[PATCH v2 02/36] KVM: arm64: gic-v3: Switch vGIC-v3 to use generated ICH_VMCR_EL2

Sascha Bischoff Sascha.Bischoff at arm.com
Wed Jan 7 02:55:18 PST 2026


On Tue, 2026-01-06 at 18:00 +0000, Jonathan Cameron wrote:
> On Fri, 19 Dec 2025 15:52:36 +0000
> Sascha Bischoff <Sascha.Bischoff at arm.com> wrote:
> 
> > From: Sascha Bischoff <Sascha.Bischoff at arm.com>
> > 
> > The VGIC-v3 code relied on hand-written definitions for the
> > ICH_VMCR_EL2 register. This register, and the associated fields, is
> > now generated as part of the sysreg framework. Move to using the
> > generated definitions instead of the hand-written ones.
> > 
> > There are no functional changes as part of this change.
> > 
> > Signed-off-by: Sascha Bischoff <sascha.bischoff at arm.com>
> Hi Sascha
> 
> Happy new year.  There is a bit in here that isn't obviously going
> to result in no functional change. I'm too lazy to chase where the
> value
> goes to check it it's a real bug or not.
> 
> Otherwise this is inconsistent on whether the _MASK or define without
> it from the sysreg generated header is used in FIELD_GET() /
> FIELD_PREP()
> 
> I'd always use the _MASK version.

Hi Jonathan,

I've updated the code to use the _MASK version.

> 
> > ---
> >  arch/arm64/include/asm/sysreg.h      | 21 ---------
> >  arch/arm64/kvm/hyp/vgic-v3-sr.c      | 64 ++++++++++++------------
> > ----
> >  arch/arm64/kvm/vgic/vgic-v3-nested.c |  8 ++--
> >  arch/arm64/kvm/vgic/vgic-v3.c        | 48 ++++++++++-----------
> >  4 files changed, 54 insertions(+), 87 deletions(-)
> > 
> > diff --git a/arch/arm64/include/asm/sysreg.h
> > b/arch/arm64/include/asm/sysreg.h
> > index 9df51accbb025..b3b8b8cd7bf1e 100644
> > --- a/arch/arm64/include/asm/sysreg.h
> > +++ b/arch/arm64/include/asm/sysreg.h
> 
> 
> > @@ -865,12 +865,12 @@ static void __vgic_v3_write_eoir(struct
> > kvm_vcpu *vcpu, u32 vmcr, int rt)
> >  
> >  static void __vgic_v3_read_igrpen0(struct kvm_vcpu *vcpu, u32
> > vmcr, int rt)
> >  {
> > -	vcpu_set_reg(vcpu, rt, !!(vmcr & ICH_VMCR_ENG0_MASK));
> > +	vcpu_set_reg(vcpu, rt, vmcr & ICH_VMCR_EL2_VENG0_MASK);
> >  }
> >  
> >  static void __vgic_v3_read_igrpen1(struct kvm_vcpu *vcpu, u32
> > vmcr, int rt)
> >  {
> > -	vcpu_set_reg(vcpu, rt, !!(vmcr & ICH_VMCR_ENG1_MASK));
> > +	vcpu_set_reg(vcpu, rt, vmcr & ICH_VMCR_EL2_VENG1_MASK);
> 
> It's more than possible it doesn't matter, but this isn't
> functionally
> equivalent.
> The original set passed 1 as the val parameter to vcpu_set_reg(), and
> now it passes 2.
> 
> Given these don't take a bool I'd use FIELD_GET() for both this and
> the veng0 one above.
> Or put back the horrible !!

Ah, that's a good catch, and might well be an unintended functional
change looking into it. I've switched to using FIELD_GET() for both.

> 
> >  }
> 
> > @@ -916,10 +916,8 @@ static void __vgic_v3_write_bpr0(struct
> > kvm_vcpu *vcpu, u32 vmcr, int rt)
> >  	if (val < bpr_min)
> >  		val = bpr_min;
> >  
> > -	val <<= ICH_VMCR_BPR0_SHIFT;
> > -	val &= ICH_VMCR_BPR0_MASK;
> > -	vmcr &= ~ICH_VMCR_BPR0_MASK;
> > -	vmcr |= val;
> > +	vmcr &= ~ICH_VMCR_EL2_VBPR0_MASK;
> > +	vmcr |= FIELD_PREP(ICH_VMCR_EL2_VBPR0, val);
> 
> You could uses FIELD_MODIFY() though that would mean using the _MASK
> define for both places.  Not sure why the sysreg script generates
> both
> (always have same actual value). I guess the idea is it is a little
> shorter if you don't want to be explicit that the intent is to use it
> as a mask.
> 
> I'd just use the _MASK defines throughout rather than trying for
> another
> consistent scheme. 

FIELD_MODIFY() is a great shout here. Done & thanks.

Yeah, I'd tried to use _MASK when explicitly using it as a mask, and
without in FIELD_x() (and still managed to be inconsistent with that).
I've now used _MASK everywhere.

