On 06.02.20 10:07, Christian Borntraeger wrote: > On 05.02.20 18:00, Thomas Huth wrote: > >>>> >>>> Uh, why the mix of a new ioctl with the existing mem_op stuff? Could you >>>> please either properly integrate this into the MEM_OP ioctl (and e.g. >>>> use gaddr as offset for the new SIDA_READ and SIDA_WRITE subcodes), or >>>> completely separate it for a new ioctl, i.e. introduce a new struct for >>>> the new ioctl instead of recycling the struct kvm_s390_mem_op here? >>>> (and in case you ask me, I'd slightly prefer to integrate everything >>>> into MEM_OP instead of introducing a new ioctl here). >>> >>> *cough* David and Christian didn't like the memop solution and it took >>> me a long time to get this to work properly in QEMU... >> >> I also don't like to re-use MEMOP_LOGICAL_READ and MEMOP_LOGICAL_WRITE >> for the SIDA like you've had it in RFC v1 ... but what's wrong with >> using KVM_S390_MEMOP_SIDA_READ and KVM_S390_MEMOP_SIDA_WRITE with the >> MEM_OP ioctl directly? >> >> Thomas >> > > In essence something like the following? > > @@ -4583,6 +4618,9 @@ static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu, > } > r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size); > break; > + case KVM_S390_MEMOP_SIDA_READ: > + case KVM_S390_MEMOP_SIDA_WRITE: > + kvm_s390_guest_sida_op(vcpu, mop); > default: FWIW, I'd prefer that. It's a special type of memory access. -- Thanks, David / dhildenb
