Skip to contentSkip to conversation

KVM: arm: fix KVM_CAP_ARM_INJECT_EXT_DABT for aarch32 guests

PATCH In review James Morse 3 participants 7 messages Jan 21, 2020 – Jan 27, 2020 KVM arm64

lore
Share & tools
lore
#0 lore raw

cover KVM: arm: fix KVM_CAP_ARM_INJECT_EXT_DABT for aarch32 guests

Beata reports that KVM_CAP_ARM_INJECT_EXT_DABT doesn't do the expected thing for aarch32 guests. We get the wrong register layout, and weren't even trying to set a 'external abort' in the first place!

Both patches are intended as fixes, but patch 2 is somewhat in the eye of the beholder. I don't know why an imp-def exception was picked...

Thanks,

James Morse (2):
KVM: arm: fix DFSR setting for non-LPAE aarch32 guests
KVM: arm: Make inject_abt32() inject an external abort instead

 virt/kvm/arm/aarch32.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
Signature · 6 lines
2.24.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
#1 lore raw

PATCH 1/2 KVM: arm: fix DFSR setting for non-LPAE aarch32 guests

Beata reports that KVM_SET_VCPU_EVENTS doesn't inject the expected exception to a non-LPAE aarch32 guest.

The host intends to inject DFSR.FS=0x14 "IMPLEMENTATION DEFINED fault (Lockdown fault)", but the guest receives DFSR.FS=0x04 "Fault on instruction cache maintenance". This fault is hooked by do_translation_fault() since ARMv6, which goes on to silently 'handle' the exception, and restart the faulting instruction.

It turns out, when TTBCR.EAE is clear DFSR is split, and FS[4] has to shuffle up to DFSR[10].

As KVM only does this in one place, fix up the static values. We now get the expected:
| Unhandled fault: lock abort (0x404) at 0x9c800f00

Fixes 74a64a981662a ("KVM: arm/arm64: Unify 32bit fault injection")
---
 virt/kvm/arm/aarch32.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
virt/kvm/arm/aarch32.c +5 -3
index c4c57ba99e90..942108b62cd3 100644
@@ -90,10 +90,12 @@ static void inject_abt32(struct kvm_vcpu *vcpu, bool is_pabt,
 
 	/* Give the guest an IMPLEMENTATION DEFINED exception */
 	is_lpae = (vcpu_cp15(vcpu, c2_TTBCR) >> 31);
-	if (is_lpae)
+	if (is_lpae) {
 		*fsr = 1 << 9 | 0x34;
-	else
-		*fsr = 0x14;
+	} else {
+		/* Surprise! DFSR's FS[4] lives in bit 10 */
+		*fsr = BIT(10) | 0x4; /* 0x14 */
+	}
 }
 
 void kvm_inject_dabt32(struct kvm_vcpu *vcpu, unsigned long addr)
Signature · 6 lines
2.24.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
#2 lore raw

PATCH 2/2 KVM: arm: Make inject_abt32() inject an external abort instead

In reply to James Morse

KVM's inject_abt64() injects an external-abort into an aarch64 guest. The KVM_CAP_ARM_INJECT_EXT_DABT is intended to do exactly this, but for an aarch32 guest inject_abt32() injects an implementation-defined exception, 'Lockdown fault'.

Change this to external abort. For non-LPAE we now get the documented:
| Unhandled fault: external abort on non-linefetch (0x008) at 0x9c800f00 and for LPAE:
| Unhandled fault: synchronous external abort (0x210) at 0x9c800f00

---
 virt/kvm/arm/aarch32.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
virt/kvm/arm/aarch32.c +7 -3
index 942108b62cd3..90433bc486d5 100644
@@ -14,6 +14,10 @@
 #include <asm/kvm_emulate.h>
 #include <asm/kvm_hyp.h>
 
+#define DFSR_FSC_EXTABT_LPAE	0x10
+#define DFSR_FSC_EXTABT_nLPAE	0x08
+#define DFSR_LPAE		BIT(9)
+
 /*
  * Table taken from ARMv8 ARM DDI0487B-B, table G1-10.
  */
@@ -91,10 +95,10 @@ static void inject_abt32(struct kvm_vcpu *vcpu, bool is_pabt,
 	/* Give the guest an IMPLEMENTATION DEFINED exception */
 	is_lpae = (vcpu_cp15(vcpu, c2_TTBCR) >> 31);
 	if (is_lpae) {
-		*fsr = 1 << 9 | 0x34;
+		*fsr = DFSR_LPAE | DFSR_FSC_EXTABT_LPAE;
 	} else {
-		/* Surprise! DFSR's FS[4] lives in bit 10 */
-		*fsr = BIT(10) | 0x4; /* 0x14 */
+		/* no need to shuffle FS[4] into DFSR[10] as its 0 */
+		*fsr = DFSR_FSC_EXTABT_nLPAE;
 	}
 }
 
Signature · 6 lines
2.24.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
maintainer
#3 lore raw
In reply to James Morse

Hi James,

Thanks for this.

Quoted 6 lines · James Morse

Beata reports that KVM_CAP_ARM_INJECT_EXT_DABT doesn't do the expected thing for aarch32 guests. We get the wrong register layout, and weren't even trying to set a 'external abort' in the first place!

Both patches are intended as fixes, but patch 2 is somewhat in the eye of the beholder. I don't know why an imp-def exception was picked...

Because we had no idea what we were doing (this is all 2013 vintage). This code comes from e82e030556e42e8 (the initial arm64 port), which didn't do the same thing as 5b3e5e5bf230f56 (the 32bit version).

Both were wrong. BTW.

