Re: [patch v3 16/16] btrace, x86: restrict to Atom

[Mark Kettenis <mark.kettenis@xs4all.nl>]

> From: "Metzger, Markus T" <markus.t.metzger@intel.com>
> Date: Mon, 24 Sep 2012 12:22:42 +0000
> 
> Ping.

Native code can only go into -nat.c files.

But I still think putting CPU model specific checks anywhere in GDB is
wrong.  If there are broken kernels out there, figure out a way to
detect them.

> > -----Original Message-----
> > From: Metzger, Markus T
> > Sent: Tuesday, August 14, 2012 3:42 PM
> > To: Mark Kettenis
> > Cc: gdb-patches@sourceware.org; markus.t.metzger@gmail.com; jan.kratochvil@redhat.com; palves@redhat.com;
> > tromey@redhat.com
> > Subject: RE: [patch v3 16/16] btrace, x86: restrict to Atom
> >
> > > -----Original Message-----
> > > From: Mark Kettenis [mailto:mark.kettenis@xs4all.nl]
> > > Sent: Tuesday, August 14, 2012 3:27 PM
> >
> > Thanks for your feedback!
> >
> >
> > > > +int
> > > > +i386_linux_supports_btrace (void)
> > > > +{
> > > > +  unsigned int cpuid, model, family;
> > > > +
> > > > +  if (!linux_supports_btrace ())
> > > > +    return 0;
> > > > +
> > > > +  __asm__ __volatile__ ("movl   $1, %%eax;"
> > > > +                 "cpuid;"
> > > > +                 : "=a" (cpuid)
> > > > +                 :: "%ebx", "%ecx", "%edx");
> > > > +
> >
> > [...]
> >
> > > You can't have this function in a -tdep.c, since this file needs to be
> > > compilable everywhere (not just on i386/amd64).  But even in a -nat.c
> > > file you shouldn't use inline assembly, since it isn't a standardized
> > > C feature.  I can only guess why you want to restrict this feature to
> > > Atom only, but I suspect that the proper solution is to query the
> > > target whether branch tracing is supported.  That means the -nat.c
> > > code should have code to query for support in the Linux kernel.
> >
> > I agree that it would be better to have this check done in the kernel. The problem is that such a patch will take very long before
> > it becomes available in distributions. And even then, we would need to handle kernels without the patch, so we would need
> > something
> > similar to this in gdb, as well.
> >
> > I took the cpuid bits from gdb/go32-nat.c. I have not seen any compilation guard around it.
> >
> > I thought that the i386-tdep files are meant to contain i386 specific code, that's why I put the code there. Where would be the
> > right place to put such code?
> >
> > Thanks,
> > Markus.
> Intel GmbH
> Dornacher Strasse 1
> 85622 Feldkirchen/Muenchen, Deutschland
> Sitz der Gesellschaft: Feldkirchen bei Muenchen
> Geschaeftsfuehrer: Peter Gleissner, Christian Lamprechter, Hannes Schwaderer, Douglas Lusk
> Registergericht: Muenchen HRB 47456
> Ust.-IdNr./VAT Registration No.: DE129385895
> Citibank Frankfurt a.M. (BLZ 502 109 00) 600119052
> 
> 
>