KVM: arm64: Fix unguarded GICv5 CPU interface accesses
Fuad Tabba
reviewerv2 cover KVM: arm64: Fix unguarded GICv5 CPU interface accesses
Hi folks,
Changes since v1 [1]:
- Reject anything that is not VGIC_V3 rather than listing the models, which drops the VGIC_V2 case. (Sascha)
- Tightened the UNDEF sentence in patch 2's commit message. No code change.
- Collected Sascha's Reviewed-by on patches 2 to 4, thanks for the reviews.
- Rebased onto Linux 7.2.
This series stops KVM reaching GICv5 CPU interface registers on hardware that does not implement them, in three places with no guard.
Under pKVM the first two are reachable from an untrusted host. EL2 copies vgic_model out of the host's struct kvm without validating it, and the nVHE world switch dispatches on that field with no cpucap guard, so a host writing KVM_DEV_TYPE_ARM_VGIC_V5 steers EL2 into ICC_ICSR_EL1 and the ICH_PPI_* registers. Separately, __vgic_v5_save_apr and __vgic_v5_restore_vmcr_apr sit in the hypercall band the de-privileged host may still call, and pKVM never registers a GICv5 vgic, so neither has a valid caller in protected mode. Without FEAT_GCIE those registers are UNDEFINED at EL2, so either path panics the hypervisor. Both need a compromised host kernel rather than host userspace, so this is hardening and not a guest-reachable hole.
I had said these paths were unreachable under pKVM because vgic_v5_probe() skips GICv5 registration in protected mode [2]. That was wrong. The skip is host-side only, and does not constrain what a malicious host can call.
The third one is not pKVM. can_access_vgic_from_kernel() excludes only the GICv3 system register interface, so on a native GICv5 system without FEAT_GCIE_LEGACY the kernel reaches EL2-only registers from EL1 under nVHE, and the world switch does the same work at EL2 anyway.
The last patch drops the VGICv3 reference from two nVHE world switch comments that cover GICv5 too. No functional change.
Tested on QEMU. I also checked the first one with a local host patch that hands EL2 a GICv5 model: it panics at __vgic_v5_restore_state before the series and boots cleanly after.
Based on Linux 7.2 (8d3ae59288f1e). It also applies cleanly to kvmarm/next and kvmarm/fixes.
Cheers,
/fuad
[1] https://lore.kernel.org/all/[email protected]/
[2] https://lore.kernel.org/all/CA%2BEHjTyGULmVCgyoya3bXG4gRj0OYFE1gnJLhNE6kvCrZFtXyQ@mail.gmail.com/
Fuad Tabba (4):
KVM: arm64: Validate the host-provided vgic model in pKVM
KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM
KVM: arm64: vgic: Do not access the GICv5 CPU interface from EL1
KVM: arm64: Fix stale VGICv3 comments in the nVHE world switch
arch/arm64/kvm/hyp/nvhe/hyp-main.c | 6 ++++++ arch/arm64/kvm/hyp/nvhe/pkvm.c | 11 +++++++++-- arch/arm64/kvm/hyp/nvhe/switch.c | 4 ++-- arch/arm64/kvm/vgic/vgic.c | 12 ++++++++---- 4 files changed, 25 insertions(+), 8 deletions(-)
Signature · 1 line
2.39.5
Fuad Tabba
reviewerPATCH v2 1/4 KVM: arm64: Validate the host-provided vgic model in pKVM
EL2 copies vgic_model from the host's struct kvm unchecked, and the nVHE world switch dispatches on it with no cpucap guard. A host writing KVM_DEV_TYPE_ARM_VGIC_V5 makes EL2 access GICv5 CPU interface registers, which are UNDEFINED without FEAT_GCIE and panic the hypervisor on any GICv3 machine.
Accept only GICv3, the one model pKVM supports, forcing anything else to 0.
