Skip to contentSkip to conversation

Target CPU=Host implementation for KVM ARM/ARM64

PATCH Superseded Anup Patel 4 participants 14 messages Sep 19, 2013 – Sep 23, 2013

Subsystems (5)
Documentation Arm (32-bit) KVM arm64 +1 more subsystems

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Anup Patel

author
#0 lore raw

v3 cover Target CPU=Host implementation for KVM ARM/ARM64

It will be very useful for user space (QEMU/KVMTOOL) if it has a method of querying VCPU target type matching to underlying Host. We can use such querying mechanism and implement machine models in user space where VCPU target type is always same-as/similar-to underlying Host (i.e. Target CPU=Host).

This patch series implements KVM_ARM_PREFERRED_TARGET vm ioclt for querying VCPU target type matching underlying host. Using this new ioctl we can implement VCPU target CPU=Host in user space.

Also, it is not mandatory to call KVM_ARM_PREFERRED_TARGET vm ioctl and the old method of trying all possible target types using the KVM_ARM_VCPU_INIT ioctl to initialize VCPU works fine.

V3:
- Return -ENODEV if no preferred target available for host
- Make KVM_ARM_PREFERRED_TARGET ioctl as vm ioctl

V2:
- Renamed the ioclt to KVM_ARM_PREFERRED_TARGET
- Return struct kvm_vcpu_init instace instead of just target type

V1:
- Initial patch-set with ioctl named as KVM_ARM_SUITABLE_TARGET

Anup Patel (4):
ARM: KVM: Implement kvm_vcpu_preferred_target() function
ARM64: KVM: Implement kvm_vcpu_preferred_target() function ARM/ARM64: KVM: Implement KVM_ARM_PREFERRED_TARGET ioctl
KVM: Add documentation for KVM_ARM_PREFERRED_TARGET ioctl

 Documentation/virtual/kvm/api.txt |   26 ++++++++++++++++++++++----
 arch/arm/include/asm/kvm_host.h   |    1 +
 arch/arm/kvm/arm.c                |   13 +++++++++++++
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 arch/arm64/kvm/guest.c            |   21 +++++++++++++++++++++
 include/uapi/linux/kvm.h          |    1 +
 7 files changed, 79 insertions(+), 4 deletions(-)
Signature · 1 line
1.7.9.5
author
#1 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +20 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 	return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+	int target = kvm_target_cpu();
+
+	if (target < 0)
+		return -ENODEV;
+
+	memset(init, 0, sizeof(*init));
+
+	/* 
+	 * For now, we return all optional features are available
+	 * for preferred target. In future, we might have features
+	 * available based on underlying host.
+	 */
+	init->target = (__u32)target;
+	init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);
+
+	return 0;
+}
+
 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
 	return -EINVAL;
arch/arm64/include/asm/kvm_host.h +1 -0
index f318c43..b609db3 100644
@@ -156,6 +156,7 @@ struct kvm_vcpu_stat {
 struct kvm_vcpu_init;
 int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 			const struct kvm_vcpu_init *init);
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init);
 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
 struct kvm_one_reg;
Signature · 1 line
1.7.9.5
#2 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 25 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +6 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 	return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+	int target = kvm_target_cpu();
+
+	if (target < 0)
+		return -ENODEV;

arch/arm/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 240

Quoted 10 lines
arch/arm/kvm/guest.c +10 -0
+
+	memset(init, 0, sizeof(*init));
+
+	/* 
+	 * For now, we return all optional features are available
+	 * for preferred target. In future, we might have features
+	 * available based on underlying host.
+	 */
+	init->target = (__u32)target;
+	init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);

I'm in two minds about this feature reporting. I see they serve a purpose, but they also duplicate capabilities, which is the standard way to advertise what KVM can do.

It means we end up having to sync two reporting mechanism, and I feel this is in general a bad idea.

Furthermore, KVM_ARM_VCPU_POWER_OFF is hardly a feature of the HW, but rather a firmware emulation thing.

Peter, Christoffer: Thoughts?

