On Thu, 23 Sep 2021 14:02:11 +0100,
Will Deacon <will@xxxxxxxxxx> wrote:
>
> On Thu, Sep 23, 2021 at 01:56:21PM +0100, Marc Zyngier wrote:
> > On Thu, 23 Sep 2021 12:22:56 +0100,
> > Will Deacon <will@xxxxxxxxxx> wrote:
[...]
> > > static void handle_host_hcall(struct kvm_cpu_context *host_ctxt)
> > > {
> > > DECLARE_REG(unsigned long, id, host_ctxt, 0);
> > > + unsigned long hcall_min = 0;
> > > hcall_t hfn;
> > >
> > > + if (static_branch_unlikely(&kvm_protected_mode_initialized))
> > > + hcall_min = __KVM_HOST_SMCCC_FUNC___pkvm_prot_finalize;
> > > +
> > > id -= KVM_HOST_SMCCC_ID(0);
> > >
> > > - if (unlikely(id >= ARRAY_SIZE(host_hcall)))
> > > + if (unlikely(id < hcall_min || id >= ARRAY_SIZE(host_hcall)))
> >
> > So I can still issue a pkvm_prot_finalize after finalisation? Seems
> > odd. As hcall_min has to be inclusive, you probably want it to be set
> > to __KVM_HOST_SMCCC_FUNC___pkvm_host_share_hyp once protected.
>
> Yeah, I ended up addresing that one in the previous patch. The problem is
> that we need to allow pkvm_prot_finalize to be called on each CPU, so I
> think we'd end up having an extra "really finalize damnit!" call to be
> issued _once_ after each CPU is done with the finalisation if we want
> to lock it down.
>
> The approach I took instead is to make pkvm_prot_finalize return -EBUSY
> if it's called on a CPU where it's already been called.
Ah, I see. Serves me right for reading patches out of order. Finalise
is of course per-CPU, and the static key global. Epic fail.
Probably deserves a comment, because I'm surely going to jump at that
again in three months.
Thanks,
M.
--
Without deviation from the norm, progress is not possible.
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
