KVM: arm64: Consistently use 'generic' vCPU target across all uarches
cover KVM: arm64: Consistently use 'generic' vCPU target across all uarches
kvm_vcpu_init::target is quite useless at this point. We don't do any uarch-specific emulation in the first place, and require userspace select the 'generic' vCPU target on all but a few implementations.
Small series to (1) clean up usage of the target value in the kernel and
(2) switch to the 'generic' target on implementations that previously had their own target values. The implementation-specific values are still tolerated, though, to avoid UAPI breakage.
Oliver Upton (4):
KVM: arm64: Delete pointless switch statement in kvm_reset_vcpu()
KVM: arm64: Remove pointless check for changed init target
KVM: arm64: Replace vCPU target with a configuration flag
KVM: arm64: Always return generic v8 as the preferred target
arch/arm64/include/asm/kvm_host.h | 5 +++-- arch/arm64/kvm/arm.c | 18 ++++++++---------- arch/arm64/kvm/guest.c | 4 +--- arch/arm64/kvm/hyp/nvhe/switch.c | 2 +- arch/arm64/kvm/reset.c | 25 ++++++++++--------------- 5 files changed, 23 insertions(+), 31 deletions(-)
base-commit: 192df2aa0113ddddee2a93e453ff46610807b425
Signature · 1 line
2.41.0.178.g377b9f9a00-goog
PATCH 1/4 KVM: arm64: Delete pointless switch statement in kvm_reset_vcpu()
The vCPU target hasn't mattered for quite a long time now. Delete the useless switch statement in kvm_reset_vcpu(), which hilariously only had a default case in it.
--- arch/arm64/kvm/reset.c | 25 ++++++++++--------------- 1 file changed, 10 insertions(+), 15 deletions(-)
arch/arm64/kvm/reset.c +10 -15
@@ -248,21 +248,16 @@ int kvm_reset_vcpu(struct kvm_vcpu *vcpu) } } -switch (vcpu->arch.target) { -default : -if (vcpu_el1_is_32bit (vcpu)) { - pstate =VCPU_RESET_PSTATE_SVC ; - }else if (vcpu_has_nv (vcpu)) { - pstate =VCPU_RESET_PSTATE_EL2 ; - }else { - pstate =VCPU_RESET_PSTATE_EL1 ; - } - -if (kvm_vcpu_has_pmu (vcpu) && !kvm_arm_support_pmu_v3 ()) { - ret = -EINVAL ; - goto out; - } -break ; +if (vcpu_el1_is_32bit (vcpu)) + pstate =VCPU_RESET_PSTATE_SVC ; +else if (vcpu_has_nv (vcpu)) + pstate =VCPU_RESET_PSTATE_EL2 ; +else + pstate =VCPU_RESET_PSTATE_EL1 ; + +if (kvm_vcpu_has_pmu (vcpu) && !kvm_arm_support_pmu_v3 ()) { + ret = -EINVAL ; + goto out; }/* Reset core registers */
Signature · 1 line
2.41.0.178.g377b9f9a00-goog
Re: [PATCH 1/4] KVM: arm64: Delete pointless switch statement in kvm_reset_vcpu()
Quoted 5 lines · Oliver Upton
The vCPU target hasn't mattered for quite a long time now. Delete the useless switch statement in
kvm_reset_vcpu(), which hilariously only had a default case in it.Signed-off-by Oliver Upton <[email protected]>
PATCH 2/4 KVM: arm64: Remove pointless check for changed init target
In reply to Oliver UptonAt any time there is only a single valid value for KVM_ARM_VCPU_INIT, depending on the current CPU implementation. In all likelihood, this will be the generic ARMv8 target.
Drop the pointless check for a changed target value between calls to KVM_ARM_VCPU_INIT and instead rely on the check against kvm_target_cpu().
