Hi Tomasz,
On Mon, Jan 22, 2018 at 01:32:57PM +0100, Tomasz Nowicki wrote:
> Please see my observations/comments below.
>
> On 20.12.2017 12:36, Christoffer Dall wrote:
> >The VGIC can now support the life-cycle of mapped level-triggered
> >interrupts, and we no longer have to read back the timer state on every
> >exit from the VM if we had an asserted timer interrupt signal, because
> >the VGIC already knows if we hit the unlikely case where the guest
> >disables the timer without ACKing the virtual timer interrupt.
> >
> >This means we rework a bit of the code to factor out the functionality
> >to snapshot the timer state from vtimer_save_state(), and we can reuse
> >this functionality in the sync path when we have an irqchip in
> >userspace, and also to support our implementation of the
> >get_input_level() function for the timer.
> >
> >This change also means that we can no longer rely on the timer's view of
> >the interrupt line to set the active state, because we no longer
> >maintain this state for mapped interrupts when exiting from the guest.
> >Instead, we only set the active state if the virtual interrupt is
> >active, and otherwise we simply let the timer fire again and raise the
> >virtual interrupt from the ISR.
> >
> >Reviewed-by: Eric Auger <eric.auger@xxxxxxxxxx>
> >Reviewed-by: Marc Zyngier <marc.zyngier@xxxxxxx>
> >Signed-off-by: Christoffer Dall <christoffer.dall@xxxxxxxxxx>
> >---
> > include/kvm/arm_arch_timer.h | 2 ++
> > virt/kvm/arm/arch_timer.c | 84 ++++++++++++++++++++------------------------
> > 2 files changed, 40 insertions(+), 46 deletions(-)
> >
> >diff --git a/include/kvm/arm_arch_timer.h b/include/kvm/arm_arch_timer.h
> >index 6e45608b2399..b1dcfde0a3ef 100644
> >--- a/include/kvm/arm_arch_timer.h
> >+++ b/include/kvm/arm_arch_timer.h
> >@@ -90,6 +90,8 @@ void kvm_timer_vcpu_put(struct kvm_vcpu *vcpu);
> > void kvm_timer_init_vhe(void);
> >+bool kvm_arch_timer_get_input_level(int vintid);
> >+
> > #define vcpu_vtimer(v) (&(v)->arch.timer_cpu.vtimer)
> > #define vcpu_ptimer(v) (&(v)->arch.timer_cpu.ptimer)
> >diff --git a/virt/kvm/arm/arch_timer.c b/virt/kvm/arm/arch_timer.c
> >index fb0533ed903d..d845d67b7062 100644
> >--- a/virt/kvm/arm/arch_timer.c
> >+++ b/virt/kvm/arm/arch_timer.c
> >@@ -343,6 +343,12 @@ static void kvm_timer_update_state(struct kvm_vcpu *vcpu)
> > phys_timer_emulate(vcpu);
> > }
> >+static void __timer_snapshot_state(struct arch_timer_context *timer)
> >+{
> >+ timer->cnt_ctl = read_sysreg_el0(cntv_ctl);
> >+ timer->cnt_cval = read_sysreg_el0(cntv_cval);
> >+}
> >+
> > static void vtimer_save_state(struct kvm_vcpu *vcpu)
> > {
> > struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
> >@@ -354,10 +360,8 @@ static void vtimer_save_state(struct kvm_vcpu *vcpu)
> > if (!vtimer->loaded)
> > goto out;
> >- if (timer->enabled) {
> >- vtimer->cnt_ctl = read_sysreg_el0(cntv_ctl);
> >- vtimer->cnt_cval = read_sysreg_el0(cntv_cval);
> >- }
> >+ if (timer->enabled)
> >+ __timer_snapshot_state(vtimer);
> > /* Disable the virtual timer */
> > write_sysreg_el0(0, cntv_ctl);
> >@@ -454,8 +458,7 @@ static void kvm_timer_vcpu_load_vgic(struct kvm_vcpu *vcpu)
> > bool phys_active;
> > int ret;
> >- phys_active = vtimer->irq.level ||
> >- kvm_vgic_map_is_active(vcpu, vtimer->irq.irq);
> >+ phys_active = kvm_vgic_map_is_active(vcpu, vtimer->irq.irq);
> > ret = irq_set_irqchip_state(host_vtimer_irq,
> > IRQCHIP_STATE_ACTIVE,
> >@@ -535,54 +538,27 @@ void kvm_timer_vcpu_put(struct kvm_vcpu *vcpu)
> > set_cntvoff(0);
> > }
> >-static void unmask_vtimer_irq(struct kvm_vcpu *vcpu)
> >+/*
> >+ * With a userspace irqchip we have to check if the guest de-asserted the
> >+ * timer and if so, unmask the timer irq signal on the host interrupt
> >+ * controller to ensure that we see future timer signals.