	M.
Quoted 19 lines
arch/arm/kvm/guest.c +3 -0
+	return 0;
+}
+
 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
 	return -EINVAL;
arch/arm64/include/asm/kvm_host.h +1 -0
index f318c43..b609db3 100644
@@ -156,6 +156,7 @@ struct kvm_vcpu_stat {
 struct kvm_vcpu_init;
 int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 			const struct kvm_vcpu_init *init);
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init);
 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
 struct kvm_one_reg;
Signature · 1 line
Jazz is not dead. It just smells funny...
#3 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 41 lines · Marc Zyngier
Quoted 25 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +6 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 	return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+	int target = kvm_target_cpu();
+
+	if (target < 0)
+		return -ENODEV;

arch/arm/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 240

Quoted 10 lines
arch/arm/kvm/guest.c +10 -0
+
+	memset(init, 0, sizeof(*init));
+
+	/* 
+	 * For now, we return all optional features are available
+	 * for preferred target. In future, we might have features
+	 * available based on underlying host.
+	 */
+	init->target = (__u32)target;
+	init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);

I'm in two minds about this feature reporting. I see they serve a purpose, but they also duplicate capabilities, which is the standard way to advertise what KVM can do.

Quoted 8 lines

It means we end up having to sync two reporting mechanism, and I feel this is in general a bad idea.

Furthermore, KVM_ARM_VCPU_POWER_OFF is hardly a feature of the HW, but rather a firmware emulation thing.

Peter, Christoffer: Thoughts?

I wanted to return the full kvm_vcpu_init instead of just a target int, so we did not have to come up with yet another ioctl if we need to return more information about the capabilities of the host CPU in the future.

But perhaps we can formulate the API, to say only the (currently empty) following list of features can only be enabled if the corresponding bit is enabled, or something along those lines.

-Christoffer

#4 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

Hi Christoffer/Marc,

Quoted 59 lines · Christoffer Dall
Quoted 41 lines · Marc Zyngier
Quoted 25 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +6 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
    return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+   int target = kvm_target_cpu();
+
+   if (target < 0)
+           return -ENODEV;

arch/arm/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 240

Quoted 10 lines
arch/arm/kvm/guest.c +10 -0
+
+   memset(init, 0, sizeof(*init));
+
+   /*
+    * For now, we return all optional features are available
+    * for preferred target. In future, we might have features
+    * available based on underlying host.
+    */
+   init->target = (__u32)target;
+   init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);

I'm in two minds about this feature reporting. I see they serve a purpose, but they also duplicate capabilities, which is the standard way to advertise what KVM can do.

Quoted 8 lines

It means we end up having to sync two reporting mechanism, and I feel this is in general a bad idea.

Furthermore, KVM_ARM_VCPU_POWER_OFF is hardly a feature of the HW, but rather a firmware emulation thing.

Peter, Christoffer: Thoughts?

I wanted to return the full kvm_vcpu_init instead of just a target int, so we did not have to come up with yet another ioctl if we need to return more information about the capabilities of the host CPU in the future.

But perhaps we can formulate the API, to say only the (currently empty) following list of features can only be enabled if the corresponding bit is enabled, or something along those lines.

Quoted 5 lines

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

Do we stick with current implementation of returning struct kvm_vcpu_init ? OR
Do we return struct kvm_vcpu_init with all features set to zero ?

Regards,
Anup

#5 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 69 lines · Anup Patel

Hi Christoffer/Marc,

Quoted 59 lines · Christoffer Dall
Quoted 41 lines · Marc Zyngier
Quoted 25 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +6 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
    return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+   int target = kvm_target_cpu();
+
+   if (target < 0)
+           return -ENODEV;

arch/arm/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 240

Quoted 10 lines
arch/arm/kvm/guest.c +10 -0
+
+   memset(init, 0, sizeof(*init));
+
+   /*
+    * For now, we return all optional features are available
+    * for preferred target. In future, we might have features
+    * available based on underlying host.
+    */
+   init->target = (__u32)target;
+   init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);

I'm in two minds about this feature reporting. I see they serve a purpose, but they also duplicate capabilities, which is the standard way to advertise what KVM can do.

Quoted 8 lines

It means we end up having to sync two reporting mechanism, and I feel this is in general a bad idea.

Furthermore, KVM_ARM_VCPU_POWER_OFF is hardly a feature of the HW, but rather a firmware emulation thing.

Peter, Christoffer: Thoughts?

I wanted to return the full kvm_vcpu_init instead of just a target int, so we did not have to come up with yet another ioctl if we need to return more information about the capabilities of the host CPU in the future.

