Re: [PATCH 2/4] KVM: arm64: maintain per VM value for CTR_EL0

[Sebastian Ott]

On Sat, 13 Apr 2024, Marc Zyngier wrote:
> On Fri, 05 Apr 2024 13:01:06 +0100,
> Sebastian Ott <sebott@xxxxxxxxxx> wrote:
> > In preparation for CTR_EL0 emulation maintain a per VM for this
> > register and use it where appropriate.
> >
> > Signed-off-by: Sebastian Ott <sebott@xxxxxxxxxx>
> > ---
> >  arch/arm64/include/asm/kvm_host.h |  1 +
> >  arch/arm64/kvm/sys_regs.c         | 22 +++++++++++++++-------
> >  2 files changed, 16 insertions(+), 7 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> > index 9e8a496fb284..481216febb46 100644
> > --- a/arch/arm64/include/asm/kvm_host.h
> > +++ b/arch/arm64/include/asm/kvm_host.h
> > @@ -318,6 +318,7 @@ struct kvm_arch {
> >
> >  	/* PMCR_EL0.N value for the guest */
> >  	u8 pmcr_n;
> > +	u64 ctr_el0;
> >
> >  	/* Iterator for idreg debugfs */
> >  	u8	idreg_debugfs_iter;
>
> Please consider the alignment of the fields. This leaves a 7 byte hole
> that could be avoided (yes, I'm on a mission to reduce the size of the
> various structures, because they are absolute pigs).

OK.

> > diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> > index 131f5b0ca2b9..4d29b1a0842d 100644
> > --- a/arch/arm64/kvm/sys_regs.c
> > +++ b/arch/arm64/kvm/sys_regs.c
> > @@ -215,13 +215,21 @@ void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
> >  /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
> >  #define CSSELR_MAX 14
> >
> > +static u64 kvm_get_ctr_el0(struct kvm *kvm)
> > +{
> > +	if (kvm->arch.ctr_el0)
> > +		return kvm->arch.ctr_el0;
>
> Is this relying on some bits not being 0?
>
> > +
> > +	return read_sanitised_ftr_reg(SYS_CTR_EL0);
>
> Why isn't the shadow value always populated?

The idea was for kvm->arch.ctr_el0 being non zero only if userspace
set it up to differ from the host value. So it can be used to decide
if we need to set up a trap for the reg access (without comparing it
to the host value again).

Thanks,
Sebastian