On Wed, Jul 29, 2020 at 10:38 AM Catalin Marinas
<catalin.marinas@xxxxxxx> wrote:
>
> On Fri, Jul 17, 2020 at 02:52:33PM -0600, Rob Herring wrote:
> > diff --git a/arch/arm64/include/asm/kvm_hyp.h b/arch/arm64/include/asm/kvm_hyp.h
> > index ce3080834bfa..ce5b0d9b12bf 100644
> > --- a/arch/arm64/include/asm/kvm_hyp.h
> > +++ b/arch/arm64/include/asm/kvm_hyp.h
> > @@ -46,6 +46,17 @@
> > #define read_sysreg_el2(r) read_sysreg_elx(r, _EL2, _EL1)
> > #define write_sysreg_el2(v,r) write_sysreg_elx(v, r, _EL2, _EL1)
> >
> > +static inline u64 __hyp_text read_sysreg_par(void)
> > +{
> > + u64 par;
> > + if (cpus_have_const_cap(ARM64_WORKAROUND_1508412))
> > + dmb(sy);
> > + par = read_sysreg(par_el1);
> > + if (cpus_have_const_cap(ARM64_WORKAROUND_1508412))
> > + dmb(sy);
> > + return par;
> > +}
>
> Even if that's not always called on a critical path, I agree with Andrew
> that we could use alternatives here for dmb(sy).
His suggestion in the KVM code was to use cpus_have_final_cap() rather
than cpus_have_const_cap. But given it's just a dmb or nop,
alternatives is a better choice for all of these?
> > diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> > index baf5ce9225ce..3f798e0f1419 100644
> > --- a/arch/arm64/kvm/sys_regs.c
> > +++ b/arch/arm64/kvm/sys_regs.c
> > @@ -94,10 +94,16 @@ static bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
> > case TPIDR_EL1: *val = read_sysreg_s(SYS_TPIDR_EL1); break;
> > case AMAIR_EL1: *val = read_sysreg_s(SYS_AMAIR_EL12); break;
> > case CNTKCTL_EL1: *val = read_sysreg_s(SYS_CNTKCTL_EL12); break;
> > - case PAR_EL1: *val = read_sysreg_s(SYS_PAR_EL1); break;
> > case DACR32_EL2: *val = read_sysreg_s(SYS_DACR32_EL2); break;
> > case IFSR32_EL2: *val = read_sysreg_s(SYS_IFSR32_EL2); break;
> > case DBGVCR32_EL2: *val = read_sysreg_s(SYS_DBGVCR32_EL2); break;
> > + case PAR_EL1:
> > + if (cpus_have_const_cap(ARM64_WORKAROUND_1508412))
> > + dmb(sy);
> > + *val = read_sysreg_s(SYS_PAR_EL1);
> > + if (cpus_have_const_cap(ARM64_WORKAROUND_1508412))
> > + dmb(sy);
> > + break;
> > default: return false;
> > }
>
> Can't we use read_sysreg_par() directly here?
I assumed read_sysreg_s() was used here for some reason instead of
read_sysreg()?
> > diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
> > index 8afb238ff335..98609532e61a 100644
> > --- a/arch/arm64/mm/fault.c
> > +++ b/arch/arm64/mm/fault.c
> > @@ -260,7 +260,17 @@ static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
> > local_irq_save(flags);
> > asm volatile("at s1e1r, %0" :: "r" (addr));
> > isb();
> > + /*
> > + * Arm Erratum 1508412 requires dmb(sy) before and after reads of
> > + * PAR_EL1.
> > + * As this location is not a hot path, just condition it on the config
> > + * option.
> > + */
> > + if (IS_ENABLED(CONFIG_ARM64_ERRATUM_1508412))
> > + dmb(sy);
> > par = read_sysreg(par_el1);
> > + if (IS_ENABLED(CONFIG_ARM64_ERRATUM_1508412))
> > + dmb(sy);
> > local_irq_restore(flags);
>
> Why not read_sysreg_par()?
Okay with read_sysreg_par() going in asm/sysreg.h instead? I was
hesitant to add it there as there didn't seem to be any other
instances of a function for a specific register there.
Rob
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
