Re: [PATCH] irqchip: gic-v3: fix unmet dependency on ARM_PSCI_FW

Hi Marc,

On 8/20/26 09:22, Marc Zyngier wrote:

> [+ SMCCC folks]
> 
> Thanks for roping me in.
> 
> This looks like the wrong fix altogether. Whatever the Broadcom stuff
> depends on is irrelevant, and has nothing to do with GICv3. Same thing
> for the NXP contraptions.
> 
> If there is something to do *for the GIC driver*, it is to make sure
> it compiles/links correctly when the platform does not select options
> that are not required. Why would NXP and Broadcom care about an
> erratum that is specific to NVIDIA?
> 
> I also disagree with what the commit message asserts, ARM_PSCI_FW is
> selected *globally* on arm64, not on a per platform basis (TFFT!).
> The NVIDIA machine being an arm64 platform, the correct dependency fix
> is to select HAVE_ARM_SMCCC_DISCOVERY on arm64.
> 
> The overall issue is that there is no fallback to do something
> sensible when HAVE_ARM_SMCCC_DISCOVERY is not selected. Random drivers
> should not have to guess or add random #ifdefs all over the shop.
> 
> I came up with the following hack, which compiles correctly on 32bit
> ARM without PSCI at all. YMMV.
> 
> Thanks,
> 
> 	M.
> 
> diff --git a/drivers/irqchip/Kconfig b/drivers/irqchip/Kconfig
> index 42f2278a702d0..937f17ec93a2e 100644
> --- a/drivers/irqchip/Kconfig
> +++ b/drivers/irqchip/Kconfig
> @@ -37,7 +37,7 @@ config ARM_GIC_V3
>  	bool
>  	select IRQ_DOMAIN_HIERARCHY
>  	select GENERIC_IRQ_EFFECTIVE_AFF_MASK if SMP
> -	select HAVE_ARM_SMCCC_DISCOVERY
> +	select HAVE_ARM_SMCCC_DISCOVERY if ARM64
>  	select IRQ_MSI_IOMMU
>  
>  config ARM_GIC_ITS_PARENT
> diff --git a/include/linux/arm-smccc.h b/include/linux/arm-smccc.h
> index 4de81848fe2ee..75f9d3f1723b5 100644
> --- a/include/linux/arm-smccc.h
> +++ b/include/linux/arm-smccc.h
> @@ -324,6 +324,7 @@ enum arm_smccc_conduit {
>  	SMCCC_CONDUIT_HVC,
>  };
>  
> +#ifdef CONFIG_HAVE_ARM_SMCCC_DISCOVERY
>  /**
>   * arm_smccc_1_1_get_conduit()
>   *
> @@ -363,6 +364,12 @@ s32 arm_smccc_get_soc_id_version(void);
>   * When ARM_SMCCC_ARCH_SOC_ID is not present, returns SMCCC_RET_NOT_SUPPORTED.
>   */
>  s32 arm_smccc_get_soc_id_revision(void);
> +#else   /* CONFIG_HAVE_ARM_SMCCC_DISCOVERY */
> +#define arm_smccc_1_1_get_conduit()	SMCCC_CONDUIT_NONE
> +#define arm_smccc_get_version()		0
> +#define arm_smccc_get_soc_id_version()	SMCCC_RET_NOT_SUPPORTED
> +#define arm_smccc_get_soc_id_revision()	SMCCC_RET_NOT_SUPPORTED
> +#endif	/* CONFIG_HAVE_ARM_SMCCC_DISCOVERY */
>  
>  #ifndef __ASSEMBLY__
>  
> 

Thank you for explaining. From my tests, your version resolves the unmet
dependency well and doesn't introduce others.

- Julian Braha