Re: [PATCH v10 37/38] compiler-gcc.h: Introduce __nostackp function attribute

[Tom Lendacky]

On 7/18/2017 4:36 AM, Ingo Molnar wrote:
> * Tom Lendacky <thomas.lendacky@xxxxxxx> wrote:
>
> > Create a new function attribute, __nostackp, that can used to turn off
> > stack protection on a per function basis.
> >
> > Signed-off-by: Tom Lendacky <thomas.lendacky@xxxxxxx>
> > ---
> >   include/linux/compiler-gcc.h | 2 ++
> >   include/linux/compiler.h     | 4 ++++
> >   2 files changed, 6 insertions(+)
> >
> > diff --git a/include/linux/compiler-gcc.h b/include/linux/compiler-gcc.h
> > index cd4bbe8..682063b 100644
> > --- a/include/linux/compiler-gcc.h
> > +++ b/include/linux/compiler-gcc.h
> > @@ -166,6 +166,8 @@
> >   
> >   #if GCC_VERSION >= 40100
> >   # define __compiletime_object_size(obj) __builtin_object_size(obj, 0)
> > +
> > +#define __nostackp	__attribute__((__optimize__("no-stack-protector")))
> >   #endif
> >   
> >   #if GCC_VERSION >= 40300
> > diff --git a/include/linux/compiler.h b/include/linux/compiler.h
> > index 219f82f..63cbca1 100644
> > --- a/include/linux/compiler.h
> > +++ b/include/linux/compiler.h
> > @@ -470,6 +470,10 @@ static __always_inline void __write_once_size(volatile void *p, void *res, int s
> >   #define __visible
> >   #endif
> >   
> > +#ifndef __nostackp
> > +#define __nostackp
> > +#endif
>
> So I changed this from the hard to read and ambiguous "__nostackp" abbreviation
> (does it mean 'no stack pointer?') to "__nostackprotector", plus added this detail
> to the changelog:
>
> | ( This is needed by the SME in-place kernel memory encryption feature,
> |   which activates encryption in its sme_enable() function and thus changes the
> |   visible value of the stack protection cookie on function return. )
>
> Agreed?

Hi Ingo,

I debugged this to needing "__nostackprotector" because sme_enable()
is called very early in the boot process before everything is properly
setup to fully support stack protection when KASLR is enabled. Without
this attribute the call to sme_enable() would fail even if encryption
was disabled with the "mem_encrypt=off" command line option.

If KASLR wasn't enabled, then everything worked fine without the
"__nostackprotector" attribute, encryption enabled or not.

The stack protection support is activated because of the 16-byte
character buffer in the sme_enable() routine.  I think we'll find that
if a character buffer greater than 8 bytes is added to, for example,
__startup_64, then this attribute will need to be added to that routine.

Thanks,
Tom

> Thanks,
>
> 	Ingo