--- arch/arm64/kvm/arm.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
arch/arm64/kvm/arm.c +1 -2
@@ -1212,8 +1212,7 @@ static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu, {unsigned long features = init->features[0 ]; -return !bitmap_equal (vcpu->arch.features, &features,KVM_VCPU_MAX_FEATURES ) || - vcpu->arch.target != init->target; +return !bitmap_equal (vcpu->arch.features, &features,KVM_VCPU_MAX_FEATURES ); } static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
Signature · 1 line
2.41.0.178.g377b9f9a00-goog
PATCH 3/4 KVM: arm64: Replace vCPU target with a configuration flag
In reply to Oliver UptonThe value of kvm_vcpu_arch::target has been used to determine if a vCPU has actually been initialized. Storing this as an integer is needless at this point, as KVM doesn't do any microarch-specific emulation in the first place. Instead, all we care about is whether or not the vCPU has been initialized.
Delete the field in favor of a vCPU configuration flag indicating if KVM_ARM_VCPU_INIT has completed for the vCPU.
--- arch/arm64/include/asm/kvm_host.h | 5 +++-- arch/arm64/kvm/arm.c | 12 +++++------- arch/arm64/kvm/hyp/nvhe/switch.c | 2 +- 3 files changed, 9 insertions(+), 10 deletions(-)
arch/arm64/include/asm/kvm_host.h +3 -2
@@ -562,8 +562,7 @@ struct kvm_vcpu_arch {/* Cache some mmu pages needed inside spinlock regions */ struct kvm_mmu_memory_cache mmu_page_cache; -/* Target CPU and feature flags */ -int target; +/* feature flags */ DECLARE_BITMAP (features,KVM_VCPU_MAX_FEATURES );/* Virtual SError ESR to restore when HCR_EL2.VSE is set */
@@ -664,6 +663,8 @@ struct kvm_vcpu_arch {#define VCPU_SVE_FINALIZED __vcpu_single_flag(cflags, BIT(1))/* PTRAUTH exposed to guest */ #define GUEST_HAS_PTRAUTH __vcpu_single_flag(cflags, BIT(2)) +/* KVM_ARM_VCPU_INIT completed */ +#define VCPU_INITIALIZED __vcpu_single_flag(cflags, BIT(3))/* Exception pending */ #define PENDING_EXCEPTION __vcpu_single_flag(iflags, BIT(0))
arch/arm64/kvm/arm.c +5 -7
@@ -360,7 +360,7 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)#endif /* Force users to call KVM_ARM_VCPU_INIT */ - vcpu->arch.target =-1 ; +vcpu_clear_flag (vcpu,VCPU_INITIALIZED );bitmap_zero (vcpu->arch.features,KVM_VCPU_MAX_FEATURES ); vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
@@ -569,7 +569,7 @@ unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)static int kvm_vcpu_initialized (struct kvm_vcpu *vcpu) { -return vcpu->arch.target >=0 ; +return vcpu_get_flag (vcpu,VCPU_INITIALIZED ); }/*
@@ -1051,7 +1051,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)* invalid. The VMM can try and fix it by issuing a * KVM_ARM_VCPU_INIT if it really wants to. */ - vcpu->arch.target =-1 ; +vcpu_clear_flag (vcpu,VCPU_INITIALIZED ); ret =ARM_EXCEPTION_IL ; }
@@ -1228,20 +1228,18 @@ static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu, !bitmap_equal(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES)) goto out_unlock; - vcpu->arch.target = init->target;bitmap_copy (vcpu->arch.features, &features,KVM_VCPU_MAX_FEATURES );/* Now we know what it is, we can reset it. */ ret =kvm_reset_vcpu (vcpu);if (ret) { - vcpu->arch.target =-1 ;bitmap_zero (vcpu->arch.features,KVM_VCPU_MAX_FEATURES ); goto out_unlock; }bitmap_copy (kvm->arch.vcpu_features, &features,KVM_VCPU_MAX_FEATURES );set_bit (KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED , &kvm->arch.flags); - +vcpu_set_flag (vcpu,VCPU_INITIALIZED ); out_unlock:mutex_unlock (&kvm->arch.config_lock);return ret;
