KVM: arm64: Clear the VM's feature bitmap when kvm_setup_vcpu() fails

__kvm_vcpu_set_target() copies the requested features into the VM-wide bitmap before kvm_setup_vcpu() runs and doesn't undo it when setup fails, so a rejected KVM_ARM_VCPU_INIT leaves the VM recording features that were never set up. With HAS_EL2 | HAS_EL2_E2H0 on a host without FEAT_NV1, kvm_vcpu_init_nested() returns -EINVAL before it allocates any nested stage-2 MMU, and vcpu_has_nv() is then true with nested_mmus_size == 0; its -ENOMEM paths do the same on a VM's first INIT.

Until one INIT has succeeded, every INIT that passes the feature check rewrites the bitmap and a failure leaves it there; once KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED is set, kvm_vcpu_init_changed() has already required the requested features to equal the bitmap, so the copy is a no-op. Clear the bitmap on failure while the flag is clear.

This showed up with the series that enables KVM_PRE_FAULT_MEMORY for arm64, since the generic kvm_vcpu_pre_fault_memory() calls vcpu_load() without checking that the vCPU has been initialised. After the rejected INIT, the first call's vcpu_load() reads hw_mmu still set to the canonical MMU and leaves it alone, but its vcpu_put() takes the vcpu_has_nv() branch into kvm_vcpu_put_hw_mmu(), which clears hw_mmu. The second call's vcpu_load() reads hw_mmu as NULL and takes the nested branch into get_s2_mmu_nested(), whose search over nested_mmus_size == 0 leaves s2_mmu NULL for the BUG_ON(atomic_read(&s2_mmu->refcnt)), under mmu_lock. On kvmarm/next with the series applied:

  Unable to handle kernel NULL pointer dereference at virtual address 0000000000000074
  Call trace:
   kvm_vcpu_load_hw_mmu (arch/arm64/kvm/nested.c:891) (P)
   kvm_arch_vcpu_load (arch/arm64/kvm/arm.c:662)
   kvm_vcpu_pre_fault_memory (virt/kvm/kvm_main.c:170 virt/kvm/kvm_main.c:4349)
   kvm_vcpu_ioctl (virt/kvm/kvm_main.c:4639)

Fixes 427733579744e ("KVM: arm64: Select default PMU in KVM_ARM_VCPU_INIT handler")

Suggested-by Oliver Upton <[email protected]>

Link https://lore.kernel.org/r/[email protected]/ open here

Signed-off-by Fuad Tabba <[email protected]>

---
v3:
- Clear the bitmap on failure instead of saving and restoring it, only while KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED is clear, since with the flag set the copy was a no-op (Oliver).
- Commit message: drop the "latent, exposed by the series" framing and state the flag invariant behind the clear; the pre-fault crash is how the bug showed up.
- Dropped Lorenzo's Reviewed-by since the code changed.

v2:
- Commit message: say what the prefault series enables, bring the two-ioctl walk and the trace up from below the fold, the trace decoded, and drop the Fixes: on 1de10b7d13a97, which had nothing fallible after the copy (Lorenzo).
- Fold Lorenzo's Reviewed-by.

Reproduced on kvmarm/next plus the series under QEMU (-cpu max with an Apple M2 MIDR, which has_nv1() denies; kvm-arm.mode=nested):
KVM_ARM_VCPU_INIT with HAS_EL2 | HAS_EL2_E2H0 returns -EINVAL, then KVM_PRE_FAULT_MEMORY twice. With the fix both calls return -ENOENT and the host is unaffected. Based on kvmarm/fixes; applies cleanly to v7.3-rc4 and to kvmarm/next.

v2: https://lore.kernel.org/r/[email protected]/
v1: https://lore.kernel.org/r/[email protected]/

 arch/arm64/kvm/arm.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

arch/arm64/kvm/arm.c +9 -1

index eaf583b771931..bd19f64ae82ef 100644
@@ -1698,8 +1698,16 @@ static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 	bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
 
 	ret = kvm_setup_vcpu(vcpu);
-	if (ret)
+	if (ret) {
+		/*
+		 * Clear the bitmap if setup fails on the first vCPU to be
+		 * initialized.
+		 */
+		if (!test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags))
+			bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
+
 		goto out_unlock;
+	}
 
 	/* Now we know what it is, we can reset it. */
 	kvm_reset_vcpu(vcpu);

base-commit: 6b1bca1b1ab77f60a62087337bfe6e2f0efb9e6d --
2.39.5