Re: [PATCH] Add VDSO time function support for x86 32-bit kernel

[Stefani Seibold <stefani@seibold.net>]

Am Mittwoch, den 12.12.2012, 15:34 -0800 schrieb H. Peter Anvin:
> On 12/12/2012 12:19 PM, stefani@seibold.net wrote:
> > diff --git a/arch/x86/vdso/vdso32/vclock_gettime.c b/arch/x86/vdso/vdso32/vclock_gettime.c
> > new file mode 100644
> > index 0000000..c9a1909
> > --- /dev/null
> > +++ b/arch/x86/vdso/vdso32/vclock_gettime.c
> > @@ -0,0 +1,7 @@
> > +/*
> > + * since vgtod layout differs between X86_64 and x86_32, it is not possible to
> > + * provide a 32 bit vclock with a 64 bit kernel
> > + */
> > +#ifdef CONFIG_X86_32
> > +#include "../vclock_gettime.c"
> > +#endif
> 
> This is where this goes fail.  Sorry, it is not acceptable to introduce 
> an ABI on x86-32 without providing it also on x86-64 in compatibility mode.
> 
> I also don't believe it is not possible... it might require some more 
> cleverness; perhaps we need to do the 32-bit vgtod in such a way that it 
> *is* compatible with 64 bits.
> 

The comment is ambiguous:

Since vgtod layout differs between X86_64 and x86_32 AND the vgtod is
not inside the accessible address space of a 32 bit program, it is
CURRENTLY not possible to provide a 32 bit vclock with a 64 bit kernel

As i understand VDSO it is an alternativ way, so if there is no support
for it, there must be a fall back to the system call interface in the
program or lib, which tries to use it.

So there is no drawback for 32 bit programs running on a 64 bit kernel. 

I think this feature is not so important and can implemented in a
subsequent patch, because a 64 bit kernel system mostly runs 64 bit
programs. Let us fix this things step by step.

- Stefani