[PATCH 1/2] Cygwin: cache IsWow64Process2 host arch in wincap.

[Corinna Vinschen]

On Nov 22 08:54, Jeremy Drake via Cygwin-patches wrote:
> On Fri, 22 Nov 2024, Corinna Vinschen wrote:
> 
> > On Nov 21 11:42, Jeremy Drake via Cygwin-patches wrote:
> > > +#if defined (__x86_64__)
> > > +      host_mach = IMAGE_FILE_MACHINE_AMD64;
> > > +#elif defined (__i386__)
> > > +      host_mach = wow64 ? IMAGE_FILE_MACHINE_AMD64 : IMAGE_FILE_MACHINE_I386;
> > > +#else
> > > +      /* this should not happen */
> >
> > It should actually result in
> >
> >   #error unimplemented for this target
> 
> No, because this is the fallback case for when IsWow64Process2 fails.
> This should only happen on OS versions that don't support it, which in
> turn don't support anything other than i386 and x86_64.  However, OS
> versions that do support it also support ARM64.  If/when Cygwin has native
> ARM64 support, this should not be a compilation error.  If anything it may
> be a runtime error (assert?)

Aaaah, yeah, I see what you mean.  In that case, just set host_mach to
IMAGE_FILE_MACHINE_AMD64 and be done with it.  There's no other choice
in that case anyway.

> > > +extern const IMAGE_DOS_HEADER
> > > +dosheader __asm__ ("__image_base__");
> > > +
> > > +const USHORT
> > > +wincapc::current_module_machine () const
> > > +{
> > > +  PIMAGE_NT_HEADERS ntheader = (PIMAGE_NT_HEADERS)((LPBYTE) &dosheader
> > > +                                                   + dosheader.e_lfanew);
> > > +  return ntheader->FileHeader.Machine;
> > >  }
> >
> > Just scratch that.  First, we're using GetModuleHandle(NULL)
> > throughout to access the image base, but apart from that,
> > the info is already available in wincap via cpu_arch().
> 
> GetModuleHandle(NULL) is the exe, __image_base__ is the cygwin dll.
> Theoretically, with ARM64EC, you can mix-and-match x86_64 and ARM64 in the
> same process,

Oh, ok, I read about this a while ago but had completely forgotten.

> so the most correct answer to the question to "are we being
> emulated" is whether the current module's architecture matches the host
> system's architecture.
> 
> Yes, due to Windows already lying in Get(Native)SystemInfo, cpu_arch()
> will tell you this, but with *different* enums, which means you'll
> need a switch somewhere to translate between them, instead of just doing
> an == or != with the host like this function lets you do.

This OTOH is well known at compile time, so this is where an #ifdef
may be in order, i .e.

  #if defined __x86_64__
    current_mach = IMAGE_FILE_MACHINE_AMD64;
  #else
    #error unimplemented for this target
  #endif

but, yeah, your code is simpler.  I just don't like the name of the
function, wouldn't something like current_machine() or cygwin_machine()
be better?


Thanks,
Corinna