Re: [PATCH v5 3/3] KVM: s390: Add opportunistic floating IRQ delivery

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] kvm: s390: timer-less wait bypasses floating IRQ distribution
--

commit f14594a6a744fbc4a4f3602b18d73cbd397d4076
Author: Janosch Frank <frankja@linux.ibm.com>

KVM: s390: Add opportunistic floating IRQ delivery

This commit attempts to distribute floating IRQs which haven't been taken by
a vCPU yet to sleeping vCPUs. This lowers the risk of floating IRQs not being
delivered.

> diff --git a/arch/s390/kvm/s390/interrupt.c b/arch/s390/kvm/s390/interrupt.c
> index 3c0e6216eee3b..1b06775e581bf 100644
> --- a/arch/s390/kvm/s390/interrupt.c
> +++ b/arch/s390/kvm/s390/interrupt.c
[ ... ]
> @@ -1294,6 +1326,8 @@ int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
>  	if (!sltime)
>  		return 0;
>  
> +	distribute_float_irqs(vcpu->kvm);
> +
>  	__set_cpu_idle(vcpu);
>  	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
>  	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);

[Severity: High]
Does this bypass distribute_float_irqs() when a guest enters an enabled wait
state with clock comparator and CPU timer interrupts disabled?

If a guest disables timer interrupts and enters a wait state, it appears
kvm_s390_handle_wait() jumps directly to the no_timer label:

arch/s390/kvm/s390/interrupt.c:kvm_s390_handle_wait() {
    ...
	if (!ckc_interrupts_enabled(vcpu) &&
	    !cpu_timer_interrupts_enabled(vcpu)) {
		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
		__set_cpu_idle(vcpu);
		goto no_timer;
	}

	sltime = __calculate_sltime(vcpu);
	if (!sltime)
		return 0;

	distribute_float_irqs(vcpu->kvm);

	__set_cpu_idle(vcpu);
	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
no_timer:
    ...
}

Because the goto completely bypasses the newly added distribute_float_irqs()
call, any pending floating IRQs that this vCPU cannot deliver might remain
undelivered when the current vCPU goes to sleep. Could this lead to guest
stalls or deadlocks?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260928123639.34226-1-frankja@linux.ibm.com?part=3