But perhaps we can formulate the API, to say only the (currently empty) following list of features can only be enabled if the corresponding bit is enabled, or something along those lines.

Quoted 5 lines

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

Quoted 4 lines

Do we stick with current implementation of returning struct kvm_vcpu_init ? OR
Do we return struct kvm_vcpu_init with all features set to zero ?

Let's give Marc a day or two to respond to this one ;)

-Christoffer

#6 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 70 lines · Christoffer Dall
Quoted 69 lines · Anup Patel

Hi Christoffer/Marc,

Quoted 64 lines · Christoffer Dall
Quoted 41 lines · Marc Zyngier
Quoted 25 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +6 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
    return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+   int target = kvm_target_cpu();
+
+   if (target < 0)
+           return -ENODEV;

arch/arm/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 240

Quoted 10 lines
arch/arm/kvm/guest.c +10 -0
+
+   memset(init, 0, sizeof(*init));
+
+   /*
+    * For now, we return all optional features are available
+    * for preferred target. In future, we might have features
+    * available based on underlying host.
+    */
+   init->target = (__u32)target;
+   init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);

I'm in two minds about this feature reporting. I see they serve a purpose, but they also duplicate capabilities, which is the standard way to advertise what KVM can do.

Quoted 8 lines

It means we end up having to sync two reporting mechanism, and I feel this is in general a bad idea.

Furthermore, KVM_ARM_VCPU_POWER_OFF is hardly a feature of the HW, but rather a firmware emulation thing.

Peter, Christoffer: Thoughts?

I wanted to return the full kvm_vcpu_init instead of just a target int, so we did not have to come up with yet another ioctl if we need to return more information about the capabilities of the host CPU in the future.

But perhaps we can formulate the API, to say only the (currently empty) following list of features can only be enabled if the corresponding bit is enabled, or something along those lines.

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

Quoted 5 lines
Quoted 4 lines

Do we stick with current implementation of returning struct kvm_vcpu_init ? OR
Do we return struct kvm_vcpu_init with all features set to zero ?

Let's give Marc a day or two to respond to this one ;)

Are you implying I'm getting slow? ;-)

To answer Anup's question, I would tend to be cautious, and not expose things for which we already have another API in place. So far, we've stuck with the KVM approach of having a capability bit for each feature we enable, and I'm quite happy with that.

So I'm in favour of Christoffer's proposal to return an empty feature set, and start adding stuff if/when the need arises.

Cheers,

	M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#7 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 79 lines · Marc Zyngier
Quoted 70 lines · Christoffer Dall
Quoted 69 lines · Anup Patel

Hi Christoffer/Marc,

Quoted 64 lines · Christoffer Dall
Quoted 41 lines · Marc Zyngier
Quoted 25 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +6 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
    return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+   int target = kvm_target_cpu();
+
+   if (target < 0)
+           return -ENODEV;

arch/arm/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 240

Quoted 10 lines
arch/arm/kvm/guest.c +10 -0
+
+   memset(init, 0, sizeof(*init));
+
+   /*
+    * For now, we return all optional features are available
+    * for preferred target. In future, we might have features
+    * available based on underlying host.
+    */
+   init->target = (__u32)target;
+   init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);

I'm in two minds about this feature reporting. I see they serve a purpose, but they also duplicate capabilities, which is the standard way to advertise what KVM can do.

Quoted 8 lines

It means we end up having to sync two reporting mechanism, and I feel this is in general a bad idea.

Furthermore, KVM_ARM_VCPU_POWER_OFF is hardly a feature of the HW, but rather a firmware emulation thing.

Peter, Christoffer: Thoughts?

I wanted to return the full kvm_vcpu_init instead of just a target int, so we did not have to come up with yet another ioctl if we need to return more information about the capabilities of the host CPU in the future.

But perhaps we can formulate the API, to say only the (currently empty) following list of features can only be enabled if the corresponding bit is enabled, or something along those lines.

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

Quoted 5 lines
Quoted 4 lines

Do we stick with current implementation of returning struct kvm_vcpu_init ? OR
Do we return struct kvm_vcpu_init with all features set to zero ?

Let's give Marc a day or two to respond to this one ;)

Are you implying I'm getting slow? ;-)

Quoted 7 lines

To answer Anup's question, I would tend to be cautious, and not expose things for which we already have another API in place. So far, we've stuck with the KVM approach of having a capability bit for each feature we enable, and I'm quite happy with that.

So I'm in favour of Christoffer's proposal to return an empty feature set, and start adding stuff if/when the need arises.

Agreed, I will send revised patch where we return zeroed-out features in struct kvm_vcpu_init (for now).

Quoted 4 lines

Cheers,

        M.
Quoted 8 lines

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

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Regards,
Anup

#8 lore raw

PATCH v3 1/4 ARM: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 72 lines · Anup Patel
Quoted 71 lines · Marc Zyngier
Quoted 70 lines · Christoffer Dall
Quoted 69 lines · Anup Patel

Hi Christoffer/Marc,

Quoted 64 lines · Christoffer Dall
Quoted 41 lines · Marc Zyngier
Quoted 25 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/kvm/guest.c              |   20 ++++++++++++++++++++
 arch/arm64/include/asm/kvm_host.h |    1 +
 2 files changed, 21 insertions(+)
arch/arm/kvm/guest.c +6 -0
index 152d036..b407e6c 100644
@@ -222,6 +222,26 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
    return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+   int target = kvm_target_cpu();
+
+   if (target < 0)
+           return -ENODEV;

arch/arm/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 240

Quoted 10 lines
arch/arm/kvm/guest.c +10 -0
+
+   memset(init, 0, sizeof(*init));
+
+   /*
+    * For now, we return all optional features are available
+    * for preferred target. In future, we might have features
+    * available based on underlying host.
+    */
+   init->target = (__u32)target;
+   init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);

I'm in two minds about this feature reporting. I see they serve a purpose, but they also duplicate capabilities, which is the standard way to advertise what KVM can do.

Quoted 8 lines

It means we end up having to sync two reporting mechanism, and I feel this is in general a bad idea.

Furthermore, KVM_ARM_VCPU_POWER_OFF is hardly a feature of the HW, but rather a firmware emulation thing.

Peter, Christoffer: Thoughts?

I wanted to return the full kvm_vcpu_init instead of just a target int, so we did not have to come up with yet another ioctl if we need to return more information about the capabilities of the host CPU in the future.

But perhaps we can formulate the API, to say only the (currently empty) following list of features can only be enabled if the corresponding bit is enabled, or something along those lines.

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

Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

Do we stick with current implementation of returning struct kvm_vcpu_init ? OR
Do we return struct kvm_vcpu_init with all features set to zero ?

Let's give Marc a day or two to respond to this one ;)

Are you implying I'm getting slow? ;-)

Wouldn't dream of it.

Quoted 6 lines
Quoted 6 lines

To answer Anup's question, I would tend to be cautious, and not expose things for which we already have another API in place. So far, we've stuck with the KVM approach of having a capability bit for each feature we enable, and I'm quite happy with that.

Quoted 6 lines
Quoted 2 lines

So I'm in favour of Christoffer's proposal to return an empty feature set, and start adding stuff if/when the need arises.

Agreed, I will send revised patch where we return zeroed-out features in struct kvm_vcpu_init (for now).

Sounds good, also remember to address this specific item in the API documentation.

-Christoffer

author
#9 lore raw

PATCH v3 2/4 ARM64: KVM: Implement kvm_vcpu_preferred_target() function

In reply to Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM64 which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/include/asm/kvm_host.h |    1 +
 arch/arm64/kvm/guest.c          |   21 +++++++++++++++++++++
 2 files changed, 22 insertions(+)
arch/arm/include/asm/kvm_host.h +1 -0
index 7d22517..76f3c19 100644
@@ -154,6 +154,7 @@ struct kvm_vcpu_stat {
 struct kvm_vcpu_init;
 int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 			const struct kvm_vcpu_init *init);
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init);
 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
 struct kvm_one_reg;
arch/arm64/kvm/guest.c +21 -0
index d7bf7d6..4da7882 100644
@@ -254,6 +254,27 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 	return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+	int target = kvm_target_cpu();
+
+	if (target < 0)
+		return -ENODEV;
+
+	memset(init, 0, sizeof(*init));
+
+	/* 
+	 * For now, we return all optional features are available
+	 * for preferred target. In future, we might have features
+	 * available based on underlying host.
+	 */
+	init->target = (__u32)target;
+	init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);
+	init->features[0] |= (1 << KVM_ARM_VCPU_EL1_32BIT);
+
+	return 0;
+}
+
 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
 	return -EINVAL;
Signature · 1 line
1.7.9.5
#10 lore raw

PATCH v3 2/4 ARM64: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 8 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM64 which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/include/asm/kvm_host.h |    1 +

This hunk belongs to patch #1, and the similar hunk in #1 belongs here.

