KVM: s390: vsie: fix Do the CRYCB validation first
PATCH 1/1 KVM: s390: vsie: fix Do the CRYCB validation first
This is a BUGFIX
Surely must be CC stable.
Pierre Morel (1):
KVM: s390: vsie: fix Do the CRYCB validation first
arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)
Signature · 1 line
2.7.4
PATCH 1/1 KVM: s390: vsie: fix Do the CRYCB validation first
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)
--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)
arch/s390/kvm/vsie.c +3 -0
@@ -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 );
Signature · 1 line
2.7.4
Quoted 8 lines · On Fri, 1 Feb 2019 10:52:05 +0100 Pierre Morel <[email protected]> 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.
Quoted 6 lines
In the case where none of
ECA_APIE,ECB3_AESorECB3_DEAare 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
s/CRYCD/CRYCB/
Quoted 4 lines
SIE for guest3 is not initialized for encryption key usage nor AP.
Fixup: d6f6959 (KVM: s390: vsie: Do the CRYCB validation first)
I think the canonical format is
Quoted 1 line
Quoted 11 lines
Signed-off-by Pierre Morel <[email protected]>Reported-by Claudio Imbrenda <[email protected]>--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +0
-0
arch/s390/kvm/vsie.c static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) · line 308
Quoted 9 lines
arch/s390/kvm/vsie.c +3
-0@@ -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 );
Looks sane to me, but I'll let vsie experts comment.
Quoted 47 lines · Cornelia Huck
Quoted 8 lines · On Fri, 1 Feb 2019 10:52:05 +0100 Pierre Morel <[email protected]> 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.
Quoted 6 lines
In the case where none of
ECA_APIE,ECB3_AESorECB3_DEAare 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
s/CRYCD/CRYCB/
Quoted 4 lines
SIE for guest3 is not initialized for encryption key usage nor AP.
Fixup: d6f6959 (KVM: s390: vsie: Do the CRYCB validation first)
I think the canonical format is
Fixes d6f6959ac587 ("KVM: s390: vsie: Do the CRYCB validation first")Quoted 1 line
CC [email protected] # 4.20+Quoted 14 lines
Signed-off-by Pierre Morel <[email protected]>Reported-by Claudio Imbrenda <[email protected]>--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +0
-0@@ -300,6 +300,9 @@ static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)if (!apie_h && !key_msk)return 0 ;
Quoted 9 lines
arch/s390/kvm/vsie.c static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) · line 308
Quoted 6 lines
arch/s390/kvm/vsie.c +3
-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 );Looks sane to me, but I'll let vsie experts comment.
Thanks
Pierre
Signature · 2 lines
Pierre Morel Linux/KVM/QEMU in Böblingen - Germany
Quoted 31 lines · Pierre Morel
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_AESorECB3_DEAare 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 <[email protected]>Reported-by Claudio Imbrenda <[email protected]>--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +0
-0
arch/s390/kvm/vsie.c static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) · line 308
Quoted 9 lines
arch/s390/kvm/vsie.c +3
-0@@ -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.
Signature · 3 lines
Thanks, David / dhildenb
Quoted 44 lines · David Hildenbrand
Quoted 31 lines · Pierre Morel
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_AESorECB3_DEAare 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 <[email protected]>Reported-by Claudio Imbrenda <[email protected]>--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +0
-0arch/s390/kvm/vsie.c static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) · line 308
Quoted 9 lines
arch/s390/kvm/vsie.c +3
-0@@ -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
Quoted 7 lines
"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:
arch/s390/kvm/vsie.c +1 -1
@@ -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.
Thanks,
Pierre
Signature · 2 lines
Pierre Morel Linux/KVM/QEMU in Böblingen - Germany
Quoted 81 lines · Pierre Morel
Quoted 42 lines · David Hildenbrand
Quoted 37 lines · Pierre Morel
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_AESorECB3_DEAare 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 <[email protected]>Reported-by Claudio Imbrenda <[email protected]>---
arch/s390/kvm/vsie.c| 3 +++1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +2
-0@@ -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);Quoted 1 line
The original patch said
Quoted 7 lines
"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_mskis defined as
intkey_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_mskis not use but for this test, so I do not understand why it is set as facility 76.so I think I better do:
arch/s390/kvm/vsie.c +1
-1@@ -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;
Quoted 2 lines
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?
