Re: [PATCH v2 3/5] KVM: s390: vsie: Allow support for a host without AP

From: Pierre Morel
Date: Thu Aug 23 2018 - 03:54:18 EST

On 23/08/2018 09:15, David Hildenbrand wrote:
On 23.08.2018 08:44, Pierre Morel wrote:
On 22/08/2018 19:06, David Hildenbrand wrote:
On 22.08.2018 18:51, Pierre Morel wrote:
Currently the CRYCB format used in the host for the
shadowed CRYCB is FORMAT2 while no check is done if
AP instructions are supported in the host.

We better use the format the host calculated for the
guest 1 as the host already tested it against its
facility set.

Signed-off-by: Pierre Morel <pmorel@xxxxxxxxxxxxx>
arch/s390/kvm/vsie.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/arch/s390/kvm/vsie.c b/arch/s390/kvm/vsie.c
index 56a9d47..0b12916 100644
--- a/arch/s390/kvm/vsie.c
+++ b/arch/s390/kvm/vsie.c
@@ -154,6 +154,7 @@ static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
unsigned long *b1, *b2;
u8 ecb3_flags;
+ unsigned long g1_fmt;
scb_s->crycbd = 0;
if (!(crycbd_o == CRYCB_FORMAT1))
@@ -180,8 +181,8 @@ static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
return set_validity_icpt(scb_s, 0x0035U);
scb_s->ecb3 |= ecb3_flags;
- scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 |
+ g1_fmt = vcpu->arch.sie_block->crycbd & 0x03;
+ scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | g1_fmt;
/* xor both blocks in one run */
b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;

This is wrong. I remember that with APXA, if FORMAT2 is available, we
should always use FORMAT2. That's why we explicitly convert it here.

You are right if FORMAT2 is available we should use FORMAT2
but the intention here is to use what KVM crypto init function did,
assuming it did the right thing.

Eventually we are running on a host without AP and we should use FORMAT1.

Isn't it correct?

Yes and no :)

No APXA -> FORMAT2 bit is ignored (and that is one of the reasons why I
am being so strict about simulating HW behavior correctly in nested code
:) )

This only holds as long as we are not using AP. Because from a MSA3
perspective, FORMAT1==FORMAT2 (apart from the length/alignment, which is
fine for us).

Once we support AP (via ECA.28), we'll properly have to create either a
Format0/Format1/Format2. Then, there is actually a semantically
difference ("different fields used").

I would have expect something more explicit in the documentation
like for firmware versions older than xxx bit 30 is ignored
instead of if APXA is not installed.
Still must learn the IBM language! :)


Pierre Morel
Linux/KVM/QEMU in BÃblingen - Germany