Re: [PATCH v2 02/15] s390: vsie: implement AP support for second level guest

From: Pierre Morel
Date: Wed Feb 28 2018 - 05:04:07 EST


On 28/02/2018 10:39, David Hildenbrand wrote:
On 27.02.2018 15:28, Tony Krowiak wrote:
Set effective masks and CRYCB format in the shadow copy of the
guest level 2 CRYCB.

Signed-off-by: Tony Krowiak <akrowiak@xxxxxxxxxxxxxxxxxx>
---
arch/s390/include/asm/kvm-ap.h | 2 +
arch/s390/kvm/kvm-ap.c | 5 +++
arch/s390/kvm/vsie.c | 71 +++++++++++++++++++++++++++++++++-------
3 files changed, 66 insertions(+), 12 deletions(-)

diff --git a/arch/s390/include/asm/kvm-ap.h b/arch/s390/include/asm/kvm-ap.h
index 4e43117..ef749e7 100644
--- a/arch/s390/include/asm/kvm-ap.h
+++ b/arch/s390/include/asm/kvm-ap.h
@@ -13,4 +13,6 @@
void kvm_ap_set_crycb_format(struct kvm *kvm, __u32 *crycbd);
+int kvm_ap_get_crycb_format(struct kvm *kvm);
+
#endif /* _ASM_KVM_AP */
diff --git a/arch/s390/kvm/kvm-ap.c b/arch/s390/kvm/kvm-ap.c
index 5305f4c..bafe63b 100644
--- a/arch/s390/kvm/kvm-ap.c
+++ b/arch/s390/kvm/kvm-ap.c
@@ -11,6 +11,11 @@
#include "kvm-s390.h"
+int kvm_ap_get_crycb_format(struct kvm *kvm)
+{
+ return kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK;
+}
Why exactly do we need this function? If there are no other users, just
do it directly in the code below.

+
static int kvm_ap_apxa_installed(void)
{
int ret;
diff --git a/arch/s390/kvm/vsie.c b/arch/s390/kvm/vsie.c
index 8961e39..93076ba 100644
--- a/arch/s390/kvm/vsie.c
+++ b/arch/s390/kvm/vsie.c
@@ -18,6 +18,7 @@
#include <asm/sclp.h>
#include <asm/nmi.h>
#include <asm/dis.h>
+#include <asm/kvm-ap.h>
#include "kvm-s390.h"
#include "gaccess.h"
@@ -137,12 +138,56 @@ static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
}
/*
+ * Set up the effective masks for the shadow copy of the crycb. The effective
+ * masks for guest 3 are set by performing a logical bitwise AND of the guest 3
+ * masks with the guest 2 masks.
+ */
+static void set_crycb_emasks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
+{
+ int fmt = kvm_ap_get_crycb_format(vcpu->kvm);
+ unsigned long *mask1, *mask2;
+
+ switch (fmt) {
+ case CRYCB_FORMAT1:
+ case CRYCB_FORMAT2:
Assume L2 gave us FORMAT0 and we are using FORMAT2. You would copy wrong
bits. See below, maybe we need conversion.
+1
and in the case we have FORTMAT1 in L2 and/or L3, you also copy them from and/or to the wrong place.


+ mask1 = (unsigned long *)vsie_page->crycb.apcb1.apm;
+ mask2 = (unsigned long *)
+ vcpu->kvm->arch.crypto.crycb->apcb1.apm;
+ bitmap_and(mask1, mask1, mask2, APCB1_MASK_SIZE);
+
+ mask1 = (unsigned long *)vsie_page->crycb.apcb1.aqm;
+ mask2 = (unsigned long *)
+ vcpu->kvm->arch.crypto.crycb->apcb1.aqm;
+ bitmap_and(mask1, mask1, mask2, APCB1_MASK_SIZE);
+
+ mask1 = (unsigned long *)vsie_page->crycb.apcb1.adm;
+ mask2 = (unsigned long *)
+ vcpu->kvm->arch.crypto.crycb->apcb1.adm;
+ bitmap_and(mask1, mask1, mask2, APCB1_MASK_SIZE);
+ break;
+ default:
+ mask1 = (unsigned long *)vsie_page->crycb.apcb0.apm;
+ mask2 = (unsigned long *)
+ vcpu->kvm->arch.crypto.crycb->apcb1.apm;
+ bitmap_and(mask1, mask1, mask2, APCB0_MASK_SIZE);
+
+ mask1 = (unsigned long *)vsie_page->crycb.apcb0.aqm;
+ mask2 = (unsigned long *)
+ vcpu->kvm->arch.crypto.crycb->apcb1.aqm;
+ bitmap_and(mask1, mask1, mask2, APCB0_MASK_SIZE);
+
+ mask1 = (unsigned long *)vsie_page->crycb.apcb0.adm;
+ mask2 = (unsigned long *)
+ vcpu->kvm->arch.crypto.crycb->apcb1.adm;
+ bitmap_and(mask1, mask1, mask2, APCB0_MASK_SIZE);
+ break;
+ }
+}
+
+/*
* Create a shadow copy of the crycb block and setup key wrapping, if
* requested for guest 3 and enabled for guest 2.
*
- * We only accept format-1 (no AP in g2), but convert it into format-2
- * There is nothing to do for format-0.
- *
* Returns: - 0 if shadowed or nothing to do
* - > 0 if control has to be given to guest 2
*/
@@ -155,9 +200,17 @@ static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
unsigned long *b1, *b2;
u8 ecb3_flags;
- scb_s->crycbd = 0;
- if (!(crycbd_o & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1))
- return 0;
+ scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb);
+ kvm_ap_set_crycb_format(vcpu->kvm, &scb_s->crycbd);
+
+ /* copy the crycb */
+ if (read_guest_real(vcpu, crycb_addr, &vsie_page->crycb,
+ sizeof(vsie_page->crycb)))
+ return set_validity_icpt(scb_s, 0x0035U);
This looks wrong. You are always copying from L2, although L2 might not
even set up a crycb for L3. You removed the important crycbd_o check.

Don't we need the following?

a) Determine the _allowed_ crycb format for L2 -> L3. (How can we tell
which format L2 is allowed to use for L3)
b) Determine the target crycb format. This is basically
vm_ap_set_crycb_format afaics.
c) Convert the given crycb to the target crycb format.

So importantly, can you clarify (as I don't have access to documentation):

- When is L2 allowed to use format-0? So when will the format not be
ignored but is actually used?

- When is L2 allowed to use fortma-2? So when will the format not be
ignored but is actually used?

- When does the SIE start interpreting AP instructions? So how can we
hinder it from doing so? Empty masks? crycb format?

+
+ /* set up the effective masks */
+ set_crycb_emasks(vcpu, vsie_page);
+
/* format-1 is supported with message-security-assist extension 3 */
if (!test_kvm_facility(vcpu->kvm, 76))
return 0;
You now already copied the wrapping keys. So they could be !0. Please
add a comment why we don't care.

/* wrapping keys are ignored without ECB3_AES / ECB3_DEA */

@@ -172,13 +225,7 @@ static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
else if (!crycb_addr)
return set_validity_icpt(scb_s, 0x0039U);
- /* copy only the wrapping keys */
- if (read_guest_real(vcpu, crycb_addr + 72, &vsie_page->crycb, 56))
- return set_validity_icpt(scb_s, 0x0035U);
-
scb_s->ecb3 |= ecb3_flags;
- scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 |
- CRYCB_FORMAT2;
/* xor both blocks in one run */
b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;



--
Pierre Morel
Linux/KVM/QEMU in BÃblingen - Germany