@@ -1259,7 +1257,7 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,if (ret)return ret; -if (vcpu->arch.target ==-1 ) +if (!vcpu_get_flag (vcpu,VCPU_INITIALIZED ))return __kvm_vcpu_set_target (vcpu, init);if (kvm_vcpu_init_changed (vcpu, init))
arch/arm64/kvm/hyp/nvhe/switch.c +1 -1
@@ -234,7 +234,7 @@ static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code)* KVM_ARM_VCPU_INIT, however, this is likely not possible for * protected VMs. */ - vcpu->arch.target =-1 ; +vcpu_clear_flag (vcpu,VCPU_INITIALIZED ); *exit_code &=BIT (ARM_EXIT_WITH_SERROR_BIT ); *exit_code |=ARM_EXCEPTION_IL ; }
Signature · 1 line
2.41.0.178.g377b9f9a00-goog
Re: [PATCH 3/4] KVM: arm64: Replace vCPU target with a configuration flag
Quoted 10 lines · Oliver Upton
The value of
kvm_vcpu_arch::targethas been used to determine if a vCPU has actually been initialized. Storing this as an integer is needless at this point, as KVM doesn't do any microarch-specific emulation in the first place. Instead, all we care about is whether or not the vCPU has been initialized.Delete the field in favor of a vCPU configuration flag indicating if
KVM_ARM_VCPU_INIThas completed for the vCPU.Signed-off-by Oliver Upton <[email protected]>
[...]
arch/arm64/kvm/arm.c static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 1260
Quoted 6 lines
arch/arm64/kvm/arm.c +1
-1@@ -1259,7 +1257,7 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,if (ret)return ret; -if (vcpu->arch.target ==-1 ) +if (!vcpu_get_flag (vcpu,VCPU_INITIALIZED ))
if (!kvm_vcpu_initialized(vcpu)) ?
Regardless,
PATCH 4/4 KVM: arm64: Always return generic v8 as the preferred target
In reply to Oliver UptonUserspace selecting an implementation-specific vCPU target has been completely useless for a very long time. Let's go whole hog and start returning the generic v8 target across all implementations as the preferred target.
Uphold the pre-existing behavior by tolerating either the generic target or an implementation-specific target if the vCPU happens to be running on one of the lucky few parts.
--- arch/arm64/kvm/arm.c | 3 ++- arch/arm64/kvm/guest.c | 4 +--- 2 files changed, 3 insertions(+), 4 deletions(-)
arch/arm64/kvm/arm.c +2 -1
@@ -1250,7 +1250,8 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, {int ret; -if (init->target !=kvm_target_cpu ()) +if (init->target !=KVM_ARM_TARGET_GENERIC_V8 && + init->target !=kvm_target_cpu ())return -EINVAL ; ret =kvm_vcpu_init_check_features (vcpu, init);
arch/arm64/kvm/guest.c +1 -3
@@ -886,8 +886,6 @@ u32 __attribute_const__ kvm_target_cpu(void)void kvm_vcpu_preferred_target (struct kvm_vcpu_init *init) { -u32 target =kvm_target_cpu (); -memset (init,0 ,sizeof (*init));/*
@@ -896,7 +894,7 @@ void kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)* specific features available for the preferred * target type. */ - init->target = (__u32 )target; + init->target =KVM_ARM_TARGET_GENERIC_V8 ; }int kvm_arch_vcpu_ioctl_get_fpu (struct kvm_vcpu *vcpu,struct kvm_fpu *fpu)
Signature · 1 line
2.41.0.178.g377b9f9a00-goog
Marc Zyngier
maintainerRe: [PATCH 4/4] KVM: arm64: Always return generic v8 as the preferred target
Quoted 44 lines · Oliver Upton
Userspace selecting an implementation-specific vCPU target has been completely useless for a very long time. Let's go whole hog and start returning the generic v8 target across all implementations as the preferred target.
Uphold the pre-existing behavior by tolerating either the generic target or an implementation-specific target if the vCPU happens to be running on one of the lucky few parts.
Signed-off-by Oliver Upton <[email protected]>--- arch/arm64/kvm/arm.c | 3 ++- arch/arm64/kvm/guest.c | 4 +--- 2 files changed, 3 insertions(+), 4 deletions(-)arch/arm64/kvm/arm.c +2
-1@@ -1250,7 +1250,8 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, {int ret; -if (init->target !=kvm_target_cpu ()) +if (init->target !=KVM_ARM_TARGET_GENERIC_V8 && + init->target !=kvm_target_cpu ())return -EINVAL ; ret =kvm_vcpu_init_check_features (vcpu, init);arch/arm64/kvm/guest.c +0
-2@@ -886,8 +886,6 @@ u32 __attribute_const__ kvm_target_cpu(void)void kvm_vcpu_preferred_target (struct kvm_vcpu_init *init) { -u32 target =kvm_target_cpu (); -memset (init,0 ,sizeof (*init));/*
arch/arm64/kvm/guest.c void kvm_vcpu_preferred_target(struct kvm_vcpu_init *init) · line 898
Quoted 6 lines
arch/arm64/kvm/guest.c +1
-1@@ -896,7 +894,7 @@ void kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)* specific features available for the preferred * target type. */ - init->target = (__u32 )target; + init->target =KVM_ARM_TARGET_GENERIC_V8 ; }
Should we take the opportunity to simply get rid of this function altogether? Something like the untested hack below.
Thanks,
M.
arch/arm64/include/asm/kvm_host.h +0 -1
@@ -890,7 +890,6 @@ struct kvm_vcpu_stat {u64 exits; }; -void 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);int kvm_arm_get_reg (struct kvm_vcpu *vcpu,const struct kvm_one_reg *reg);
arch/arm64/kvm/arm.c +3 -3
@@ -1588,9 +1588,9 @@ int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)return kvm_vm_ioctl_set_device_addr (kvm, &dev_addr); }case KVM_ARM_PREFERRED_TARGET : { -struct kvm_vcpu_init init; - -kvm_vcpu_preferred_target (&init); +struct kvm_vcpu_init init = { + .target = KVM_ARM_TARGET_GENERIC_V8; + };if (copy_to_user (argp, &init,sizeof (init)))return -EFAULT ;
arch/arm64/kvm/guest.c +0 -15
@@ -884,21 +884,6 @@ u32 __attribute_const__ kvm_target_cpu(void)return KVM_ARM_TARGET_GENERIC_V8 ; } -void kvm_vcpu_preferred_target (struct kvm_vcpu_init *init) -{ -u32 target =kvm_target_cpu (); - -memset (init,0 ,sizeof (*init)); - -/* -* For now, we don't return any features. -* In future, we might use features to return target -* specific features available for the preferred -* target type. -*/ - init->target = (__u32 )target; -} -int kvm_arch_vcpu_ioctl_get_fpu (struct kvm_vcpu *vcpu,struct kvm_fpu *fpu) {return -EINVAL ;
Signature · 1 line
Without deviation from the norm, progress is not possible.
Re: [PATCH 4/4] KVM: arm64: Always return generic v8 as the preferred target
Quoted 44 lines · Marc Zyngier
Quoted 42 lines · Oliver Upton
Userspace selecting an implementation-specific vCPU target has been completely useless for a very long time. Let's go whole hog and start returning the generic v8 target across all implementations as the preferred target.
Uphold the pre-existing behavior by tolerating either the generic target or an implementation-specific target if the vCPU happens to be running on one of the lucky few parts.
Signed-off-by Oliver Upton <[email protected]>--- arch/arm64/kvm/arm.c | 3 ++- arch/arm64/kvm/guest.c | 4 +--- 2 files changed, 3 insertions(+), 4 deletions(-)arch/arm64/kvm/arm.c +2
-1@@ -1250,7 +1250,8 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, {int ret; -if (init->target !=kvm_target_cpu ()) +if (init->target !=KVM_ARM_TARGET_GENERIC_V8 && + init->target !=kvm_target_cpu ())return -EINVAL ; ret =kvm_vcpu_init_check_features (vcpu, init);arch/arm64/kvm/guest.c +0
-2@@ -886,8 +886,6 @@ u32 __attribute_const__ kvm_target_cpu(void)void kvm_vcpu_preferred_target (struct kvm_vcpu_init *init) { -u32 target =kvm_target_cpu (); -memset (init,0 ,sizeof (*init));
Quoted 11 lines
arch/arm64/kvm/guest.c void kvm_vcpu_preferred_target(struct kvm_vcpu_init *init) · line 898
Quoted 8 lines
arch/arm64/kvm/guest.c +1
-1/* @@ -896,7 +894,7 @@ void kvm_vcpu_preferred_target(struct kvm_vcpu_init *init)* specific features available for the preferred * target type. */ - init->target = (__u32 )target; + init->target =KVM_ARM_TARGET_GENERIC_V8 ; }Should we take the opportunity to simply get rid of this function altogether? Something like the untested hack below.
