I forgot to mention, the old

Nick3 · ‎08-06-2018

Looking at the version of kernel32.f90 in C:\Program Files (x86)\Intel\Composer XE 2015\compiler\include

INTERFACE
FUNCTION VirtualProtect( &
        lpAddress, &
        dwSize, &
        flNewProtect, &
        lpflOldProtect)
import
integer(BOOL) :: VirtualProtect ! BOOL
    !DEC$ ATTRIBUTES DEFAULT, STDCALL, DECORATE, ALIAS:'VirtualProtect' :: VirtualProtect
integer(LPVOID) lpAddress ! LPVOID lpAddress
integer(SIZE_T) dwSize ! SIZE_T dwSize
integer(DWORD) flNewProtect ! DWORD flNewProtect
integer(LPDWORD) lpflOldProtect ! PDWORD lpflOldProtect
END FUNCTION
END INTERFACE

And then the version in C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2018.3.210\windows\compiler\include

INTERFACE
FUNCTION VirtualProtect( &
        lpAddress, &
        dwSize, &
        flNewProtect, &
        lpflOldProtect)
import
integer(BOOL) :: VirtualProtect ! BOOL
    !DEC$ ATTRIBUTES DEFAULT, STDCALL, DECORATE, ALIAS:'VirtualProtect' :: VirtualProtect
integer(LPVOID) lpAddress ! LPVOID lpAddress
integer(SIZE_T) dwSize ! SIZE_T dwSize
integer(DWORD) flNewProtect ! DWORD flNewProtect
integer(DWORD) lpflOldProtect ! PDWORD lpflOldProtect
!DEC$ ATTRIBUTES REFERENCE, IGNORE_LOC :: lpflOldProtect
END FUNCTION
END INTERFACE

Running a 64-bit executable, the call in the older ifort version succeeds. The call in the newer version fails. May I suggest, from:

https://msdn.microsoft.com/en-us/library/windows/desktop/aa366898(v=vs.85).aspx

That the correct declaration should be:

INTERFACE

FUNCTION VirtualProtect(

. lpAddress,

. dwSize,

. flNewProtect,

. pflOldProtect)

import

integer(BOOL) :: VirtualProtect ! BOOL

!DEC$ ATTRIBUTES DEFAULT, STDCALL, DECORATE :: VirtualProtect

!DEC$ ATTRIBUTES ALIAS:'VirtualProtect' :: VirtualProtect

integer(LPVOID) lpAddress ! LPVOID lpAddress

integer(SIZE_T) dwSize ! SIZE_T dwSize

integer(DWORD) flNewProtect ! DWORD flNewProtect

integer(PDWORD) pflOldProtect ! PDWORD lpflOldProtect

END FUNCTION

END INTERFACE

Steve_Lionel · ‎08-06-2018

Unless I am misreading the documentation, which I suppose is possible, the new interface is correct and should work with old sources that passed LOC(old-protect-variable). The goal of the change, and there are a lot of these in the modules, is to replace "address by value" with "correctly-typed-thing by reference". The IGNORE_LOC allows old code to continue to work.

Can you provide a short but complete example showing what fails (and please elaborate on "fails")?

Nick3 · ‎08-06-2018

This should work with any DLL and function, I picked out dbghelp from v7.1 Windows API (I believe the error is in 4-byte vs 8-byte):

      program main
      USE KERNEL32
      USE IFWINTY
      USE ISO_C_BINDING
      
      INTEGER(HANDLE)hLib
      TYPE(C_PTR) PATCHLOC1
      INTEGER(DWORD)iGetLastError
      CHARACTER(100)SLIB
    
      SLIB=
     & "C:\\Program Files\\Debugging Tools for Windows (x64)\\"//
     & "dbghelp.dll"c
      hLib=LoadLibrary(SLIB)
      PATCHLOC1=TRANSFER(GetProcAddress(hLib,"StackWalk64"C),PATCHLOC1)
      CALL DoCrash(PATCHLOC1)
      END
      
