Re: [PATCH 3/7] KVM: arm64: Add save/restore support for FPMR

[Marc Zyngier <maz@kernel.org>]

On Mon, 08 Jul 2024 18:47:45 +0100,
Marc Zyngier <maz@kernel.org> wrote:
> 
> On Mon, 08 Jul 2024 18:34:36 +0100,
> Mark Brown <broonie@kernel.org> wrote:
> > 
> > [1  <text/plain; us-ascii (7bit)>]
> > On Mon, Jul 08, 2024 at 04:44:34PM +0100, Marc Zyngier wrote:
> > > Just like the rest of the FP/SIMD state, FPMR needs to be context
> > > switched.
> > 
> > > The only interesting thing here is that we need to treat the pKVM
> > > part a bit differently, as the host FP state is never written back
> > > to the vcpu thread, but instead stored locally and eagerly restored.
> > 
> > > Signed-off-by: Marc Zyngier <maz@kernel.org>
> > > ---
> > >  arch/arm64/include/asm/kvm_host.h  | 10 ++++++++++
> > >  arch/arm64/kvm/fpsimd.c            |  1 +
> > >  arch/arm64/kvm/hyp/nvhe/hyp-main.c |  4 ++++
> > >  arch/arm64/kvm/hyp/nvhe/switch.c   | 10 ++++++++++
> > >  arch/arm64/kvm/hyp/vhe/switch.c    |  4 ++++
> > >  5 files changed, 29 insertions(+)
> > 
> > I'm possibly missing something here but I'm not seeing where we load the
> > state for the guest, especially in the VHE case.  I would expect to see
> > a change in kvm_hyp_handle_fpsimd() to load FPMR for guests with the
> > feature (it needs to be in there to keep in sync with the ownership
> > tracking for the rest of the FP state, and to avoid loading needlessly
> > in cases where the guest never touches FP).
> > 
> > Saving for the guest was handled in the previous patch.
> > 
> > > diff --git a/arch/arm64/kvm/hyp/vhe/switch.c b/arch/arm64/kvm/hyp/vhe/switch.c
> > > index 77010b76c150f..a307c1d5ac874 100644
> > > --- a/arch/arm64/kvm/hyp/vhe/switch.c
> > > +++ b/arch/arm64/kvm/hyp/vhe/switch.c
> > > @@ -312,6 +312,10 @@ static bool kvm_hyp_handle_eret(struct kvm_vcpu *vcpu, u64 *exit_code)
> > >  static void kvm_hyp_save_fpsimd_host(struct kvm_vcpu *vcpu)
> > >  {
> > >  	__fpsimd_save_state(*host_data_ptr(fpsimd_state));
> > > +
> > > +	if (system_supports_fpmr() &&
> > > +	    kvm_has_feat(vcpu->kvm, ID_AA64PFR2_EL1, FPMR, IMP))
> > > +		**host_data_ptr(fpmr_ptr) = read_sysreg_s(SYS_FPMR);
> > >  }
> > 
> > That's only saving the host state, it doesn't load the guest state.
> 
> Ah, I forgot to cherry-pick the fixes. Fsck knows what else I forgot.
> Thanks for reminding me.
> 
> 	M.
> 
> diff --git a/arch/arm64/kvm/hyp/include/hyp/switch.h b/arch/arm64/kvm/hyp/include/hyp/switch.h
> index f59ccfe11ab9a..52c7dc8446f16 100644
> --- a/arch/arm64/kvm/hyp/include/hyp/switch.h
> +++ b/arch/arm64/kvm/hyp/include/hyp/switch.h
> @@ -404,6 +404,10 @@ static bool kvm_hyp_handle_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
>  	else
>  		__fpsimd_restore_state(&vcpu->arch.ctxt.fp_regs);
>  
> +	if (system_supports_fpmr() &&
> +	    kvm_has_feat(kern_hyp_va(vcpu->kvm), ID_AA64PFR2_EL1, FPMR, IMP))
> +		write_sysreg_s(__vcpu_sys_reg(vcpu, FPMR), SYS_FPMR);
> +
>  	/* Skip restoring fpexc32 for AArch64 guests */
>  	if (!(read_sysreg(hcr_el2) & HCR_RW))
>  		write_sysreg(__vcpu_sys_reg(vcpu, FPEXC32_EL2), fpexc32_el2);
> 

And thinking of it, that's not enough. I'm missing the pKVM angle that
needs to be special cased. Drat.

	M.

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