Re: [PATCH v5 01/27] arm64: cpufeature: Set SYSREG_GIC_CPUIF as a boot system feature

[Marc Zyngier]

On 25/09/18 04:10, Yao Lihua wrote:
> Hi Marc, Julienï
>
>
> On 09/21/2018 11:56 PM, Marc Zyngier wrote:
> > On Tue, 28 Aug 2018 16:51:11 +0100,
> > Julien Thierry <julien.thierry@xxxxxxx> wrote:
> > > Signed-off-by: Julien Thierry <julien.thierry@xxxxxxx>
> > > Suggested-by: Daniel Thompson <daniel.thompson@xxxxxxxxxx>
> > > Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> > > Cc: Will Deacon <will.deacon@xxxxxxx>
> > > Cc: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
> > > Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
> > > ---
> > >   arch/arm64/kernel/cpufeature.c | 2 +-
> > >   1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
> > > index e238b79..1e433ac 100644
> > > --- a/arch/arm64/kernel/cpufeature.c
> > > +++ b/arch/arm64/kernel/cpufeature.c
> > > @@ -1039,7 +1039,7 @@ static void cpu_has_fwb(const struct arm64_cpu_capabilities *__unused)
> > >   	{
> > >   		.desc = "GIC system register CPU interface",
> > >   		.capability = ARM64_HAS_SYSREG_GIC_CPUIF,
> > > -		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
> > > +		.type = ARM64_CPUCAP_STRICT_BOOT_CPU_FEATURE,
> > >   		.matches = has_useable_gicv3_cpuif,
> > >   		.sys_reg = SYS_ID_AA64PFR0_EL1,
> > >   		.field_pos = ID_AA64PFR0_GIC_SHIFT,
> > > --
> > > 1.9.1
> > This definitely deserves a commit message, such as:
> >
> > "We do not support systems where some CPUs have an operational GICv3
> >   CPU interface, and some don't. Let's make this requirement obvious by
> >   flagging the GICv3 capability as being strict."
> May I ask if it is possible to implement psedue-NMI on a arm64 SoC with GIC-400?

In theory, yes. In practice, this is likely to be both hard to implement 
(you need to discover the GIC CPU interface address very early so that 
you can patch the the PMR flipping code at the right time), and pretty 
bad from a performance point of view (MMIO accesses are likely to be slow).

Given the above, the incentive to support such a configuration is close 
to zero.

Thanks,

	M.
--
Jazz is not dead. It just smells funny...