Re: [PATCH] KVM: VMX: Correct asm constraint in vmcs_load()

[Avi Kivity <avi@redhat.com>]

On 12/20/2010 08:22 PM, Randy Dunlap wrote:
> On 12/20/10 09:58, Avi Kivity wrote:
> >  On 12/20/2010 07:57 PM, Avi Kivity wrote:
> >>  On 12/20/2010 07:53 PM, Randy Dunlap wrote:
> >>>  On 12/20/10 09:47, Avi Kivity wrote:
> >>>  >   On 12/20/2010 06:56 PM, Randy Dunlap wrote:
> >>>  >>   On 12/20/10 01:06, Avi Kivity wrote:
> >>>  >>  >    'error' is byte sized, so use a byte register constraint.
> >>>  >>
> >>>  >>   Hm, OK, but I still get the same build error.
> >>>  >
> >>>  >   Not okay then.  Can you verify that by removing the content of the
> >>>  >   function the problem goes away?
> >>>
> >>>  Nope.  I put #if 0 / #endif around the asm() and if (error) lines
> >>>  but I still get the error message:
> >>>
> >>>  linux-next-20101217/arch/x86/kvm/vmx.c: Assembler messages:
> >>>  linux-next-20101217/arch/x86/kvm/vmx.c:488: Error: bad register name
> >>>  `%sil'
> >>>
> >>
> >>  And line 488 was one of those under #if 0?
> >>
> >>>  which is on the "u8 error;" line below:
> >>>
> >>>  static void vmcs_load(struct vmcs *vmcs)
> >>>  {
> >>>      u64 phys_addr = __pa(vmcs);
> >>>      u8 error;
> >>>
> >
> >  Er, you just answered me.  Are you sure gcc saw your update?  what
> >  happens if you comment that line as well?  assign to 'error' instead?
>
> >  pwd
> /lnx/src/NEXT/linux-next-20101217
>
> >  quilt ser -v
> = kvm-asm-constraint12.patch
>
> >  make ARCH=i386 O=X32 all 2>&1 | tee kvm7.out
> <kconfig messages deleted>
>    Using /lnx/src/NEXT/linux-next-20101217 as source for kernel
>    GEN     /lnx/src/NEXT/linux-next-20101217/X32/Makefile
>    CHK     include/linux/version.h
>    CHK     include/generated/utsrelease.h
>    CALL    /lnx/src/NEXT/linux-next-20101217/scripts/checksyscalls.sh
>    CHK     include/generated/compile.h
>    CC      arch/x86/kvm/vmx.o
> /lnx/src/NEXT/linux-next-20101217/arch/x86/kvm/vmx.c: Assembler messages:
> /lnx/src/NEXT/linux-next-20101217/arch/x86/kvm/vmx.c:488: Error: bad register name `%sil'
> make[3]: *** [arch/x86/kvm/vmx.o] Error 1
> make[2]: *** [arch/x86/kvm] Error 2
> make[1]: *** [arch/x86] Error 2
> make: *** [sub-make] Error 2
>
>
> This is with the entire function being empty (no code, no data).
>
> >  Wierd.
>
> Definitely.  I can't explain it.
>

Maybe it's the vmcs_clear() above.  Try the same treatment?

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.