--- arch/arm64/kvm/hyp/nvhe/pkvm.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
arch/arm64/kvm/hyp/nvhe/pkvm.c +9 -2
@@ -340,13 +340,20 @@ static void pkvm_init_features_from_host(struct pkvm_hyp_vm *hyp_vm, const struc {struct kvm *kvm = &hyp_vm->kvm;unsigned long host_arch_flags =READ_ONCE (host_kvm->arch.flags); +u32 vgic_model =READ_ONCE (host_kvm->arch.vgic.vgic_model);DECLARE_BITMAP (allowed_features,KVM_VCPU_MAX_FEATURES );/* CTR_EL0 is always under host control, even for protected VMs. */ hyp_vm->kvm.arch.ctr_el0 = host_kvm->arch.ctr_el0; -/* Preserve the vgic model so that GICv3 emulation works */ - hyp_vm->kvm.arch.vgic.vgic_model = host_kvm->arch.vgic.vgic_model; +/* +* GICv3 is the only model pKVM runs, and the GICv5 world switch +* touches registers UNDEFINED at EL2 without FEAT_GCIE. 0 is not a +* valid kvm_device_type: "no vgic". +*/ +if (vgic_model !=KVM_DEV_TYPE_ARM_VGIC_V3 ) + vgic_model =0 ; + hyp_vm->kvm.arch.vgic.vgic_model = vgic_model;/* No restrictions for non-protected VMs. */ if (!kvm_vm_is_protected (kvm)) {
Signature · 1 line
2.39.5
Fuad Tabba
reviewerPATCH v2 2/4 KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM
In reply to Fuad Tabba__vgic_v5_save_apr() and __vgic_v5_restore_vmcr_apr() remain callable after pKVM finalises. pKVM never registers a GICv5 vgic, so neither has a valid caller in protected mode, and on a GICv3 machine both reach a register that is UNDEFINED at EL2 and panic the hypervisor.
Reject both when protected mode is enabled.
--- arch/arm64/kvm/hyp/nvhe/hyp-main.c | 6 ++++++ 1 file changed, 6 insertions(+)
arch/arm64/kvm/hyp/nvhe/hyp-main.c +6 -0
@@ -695,6 +695,9 @@ static void handle___vgic_v5_save_apr(struct kvm_cpu_context *host_ctxt) { DECLARE_REG(struct vgic_v5_cpu_if *, cpu_if, host_ctxt, 1); +if (unlikely (is_protected_kvm_enabled ())) +return ; +__vgic_v5_save_apr (kern_hyp_va (cpu_if)); }
@@ -702,6 +705,9 @@ static void handle___vgic_v5_restore_vmcr_apr(struct kvm_cpu_context *host_ctxt) { DECLARE_REG(struct vgic_v5_cpu_if *, cpu_if, host_ctxt, 1); +if (unlikely (is_protected_kvm_enabled ())) +return ; +__vgic_v5_restore_vmcr_apr (kern_hyp_va (cpu_if)); }
Signature · 1 line
2.39.5
Fuad Tabba
reviewerPATCH v2 3/4 KVM: arm64: vgic: Do not access the GICv5 CPU interface from EL1
In reply to Fuad Tabbacan_access_vgic_from_kernel() excludes only the GICv3 system register interface, so on a native GICv5 system without FEAT_GCIE_LEGACY it returns true under nVHE. The kernel then saves and restores the CPU interface from EL1, where ICH_VMCR_EL2 and the ICH_PPI_* registers are UNDEFINED, and the nVHE world switch already does that work at EL2.
Require VHE for GICv5 as for GICv3.
--- arch/arm64/kvm/vgic/vgic.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)
arch/arm64/kvm/vgic/vgic.c +8 -4
@@ -1042,11 +1042,15 @@ static void vgic_flush_lr_state(struct kvm_vcpu *vcpu)static inline bool can_access_vgic_from_kernel (void ) {/* -* GICv2 can always be accessed from the kernel because it is -* memory-mapped, and VHE systems can access GICv3 EL2 system -* registers. +* GICv3 and GICv5 drive the CPU interface through EL2 system +* registers, so only VHE reaches them from the kernel. GICv2 is +* memory-mapped and always reachable. */ -return !static_branch_unlikely (&kvm_vgic_global_state.gicv3_cpuif) ||has_vhe (); +if (kvm_vgic_global_state.type ==VGIC_V5 || +static_branch_unlikely (&kvm_vgic_global_state.gicv3_cpuif)) +return has_vhe (); + +return true; }static inline void vgic_save_state (struct kvm_vcpu *vcpu)
Signature · 1 line
2.39.5
Fuad Tabba
reviewerPATCH v2 4/4 KVM: arm64: Fix stale VGICv3 comments in the nVHE world switch
In reply to Fuad Tabba__hyp_vgic_save_state() and __hyp_vgic_restore_state() handle GICv5 as well as GICv3, but their comments name VGICv3 only.
No functional change intended.
