Skip to contentSkip to conversation

kvm: x86: Do proper cleanup if kvm_x86_ops->vm_init() fails

PATCH Superseded Junaid Shahid 2 participants 3 messages Jul 29, 2022

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kvm: x86: Do proper cleanup if kvm_x86_ops->vm_init() fails

If vm_init() fails [which can happen, for instance, if a memory allocation fails during avic_vm_init()], we need to cleanup some state in order to avoid resource leaks.

---
 arch/x86/kvm/x86.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
arch/x86/kvm/x86.c +7 -1
index f389691d8c04..ef5fd2f05c79 100644
@@ -12064,8 +12064,14 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 	kvm_hv_init_vm(kvm);
 	kvm_xen_init_vm(kvm);
 
-	return static_call(kvm_x86_vm_init)(kvm);
+	ret = static_call(kvm_x86_vm_init)(kvm);
+	if (ret)
+		goto out_uninit_mmu;
 
+	return 0;
+
+out_uninit_mmu:
+	kvm_mmu_uninit_vm(kvm);
 out_page_track:
 	kvm_page_track_cleanup(kvm);
 out:

base-commit: a4850b5590d01bf3fb19fda3fc5d433f7382a974

Signature · 1 line
2.37.1.455.g008518b4e5-goog
#1 lore raw
Quoted 16 lines · Junaid Shahid

If vm_init() fails [which can happen, for instance, if a memory allocation fails during avic_vm_init()], we need to cleanup some state in order to avoid resource leaks.

---
 arch/x86/kvm/x86.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
arch/x86/kvm/x86.c +0 -0
index f389691d8c04..ef5fd2f05c79 100644
@@ -12064,8 +12064,14 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 	kvm_hv_init_vm(kvm);
 	kvm_xen_init_vm(kvm);

arch/x86/kvm/x86.c int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) · line 12074

Quoted 10 lines
arch/x86/kvm/x86.c +7 -1
 
-	return static_call(kvm_x86_vm_init)(kvm);
+	ret = static_call(kvm_x86_vm_init)(kvm);
+	if (ret)
+		goto out_uninit_mmu;
 
+	return 0;
+
+out_uninit_mmu:
+	kvm_mmu_uninit_vm(kvm);

Hrm, this works for now (I think), but I really don't like that kvm_apicv_init(), kvm_hv_init_vm(), and kvm_xen_init_vm() all do something without that something being unwound on failure. E.g. both Hyper-V and Xen have a paired "destroy" function, it just so happens that their destroy paths are guaranteed nops in this case.

AFAICT, there are no dependencies on doing vendor init at the end, so what if we hoist it up so that all paths that can fail are at the top?

arch/x86/kvm/x86.c +7 -1
index 5366f884e9a7..7e749be356b2 100644
@@ -12042,6 +12042,10 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
        if (ret)
                goto out_page_track;
 
+       ret = static_call(kvm_x86_vm_init)(kvm);
+       if (ret)
+               goto out_uninit_mmu;
+
        INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
        INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
        atomic_set(&kvm->arch.noncoherent_dma_count, 0);
@@ -12077,8 +12081,10 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
        kvm_hv_init_vm(kvm);
        kvm_xen_init_vm(kvm);
 
-       return static_call(kvm_x86_vm_init)(kvm);
+       return 0;
 
+out_uninit_mmu:
+       kvm_mmu_uninit_vm(kvm);
 out_page_track:
        kvm_page_track_cleanup(kvm);
 out:

Calling kvm_apicv_init() after avic_vm_init() is somewhat odd. If we really want to avoid that, we could add a dedicated kvm_x86_ops to initialize APICv and then make kvm_x86_ops.vm_init() a void return e.g.

static int kvm_apicv_init(struct kvm *kvm)

{
	unsigned long *inhibits = &kvm->arch.apicv_inhibit_reasons;

	init_rwsem(&kvm->arch.apicv_update_lock);

	set_or_clear_apicv_inhibit(inhibits, APICV_INHIBIT_REASON_ABSENT, true);

	if (!enable_apicv) {
		set_or_clear_apicv_inhibit(inhibits,
					   APICV_INHIBIT_REASON_DISABLE, true);
		return 0;
	}

	return static_call(kvm_x86_apicv_init(kvm));
}
#2 lore raw
Quoted 34 lines · Sean Christopherson
Quoted 16 lines · Junaid Shahid

If vm_init() fails [which can happen, for instance, if a memory allocation fails during avic_vm_init()], we need to cleanup some state in order to avoid resource leaks.

---
  arch/x86/kvm/x86.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
arch/x86/kvm/x86.c +0 -0
index f389691d8c04..ef5fd2f05c79 100644
@@ -12064,8 +12064,14 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
  	kvm_hv_init_vm(kvm);
  	kvm_xen_init_vm(kvm);

arch/x86/kvm/x86.c int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) · line 12074

Quoted 10 lines
arch/x86/kvm/x86.c +7 -1
 
-	return static_call(kvm_x86_vm_init)(kvm);
+	ret = static_call(kvm_x86_vm_init)(kvm);
+	if (ret)
+		goto out_uninit_mmu;
 
+	return 0;
+
+out_uninit_mmu:
+	kvm_mmu_uninit_vm(kvm);

Hrm, this works for now (I think), but I really don't like that kvm_apicv_init(), kvm_hv_init_vm(), and kvm_xen_init_vm() all do something without that something being unwound on failure. E.g. both Hyper-V and Xen have a paired "destroy" function, it just so happens that their destroy paths are guaranteed nops in this case.

Quoted 3 lines

AFAICT, there are no dependencies on doing vendor init at the end, so what if we hoist it up so that all paths that can fail are at the top?

Ack. I'll move the vendor init call up. I think I'll skip breaking out a separate kvm_x86_op for apicv_init() for now.

Thanks,
Junaid