arm64: KVM: Support Cortex-A57 guest CPU on APM X-Gene host

PATCH New Anup Patel 4 participants 6 messages Jul 3, 2013

Subsystems (2)

arm64 KVM arm64

lore

Share & tools

lore

Anup Patel

author

2013-07-03 08:42 UTC #0 lore raw

PATCH 1/1 arm64: KVM: Support Cortex-A57 guest CPU on APM X-Gene host

Update kvm_target_cpu() to allow Cortex-A57 guest CPU on APM X-Gene.

Signed-off-by Anup Patel <[email protected]>

Signed-off-by Pranavkumar Sawargaonkar <[email protected]>

---
 arch/arm64/kvm/guest.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

arch/arm64/kvm/guest.c +8 -2

index 2c3ff67..765f56f 100644
@@ -207,8 +207,13 @@ int __attribute_const__ kvm_target_cpu(void)
 	unsigned long implementor = read_cpuid_implementor();
 	unsigned long part_number = read_cpuid_part_number();
 
-	if (implementor != ARM_CPU_IMP_ARM)
-		return -EINVAL;
+	switch (implementor) {
+	case ARM_CPU_IMP_ARM:
+	case ARM_CPU_IMP_APM:
+		break;
+	default:
+ 		return -EINVAL;
+	}
 
 	switch (part_number) {
 	case ARM_CPU_PART_AEM_V8:

@@ -216,6 +221,7 @@ int __attribute_const__ kvm_target_cpu(void)
 	case ARM_CPU_PART_FOUNDATION:
 		return KVM_ARM_TARGET_FOUNDATION_V8;
 	case ARM_CPU_PART_CORTEX_A57:
+	case APM_CPU_PART_POTENZA:
 		/* Currently handled by the generic backend */
 		return KVM_ARM_TARGET_CORTEX_A57;
 	default:

Signature · 1 line

1.7.9.5

Anup Patel

author

2013-07-03 08:46 UTC #1 lore raw

PATCH 1/1 arm64: KVM: Support Cortex-A57 guest CPU on APM X-Gene host

Quoted 37 lines · Anup Patel

Update kvm_target_cpu() to allow Cortex-A57 guest CPU on APM X-Gene.

Signed-off-by Anup Patel <[email protected]>

Signed-off-by Pranavkumar Sawargaonkar <[email protected]>

---
 arch/arm64/kvm/guest.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

arch/arm64/kvm/guest.c +8 -3

index 2c3ff67..765f56f 100644
@@ -207,8 +207,13 @@ int __attribute_const__ kvm_target_cpu(void)
        unsigned long implementor = read_cpuid_implementor();
        unsigned long part_number = read_cpuid_part_number();
 
-       if (implementor != ARM_CPU_IMP_ARM)
-               return -EINVAL;
+       switch (implementor) {
+       case ARM_CPU_IMP_ARM:
+       case ARM_CPU_IMP_APM:
+               break;
+       default:
+               return -EINVAL;
+       }
 
        switch (part_number) {
        case ARM_CPU_PART_AEM_V8:

@@ -216,6 +221,7 @@ int __attribute_const__ kvm_target_cpu(void)
        case ARM_CPU_PART_FOUNDATION:
                return KVM_ARM_TARGET_FOUNDATION_V8;
        case ARM_CPU_PART_CORTEX_A57:
+       case APM_CPU_PART_POTENZA:
                /* Currently handled by the generic backend */
                return KVM_ARM_TARGET_CORTEX_A57;
        default:
--

Quoted 2 lines

1.7.9.5

Hi Catalin/Marc,

This patch depends on the patch titled "arm64: Add defines for APM ARMv8 implementation" from Vinyak Kale.

Regards,
Anup

Peter Maydell

2013-07-03 10:54 UTC #2 lore raw

PATCH 1/1 arm64: KVM: Support Cortex-A57 guest CPU on APM X-Gene host

In reply to Anup Patel

Quoted 15 lines · Anup Patel

Update kvm_target_cpu() to allow Cortex-A57 guest CPU on APM X-Gene.
Signed-off-by Anup Patel <[email protected]>
Signed-off-by Pranavkumar Sawargaonkar <[email protected]>
---
 arch/arm64/kvm/guest.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
arch/arm64/kvm/guest.c +0 -0
index 2c3ff67..765f56f 100644
@@ -207,8 +207,13 @@ int __attribute_const__ kvm_target_cpu(void)
        unsigned long implementor = read_cpuid_implementor();
        unsigned long part_number = read_cpuid_part_number();

arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 216

Quoted 10 lines

arch/arm64/kvm/guest.c +7 -2

 
-       if (implementor != ARM_CPU_IMP_ARM)
-               return -EINVAL;
+       switch (implementor) {
+       case ARM_CPU_IMP_ARM:
+       case ARM_CPU_IMP_APM:
+               break;
+       default:
+               return -EINVAL;
+       }