Yeah, I should've thought about it a little more and done as you suggest. Thanks for the diff, I'll use it in v2.
--
Thanks,
Oliver
Re: [PATCH 4/4] KVM: arm64: Always return generic v8 as the preferred target
In reply to Oliver UptonHi Oliver,
Quoted 10 lines · Oliver Upton
Userspace selecting an implementation-specific vCPU target has been completely useless for a very long time. Let's go whole hog and start returning the generic v8 target across all implementations as the preferred target.
Uphold the pre-existing behavior by tolerating either the generic target or an implementation-specific target if the vCPU happens to be running on one of the lucky few parts.
Signed-off-by Oliver Upton <[email protected]>
I love the idea.
Quoted 10 lines
--- arch/arm64/kvm/arm.c | 3 ++- arch/arm64/kvm/guest.c | 4 +--- 2 files changed, 3 insertions(+), 4 deletions(-)arch/arm64/kvm/arm.c +0
-0
arch/arm64/kvm/arm.c static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 1255
Quoted 7 lines
arch/arm64/kvm/arm.c +2
-1@@ -1250,7 +1250,8 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, {int ret; -if (init->target !=kvm_target_cpu ()) +if (init->target !=KVM_ARM_TARGET_GENERIC_V8 && + init->target !=kvm_target_cpu ())return -EINVAL ;
I can see that user-space can now select GENERIC_V8 or CORTEX_A53 (for example) between two KVM_ARM_VCPU_INIT calls on an A53 host, which was forbidden before this change. I don't think it's a big deal though.
Thanks,
Zenghui
Re: [PATCH 4/4] KVM: arm64: Always return generic v8 as the preferred target
Quoted 12 lines · Zenghui Yu
Hi Oliver,
Quoted 9 lines · Oliver Upton
Userspace selecting an implementation-specific vCPU target has been completely useless for a very long time. Let's go whole hog and start returning the generic v8 target across all implementations as the preferred target.
Uphold the pre-existing behavior by tolerating either the generic target or an implementation-specific target if the vCPU happens to be running on one of the lucky few parts.
Quoted 5 lines
Quoted 1 line
Signed-off-by Oliver Upton <[email protected]>I love the idea.
Acked-by Zenghui Yu <[email protected]>
Thanks!
Quoted 5 lines
Quoted 5 lines
--- arch/arm64/kvm/arm.c | 3 ++- arch/arm64/kvm/guest.c | 4 +--- 2 files changed, 3 insertions(+), 4 deletions(-)
Quoted 15 lines
arch/arm64/kvm/arm.c static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, · line 1254
Quoted 11 lines
arch/arm64/kvm/arm.c +2
-1@@ -1250,7 +1250,8 @@ static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, {int ret; -if (init->target !=kvm_target_cpu ()) +if (init->target !=KVM_ARM_TARGET_GENERIC_V8 && + init->target !=kvm_target_cpu ())return -EINVAL ;I can see that user-space can now select
GENERIC_V8orCORTEX_A53(for example) between twoKVM_ARM_VCPU_INITcalls on an A53 host, which was forbidden before this change. I don't think it's a big deal though.
Agreed, I thought the slight relaxation on invariance was OK. If needed, we can always change the rules for a new target value, but I doubt we will need to do that anytime soon.
Signature · 2 lines
Thanks, Oliver