING that
;           rbp-80h thru rbp-09h is unused.
;
; Modifies: rbp
;
BEGINPROC NtWrapLocateMarkerHelper
        ;
        ; Prolog. Save volatile regs and reserve callee space.
        ;
        sub     rsp, 20h                ; For IoGetStackLimits().
        mov     [rdi - 80h], rax
        mov     [rdi - 78h], rcx
        mov     [rdi - 70h], rdx
        mov     [rdi - 68h], r8
        mov     [rdi - 60h], r9

        ;
        ; Call VOID IoGetStackLimits(OUT PULONG_PTR LowLimit, OUT PULONG_PTR HighLimit);
        ;
        ; Use rdi-40h for the high limit and rdi-50h for the low one, we're only
        ; interested in the high one.
        ;
        lea     rcx, [rdi - 40h]        ; arg #1 LowLimit
        lea     rdx, [rdi - 50h]        ; arg #2 HighLimit
        mov     [rdx], eax              ; paranoia - init to end of current search.
        call    IoGetStackLimits

        ;
        ; Move the top address into r10, restore rax and continue
        ; the search. Check that r10 is less than 3 pages from rax.
        ;
        mov     rax, [rdi - 80h]        ; Restore eax (see prolog)
        mov     r10, [rdi - 50h]        ; HighLimit
        and     r10, ~1fh               ; 32-byte align it (downwards)
        sub     r10, rax
        jz      .not_found              ; If already at the top of the stack.
        cmp     r10, 3000h
        jae     .out_of_bounds          ; If too far away, something is busted.
        shr     r10, 5                  ; /= 32.

        ; The loop body.
.search_loop:
        cmp     dword [rax      ], 0x20080901
        je      .candidate
.continue_searching:
        dec     r10
        jz      .not_found
        lea     rax, [rax + 20h]
        jmp     .search_loop

        ; Found the first marker, check for the rest.
.candidate:
        cmp     dword [rax + 04h], 0x20080902
        jne     .continue_searching
        cmp     dword [rax + 10h], 0x20080903
        jne     .continue_searching
        cmp     dword [rax + 14h], 0x20080904
        jne     .continue_searching
        mov     r11,  [rax + 08h]
        cmp     r11,  [rax + 18h]
        jne     .continue_searching

        ; found it, change rbp.
        mov     rbp, r11

        ;
        ; Restore registers and pop the stack frame.
        ;
.epilog:
        mov     r9,  [rdi - 60h]
        mov     r8,  [rdi - 68h]
        mov     rdx, [rdi - 70h]
        mov     rcx, [rdi - 78h]
        ; mov     rax, [rdi - 80h]
        add     rsp, 20h
        ret

        ;
        ; Needless to say, this isn't supposed to happen. Thus the int3.
        ; Note down r10 and rax.
        ;
.out_of_bounds:
%ifdef DEBUG
        int3
%endif
.not_found:
%ifdef DEBUG
        int3
%endif
        jmp     .epilog
ENDPROC   NtWrapLocateMarkerHelper

%endif ; !IN_DYNAMICLOAD_CODE

%endif ; RT_ARCH_AMD64
%endif ; RT_WITH_W64_UNWIND_HACK

%endif ; !___iprt_ntwrap_mac