Re: [PATCH v2 03/11] KVM: arm64: Encapsulate reset request logic in a helper function

[Reiji Watanabe <reijiw@google.com>]

On Tue, Oct 5, 2021 at 12:02 PM Andrew Jones <drjones@redhat.com> wrote:
>
> On Tue, Oct 05, 2021 at 08:05:02AM -0700, Oliver Upton wrote:
> > Hi folks,
> >
> > On Tue, Oct 5, 2021 at 6:33 AM Andrew Jones <drjones@redhat.com> wrote:
> > >
> > > On Fri, Oct 01, 2021 at 09:10:14AM -0700, Oliver Upton wrote:
> > > > On Thu, Sep 30, 2021 at 11:05 PM Reiji Watanabe <reijiw@google.com> wrote:
> > > > >
> > > > > On Thu, Sep 23, 2021 at 12:16 PM Oliver Upton <oupton@google.com> wrote:
> > > > > >
> > > > > > In its implementation of the PSCI function, KVM needs to request that a
> > > > > > target vCPU resets before its next entry into the guest. Wrap the logic
> > > > > > for requesting a reset in a function for later use by other implemented
> > > > > > PSCI calls.
> > > > > >
> > > > > > No functional change intended.
> > > > > >
> > > > > > Signed-off-by: Oliver Upton <oupton@google.com>
> > > > > > ---
> > > > > >  arch/arm64/kvm/psci.c | 59 +++++++++++++++++++++++++------------------
> > > > > >  1 file changed, 35 insertions(+), 24 deletions(-)
> > > > > >
> > > > > > diff --git a/arch/arm64/kvm/psci.c b/arch/arm64/kvm/psci.c
> > > > > > index 310b9cb2b32b..bb59b692998b 100644
> > > > > > --- a/arch/arm64/kvm/psci.c
> > > > > > +++ b/arch/arm64/kvm/psci.c
> > > > > > @@ -64,9 +64,40 @@ static inline bool kvm_psci_valid_affinity(unsigned long affinity)
> > > > > >         return !(affinity & ~MPIDR_HWID_BITMASK);
> > > > > >  }
> > > > > >
> > > > > > -static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
> > > > > > +static void kvm_psci_vcpu_request_reset(struct kvm_vcpu *vcpu,
> > > > > > +                                       unsigned long entry_addr,
> > > > > > +                                       unsigned long context_id,
> > > > > > +                                       bool big_endian)
> > > > > >  {
> > > > > >         struct vcpu_reset_state *reset_state;
> > > > > > +
> > > > > > +       lockdep_assert_held(&vcpu->kvm->lock);
> > > > > > +
> > > > > > +       reset_state = &vcpu->arch.reset_state;
> > > > > > +       reset_state->pc = entry_addr;
> > > > > > +
> > > > > > +       /* Propagate caller endianness */
> > > > > > +       reset_state->be = big_endian;
> > > > > > +
> > > > > > +       /*
> > > > > > +        * NOTE: We always update r0 (or x0) because for PSCI v0.1
> > > > > > +        * the general purpose registers are undefined upon CPU_ON.
> > > > > > +        */
> > > > > > +       reset_state->r0 = context_id;
> > > > > > +
> > > > > > +       WRITE_ONCE(reset_state->reset, true);
> > > > > > +       kvm_make_request(KVM_REQ_VCPU_RESET, vcpu);
> > > > > > +
> > > > > > +       /*
> > > > > > +        * Make sure the reset request is observed if the change to
> > > > > > +        * power_state is observed.
> > > > > > +        */
> > > > > > +       smp_wmb();
> > > > > > +       vcpu->arch.power_off = false;
> > > > > > +}
> > > > > > +
> > > > > > +static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
> > > > > > +{
> > > > > >         struct kvm *kvm = source_vcpu->kvm;
> > > > > >         struct kvm_vcpu *vcpu = NULL;
> > > > > >         unsigned long cpu_id;
> > > > > > @@ -90,29 +121,9 @@ static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
> > > > > >                         return PSCI_RET_INVALID_PARAMS;
> > > > > >         }
> > > > > >
> > > > > > -       reset_state = &vcpu->arch.reset_state;
> > > > > > -
> > > > > > -       reset_state->pc = smccc_get_arg2(source_vcpu);
> > > > > > -
> > > > > > -       /* Propagate caller endianness */
> > > > > > -       reset_state->be = kvm_vcpu_is_be(source_vcpu);
> > > > > > -
> > > > > > -       /*
> > > > > > -        * NOTE: We always update r0 (or x0) because for PSCI v0.1
> > > > > > -        * the general purpose registers are undefined upon CPU_ON.
> > > > > > -        */
> > > > > > -       reset_state->r0 = smccc_get_arg3(source_vcpu);
> > > > > > -
> > > > > > -       WRITE_ONCE(reset_state->reset, true);
> > > > > > -       kvm_make_request(KVM_REQ_VCPU_RESET, vcpu);
> > > > > > -
> > > > > > -       /*
> > > > > > -        * Make sure the reset request is observed if the change to
> > > > > > -        * power_state is observed.
> > > > > > -        */
> > > > > > -       smp_wmb();
> > > > > > -
> > > > > > -       vcpu->arch.power_off = false;
> > > > > > +       kvm_psci_vcpu_request_reset(vcpu, smccc_get_arg2(source_vcpu),
> > > > > > +                                   smccc_get_arg3(source_vcpu),
> > > > > > +                                   kvm_vcpu_is_be(source_vcpu));
> > > > > >         kvm_vcpu_wake_up(vcpu);
> > > > > >
> > > > > >         return PSCI_RET_SUCCESS;
> > > > > > --
> > > > > > 2.33.0.685.g46640cef36-goog
> > > > >
> > > > > Reviewed-by: Reiji Watanabe <reijiw@google.com>
> > > > >
> > > > > Not directly related to the patch, but the (original) code doesn't
> > > > > do any sanity checking for the entry address although the PSCI spec says:
> > > > >
> > > > > "INVALID_ADDRESS is returned when the entry point address is known
> > > > > by the implementation to be invalid, because it is in a range that
> > > > > is known not to be available to the caller."
> > > >
> > > > Right, I had noticed the same but was a tad too lazy to address in
> > > > this series :) Thanks for the review, Reji!
> > > >
> > >
> > > KVM doesn't reserve any subrange within [0 - max_ipa), afaik. So all
> > > we need to do is check 'entry_addr < max_ipa', right?
> > >
> >
> > We could be a bit more pedantic and check if the IPA exists in a
> > memory slot, seems like kvm_vcpu_is_visible_gfn() should do the trick.
> >
> > Thoughts?
>
> Are we sure that all emulated devices, nvram, etc. will always use a
> memslot for regions that contain executable code? If there's any doubt,
> then we can't be sure about non-memslot regions within the max_ipa range.
> That'd be up to userspace.

I'm sorry for the late response.
IMHO, I would prefer Andrew's suggestion (check with the max_ipa).

It looks like instructions must always be in memslot for KVM/ARM looking
at the current implementation (especially kvm_handle_guest_abort()).
But, it doesn't necessarily mean the address is not invalid for the
guest (could be valid for load/store) and it might be changed in
the future.


Thanks,
Reiji
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm