* Re: [PATCH] KVM: arm64: nv: Fix NULL dereference in kvm_inject_nested()
2026-11-16 13:40 [PATCH] KVM: arm64: nv: Fix NULL dereference in kvm_inject_nested() Dan Carpenter
@ 2026-11-16 18:11 ` Marc Zyngier
0 siblings, 0 replies; 2+ messages in thread
From: Marc Zyngier @ 2026-11-16 18:11 UTC (permalink / raw)
To: Dan Carpenter
Cc: Oliver Upton, Fuad Tabba, Joey Gouly, Steffen Eiden,
Suzuki K Poulose, Zenghui Yu, Catalin Marinas, Will Deacon,
Mark Rutland, linux-arm-kernel, kvmarm, linux-kernel,
Karthik Bhargavan, Clément Blaudeau
On Mon, 16 Nov 2026 13:40:34 +0000,
Dan Carpenter <dan@pereq.eu> wrote:
>
> There is a NULL dereference of vcpu->arch.hw_mmu in kvm_inject_nested()
>
> Unable to handle kernel NULL pointer dereference at virtual address 0000000000000084
> kvm_vcpu_put_hw_mmu (arch/arm64/kvm/nested.c:1046) (P)
> vhe_vcpu_put (arch/arm64/kvm/arm.c:827)
> kvm_arch_vcpu_put (arch/arm64/kvm/arm.c:860)
> kvm_inject_nested (arch/arm64/kvm/emulate-nested.c:2903)
> kvm_inject_nested_serror (arch/arm64/kvm/emulate-nested.c:2976)
> kvm_inject_serror_esr (arch/arm64/kvm/inject_fault.c:365)
> __kvm_arm_vcpu_set_events (arch/arm64/kvm/guest.c:827)
> kvm_arch_vcpu_ioctl (arch/arm64/kvm/arm.c:2017 arch/arm64/kvm/arm.c:2157)
> kvm_vcpu_ioctl (arch/arm64/kvm/../../../virt/kvm/kvm_main.c:4772)
>
> KVM_SET_VCPU_EVENTS does not load the vcpu, but it injects a pending
> SError or external abort straight away. If the vcpu is in L2 context
> and L1 has asked for the exception, kvm_inject_nested() puts and loads
> a vcpu that is not loaded. vcpu->arch.hw_mmu is NULL when the vcpu is
> not loaded, so kvm_vcpu_put_hw_mmu() dereferences a NULL pointer.
>
> To fix this bug, only put and load the vcpu when it is loaded, similar
> to what kvm_reset_vcpu() does. When the vcpu is not loaded there is
> nothing to switch: the next kvm_arch_vcpu_load() loads L1's context.
>
> Fixes: 9aba641b9ec2 ("KVM: arm64: nv: Respect exception routing rules for SEAs")
> Fixes: 77ee70a07357 ("KVM: arm64: nv: Honor SError exception routing / masking")
> Cc: stable@vger.kernel.org
> Signed-off-by: Dan Carpenter <dan@pereq.eu>
> ---
> This was found through Pereq's formal analysis but the stack trace
> comes from real life testing.
>
> x86 implements kvm_arch_vcpu_ioctl() differently and puts vcpu_load()
> at the start of the ioctl. But the way arm64 does it is deliberately
> and just something that people working with the code need to be aware
> of, I guess.
>
> arch/arm64/kvm/emulate-nested.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm64/kvm/emulate-nested.c b/arch/arm64/kvm/emulate-nested.c
> index 2bec81668361..22ca735a34f6 100644
> --- a/arch/arm64/kvm/emulate-nested.c
> +++ b/arch/arm64/kvm/emulate-nested.c
> @@ -2862,7 +2862,7 @@ static int kvm_inject_nested(struct kvm_vcpu *vcpu, u64 esr_el2,
> enum exception_type type)
> {
> u64 pstate, mode;
> - bool direct_inject;
> + bool direct_inject, loaded;
>
> if (!vcpu_has_nv(vcpu)) {
> kvm_err("Unexpected call to %s for the non-nesting configuration\n",
> @@ -2891,6 +2891,7 @@ static int kvm_inject_nested(struct kvm_vcpu *vcpu, u64 esr_el2,
> }
>
> preempt_disable();
> + loaded = (vcpu->cpu != -1);
>
> vcpu_set_flag(vcpu, IN_NESTED_EXCEPTION);
>
> @@ -2900,7 +2901,8 @@ static int kvm_inject_nested(struct kvm_vcpu *vcpu, u64 esr_el2,
> */
> __kvm_adjust_pc(vcpu);
>
> - kvm_arch_vcpu_put(vcpu);
> + if (loaded)
> + kvm_arch_vcpu_put(vcpu);
>
> kvm_inject_el2_exception(vcpu, esr_el2, type);
>
> @@ -2915,7 +2917,8 @@ static int kvm_inject_nested(struct kvm_vcpu *vcpu, u64 esr_el2,
> */
> __kvm_adjust_pc(vcpu);
>
> - kvm_arch_vcpu_load(vcpu, smp_processor_id());
> + if (loaded)
> + kvm_arch_vcpu_load(vcpu, smp_processor_id());
> vcpu_clear_flag(vcpu, IN_NESTED_EXCEPTION);
>
> preempt_enable();
I wonder if a marginally better approach is to add an unconditional
vcpu_load() in the event injection ioctl:
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index 5959bbbfa026f..21cd872ea394b 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -2007,7 +2007,7 @@ static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
struct kvm_vcpu_events *events)
{
- int i;
+ int i, ret;
/* check whether the reserved field is zero */
for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
@@ -2019,7 +2019,11 @@ static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
if (events->exception.pad[i])
return -EINVAL;
- return __kvm_arm_vcpu_set_events(vcpu, events);
+ vcpu_load(vcpu);
+ ret = __kvm_arm_vcpu_set_events(vcpu, events);
+ vcpu_put(vcpu);
+
+ return ret;
}
/*
and be done with it, avoiding the fast-path bloating altogether.
Unless you can spot another path where the vcpu is not loaded, but
still requires the injection of an exception?
Thanks,
M.
--
Without deviation from the norm, progress is not possible.
^ permalink raw reply related [flat|nested] 2+ messages in thread