[PATCH v3 2/5] KVM: arm64: Compile stacktrace.nvhe.o

[Kalesh Singh]

On Wed, Jun 8, 2022 at 12:33 AM Marc Zyngier <maz at kernel.org> wrote:
>
> On Tue, 07 Jun 2022 17:50:44 +0100,
> Kalesh Singh <kaleshsingh at google.com> wrote:
> >
> > Recompile stack unwinding code for use with the nVHE hypervisor. This is
> > a preparatory patch that will allow reusing most of the kernel unwinding
> > logic in the nVHE hypervisor.
> >
> > Suggested-by: Mark Rutland <mark.rutland at arm.com>
> > Signed-off-by: Kalesh Singh <kaleshsingh at google.com>
> > Reviewed-by: Mark Brown <broonie at kernel.org>
> > ---
> >
> > Changes in v3:
> >   - Add Mark's Reviewed-by tag
> >
> > Changes in v2:
> >   - Split out refactoring of common unwinding logic into a separate patch,
> >     per Mark Brown
> >
> >  arch/arm64/include/asm/stacktrace.h | 18 +++++++++-----
> >  arch/arm64/kernel/stacktrace.c      | 37 ++++++++++++++++-------------
> >  arch/arm64/kvm/hyp/nvhe/Makefile    |  3 ++-
> >  3 files changed, 35 insertions(+), 23 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/stacktrace.h b/arch/arm64/include/asm/stacktrace.h
> > index aec9315bf156..f5af9a94c5a6 100644
> > --- a/arch/arm64/include/asm/stacktrace.h
> > +++ b/arch/arm64/include/asm/stacktrace.h
> > @@ -16,12 +16,14 @@
> >  #include <asm/sdei.h>
> >
> >  enum stack_type {
> > -     STACK_TYPE_UNKNOWN,
> > +#ifndef __KVM_NVHE_HYPERVISOR__
> >       STACK_TYPE_TASK,
> >       STACK_TYPE_IRQ,
> >       STACK_TYPE_OVERFLOW,
> >       STACK_TYPE_SDEI_NORMAL,
> >       STACK_TYPE_SDEI_CRITICAL,
> > +#endif /* !__KVM_NVHE_HYPERVISOR__ */
> > +     STACK_TYPE_UNKNOWN,
>
> What is the reason for this reordering? I have the sinking feeling
> that this could play badly with the logic that assumes that it is
> legal to switch from a lesser stack type to a higher one, and could
> allow switching to a duff stack.

HI Marc. Thanks for reviewing.

I only reordered the enum to group the common types. But I don't have
a strong opinion on it. The unwinding doesn't depend on the ordering
in this enum. When we transition form stack 'A'-->'B', we set the
stack_done bit for stack A so that we never transition back to 'A', as
it's not valid to transition back to a previous stack. But the order
of the sequence itself is not something enforced.

>
> I would at least like to see a justification of why this isn't less
> safe than the current code.
>
> [...]
>
> > index f9fe4dc21b1f..c0ff0d6fc403 100644
> > --- a/arch/arm64/kvm/hyp/nvhe/Makefile
> > +++ b/arch/arm64/kvm/hyp/nvhe/Makefile
> > @@ -14,7 +14,8 @@ lib-objs := $(addprefix ../../../lib/, $(lib-objs))
> >
> >  obj-y := timer-sr.o sysreg-sr.o debug-sr.o switch.o tlb.o hyp-init.o host.o \
> >        hyp-main.o hyp-smp.o psci-relay.o early_alloc.o page_alloc.o \
> > -      cache.o setup.o mm.o mem_protect.o sys_regs.o pkvm.o
> > +      cache.o setup.o mm.o mem_protect.o sys_regs.o pkvm.o \
> > +      ../../../kernel/stacktrace.o
>
> This, I positively hate. It is only a marginally better than the
> cross-arch references we used to have with arch/arm/kvm. I'd be much
> more happy with an include file containing the shared code. It would
> also allow the removal of some of the #ifdeferry. Note that this is
> the approach that we ended up adopting for the VHE/nVHE split.
>

Also thought about moving stuff to some header file, but I thought
this might be less intrusive. Let me prototype to see how they
compare.

Thanks,
Kalesh

> Thanks,
>
>         M.
>
> --
> Without deviation from the norm, progress is not possible.
>
> --
> To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe at android.com.
>