Re: [PATCH v3 06/22] KVM: arm64: Simplify the CPUHP logic

[Marc Zyngier <maz@kernel.org>]

On 2022-09-05 10:29, Marc Zyngier wrote:
> On Mon, 05 Sep 2022 08:05:09 +0100,
> Yuan Yao <yuan.yao@linux.intel.com> wrote:
>> 
>> On Thu, Sep 01, 2022 at 07:17:41PM -0700, isaku.yamahata@intel.com 
>> wrote:
>> > From: Marc Zyngier <maz@kernel.org>
>> >
>> > For a number of historical reasons, the KVM/arm64 hotplug setup is pretty
>> > complicated, and we have two extra CPUHP notifiers for vGIC and timers.
>> >
>> > It looks pretty pointless, and gets in the way of further changes.
>> > So let's just expose some helpers that can be called from the core
>> > CPUHP callback, and get rid of everything else.
>> >
>> > This gives us the opportunity to drop a useless notifier entry,
>> > as well as tidy-up the timer enable/disable, which was a bit odd.
>> >
>> > Signed-off-by: Marc Zyngier <maz@kernel.org>
>> > Signed-off-by: Chao Gao <chao.gao@intel.com>
>> > Reviewed-by: Oliver Upton <oupton@google.com>
>> > Link: https://lore.kernel.org/r/20220216031528.92558-5-chao.gao@intel.com
>> > Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
>> > ---
>> >  arch/arm64/kvm/arch_timer.c     | 27 ++++++++++-----------------
>> >  arch/arm64/kvm/arm.c            |  4 ++++
>> >  arch/arm64/kvm/vgic/vgic-init.c | 19 ++-----------------
>> >  include/kvm/arm_arch_timer.h    |  4 ++++
>> >  include/kvm/arm_vgic.h          |  4 ++++
>> >  include/linux/cpuhotplug.h      |  3 ---
>> >  6 files changed, 24 insertions(+), 37 deletions(-)
>> >
>> > diff --git a/arch/arm64/kvm/arch_timer.c b/arch/arm64/kvm/arch_timer.c
>> > index bb24a76b4224..33fca1a691a5 100644
>> > --- a/arch/arm64/kvm/arch_timer.c
>> > +++ b/arch/arm64/kvm/arch_timer.c
>> > @@ -811,10 +811,18 @@ void kvm_timer_vcpu_init(struct kvm_vcpu *vcpu)
>> >  	ptimer->host_timer_irq_flags = host_ptimer_irq_flags;
>> >  }
>> >
>> > -static void kvm_timer_init_interrupt(void *info)
>> > +void kvm_timer_cpu_up(void)
>> >  {
>> >  	enable_percpu_irq(host_vtimer_irq, host_vtimer_irq_flags);
>> > -	enable_percpu_irq(host_ptimer_irq, host_ptimer_irq_flags);
>> > +	if (host_ptimer_irq)
>> > +		enable_percpu_irq(host_ptimer_irq, host_ptimer_irq_flags);
>> > +}
>> > +
>> > +void kvm_timer_cpu_down(void)
>> > +{
>> > +	disable_percpu_irq(host_vtimer_irq);
>> > +	if (host_ptimer_irq)
>> > +		disable_percpu_irq(host_ptimer_irq);
>> >  }
>> 
>> Should "host_vtimer_irq" be checked yet as host_ptimer_irq ?
> 
> No, because although the ptimer interrupt is optional (on older
> systems, we fully emulate that timer, including the interrupt), the
> vtimer interrupt is always present and can be used unconditionally.
> 
>> Because
>> the host_{v,p}timer_irq is set in same function kvm_irq_init() which
>> called AFTER the on_each_cpu(_kvm_arch_hardware_enable, NULL, 1) from
>> init_subsystems():
>> 
>> kvm_init()
>>   kvm_arch_init()
>>     init_subsystems()
>>       on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
>>       kvm_timer_hyp_init()
>>         kvm_irq_init()
>>           host_vtimer_irq = info->virtual_irq;
>>           host_ptimer_irq = info->physical_irq;
>>   hardware_enable_all()
> 
> This, however, is a very nice catch. I doubt this results in anything
> really bad (the interrupt enable will fail as the interrupt number
> is 0, and the disable will also fail due to no prior enable), but
> that's extremely ugly anyway.
> 
> The best course of action AFAICS is to differentiate between the
> arm64-specific initialisation (which is a one-off) and the runtime
> stuff. Something like the hack below, that I haven't tested yet:
> 
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index 32c1022eb4b3..65d03c28f32a 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -1671,23 +1671,27 @@ static void _kvm_arch_hardware_enable(void 
> *discard)
>  {
>  	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
>  		cpu_hyp_reinit();
> -		kvm_vgic_cpu_up();
> -		kvm_timer_cpu_up();
>  		__this_cpu_write(kvm_arm_hardware_enabled, 1);
>  	}
>  }
> 
>  int kvm_arch_hardware_enable(void)
>  {
> +	int was_enabled = __this_cpu_read(kvm_arm_hardware_enabled);
> +
>  	_kvm_arch_hardware_enable(NULL);
> +
> +	if (!was_enabled) {
> +		kvm_vgic_cpu_up();
> +		kvm_timer_cpu_up();
> +	}
> +
>  	return 0;
>  }
> 
>  static void _kvm_arch_hardware_disable(void *discard)
>  {
>  	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
> -		kvm_timer_cpu_down();
> -		kvm_vgic_cpu_down();
>  		cpu_hyp_reset();
>  		__this_cpu_write(kvm_arm_hardware_enabled, 0);
>  	}
> @@ -1695,6 +1699,11 @@ static void _kvm_arch_hardware_disable(void 
> *discard)
> 
>  void kvm_arch_hardware_disable(void)
>  {
> +	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
> +		kvm_timer_cpu_down();
> +		kvm_vgic_cpu_down();
> +	}
> +
>  	if (!is_protected_kvm_enabled())
>  		_kvm_arch_hardware_disable(NULL);
>  }

OK, this seems to work here, at least based on a sample of 2
systems, bringing CPUs up and down whist a VM is pinned to
these CPUs.

Isaku, can you please squash this into the original patch
and drop Oliver's Reviewed-by: tag, as this significantly
changes the logic?

Alternatively, I can repost this patch as a standalone change.

Thanks,

         M.
-- 
Jazz is not dead. It just smells funny...