> 
> 
> 
> 
> >  
> >  	__vgic_v3_write_vmcr(vmcr);
> >  }
> > @@ -929,17 +927,15 @@ static void __vgic_v3_write_bpr1(struct
> > kvm_vcpu *vcpu, u32 vmcr, int rt)
> >  	u64 val = vcpu_get_reg(vcpu, rt);
> >  	u8 bpr_min = __vgic_v3_bpr_min();
> >  
> > -	if (vmcr & ICH_VMCR_CBPR_MASK)
> > +	if (FIELD_GET(ICH_VMCR_EL2_VCBPR_MASK, val))
> >  		return;
> >  
> >  	/* Enforce BPR limiting */
> >  	if (val < bpr_min)
> >  		val = bpr_min;
> >  
> > -	val <<= ICH_VMCR_BPR1_SHIFT;
> > -	val &= ICH_VMCR_BPR1_MASK;
> > -	vmcr &= ~ICH_VMCR_BPR1_MASK;
> > -	vmcr |= val;
> > +	vmcr &= ~ICH_VMCR_EL2_VBPR1_MASK;
> > +	vmcr |= FIELD_PREP(ICH_VMCR_EL2_VBPR1, val);
> 
> As above, FIELD_MODIFY() makes this a one liner.
> 

Done.

> >  
> >  	__vgic_v3_write_vmcr(vmcr);
> >  }
> > @@ -1029,19 +1025,15 @@ static void __vgic_v3_read_hppir(struct
> > kvm_vcpu *vcpu, u32 vmcr, int rt)
> >  
> >  static void __vgic_v3_read_pmr(struct kvm_vcpu *vcpu, u32 vmcr,
> > int rt)
> >  {
> > -	vmcr &= ICH_VMCR_PMR_MASK;
> > -	vmcr >>= ICH_VMCR_PMR_SHIFT;
> > -	vcpu_set_reg(vcpu, rt, vmcr);
> > +	vcpu_set_reg(vcpu, rt, FIELD_GET(ICH_VMCR_EL2_VPMR,
> > vmcr));
> >  }
> >  
> >  static void __vgic_v3_write_pmr(struct kvm_vcpu *vcpu, u32 vmcr,
> > int rt)
> >  {
> >  	u32 val = vcpu_get_reg(vcpu, rt);
> >  
> > -	val <<= ICH_VMCR_PMR_SHIFT;
> > -	val &= ICH_VMCR_PMR_MASK;
> > -	vmcr &= ~ICH_VMCR_PMR_MASK;
> > -	vmcr |= val;
> > +	vmcr &= ~ICH_VMCR_EL2_VPMR_MASK;
> > +	vmcr |= FIELD_PREP(ICH_VMCR_EL2_VPMR, val);
> 
> FIELD_MODIFY() should be fine here I think.
> 

Done.

> >  
> >  	write_gicreg(vmcr, ICH_VMCR_EL2);
> >  }
> > @@ -1064,9 +1056,9 @@ static void __vgic_v3_read_ctlr(struct
> > kvm_vcpu *vcpu, u32 vmcr, int rt)
> >  	/* A3V */
> >  	val |= ((vtr >> 21) & 1) << ICC_CTLR_EL1_A3V_SHIFT;
> >  	/* EOImode */
> > -	val |= ((vmcr & ICH_VMCR_EOIM_MASK) >>
> > ICH_VMCR_EOIM_SHIFT) << ICC_CTLR_EL1_EOImode_SHIFT;
> > +	val |= FIELD_GET(ICH_VMCR_EL2_VEOIM, vmcr) <<
> > ICC_CTLR_EL1_EOImode_SHIFT;
> 
> Bit ugly to mix and match styles.
> ICC_CTRL_EL1_EOImode_MASK is defined so you could do a FIELD_PREP()

Done.

> 
> >  	/* CBPR */
> > -	val |= (vmcr & ICH_VMCR_CBPR_MASK) >> ICH_VMCR_CBPR_SHIFT;
> > +	val |= FIELD_GET(ICH_VMCR_EL2_VCBPR, vmcr);
> >  
> >  	vcpu_set_reg(vcpu, rt, val);
> >  }
> > @@ -1076,14 +1068,14 @@ static void __vgic_v3_write_ctlr(struct
> > kvm_vcpu *vcpu, u32 vmcr, int rt)
> >  	u32 val = vcpu_get_reg(vcpu, rt);
> >  
> >  	if (val & ICC_CTLR_EL1_CBPR_MASK)
> > -		vmcr |= ICH_VMCR_CBPR_MASK;
> > +		vmcr |= ICH_VMCR_EL2_VCBPR_MASK;
> >  	else
> > -		vmcr &= ~ICH_VMCR_CBPR_MASK;
> > +		vmcr &= ~ICH_VMCR_EL2_VCBPR_MASK;
> These could be something like
> 
> 	FIELD_MODIFY(ICH_VMCR_EL2_VCBPR_MASK, &vmcr,
> 		     FIELD_GET(ICC_CTRL_EL1_CBPR_MASK, val));
> 
> More compact. Whether more readable is a little bit more debatable.

I've gone with this for now. I think it is sufficiently readable.

> 
> >  
> >  	if (val & ICC_CTLR_EL1_EOImode_MASK)
> > -		vmcr |= ICH_VMCR_EOIM_MASK;
> > +		vmcr |= ICH_VMCR_EL2_VEOIM_MASK;
> >  	else
> > -		vmcr &= ~ICH_VMCR_EOIM_MASK;
> > +		vmcr &= ~ICH_VMCR_EL2_VEOIM_MASK;
> >  
> >  	write_gicreg(vmcr, ICH_VMCR_EL2);
> >  }
> 
> Thanks,
> 
> Jonathan
> 

Thanks a lot!
Sascha


More information about the linux-arm-kernel mailing list