Doesn't this change mean we now accept the below part numbers for all implementors? That doesn't look right.

arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 227

Quoted 10 lines

arch/arm64/kvm/guest.c +1 -0

        switch (part_number) {
        case ARM_CPU_PART_AEM_V8:

@@ -216,6 +221,7 @@ int __attribute_const__ kvm_target_cpu(void)
        case ARM_CPU_PART_FOUNDATION:
                return KVM_ARM_TARGET_FOUNDATION_V8;
        case ARM_CPU_PART_CORTEX_A57:
+       case APM_CPU_PART_POTENZA:
                /* Currently handled by the generic backend */
                return KVM_ARM_TARGET_CORTEX_A57;
        default:

Do we really model all the system registers and so on correctly sufficiently to be able to present the guest with an A57 vcpu on an APM X-Gene host? (ie without accidentally leaking the host ID registers/system registers/impdef registers to the guest).

thanks
-- PMM

Marc Zyngier

2013-07-03 13:59 UTC #3 lore raw

PATCH 1/1 arm64: KVM: Support Cortex-A57 guest CPU on APM X-Gene host

Quoted 30 lines · Peter Maydell

Quoted 15 lines · Anup Patel
Update kvm_target_cpu() to allow Cortex-A57 guest CPU on APM X-Gene.
Signed-off-by Anup Patel <[email protected]>
Signed-off-by Pranavkumar Sawargaonkar <[email protected]>
---
 arch/arm64/kvm/guest.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
arch/arm64/kvm/guest.c +0 -0
index 2c3ff67..765f56f 100644
@@ -207,8 +207,13 @@ int __attribute_const__ kvm_target_cpu(void)
        unsigned long implementor = read_cpuid_implementor();
        unsigned long part_number = read_cpuid_part_number();
arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 216
Quoted 10 lines
arch/arm64/kvm/guest.c +7 -2
 
-       if (implementor != ARM_CPU_IMP_ARM)
-               return -EINVAL;
+       switch (implementor) {
+       case ARM_CPU_IMP_ARM:
+       case ARM_CPU_IMP_APM:
+               break;
+       default:
+               return -EINVAL;
+       }
Doesn't this change mean we now accept the below part numbers for all implementors? That doesn't look right.

Quoted 16 lines

arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 227
Quoted 10 lines
arch/arm64/kvm/guest.c +1 -0
        switch (part_number) {
        case ARM_CPU_PART_AEM_V8:
@@ -216,6 +221,7 @@ int __attribute_const__ kvm_target_cpu(void)
        case ARM_CPU_PART_FOUNDATION:
                return KVM_ARM_TARGET_FOUNDATION_V8;
        case ARM_CPU_PART_CORTEX_A57:
+       case APM_CPU_PART_POTENZA:
                /* Currently handled by the generic backend */
                return KVM_ARM_TARGET_CORTEX_A57;
        default:
Do we really model all the system registers and so on correctly sufficiently to be able to present the guest with an A57 vcpu on an APM X-Gene host? (ie without accidentally leaking the host ID registers/system registers/impdef registers to the guest).

There is no such thing in the KVM/arm64 code. The guest sees the real host, and I really don't lie the idea of lying to userspace by pretending that we're going to emulate an A57.

Anup: I suggest you rework that patch to present a X-Gene vcpu. You can implement it by declaring a new target and implementing it in terms of the generic backend one if that's convenient. But just pretending this is an A57 is not going to fly, sorry.

Cheers,

M.

Signature · 1 line

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

Anup Patel

2013-07-03 14:06 UTC #4 lore raw

PATCH 1/1 arm64: KVM: Support Cortex-A57 guest CPU on APM X-Gene host

Quoted 48 lines · Marc Zyngier

Quoted 30 lines · Peter Maydell
Quoted 15 lines · Anup Patel
Update kvm_target_cpu() to allow Cortex-A57 guest CPU on APM X-Gene.
Signed-off-by Anup Patel <[email protected]>
Signed-off-by Pranavkumar Sawargaonkar <[email protected]>
---
 arch/arm64/kvm/guest.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
arch/arm64/kvm/guest.c +0 -0
index 2c3ff67..765f56f 100644
@@ -207,8 +207,13 @@ int __attribute_const__ kvm_target_cpu(void)
        unsigned long implementor = read_cpuid_implementor();
        unsigned long part_number = read_cpuid_part_number();
arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 216
Quoted 10 lines
arch/arm64/kvm/guest.c +7 -2
 
