KVM: Move initialization of preempt notifier to kvm_vcpu_init()
authorSean Christopherson <sean.j.christopherson@intel.com>
Wed, 18 Dec 2019 21:55:17 +0000 (13:55 -0800)
committerPaolo Bonzini <pbonzini@redhat.com>
Mon, 27 Jan 2020 18:59:25 +0000 (19:59 +0100)
commitd5c48debcc4fbdd09e3e0871980b7ebf06ec6e07
tree6b86dda12341d08840f899deee7881003e48b3cc
parentaaba298c6bca8d8625880a8016e5b80adc8a11af
KVM: Move initialization of preempt notifier to kvm_vcpu_init()

Initialize the preempt notifier immediately in kvm_vcpu_init() to pave
the way for removing kvm_arch_vcpu_setup(), i.e. to allow arch specific
code to call vcpu_load() during kvm_arch_vcpu_create().

Back when preemption support was added, the location of the call to init
the preempt notifier was perfectly sane.  The overall vCPU creation flow
featured a single arch specific hook and the preempt notifer was used
immediately after its initialization (by vcpu_load()).  E.g.:

        vcpu = kvm_arch_ops->vcpu_create(kvm, n);
        if (IS_ERR(vcpu))
                return PTR_ERR(vcpu);

        preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);

        vcpu_load(vcpu);
        r = kvm_mmu_setup(vcpu);
        vcpu_put(vcpu);
        if (r < 0)
                goto free_vcpu;

Today, the call to preempt_notifier_init() is sandwiched between two
arch specific calls, kvm_arch_vcpu_create() and kvm_arch_vcpu_setup(),
which needlessly forces x86 (and possibly others?) to split its vCPU
creation flow.  Init the preempt notifier prior to any arch specific
call so that each arch can independently decide how best to organize
its creation flow.

Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
virt/kvm/kvm_main.c