	M.
Quoted 48 lines
 arch/arm64/kvm/guest.c          |   21 +++++++++++++++++++++
 2 files changed, 22 insertions(+)
arch/arm/include/asm/kvm_host.h +1 -0
index 7d22517..76f3c19 100644
@@ -154,6 +154,7 @@ struct kvm_vcpu_stat {
 struct kvm_vcpu_init;
 int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 			const struct kvm_vcpu_init *init);
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init);
 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
 struct kvm_one_reg;
arch/arm64/kvm/guest.c +21 -0
index d7bf7d6..4da7882 100644
@@ -254,6 +254,27 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 	return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+	int target = kvm_target_cpu();
+
+	if (target < 0)
+		return -ENODEV;
+
+	memset(init, 0, sizeof(*init));
+
+	/* 
+	 * For now, we return all optional features are available
+	 * for preferred target. In future, we might have features
+	 * available based on underlying host.
+	 */
+	init->target = (__u32)target;
+	init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);
+	init->features[0] |= (1 << KVM_ARM_VCPU_EL1_32BIT);
+
+	return 0;
+}
+
 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
 	return -EINVAL;
Signature · 1 line
Jazz is not dead. It just smells funny...
#11 lore raw

PATCH v3 2/4 ARM64: KVM: Implement kvm_vcpu_preferred_target() function

Quoted 11 lines · Marc Zyngier
Quoted 8 lines · Anup Patel

This patch implements kvm_vcpu_preferred_target() function for KVM ARM64 which will help us implement KVM_ARM_PREFERRED_TARGET ioctl for user space.

---
 arch/arm/include/asm/kvm_host.h |    1 +

This hunk belongs to patch #1, and the similar hunk in #1 belongs here.

UHH, How did I miss this one !!!

Thanks for pointing.

--Anup

Quoted 61 lines
        M.
Quoted 38 lines
 arch/arm64/kvm/guest.c          |   21 +++++++++++++++++++++
 2 files changed, 22 insertions(+)
arch/arm/include/asm/kvm_host.h +1 -0
index 7d22517..76f3c19 100644
@@ -154,6 +154,7 @@ struct kvm_vcpu_stat {
 struct kvm_vcpu_init;
 int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
                      const struct kvm_vcpu_init *init);
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init);
 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
 struct kvm_one_reg;
arch/arm64/kvm/guest.c +14 -0
index d7bf7d6..4da7882 100644
@@ -254,6 +254,27 @@ int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
      return kvm_reset_vcpu(vcpu);
 }
 
+int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)
+{
+     int target = kvm_target_cpu();
+
+     if (target < 0)
+             return -ENODEV;
+
+     memset(init, 0, sizeof(*init));
+
+     /*
+      * For now, we return all optional features are available
+      * for preferred target. In future, we might have features
+      * available based on underlying host.
+      */

arch/arm64/kvm/guest.c int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 280

Quoted 10 lines
arch/arm64/kvm/guest.c +7 -0
+     init->target = (__u32)target;
+     init->features[0] |= (1 << KVM_ARM_VCPU_POWER_OFF);
+     init->features[0] |= (1 << KVM_ARM_VCPU_EL1_32BIT);
+
+     return 0;
+}
+
 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
      return -EINVAL;

--
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

author
#12 lore raw

PATCH v3 3/4 ARM/ARM64: KVM: Implement KVM_ARM_PREFERRED_TARGET ioctl

In reply to Anup Patel

For implementing CPU=host, we need a mechanism for querying preferred VCPU target type on underlying Host.

This patch implements KVM_ARM_PREFERRED_TARGET vm ioctl which returns struct kvm_vcpu_init instance containing information about preferred VCPU target type and available optional VCPU features to user space.

---
 arch/arm/kvm/arm.c       |   13 +++++++++++++
 include/uapi/linux/kvm.h |    1 +
 2 files changed, 14 insertions(+)
arch/arm/kvm/arm.c +13 -0
index 9c697db..cc5adb9 100644
@@ -797,6 +797,19 @@ long kvm_arch_vm_ioctl(struct file *filp,
 			return -EFAULT;
 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 	}
+	case KVM_ARM_PREFERRED_TARGET: {
+		int err;
+		struct kvm_vcpu_init init;
+
+		err = kvm_vcpu_preferred_target(&init);
+		if (err)
+			return err;
+
+		if (copy_to_user(argp, &init, sizeof(init)))
+			return -EFAULT;
+
+		return 0;
+	}
 	default:
 		return -EINVAL;
 	}
include/uapi/linux/kvm.h +1 -0
index 99c2533..e32e776 100644
@@ -1012,6 +1012,7 @@ struct kvm_s390_ucas_mapping {
 /* VM is being stopped by host */
 #define KVM_KVMCLOCK_CTRL	  _IO(KVMIO,   0xad)
 #define KVM_ARM_VCPU_INIT	  _IOW(KVMIO,  0xae, struct kvm_vcpu_init)
+#define KVM_ARM_PREFERRED_TARGET  _IOR(KVMIO,  0xaf, struct kvm_vcpu_init)
 #define KVM_GET_REG_LIST	  _IOWR(KVMIO, 0xb0, struct kvm_reg_list)
 
 #define KVM_DEV_ASSIGN_ENABLE_IOMMU	(1 << 0)
Signature · 1 line
1.7.9.5
author
#13 lore raw

PATCH v3 4/4 KVM: Add documentation for KVM_ARM_PREFERRED_TARGET ioctl

In reply to Anup Patel

To implement CPU=Host we have added KVM_ARM_PREFERRED_TARGET vm ioctl which provides information to user space required for creating VCPU matching underlying Host.

This patch adds info related to this new KVM_ARM_PREFERRED_TARGET vm ioctl in the KVM API documentation.

---
 Documentation/virtual/kvm/api.txt |   26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)
Documentation/virtual/kvm/api.txt +22 -4
index 858aecf..d513cc5 100644
@@ -2303,8 +2303,27 @@ Possible features:
 	- KVM_ARM_VCPU_EL1_32BIT: Starts the CPU in a 32bit mode.
 	  Depends on KVM_CAP_ARM_EL1_32BIT (arm64 only).
 
+4.83 KVM_ARM_PREFERRED_TARGET
 
-4.83 KVM_GET_REG_LIST
+Capability: basic
+Architectures: arm, arm64
+Type: vm ioctl
+Parameters: struct struct kvm_vcpu_init (out)
+Returns: 0 on success; -1 on error
+Errors:
+ ?ENODEV: ???no preferred target available for the host
+
+This queries KVM for preferred CPU target type which can be emulated
+by KVM on underlying host.
+
+The ioctl returns struct kvm_vcpu_init instance containing information
+about preferred CPU target type and optional features available for it.
+
+The information returned by this ioctl can be used to prepare instance
+of struct kvm_vcpu_init for KVM_ARM_VCPU_INIT ioctl which will result
+in VCPU matching underlying host.
+
+4.84 KVM_GET_REG_LIST
 
 Capability: basic
 Architectures: arm, arm64
@@ -2323,8 +2342,7 @@ struct kvm_reg_list {
 This ioctl returns the guest registers that are supported for the
 KVM_GET_ONE_REG/KVM_SET_ONE_REG calls.
 
-
-4.84 KVM_ARM_SET_DEVICE_ADDR
+4.85 KVM_ARM_SET_DEVICE_ADDR
 
 Capability: KVM_CAP_ARM_SET_DEVICE_ADDR
 Architectures: arm, arm64
@@ -2362,7 +2380,7 @@ must be called after calling KVM_CREATE_IRQCHIP, but before calling
 KVM_RUN on any of the VCPUs.  Calling this ioctl twice for any of the
 base addresses will return -EEXIST.
 
-4.85 KVM_PPC_RTAS_DEFINE_TOKEN
+4.86 KVM_PPC_RTAS_DEFINE_TOKEN
 
 Capability: KVM_CAP_PPC_RTAS
 Architectures: ppc
Signature · 1 line
1.7.9.5