--- arch/arm64/kvm/hyp/nvhe/switch.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/arm64/kvm/hyp/nvhe/switch.c +2 -2
@@ -110,7 +110,7 @@ static void __deactivate_traps(struct kvm_vcpu *vcpu)write_sysreg (__kvm_hyp_host_vector, vbar_el2); } -/* Save VGICv3 state on non-VHE systems */ +/* Save vgic state on non-VHE systems */ static void __hyp_vgic_save_state (struct kvm_vcpu *vcpu) {if (vgic_is_v5 (kern_hyp_va (vcpu->kvm))) {
@@ -125,7 +125,7 @@ static void __hyp_vgic_save_state(struct kvm_vcpu *vcpu) } } -/* Restore VGICv3 state on non-VHE systems */ +/* Restore vgic state on non-VHE systems */ static void __hyp_vgic_restore_state (struct kvm_vcpu *vcpu) {if (vgic_is_v5 (kern_hyp_va (vcpu->kvm))) {
Signature · 1 line
2.39.5
[email protected]
AI reviewRe: [PATCH v2 2/4] KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM
In reply to Fuad TabbaThank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
New issues:
- [Medium] The patch manually adds runtime mode checks inside GICv5 handlers instead of correctly categorizing the hypercall IDs in the pKVM availability bands, resulting in an incorrect SUCCESS return for an unsupported operation.
Pre-existing issues:
- [Critical] Always-on legacy hypercalls (like handle___vgic_v3_save_aprs and handle___kvm_adjust_pc) dereference host-provided pointers using kern_hyp_va() without ownership validation, enabling a malicious host to arbitrarily corrupt hypervisor memory. --
--- Patch [2]: [PATCH v2 2/4] KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM --- commit b55739e7ec2cd459ba47129f61f967475ca69c49
Author: Fuad Tabba <[email protected]>
KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM
__vgic_v5_save_apr() and __vgic_v5_restore_vmcr_apr() remain callable after pKVM finalises. pKVM never registers a GICv5 vgic, so neither has a valid caller in protected mode, and on a GICv3 machine both reach a register that is UNDEFINED at EL2 and panic the hypervisor.
arch/arm64/kvm/hyp/nvhe/hyp-main.c static void handle___vgic_v5_save_apr(struct kvm_cpu_context *host_ctxt) · line 700
Quoted 11 lines
arch/arm64/kvm/hyp/nvhe/hyp-main.c +3
-0@@ -695,6 +695,9 @@ static void handle___vgic_v5_save_apr(struct kvm_cpu_context *host_ctxt) { DECLARE_REG(struct vgic_v5_cpu_if *, cpu_if, host_ctxt, 1); +if (unlikely (is_protected_kvm_enabled ())) +return ; +
[Severity: Medium]
Does this early return leave SMCCC_RET_SUCCESS in the return register instead of correctly rejecting the call?
Because the hypercall ID remains in the always-on band in enum __kvm_host_smccc_func, the dispatcher will preload SMCCC_RET_SUCCESS and route it here. Returning early means the host receives a success code for an unsupported operation.
Would it be better to move these hypercall IDs before __KVM_HOST_SMCCC_FUNC_MIN_PKVM in asm/kvm_asm.h so the dispatcher rejects them automatically with SMCCC_RET_NOT_SUPPORTED?
arch/arm64/kvm/hyp/nvhe/hyp-main.c static void handle___vgic_v5_restore_vmcr_apr(struct kvm_cpu_context *host_ctxt) · line 712
Quoted 12 lines
arch/arm64/kvm/hyp/nvhe/hyp-main.c +3
-0__vgic_v5_save_apr (kern_hyp_va (cpu_if)); }@@ -702,6 +705,9 @@ static void handle___vgic_v5_restore_vmcr_apr(struct kvm_cpu_context *host_ctxt) { DECLARE_REG(struct vgic_v5_cpu_if *, cpu_if, host_ctxt, 1); +if (unlikely (is_protected_kvm_enabled ())) +return ; +__vgic_v5_restore_vmcr_apr (kern_hyp_va (cpu_if)); }
[Severity: Critical]
This is a pre-existing issue, but while looking at how kern_hyp_va() is used in this file, could a host use the always-on legacy hypercalls to modify hypervisor memory?
In handle___vgic_v3_save_aprs():
static void handle___vgic_v3_save_aprs(struct kvm_cpu_context *host_ctxt)
{
DECLARE_REG(struct vgic_v3_cpu_if *, cpu_if, host_ctxt, 1);
__vgic_v3_save_aprs(kern_hyp_va(cpu_if));
}And similarly in handle___kvm_adjust_pc():
static void handle___kvm_adjust_pc(struct kvm_cpu_context *host_ctxt)
{
DECLARE_REG(struct kvm_vcpu *, vcpu, host_ctxt, 1);
__kvm_adjust_pc(kern_hyp_va(vcpu));
}Since kern_hyp_va() transforms the pointer without validating page ownership, what happens if the host passes a pointer that resolves to an EL2-owned physical page? It appears this might give the host an arbitrary write or bit-clear primitive against hypervisor memory. Should these always-on handlers validate the host pointer before dereferencing it?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
Fuad Tabba
reviewerRe: [PATCH v2 2/4] KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM
It's been a while...
