Skip to contentSkip to conversation

KVM: arm64: Clarify vcpu reset behaviour

PATCH In review Marc Zyngier 3 participants 7 messages Apr 6, 2021 – Apr 8, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 KVM: arm64: Clarify vcpu reset behaviour

Although the KVM_ARM_VCPU_INIT documentation mention that the registers are reset to their "initial values", it doesn't describe what these values are.

Describe this state explicitly.

---
 Documentation/virt/kvm/api.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
Documentation/virt/kvm/api.rst +10 -0
index 38e327d4b479..e2237e4e10ba 100644
@@ -3115,6 +3115,16 @@ optional features it should have.  This will cause a reset of the cpu
 registers to their initial values.  If this is not called, KVM_RUN will
 return ENOEXEC for that vcpu.
 
+The initial values are defined as:
+	- Processor state:
+		* AArch64: EL1h, D, A, I and F bits set
+		* Aarch32: SVC, D, A, I and F bits set
+	- General Purpose registers, including PC and SP: set to 0
+	- FPSIMD/NEON registers: set to 0
+	- SVE registers: set to 0
+	- System registers: Reset to their architecturally defined
+	  values as for a warm reset to EL1 (resp. SVC)
+
 Note that because some registers reflect machine topology, all vcpus
 should be created before this ioctl is invoked.
 
Signature · 7 lines
2.29.2


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Reviewed-by on v1 1/1
#1 lore raw

Hi Marc,

Quoted 5 lines · Marc Zyngier

Although the KVM_ARM_VCPU_INIT documentation mention that the registers are reset to their "initial values", it doesn't describe what these values are.

Describe this state explicitly.

This is a good idea.

Quoted 11 lines
---
 Documentation/virt/kvm/api.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
Documentation/virt/kvm/api.rst +0 -0
index 38e327d4b479..e2237e4e10ba 100644

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3120

Quoted 6 lines
Documentation/virt/kvm/api.rst +3 -0
@@ -3115,6 +3115,16 @@ optional features it should have.  This will cause a reset of the cpu
 registers to their initial values.  If this is not called, KVM_RUN will
 return ENOEXEC for that vcpu.
 
+The initial values are defined as:
+	- Processor state:
+		* AArch64: EL1h, D, A, I and F bits set

That value matches the macro VCPU_RESET_STATE_EL1.

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3121

Quoted 1 line
Documentation/virt/kvm/api.rst +1 -0
+		* Aarch32: SVC, D, A, I and F bits set

VCPU_RESET_STATE_SVC doesn't have a D bit; according to ARM DDI 0487G.a, page G1-5965, CPSR doesn't have a D bit either (I might be reading it wrong).

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3122

Quoted 1 line
Documentation/virt/kvm/api.rst +1 -0
+	- General Purpose registers, including PC and SP: set to 0

They are zero'ed explicitly in kvm_reset_vcpu().

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3123

Quoted 1 line
Documentation/virt/kvm/api.rst +1 -0
+	- FPSIMD/NEON registers: set to 0

I haven't been able to find where they are initialized explicitly, I assume they aren't. They are zero because the kvm_vcpu struct is allocated via kmem_cache_zalloc() in kvm_vm_ioctl_create_vcpu(), so this is correct.

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3124

Quoted 1 line
Documentation/virt/kvm/api.rst +1 -0
+	- SVE registers: set to 0

They are zero'ed explicitly in kvm_vcpu_reset_sve().

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3126

Quoted 2 lines
Documentation/virt/kvm/api.rst +2 -0
+	- System registers: Reset to their architecturally defined
+	  values as for a warm reset to EL1 (resp. SVC)

This is done in kvm_reset_vcpu() -> kvm_reset_sys_regs(), which from what I can tell does the right thing, but I haven't checked each and every register.

Assuming the D bit for CPSR/SPSR was a typo and with it removed:

Thanks,

Alex

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3130

Quoted 4 lines
Documentation/virt/kvm/api.rst +1 -0
+
 Note that because some registers reflect machine topology, all vcpus
 should be created before this ioctl is invoked.
 

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#2 lore raw
Quoted 11 lines · Alexandru Elisei

Hi Marc,

Quoted 5 lines · Marc Zyngier

Although the KVM_ARM_VCPU_INIT documentation mention that the registers are reset to their "initial values", it doesn't describe what these values are.

