On Fri, Sep 11, 2020 at 11:37:14AM +0100, Catalin Marinas wrote:
> On Wed, Sep 09, 2020 at 05:13:10PM -0600, Rob Herring wrote:
> > diff --git a/arch/arm64/include/asm/sysreg.h b/arch/arm64/include/asm/sysreg.h
> > index 554a7e8ecb07..55dfff8ca466 100644
> > --- a/arch/arm64/include/asm/sysreg.h
> > +++ b/arch/arm64/include/asm/sysreg.h
> > @@ -943,6 +943,7 @@
> >
> > #include <linux/build_bug.h>
> > #include <linux/types.h>
> > +#include <asm/alternative.h>
> >
> > #define __DEFINE_MRS_MSR_S_REGNUM \
> > " .irp num,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30\n" \
> > @@ -1024,6 +1025,17 @@
> > write_sysreg(__scs_new, sysreg); \
> > } while (0)
> >
> > +#define read_sysreg_par() ({ \
> > + unsigned long flags; \
> > + u64 par; \
> > + local_irq_save(flags); \
> > + asm(ALTERNATIVE("nop", "dmb sy", ARM64_WORKAROUND_1508412)); \
> > + par = read_sysreg(par_el1); \
> > + asm(ALTERNATIVE("nop", "dmb sy", ARM64_WORKAROUND_1508412)); \
> > + local_irq_restore(flags); \
> > + par; \
> > +})
>
> As an alternative to local_irq_save/restore, we could have added a dmb
> in the kernel_exit macro. The minor nit here is that we always disable
> interrupts even when the erratum doesn't apply.
>
> The EL1 code accessing PAR_EL1 already runs with interrupts disabled
> (which covers the prior AT instruction). If that's the case for KVM as
> well, we could drop the local_irq_* entirely and just leave the DMB in
> the exit to guest code.
I wonder if that's actually a more robust approach in the case of psuedo
NMIs using ARM64_HAS_IRQ_PRIO_MASKING?
Will
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
