Re: [PATCH 08/13] KVM: arm64: Move CNT*CT_EL0 userspace accessors to generic infrastructure

[Marc Zyngier <maz@kernel.org>]

On Tue, 30 Sep 2025 11:45:52 +0100,
Joey Gouly <joey.gouly@arm.com> wrote:
> 
> Observation below
> 
>   |
>   v
> 
> On Mon, Sep 29, 2025 at 05:04:52PM +0100, Marc Zyngier wrote:
> > Moving the counter registers is a bit more involved than for the control
> > and comparator (there is no shadow data for the counter), but still
> > pretty manageable.
> > 
> > Signed-off-by: Marc Zyngier <maz@kernel.org>
> > ---
> >  arch/arm64/kvm/guest.c    |  7 -------
> >  arch/arm64/kvm/sys_regs.c | 34 +++++++++++++++++++++++++++++++---
> >  2 files changed, 31 insertions(+), 10 deletions(-)
> > 
> > diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c
> > index c23ec9be4ce27..138e5e2dc10c8 100644
> > --- a/arch/arm64/kvm/guest.c
> > +++ b/arch/arm64/kvm/guest.c
> > @@ -592,19 +592,12 @@ static unsigned long num_core_regs(const struct kvm_vcpu *vcpu)
> >  }
> >  
> >  static const u64 timer_reg_list[] = {
> > -	KVM_REG_ARM_TIMER_CNT,
> > -	KVM_REG_ARM_PTIMER_CNT,
> >  };
> >  
> >  #define NUM_TIMER_REGS ARRAY_SIZE(timer_reg_list)
> >  
> >  static bool is_timer_reg(u64 index)
> >  {
> > -	switch (index) {
> > -	case KVM_REG_ARM_TIMER_CNT:
> > -	case KVM_REG_ARM_PTIMER_CNT:
> > -		return true;
> > -	}
> >  	return false;
> >  }
> >  
> > diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> > index 68e88d5c0dfb5..e67eb39ddc118 100644
> > --- a/arch/arm64/kvm/sys_regs.c
> > +++ b/arch/arm64/kvm/sys_regs.c
> > @@ -1605,12 +1605,38 @@ static int arch_timer_set_user(struct kvm_vcpu *vcpu,
> >  	case SYS_CNTHP_CTL_EL2:
> >  		val &= ~ARCH_TIMER_CTRL_IT_STAT;
> >  		break;
> > +	case SYS_CNTVCT_EL0:
> > +		if (!test_bit(KVM_ARCH_FLAG_VM_COUNTER_OFFSET, &vcpu->kvm->arch.flags))
> > +			timer_set_offset(vcpu_vtimer(vcpu), kvm_phys_timer_read() - val);
> > +		return 0;
> > +	case SYS_CNTPCT_EL0:
> > +		if (!test_bit(KVM_ARCH_FLAG_VM_COUNTER_OFFSET, &vcpu->kvm->arch.flags))
> > +			timer_set_offset(vcpu_ptimer(vcpu), kvm_phys_timer_read() - val);
> > +		return 0;
> >  	}
> >  
> >  	__vcpu_assign_sys_reg(vcpu, rd->reg, val);
> >  	return 0;
> >  }
> >  
> > +static int arch_timer_get_user(struct kvm_vcpu *vcpu,
> > +			       const struct sys_reg_desc *rd,
> > +			       u64 *val)
> > +{
> > +	switch (reg_to_encoding(rd)) {
> > +	case SYS_CNTVCT_EL0:
> > +		*val = kvm_phys_timer_read() - timer_get_offset(vcpu_vtimer(vcpu));
> > +		break;
> > +	case SYS_CNTPCT_EL0:
> > +		*val = kvm_phys_timer_read() - timer_get_offset(vcpu_ptimer(vcpu));
> > +		break;
> > +	default:
> > +		*val = __vcpu_sys_reg(vcpu, rd->reg);
> 
> Unsure if this is actually an issue but for the _CTL registers, via
> access_arch_timer() (kvm_arm_timer_read_sysreg() -> .. -> read_timer_ctl()),
> the ARCH_TIMER_CTRL_IT_STAT bit will be set if the timer expired, but that's
> not done here.

Indeed, but I don't think this really matters, at least not for
save/restore. We always clear the ISTATUS bit on restore, and
snapshoting CTL is always a racy process.

	M.

-- 
Without deviation from the norm, progress is not possible.