      subroutine DoCrash(PFORIGINAL)
      use kernel32
      USE ISO_C_BINDING
C
      TYPE(C_PTR), INTENT(IN) :: PFORIGINAL
      INTEGER(LPVOID) FPFORIGINAL
      TYPE(C_PTR) CPTR_OLDPROTECT
      INTEGER(DWORD) OLDPROTECT
      INTEGER(PDWORD) FPTR_OLDPROTECT
      INTEGER(BOOL) BSUCCESS
      INTEGER(SIZE_T) SIZE
C
      FPFORIGINAL = TRANSFER(PFORIGINAL,FPFORIGINAL)
      CPTR_OLDPROTECT = C_LOC(OLDPROTECT)
      FPTR_OLDPROTECT = TRANSFER(CPTR_OLDPROTECT,FPTR_OLDPROTECT)
      SIZE = 50
      BSUCCESS = VirtualProtect(FPFORIGINAL,SIZE,PAGE_READWRITE,
     &                          FPTR_OLDPROTECT)
      IF (BSUCCESS == 0) THEN
        WRITE(*,*)"Uh oh, failed!"
      ENDIF
C Now, Undo
      BSUCCESS = VirtualProtect(FPFORIGINAL,SIZE,OLDPROTECT,
     &                          FPTR_OLDPROTECT)
      IF (BSUCCESS == 0) THEN
        WRITE(*,*)"Uh oh, failed!"
      ENDIF
      end subroutine

Nick3 · ‎08-06-2018

I forgot to mention, the old compiler executes this with both BSUCCESS values being 1, while the new compiler sets BSUCCESS = 0 on the second call.

Steve_Lionel · ‎08-06-2018

This fails for you because you're constructing the OLDPROTECT pointer manually, which is unnecessary and not typical. This works with the new interface.

      subroutine DoCrash(PFORIGINAL)
      use kernel32
      USE ISO_C_BINDING
C
      TYPE(C_PTR), INTENT(IN) :: PFORIGINAL
      INTEGER(LPVOID) FPFORIGINAL
!      TYPE(C_PTR) CPTR_OLDPROTECT
      INTEGER(DWORD) OLDPROTECT
      INTEGER(BOOL) BSUCCESS
      INTEGER(SIZE_T) SIZE
C
      FPFORIGINAL = TRANSFER(PFORIGINAL,FPFORIGINAL)
      !CPTR_OLDPROTECT = C_LOC(OLDPROTECT)
      !FPTR_OLDPROTECT = TRANSFER(CPTR_OLDPROTECT,FPTR_OLDPROTECT)
      SIZE = 50
      BSUCCESS = VirtualProtect(FPFORIGINAL,SIZE,PAGE_READWRITE,
     &                          OLDPROTECT)
      IF (BSUCCESS == 0) THEN
        WRITE(*,*)"Uh oh, failed!"
      ENDIF
C Now, Undo
      BSUCCESS = VirtualProtect(FPFORIGINAL,SIZE,OLDPROTECT,
     &                          OLDPROTECT)
      IF (BSUCCESS == 0) THEN
        WRITE(*,*)"Uh oh, failed!"
      ENDIF
      end subroutine

And this (snippet) works with both old and new:

      BSUCCESS = VirtualProtect(FPFORIGINAL,SIZE,PAGE_READWRITE,
     &                          LOC(OLDPROTECT))
      IF (BSUCCESS == 0) THEN
        WRITE(*,*)"Uh oh, failed!"
      ENDIF
C Now, Undo
      BSUCCESS = VirtualProtect(FPFORIGINAL,SIZE,OLDPROTECT,
     &                          LOC(OLDPROTECT))

Note that you get a compile error with the way you had it, which would be a clue that something needs to change.

VirtualProtect in kernel32.f90