Quoted 84 lines · Christian Borntraeger
Quoted 81 lines · Pierre Morel
Quoted 42 lines · David Hildenbrand
Quoted 30 lines · Pierre Morel
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_AESorECB3_DEAare 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 <[email protected]>Reported-by Claudio Imbrenda <[email protected]>--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +0
-0arch/s390/kvm/vsie.c static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) · line 307
Quoted 8 lines
arch/s390/kvm/vsie.c +2
-0@@ -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
Quoted 7 lines
"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_mskis defined as
intkey_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_mskis not use but for this test, so I do not understand why it is set as facility 76.so I think I better do:
arch/s390/kvm/vsie.c +1
-1@@ -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;
Quoted 6 lines
Quoted 2 lines
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?
Yes I will do.
Seems I need more analyzing of the problem
Regards,
Pierre
Signature · 2 lines
Pierre Morel Linux/KVM/QEMU in Böblingen - Germany
Quoted 85 lines · Pierre Morel
Quoted 44 lines · David Hildenbrand
Quoted 31 lines · Pierre Morel
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_AESorECB3_DEAare 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 <[email protected]>Reported-by Claudio Imbrenda <[email protected]>--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +0
-0arch/s390/kvm/vsie.c static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) · line 308
Quoted 9 lines
arch/s390/kvm/vsie.c +3
-0@@ -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
Quoted 7 lines
"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_mskis defined asint 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_mskis not use but for this test, so I do not understand why it is set as facility 76.so I think I better do:
arch/s390/kvm/vsie.c +1
-1@@ -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;
Quoted 3 lines
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.
Without documentation at hand I cannot really judge what is the right approach. You have to read the architecture description very carefully and thimk about the different scenarios. I cannot help here.
test_kvm_facility(vcpu->kvm, 76) means "hardware knows key wrapping exists"
scb_o->ecb3 & (ECB3_AES | ECB3_DEA) means "key wrapping was actually enabled for AES or DEA"
Quoted 4 lines
Thanks,
Pierre
Signature · 3 lines
Thanks, David / dhildenb
Quoted 94 lines · David Hildenbrand
Quoted 85 lines · Pierre Morel
Quoted 44 lines · David Hildenbrand
Quoted 31 lines · Pierre Morel
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_AESorECB3_DEAare 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 <[email protected]>Reported-by Claudio Imbrenda <[email protected]>--- arch/s390/kvm/vsie.c | 3 +++ 1 file changed, 3 insertions(+)arch/s390/kvm/vsie.c +0
-0arch/s390/kvm/vsie.c static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) · line 308
Quoted 9 lines
arch/s390/kvm/vsie.c +3
-0@@ -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
Quoted 7 lines
"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_mskis defined asint 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_mskis not use but for this test, so I do not understand why it is set as facility 76.so I think I better do:
arch/s390/kvm/vsie.c +1
-1@@ -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;Quoted 3 lines
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.Without documentation at hand I cannot really judge what is the right approach. You have to read the architecture description very carefully and thimk about the different scenarios. I cannot help here.
Quoted 4 lines
test_kvm_facility(vcpu->kvm, 76)means "hardware knows key wrapping exists"
scb_o->ecb3& (ECB3_AES|ECB3_DEA) means "key wrapping was actually enabled for AES or DEA"
Right.
We also must take care of the format. So not sure my propositions are right. And helas I have the documentation :) So ... I have to read it again again.
Regards,
Pierre
Quoted 7 lines
Quoted 5 lines
Thanks,
Pierre
Signature · 2 lines
Pierre Morel Linux/KVM/QEMU in Böblingen - Germany