Re: [PATCH] x86/headers: Fix (old) header file dependency bug in uapi/asm/sigcontext32.h

[Ingo Molnar]

* Mikko Rapeli <mikko.rapeli@xxxxxx> wrote:

> On Thu, Sep 03, 2015 at 09:42:32AM +0200, Ingo Molnar wrote:
> > 
> > * Mikko Rapeli <mikko.rapeli@xxxxxx> wrote:
> > 
> > > Hi Ingo, Thomas and Peter,
> > > 
> > > Do you have any pointers for me how arch/x86/include/uapi/asm/sigcontext32.h
> > > could be fixed to also compile in userspace? Is definition of _fpx_sw_bytes
> > > or _fpstate_ia32 even needed there?
> > > 
> > > Instructions to reproduce the userspace build failure are here:
> > > https://lkml.org/lkml/2015/5/30/98
> > 
> > Yeah, so this is a real bug in the headers, good find.
> > 
> > Also note that somewhat luckily your testcase is artificial, no real user-space 
> > code can contain that code at the moment because the header never built standalone 
> > in the past either AFAICS.
> > 
> > So it's an old header file dependency bug, to make it build in user-space you need 
> > to do this workaround:
> > 
> >  #include <asm/sigcontext.h>
> >  #include <asm/sigcontext32.h>
> > 
> > If you only include sigcontext32.h:
> > 
> >  #include <asm/sigcontext32.h>
> > 
> > it will fail to build.
> > 
> > To fix the bug:
> > 
> > > > --- a/arch/x86/include/uapi/asm/sigcontext32.h
> > > > +++ b/arch/x86/include/uapi/asm/sigcontext32.h
> > > > @@ -45,7 +45,6 @@ struct _fpstate_ia32 {
> > > >  	__u32	padding[44];
> > > >  	union {
> > > >  		__u32 padding2[12];
> > > > -		struct _fpx_sw_bytes sw_reserved;
> > > >  	};
> > > >  };
> > 
> > No, those fields are real and compat sigframe handling user-space might be relying 
> > on them.
> >
> > Does the fix below work for you?
> 
> Unfortunately no:
> 
> cc -Wall -c -nostdinc -I /usr/lib/gcc/i586-linux-gnu/5/include -I /usr/lib/gcc/i
> 586-linux-gnu/5/include-fixed -I . -I ../headers_compile_test_include -I ../head
> ers_compile_test_include/i586-linux-gnu ./asm/sigcontext32.c
> In file included from ./asm/sigcontext32.c:1:0:
> ./asm/sigcontext32.h:12:8: error: redefinition of âstruct _fpregâ
>  struct _fpreg {
>         ^
> In file included from ./asm/sigcontext32.h:6:0,
>                  from ./asm/sigcontext32.c:1:
> ./asm/sigcontext.h:56:8: note: originally defined here
>  struct _fpreg {
>         ^
> In file included from ./asm/sigcontext32.c:1:0:
> ./asm/sigcontext32.h:17:8: error: redefinition of âstruct _fpxregâ
>  struct _fpxreg {
>         ^
> In file included from ./asm/sigcontext32.h:6:0,
>                  from ./asm/sigcontext32.c:1:
> ./asm/sigcontext.h:61:8: note: originally defined here
>  struct _fpxreg {
>         ^
> In file included from ./asm/sigcontext32.c:1:0:
> ./asm/sigcontext32.h:23:8: error: redefinition of âstruct _xmmregâ
>  struct _xmmreg {
>         ^
> In file included from ./asm/sigcontext32.h:6:0,
>                  from ./asm/sigcontext32.c:1:
> ./asm/sigcontext.h:67:8: note: originally defined here
>  struct _xmmreg {
>         ^
> FAILED: ./asm/sigcontext32.h
> 
> I guess this was the reason why I ended with the hack.

sigcontext32.h is a 64-bit only header (for compat functionality), and you are 
trying to build this on i386, right?

In theory we could unify them mostly, the two structures match mostly, except that 
'_fpstate_ia32::padding' is named 'padding1' in _fpstate, and there's also some 
legacy uglies wrt. user-space sigcontext definitions.

So this either gets cleaned up properly, with the legacy hacks/wrappers clearly 
isolated, or we could state that you shouldn't build 64-bit headers in 32-bit 
environments.

Thanks,

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/