s390/kvm: handle diagnose 318 instruction call
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 51 lines · Collin Walling
Quoted 13 lines · Collin Walling
Quoted 12 lines · David Hildenbrand
Quoted 11 lines · Collin Walling
Diagnose 318 is a privileged instruction that must be interpreted by SIE and handled via KVM.
The control program name and version codes (CPNC and CPVC) set by this instruction are saved to the
kvm->archstruct. The CPNC is also set in the SIE control block of all VCPUs. The newkvm_s390_set_miscinterface is introduced for migration.Signed-off-by Collin Walling <[email protected]>Reviewed-by Janosch Frank <[email protected]>---
[...]
Quoted 23 lines
Quoted 18 lines
Quoted 4 lines
(quoted diff) +3
-0+void kvm_s390_set_cpc(struct kvm *kvm, u64 cpc) +{ + struct kvm_vcpu *vcpu;Review
Quoted 10 lines
(quoted diff) +10
-0+ int i; + + mutex_lock(&kvm->lock); + kvm->arch.diag318_info.cpc = cpc; + + VM_EVENT(kvm, 3, "SET: cpnc: 0x%x cpvc: 0x%llx", + (u8)kvm->arch.diag318_info.cpnc, (u64)kvm->arch.diag318_info.cpvc); + + kvm_for_each_vcpu(i, vcpu, kvm) { + vcpu->arch.sie_block->cpnc = kvm->arch.diag318_info.cpnc;No, that does not look completely right. The other VCPUs could be running in the SIE. Can you update that via a sync request instead?
Quoted 4 lines
(after a VCPU updated the CPNC, other VCPUS could read for some time the old value, as values in this part of the
sie_blockmight be cached while a CPU is in the SIE)Quoted 6 lines
True. There are different name codes for other control programs (v/ZM, z/OS, etc). We should make sure that the name code is properly updated for all VCPUs if we are running a non-linux environment on top of KVM.
Thanks for the heads up!
I've been reading up on what the other requests (KVM_REQ_*) are used for, and none of them seem to make sense for what we want to do here (update a field in the vcpu). So let's see if I fully understand how this should be implemented:
Quoted 5 lines
1) create a new s390 specific request that will handle updating the cpc
2) call async_broadcastusing the new request withinset_cpcAlternatively, I think a vcpu block would work as well (you mentioned this in the RFC back at the end of August / beginning of September).
Yes it would as well. E.g. see kvm_arch_crypto_clear_masks()
mutex_lock(&kvm->lock);kvm_s390_vcpu_block_all(kvm);
/* do stuff to other VCPUs */
kvm_s390_vcpu_unblock_all(kvm);mutex_unlock(&kvm->lock);
Quoted 5 lines
Am I headed in the right direction?
[...]
Signature · 3 lines
Thanks, David / dhildenb