-       if (implementor != ARM_CPU_IMP_ARM)
-               return -EINVAL;
+       switch (implementor) {
+       case ARM_CPU_IMP_ARM:
+       case ARM_CPU_IMP_APM:
+               break;
+       default:
+               return -EINVAL;
+       }
Doesn't this change mean we now accept the below part numbers for all implementors? That doesn't look right.
Quoted 16 lines
arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 227
Quoted 10 lines
arch/arm64/kvm/guest.c +1 -0
        switch (part_number) {
        case ARM_CPU_PART_AEM_V8:
@@ -216,6 +221,7 @@ int __attribute_const__ kvm_target_cpu(void)
        case ARM_CPU_PART_FOUNDATION:
                return KVM_ARM_TARGET_FOUNDATION_V8;
        case ARM_CPU_PART_CORTEX_A57:
+       case APM_CPU_PART_POTENZA:
                /* Currently handled by the generic backend */
                return KVM_ARM_TARGET_CORTEX_A57;
        default:
Do we really model all the system registers and so on correctly sufficiently to be able to present the guest with an A57 vcpu on an APM X-Gene host? (ie without accidentally leaking the host ID registers/system registers/impdef registers to the guest).

Quoted 8 lines

There is no such thing in the KVM/arm64 code. The guest sees the real host, and I really don't lie the idea of lying to userspace by pretending that we're going to emulate an A57.
Anup: I suggest you rework that patch to present a X-Gene vcpu. You can implement it by declaring a new target and implementing it in terms of the generic backend one if that's convenient. But just pretending this is an A57 is not going to fly, sorry.

Sure, I will add separate target vcpu for X-Gene and try to re-use your generic backend.

Quoted 4 lines

Cheers,

M.

Quoted 8 lines

--
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm at lists.cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm

--Anup

Anup Patel

2013-07-03 14:00 UTC #5 lore raw

PATCH 1/1 arm64: KVM: Support Cortex-A57 guest CPU on APM X-Gene host

In reply to Peter Maydell

Quoted 17 lines · Peter Maydell

Quoted 16 lines · Anup Patel
Update kvm_target_cpu() to allow Cortex-A57 guest CPU on APM X-Gene.
Signed-off-by Anup Patel <[email protected]>
Signed-off-by Pranavkumar Sawargaonkar <[email protected]>
---
 arch/arm64/kvm/guest.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
arch/arm64/kvm/guest.c +0 -0
index 2c3ff67..765f56f 100644
@@ -207,8 +207,13 @@ int __attribute_const__ kvm_target_cpu(void)
        unsigned long implementor = read_cpuid_implementor();
        unsigned long part_number = read_cpuid_part_number();
 

Quoted 12 lines

arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 216
Quoted 9 lines
arch/arm64/kvm/guest.c +7 -2
-       if (implementor != ARM_CPU_IMP_ARM)
-               return -EINVAL;
+       switch (implementor) {
+       case ARM_CPU_IMP_ARM:
+       case ARM_CPU_IMP_APM:
+               break;
+       default:
+               return -EINVAL;
+       }
Doesn't this change mean we now accept the below part numbers for all implementors? That doesn't look right.

Sure, we can use nested switch case instead of two separate switch cases.

Quoted 17 lines

arch/arm64/kvm/guest.c int __attribute_const__ kvm_target_cpu(void) · line 227
Quoted 11 lines
arch/arm64/kvm/guest.c +1 -0
        switch (part_number) {
        case ARM_CPU_PART_AEM_V8:
@@ -216,6 +221,7 @@ int __attribute_const__ kvm_target_cpu(void)
        case ARM_CPU_PART_FOUNDATION:
                return KVM_ARM_TARGET_FOUNDATION_V8;
        case ARM_CPU_PART_CORTEX_A57:
+       case APM_CPU_PART_POTENZA:
                /* Currently handled by the generic backend */
                return KVM_ARM_TARGET_CORTEX_A57;
        default:
Do we really model all the system registers and so on correctly sufficiently to be able to present the guest with an A57 vcpu on an APM X-Gene host? (ie without accidentally leaking the host ID registers/system registers/impdef registers to the guest).

Yes, there are very few APM impdef registers. For now A57 vcpu is not allowed to access these impdef registers.

For rest of the registers, A57 vcpu will see:
1. MIDR and MPIDR as expected on a A57 vcpu.
2. Cache configuration related registers will be same as that of underlying host so that cache operations work fine in guest

Quoted 7 lines

thanks
-- PMM
_______________________________________________
kvmarm mailing list
kvmarm at lists.cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm

--Anup