Quoted 5 lines · On Mon, 17 Aug 2026 at 11:47, <[email protected]> wrote:
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
New issues:
- [Medium] The patch manually adds runtime mode checks inside GICv5 handlers instead of correctly categorizing the hypercall IDs in the pKVM availability bands, resulting in an incorrect SUCCESS return for an unsupported operation.
pKVM does not support GICv5 yet but is expected to, so these are gated off for now, not rejected outright.
Quoted 2 lines
Pre-existing issues:
- [Critical] Always-on legacy hypercalls (likehandle___vgic_v3_save_aprsandhandle___kvm_adjust_pc) dereference host-provided pointers usingkern_hyp_va()without ownership validation, enabling a malicious host to arbitrarily corrupt hypervisor memory.
The _proper_ fix for this one is coming soon to a series near you (i.e., proper pVM support upstream)!
/fuad
Quoted 80 lines
--
--- Patch [2]: [PATCH v2 2/4] KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM --- commit
b55739e7ec2cd459ba47129f61f967475ca69c49
Author: Fuad Tabba <[email protected]>KVM: arm64: Reject the GICv5 CPU interface hypercalls under pKVM
__vgic_v5_save_apr()and__vgic_v5_restore_vmcr_apr()remain callable after pKVM finalises. pKVM never registers a GICv5 vgic, so neither has a valid caller in protected mode, and on a GICv3 machine both reach a register that is UNDEFINED at EL2 and panic the hypervisor.arch/arm64/kvm/hyp/nvhe/hyp-main.c static void handle___vgic_v5_save_apr(struct kvm_cpu_context *host_ctxt) · line 700
Quoted 11 lines
arch/arm64/kvm/hyp/nvhe/hyp-main.c +3
-0@@ -695,6 +695,9 @@ static void handle___vgic_v5_save_apr(struct kvm_cpu_context *host_ctxt) { DECLARE_REG(struct vgic_v5_cpu_if *, cpu_if, host_ctxt, 1); +if (unlikely (is_protected_kvm_enabled ())) +return ; +[Severity: Medium]
Does this early return leaveSMCCC_RET_SUCCESSin the return register instead of correctly rejecting the call?Because the hypercall ID remains in the always-on band in enum
__kvm_host_smccc_func, the dispatcher will preloadSMCCC_RET_SUCCESSand route it here. Returning early means the host receives a success code for an unsupported operation.Would it be better to move these hypercall IDs before
__KVM_HOST_SMCCC_FUNC_MIN_PKVMinasm/kvm_asm.hso the dispatcher rejects them automatically withSMCCC_RET_NOT_SUPPORTED?arch/arm64/kvm/hyp/nvhe/hyp-main.c static void handle___vgic_v5_restore_vmcr_apr(struct kvm_cpu_context *host_ctxt) · line 712
Quoted 12 lines
arch/arm64/kvm/hyp/nvhe/hyp-main.c +3
-0__vgic_v5_save_apr (kern_hyp_va (cpu_if)); }@@ -702,6 +705,9 @@ static void handle___vgic_v5_restore_vmcr_apr(struct kvm_cpu_context *host_ctxt) { DECLARE_REG(struct vgic_v5_cpu_if *, cpu_if, host_ctxt, 1); +if (unlikely (is_protected_kvm_enabled ())) +return ; +__vgic_v5_restore_vmcr_apr (kern_hyp_va (cpu_if)); }[Severity: Critical]
This is a pre-existing issue, but while looking at howkern_hyp_va()is used in this file, could a host use the always-on legacy hypercalls to modify hypervisor memory?In
handle___vgic_v3_save_aprs():static void
handle___vgic_v3_save_aprs(struct kvm_cpu_context *host_ctxt){ DECLARE_REG(struct vgic_v3_cpu_if *, cpu_if, host_ctxt, 1); __vgic_v3_save_aprs(kern_hyp_va(cpu_if)); }And similarly in
handle___kvm_adjust_pc():static void
handle___kvm_adjust_pc(struct kvm_cpu_context *host_ctxt){ DECLARE_REG(struct kvm_vcpu *, vcpu, host_ctxt, 1); __kvm_adjust_pc(kern_hyp_va(vcpu)); }Since
kern_hyp_va()transforms the pointer without validating page ownership, what happens if the host passes a pointer that resolves to an EL2-owned physical page? It appears this might give the host an arbitrary write or bit-clear primitive against hypervisor memory. Should these always-on handlers validate the host pointer before dereferencing it?--
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2