Re: [PATCH 3/3] xen/arm: stack check instrumentation

[Julien Grall <julien@xxxxxxx>]

Hi,

On 29/07/2024 20:40, Stewart Hildebrand wrote:
> On 7/29/24 14:50, Julien Grall wrote:
> > Hi again,
> >
> > On 29/07/2024 15:24, Stewart Hildebrand wrote:
> > > diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
> > > index aac6c599f878..b4890eec7ec4 100644
> > > --- a/xen/arch/arm/traps.c
> > > +++ b/xen/arch/arm/traps.c
> > > @@ -2325,6 +2325,51 @@ void asmlinkage leave_hypervisor_to_guest(void)
> > >            arm_smccc_1_1_smc(ARM_SMCCC_ARCH_WORKAROUND_2_FID, 0, NULL);
> > >    }
> > >    +#ifdef CONFIG_DEBUG
> > > +DEFINE_PER_CPU(unsigned int, stack_check_nesting);
> > > +DEFINE_PER_CPU(unsigned char *, stack_base);
> > While looking at the code, I just realized that this should be equivalent to 
> > current->arch.base.
> current->arch.stack
>
> That's true only after the idle vcpu stacks have been allocated.
>
> > So do we actually need stack_base?
> This is a way to enable stack instrumentation much earlier during boot
> when we are still using the static boot stack.
Sure. But we are only partially checking the static boot stack. I am not 
entirely sure if there is any value for that because at that point the 
stack would be barely used.
Anyway, this is only for debug build so far, so I am ok to keep it.

Cheers,

--
Julien Grall