Describe this state explicitly.

This is a good idea.

Yeah, it turns out this is something we need for pKVM, because the reset state is slightly different.

Quoted 21 lines
Quoted 12 lines
---
 Documentation/virt/kvm/api.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
Documentation/virt/kvm/api.rst +0 -0
index 38e327d4b479..e2237e4e10ba 100644

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3120

Quoted 6 lines
Documentation/virt/kvm/api.rst +3 -0
@@ -3115,6 +3115,16 @@ optional features it should have.  This will cause a reset of the cpu
 registers to their initial values.  If this is not called, KVM_RUN will
 return ENOEXEC for that vcpu.
 
+The initial values are defined as:
+	- Processor state:
+		* AArch64: EL1h, D, A, I and F bits set

That value matches the macro VCPU_RESET_STATE_EL1.

Quoted 3 lines

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3121

Quoted 1 line
Documentation/virt/kvm/api.rst +1 -0
+		* Aarch32: SVC, D, A, I and F bits set

VCPU_RESET_STATE_SVC doesn't have a D bit; according to ARM DDI 0487G.a, page G1-5965, CPSR doesn't have a D bit either (I might be reading it wrong).

Meh, copy paste. Thanks for that,

Quoted 10 lines

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3122

Quoted 1 line
Documentation/virt/kvm/api.rst +1 -0
+	- General Purpose registers, including PC and SP: set to 0

They are zero'ed explicitly in kvm_reset_vcpu().

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3123

Quoted 1 line
Documentation/virt/kvm/api.rst +1 -0
+	- FPSIMD/NEON registers: set to 0

I haven't been able to find where they are initialized explicitly, I assume they aren't. They are zero because the kvm_vcpu struct is allocated via kmem_cache_zalloc() in kvm_vm_ioctl_create_vcpu(), so this is correct.

Indeed. However, this outlines an interesting buglet. Userspace is allowed to call KVM_ARM_VCPU_INIT at any point, and get the vcpu back to the reset state.

The lack of explicit wipeout of the FPSIMD file is on its own a bug which crept in with e47c2055c6 ("KVM: arm64: Make struct kvm_regs userspace-only"), when kvm_cpu_context started to use user_pt_regs directly.

I'll fix that separately, thanks for pushing me to have a closer look.

Quoted 12 lines

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3124

Quoted 2 lines
Documentation/virt/kvm/api.rst +1 -0
+	- SVE registers: set to 0

They are zero'ed explicitly in kvm_vcpu_reset_sve().

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3126

Quoted 2 lines
Documentation/virt/kvm/api.rst +2 -0
+	- System registers: Reset to their architecturally defined
+	  values as for a warm reset to EL1 (resp. SVC)

This is done in kvm_reset_vcpu() -> kvm_reset_sys_regs(), which from what I can tell does the right thing, but I haven't checked each and every register.

Quoted 3 lines

Assuming the D bit for CPSR/SPSR was a typo and with it removed:

Thanks for that. I'll send the other fix separately.

	M.
Signature · 6 lines
Without deviation from the norm, progress is not possible.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#3 lore raw
In reply to Marc Zyngier
Quoted 16 lines · Marc Zyngier

Although the KVM_ARM_VCPU_INIT documentation mention that the registers are reset to their "initial values", it doesn't describe what these values are.

Describe this state explicitly.

---
 Documentation/virt/kvm/api.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
Documentation/virt/kvm/api.rst +0 -0
index 38e327d4b479..e2237e4e10ba 100644

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3121

Quoted 7 lines
Documentation/virt/kvm/api.rst +4 -0
@@ -3115,6 +3115,16 @@ optional features it should have.  This will cause a reset of the cpu
 registers to their initial values.  If this is not called, KVM_RUN will
 return ENOEXEC for that vcpu.
 
+The initial values are defined as:
+	- Processor state:
+		* AArch64: EL1h, D, A, I and F bits set
+		* Aarch32: SVC, D, A, I and F bits set

nit, but capitalisation should be "AArch32"

Is "SVC" sufficient, or do we also need to describe Arm vs Thumb?

