Re: [RFC patch 14/15] workpending: Provide infrastructure for work before entering a guest

[Miroslav Benes]

> --- a/include/linux/entry-common.h
> +++ b/include/linux/entry-common.h

[...]

> +#define EXIT_TO_GUESTMODE_WORK						\
> +	(_TIF_NEED_RESCHED | _TIF_SIGPENDING | _TIF_NOTIFY_RESUME |	\
> +	 ARCH_EXIT_TO_GUESTMODE_WORK)

[...]

> --- a/kernel/entry/common.c
> +++ b/kernel/entry/common.c
>
> +int core_exit_to_guestmode_work(struct kvm *kvm, struct kvm_vcpu *vcpu,
> +				unsigned long ti_work)
> +{
> +	/*
> +	 * Before returning to guest mode handle all pending work
> +	 */
> +	if (ti_work & _TIF_SIGPENDING) {
> +		vcpu->run->exit_reason = KVM_EXIT_INTR;
> +		vcpu->stat.signal_exits++;
> +		return -EINTR;
> +	}
> +
> +	if (ti_work & _TIF_NEED_RESCHED) {
> +		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
> +		schedule();
> +		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
> +	}
> +
> +	if (ti_work & _TIF_PATCH_PENDING) {
> +		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
> +		klp_update_patch_state(current);
> +		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
> +	}

If I am reading the code correctly, _TIF_PATCH_PENDING is not a part of 
EXIT_TO_GUESTMODE_WORK, so the handling code here would not be called on 
any arch as of now.

I also think that _TIF_PATCH_PENDING must not be handled here generally. 
It could break consistency guarantees when live patching KVM (and we do 
that from time to time).

Adding live-patching ML to CC.

Miroslav

> +	if (ti_work & _TIF_NOTIFY_RESUME) {
> +		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
> +		clear_thread_flag(TIF_NOTIFY_RESUME);
> +		tracehook_notify_resume(NULL);
> +		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
> +	}
> +
> +	/* Any extra architecture specific work */
> +	return arch_exit_to_guestmode_work(kvm, vcpu, ti_work);
> +}