> >+ */
> >+static void unmask_vtimer_irq_user(struct kvm_vcpu *vcpu)
> > {
> > struct arch_timer_context *vtimer = vcpu_vtimer(vcpu);
> > if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
> >- kvm_vtimer_update_mask_user(vcpu);
> >- return;
> >- }
> >-
> >- /*
> >- * If the guest disabled the timer without acking the interrupt, then
> >- * we must make sure the physical and virtual active states are in
> >- * sync by deactivating the physical interrupt, because otherwise we
> >- * wouldn't see the next timer interrupt in the host.
> >- */
> >- if (!kvm_vgic_map_is_active(vcpu, vtimer->irq.irq)) {
> >- int ret;
> >- ret = irq_set_irqchip_state(host_vtimer_irq,
> >- IRQCHIP_STATE_ACTIVE,
> >- false);
> >- WARN_ON(ret);
> >+ __timer_snapshot_state(vtimer);
> >+ if (!kvm_timer_should_fire(vtimer)) {
> >+ kvm_timer_update_irq(vcpu, false, vtimer);
> >+ kvm_vtimer_update_mask_user(vcpu);
> >+ }
> > }
> > }
> >-/**
> >- * kvm_timer_sync_hwstate - sync timer state from cpu
> >- * @vcpu: The vcpu pointer
> >- *
> >- * Check if any of the timers have expired while we were running in the guest,
> >- * and inject an interrupt if that was the case.
> >- */
> > void kvm_timer_sync_hwstate(struct kvm_vcpu *vcpu)
> > {
> >- struct arch_timer_context *vtimer = vcpu_vtimer(vcpu);
> >-
> >- /*
> >- * If we entered the guest with the vtimer output asserted we have to
> >- * check if the guest has modified the timer so that we should lower
> >- * the line at this point.
> >- */
> >- if (vtimer->irq.level) {
> >- vtimer->cnt_ctl = read_sysreg_el0(cntv_ctl);
> >- vtimer->cnt_cval = read_sysreg_el0(cntv_cval);
> >- if (!kvm_timer_should_fire(vtimer)) {
> >- kvm_timer_update_irq(vcpu, false, vtimer);
> >- unmask_vtimer_irq(vcpu);
> >- }
> >- }
> >+ unmask_vtimer_irq_user(vcpu);
> > }
>
> While testing your VHE optimization series [1] I noticed massive WFI exits
> on ThunderX2 so I bisected 'vhe-optimize-v3-with-fixes' branch of [2] down
> to this commit.
Thanks for doing that!
>
> The host traps on WFI so that VCPU thread should be scheduled out for some
> time. However, this is not happening because kvm_vcpu_block() ->
> kvm_vcpu_check_block() gives negative value (vtimer->irq.level is asserted)
> and VCPU is woken up immediately. Nobody takes care about lowering the
> vtimer line in this case.
Indeed, this is a problem. I thought this was properly handled, but I
think this part changed at some version of these patches.
>
> I reverted shape of above kvm_timer_sync_hwstate() and things are good now.
> Before I start digging I wanted to check with you. Can you please check on
> your side?
I'd still like to not have kvm_timer_sync_hwstate() do any work, and I
think this can be accomplished by updating vtimer->irq.level in
vtimer_save_state().
I didn't observe that a guest couldn't sleep on WFI when I tested this
series, but I may have simply not noticed it on any of the platforms I
used for testing. I'll investigate and come back to you.
Thanks,
-Christoffer