Will

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#4 lore raw
Quoted 22 lines · Will Deacon
Quoted 21 lines · Marc Zyngier

Although the KVM_ARM_VCPU_INIT documentation mention that the registers are reset to their "initial values", it doesn't describe what these values are.

Describe this state explicitly.

---
 Documentation/virt/kvm/api.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
Documentation/virt/kvm/api.rst +0 -0
index 38e327d4b479..e2237e4e10ba 100644
@@ -3115,6 +3115,16 @@ optional features it should have.  This will 

cause a reset of the cpu
registers to their initial values.  If this is not called, KVM_RUN
will
return ENOEXEC for that vcpu.

Quoted 8 lines

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3121

Quoted 4 lines
Documentation/virt/kvm/api.rst +4 -0
+The initial values are defined as:
+	- Processor state:
+		* AArch64: EL1h, D, A, I and F bits set
+		* Aarch32: SVC, D, A, I and F bits set

nit, but capitalisation should be "AArch32"

Is "SVC" sufficient, or do we also need to describe Arm vs Thumb?

SVC doesn't describe the instruction set, but we don't say that PSR.T is set, which implies ARM.

I can add something like "All other bits are set to 0", if that makes it clearer.

Thanks,

        M.
Signature · 6 lines
Jazz is not dead. It just smells funny...

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Acked-by
#5 lore raw
Quoted 31 lines · Marc Zyngier
Quoted 30 lines · Will Deacon
Quoted 22 lines · Marc Zyngier

Although the KVM_ARM_VCPU_INIT documentation mention that the registers are reset to their "initial values", it doesn't describe what these values are.

Describe this state explicitly.

---
 Documentation/virt/kvm/api.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
Documentation/virt/kvm/api.rst +0 -0
index 38e327d4b479..e2237e4e10ba 100644
@@ -3115,6 +3115,16 @@ optional features it should have.  This will

cause a reset of the cpu
registers to their initial values.  If this is not called, KVM_RUN
will
return ENOEXEC for that vcpu.

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3121

Quoted 4 lines
Documentation/virt/kvm/api.rst +4 -0
+The initial values are defined as:
+	- Processor state:
+		* AArch64: EL1h, D, A, I and F bits set
+		* Aarch32: SVC, D, A, I and F bits set

nit, but capitalisation should be "AArch32"

Quoted 4 lines
Quoted 1 line

Is "SVC" sufficient, or do we also need to describe Arm vs Thumb?

SVC doesn't describe the instruction set, but we don't say that PSR.T is set, which implies ARM.

Ah yes, thanks.

Quoted 2 lines

I can add something like "All other bits are set to 0", if that makes it clearer.

I think that would be worth adding. With that and the minor typo fix:

Will

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#6 lore raw
Quoted 38 lines · Will Deacon
Quoted 31 lines · Marc Zyngier
Quoted 30 lines · Will Deacon
Quoted 22 lines · Marc Zyngier

Although the KVM_ARM_VCPU_INIT documentation mention that the registers are reset to their "initial values", it doesn't describe what these values are.

Describe this state explicitly.

---
 Documentation/virt/kvm/api.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
Documentation/virt/kvm/api.rst +0 -0
index 38e327d4b479..e2237e4e10ba 100644
@@ -3115,6 +3115,16 @@ optional features it should have.  This will

cause a reset of the cpu
registers to their initial values.  If this is not called, KVM_RUN
will
return ENOEXEC for that vcpu.

Documentation/virt/kvm/api.rst optional features it should have.  This will cause a reset of the cpu · line 3121

Quoted 4 lines
Documentation/virt/kvm/api.rst +4 -0
+The initial values are defined as:
+	- Processor state:
+		* AArch64: EL1h, D, A, I and F bits set
+		* Aarch32: SVC, D, A, I and F bits set

nit, but capitalisation should be "AArch32"

Quoted 4 lines
Quoted 1 line

Is "SVC" sufficient, or do we also need to describe Arm vs Thumb?

SVC doesn't describe the instruction set, but we don't say that PSR.T is set, which implies ARM.

Quoted 8 lines

Ah yes, thanks.

Quoted 2 lines

I can add something like "All other bits are set to 0", if that makes it clearer.

I think that would be worth adding. With that and the minor typo fix:

Now fixed.

Thanks,

	M.
Signature · 6 lines
Without deviation from the norm, progress is not possible.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel