Re: [PATCH v5 11/37] mm: Define VM_SHADOW_STACK for arm64 when we support GCS

[Edgecombe, Rick P]

On Tue, 2023-08-22 at 16:41 +0100, Mark Brown wrote:
> On Tue, Aug 22, 2023 at 05:21:09PM +0200, David Hildenbrand wrote:
> > On 22.08.23 15:56, Mark Brown wrote:
> 
> > > @@ -372,7 +372,17 @@ extern unsigned int kobjsize(const void
> > > *objp);
> > >     * having a PAGE_SIZE guard gap.
> > >     */
> > >    # define VM_SHADOW_STACK	VM_HIGH_ARCH_5
> > > -#else
> > > +#endif
> > > +
> > > +#if defined(CONFIG_ARM64_GCS)
> > > +/*
> > > + * arm64's Guarded Control Stack implements similar
> > > functionality and
> > > + * has similar constraints to shadow stacks.
> > > + */
> > > +# define VM_SHADOW_STACK	VM_HIGH_ARCH_5
> > > +#endif
> 
> > Shouldn't that all just merged with the previous define(s)?
> 
> > Also, I wonder if we now want to have CONFIG_HAVE_ARCH_SHADOW_STACK
> > or
> > similar.
> 
> I can certainly update it to do that, I was just trying to fit in
> with
> how the code was written on the basis that there was probably a good
> reason for it that had been discussed somewhere.  I can send an
> incremental patch for this on top of the x86 patches assuming they go
> in
> during the merge window.

There was something like that on the x86 series way back, but it was
dropped[0]. IIRC risc-v was going to try to do something other than
VM_SHADOW_STACK, so they may conflict some day. But in the meantime,
adding a CONFIG_HAVE_ARCH_SHADOW_STACK here in the arm series makes
sense to me.

[0]
https://lore.kernel.org/lkml/d09e952d8ae696f687f0787dfeb7be7699c02913.camel@xxxxxxxxx/