On Fri, Oct 04, 2019 at 10:13:40AM +0100, Steven Price wrote: > On 04/10/2019 08:03, Andrew Jones wrote: > > On Thu, Oct 03, 2019 at 03:22:35PM +0200, Andrew Jones wrote: > >> On Wed, Oct 02, 2019 at 03:50:32PM +0100, Steven Price wrote: > >>> +int kvm_update_stolen_time(struct kvm_vcpu *vcpu, bool init) > >>> +{ > >>> + struct kvm *kvm = vcpu->kvm; > >>> + u64 steal; > >>> + u64 steal_le; > >>> + u64 offset; > >>> + int idx; > >>> + u64 base = vcpu->arch.steal.base; > >>> + > >>> + if (base == GPA_INVALID) > >>> + return -ENOTSUPP; > >>> + > >>> + /* Let's do the local bookkeeping */ > >>> + steal = vcpu->arch.steal.steal; > >>> + steal += current->sched_info.run_delay - vcpu->arch.steal.last_steal; > >>> + vcpu->arch.steal.last_steal = current->sched_info.run_delay; > >>> + vcpu->arch.steal.steal = steal; > >>> + > >>> + steal_le = cpu_to_le64(steal); > >> > >> Agreeing on a byte order for this interface makes sense, but I don't see > >> it documented anywhere. Is this an SMCCC thing? Because I skimmed some > >> of those specs and other users too but didn't see anything obvious. Anyway > >> even if everybody but me knows that all data returned from SMCCC calls > >> should be LE, it might be nice to document that in the pvtime doc. > > A very good point - I'll document this in the Linux document and feed > that back for DEN0057A. > > > > > I have another [potentially dumb] SMCCC byte order question. If we need > > to worry about using LE for the members of this structure, then why don't > > we need to worry about the actual return values of the SMCCC calls? Like > > the IPA of the structure? > > The SMCCC calls pass values in registers. It's only when reading/writing > these values from/to memory that the endianness actually has any meaning. > Ah yes, of course. Thanks, drew