Quoted 9 lines

Thanks,

James Morse (2):
KVM: arm: fix DFSR setting for non-LPAE aarch32 guests
KVM: arm: Make inject_abt32() inject an external abort instead

 virt/kvm/arm/aarch32.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

It all looks good to me. I'm minded to add a

Fixes 74a64a981662a ("KVM: arm/arm64: Unify 32bit fault injection")

to the second patch as well, so that KVM_CAP_ARM_INJECT_EXT_DABT does what it says on the tin.

Thanks,

        M.
Signature · 5 lines
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
#4 lore raw
In reply to James Morse

Hi James,

Thanks for the fixes - they work like a charm.

Quoted 8 lines · James Morse

Beata reports that KVM_CAP_ARM_INJECT_EXT_DABT doesn't do the expected thing for aarch32 guests. We get the wrong register layout, and weren't even trying to set a 'external abort' in the first place!

Both patches are intended as fixes, but patch 2 is somewhat in the eye of the beholder. I don't know why an imp-def exception was picked...

On a side note - currently KVM exposes capability that is not fully supported (till the fix gets applied) and there is no easy way for the user space to determine whether the injection will work as expected and whether it is safe to use it or not. Although this is addressing a problem that is not that common (I suppose) but still it might be worth to add a way for the kernel to inform the user-space that it is all good to go? There has been a 'similar' case in the past with KVM_SET_USER_MEMORY_REGION, where fixes where needed
and those were announced through new caps. Now, I'm not sure if adding new capability would be the best approach here though it seems that there is not much of a choice?

Best Regards
Beata

Quoted 6 lines

Thanks,

James Morse (2):
KVM: arm: fix DFSR setting for non-LPAE aarch32 guests
KVM: arm: Make inject_abt32() inject an external abort instead

Quoted 6 lines
 virt/kvm/arm/aarch32.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

--
2.24.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

maintainer
#5 lore raw

On Fri, 24 Jan 2020 15:39:29 +0000
Beata Michalska <[email protected]> wrote:

Hi Beata,

Quoted 13 lines

Hi James,

Thanks for the fixes - they work like a charm.

Quoted 8 lines · James Morse

Beata reports that KVM_CAP_ARM_INJECT_EXT_DABT doesn't do the expected thing for aarch32 guests. We get the wrong register layout, and weren't even trying to set a 'external abort' in the first place!

Both patches are intended as fixes, but patch 2 is somewhat in the eye of the beholder. I don't know why an imp-def exception was picked...

Quoted 10 lines

On a side note - currently KVM exposes capability that is not fully supported (till the fix gets applied) and there is no easy way for the user space to determine whether the injection will work as expected and whether it is safe to use it or not. Although this is addressing a problem that is not that common (I suppose) but still it might be worth to add a way for the kernel to inform the user-space that it is all good to go? There has been a 'similar' case in the past with KVM_SET_USER_MEMORY_REGION, where fixes where needed
and those were announced through new caps. Now, I'm not sure if adding new capability would be the best approach here though it seems that there is not much of a choice?

My take on this particular issue is that although the functionality is not perfectly working (far from it), it isn't completely broken (the guest does get some form of abort). Furthermore, we tend to add this kind of discovery mechanism when the userspace interface is broken, not when we have an implementation defect in the CPU emulation.

The real question is whether there anything out there that would depend on such broken behaviour?

Thanks,

	M.
Signature · 5 lines
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
#6 lore raw

Hi Marc,

Quoted 30 lines · Marc Zyngier

On Fri, 24 Jan 2020 15:39:29 +0000
Beata Michalska <[email protected]> wrote:

Hi Beata,

Quoted 13 lines

Hi James,

Thanks for the fixes - they work like a charm.

Quoted 8 lines · James Morse

Beata reports that KVM_CAP_ARM_INJECT_EXT_DABT doesn't do the expected thing for aarch32 guests. We get the wrong register layout, and weren't even trying to set a 'external abort' in the first place!

Both patches are intended as fixes, but patch 2 is somewhat in the eye of the beholder. I don't know why an imp-def exception was picked...

Quoted 10 lines

On a side note - currently KVM exposes capability that is not fully supported (till the fix gets applied) and there is no easy way for the user space to determine whether the injection will work as expected and whether it is safe to use it or not. Although this is addressing a problem that is not that common (I suppose) but still it might be worth to add a way for the kernel to inform the user-space that it is all good to go? There has been a 'similar' case in the past with KVM_SET_USER_MEMORY_REGION, where fixes where needed
and those were announced through new caps. Now, I'm not sure if adding new capability would be the best approach here though it seems that there is not much of a choice?

Quoted 6 lines

My take on this particular issue is that although the functionality is not perfectly working (far from it), it isn't completely broken (the guest does get some form of abort). Furthermore, we tend to add this kind of discovery mechanism when the userspace interface is broken, not when we have an implementation defect in the CPU emulation.

Indeed, the guest will get 'the' abort with a small catch though:
the fault handler will actually manage to handle it leading to the faulting instruction being restarted, trapping thereby the guest in a vicious circle. This is trading in abnormal termination of the guest (aka 'no idea what has just happened' ) for rather meaningless back-and-forth with the host kernel.

Quoted 3 lines

The real question is whether there anything out there that would depend on such broken behaviour?

AFAICT, that would be Qemu trying to handle 'nicely' all the unexpected cases, where guest triggers the DABT with no valid instruction syndrome (changes on the way). Now, I agree this is not the most common case ever, but still. Currently the guest will keep on repeating the same mistake ... until it gets told
it's high time to stop trying. And that becomes arbitrary without implementing some additional logic behind it - a bit of an overkill.

BR

Beata

Quoted 5 lines

Thanks,

        M.

--
Jazz is not dead. It just smells funny...

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm