On 01.02.2019 14:37, Pierre Morel wrote: > On 01/02/2019 11:56, David Hildenbrand wrote: >> On 01.02.19 10:52, Pierre Morel wrote: >>> The case when the SIE for guest3 is not setup for using >>> encryption keys nor Adjunct processor but the guest2 >>> does use these features was not properly handled. >>> >>> This leads SIE entry for guest3 to crash with validity intercept >>> because the guest2, not having the use of encryption keys nor >>> Adjunct Processor did not initialize the CRYCB designation. >>> >>> In the case where none of ECA_APIE, ECB3_AES or ECB3_DEA >>> are set in guest3 a format 0 CRYCB is allowed for guest3 >>> and the CRYCB designation in the SIE for guest3 is not checked >>> on SIE entry. >>> >>> Let's allow the CRYCD designation to be ignored when the >>> SIE for guest3 is not initialized for encryption key usage >>> nor AP. >>> >>> Fixup: d6f6959 (KVM: s390: vsie: Do the CRYCB validation first) >>> >>> Signed-off-by: Pierre Morel <pmorel@xxxxxxxxxxxxx> >>> Reported-by: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx> >>> --- >>> arch/s390/kvm/vsie.c | 3 +++ >>> 1 file changed, 3 insertions(+) >>> >>> diff --git a/arch/s390/kvm/vsie.c b/arch/s390/kvm/vsie.c >>> index a153257..a748f76 100644 >>> --- a/arch/s390/kvm/vsie.c >>> +++ b/arch/s390/kvm/vsie.c >>> @@ -300,6 +300,9 @@ static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) >>> if (!apie_h && !key_msk) >>> return 0; >>> + if (!(scb_o->eca & ECA_APIE) && !(scb_o->ecb3 & (ECB3_AES | ECB3_DEA))) >>> + return 0; >>> + >>> if (!crycb_addr) >>> return set_validity_icpt(scb_s, 0x0039U); >>> >> >> The original patch said >> >> "We need to handle the validity checks for the crycb, no matter what the >> settings for the keywrappings are. So lets move the keywrapping checks >> after we have done the validy checks." >> >> Can you explain why keywrapping now is important? These patches seem to >> contradict. >> > > No it does not, having the flags set or not is part of the validity check. > but, I acted too fast. > > The problem seems to be even clearer: > key_msk is defined as > int key_msk = test_kvm_facility(vcpu->kvm, 76); > > If it is defined, as it should for a mask, as > (scb_o->ecb3 & (ECB3_AES | ECB3_DEA)) > > all is clear..., key_msk is not use but for this test, so I do not understand why it is set as facility 76. > > so I think I better do: > > > diff --git a/arch/s390/kvm/vsie.c b/arch/s390/kvm/vsie.c > index a153257..30843a8 100644 > --- a/arch/s390/kvm/vsie.c > +++ b/arch/s390/kvm/vsie.c > @@ -289,7 +289,7 @@ static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) > unsigned long *b1, *b2; > u8 ecb3_flags; > int apie_h; > - int key_msk = test_kvm_facility(vcpu->kvm, 76); > + int key_msk = scb_o->ecb3 & (ECB3_AES | ECB3_DEA); > int fmt_o = crycbd_o & CRYCB_FORMAT_MASK; > int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK; > int ret = 0; > > > So just define a mask a mask. > I verify the functionality and test on Monday and if in between it seems better to you so too I post the patch. Can you spin a v2 with an improved patch description outlining the problem according to this discussion?