Re: [PATCH] irqchip/irq-gic-v3-its: Add the checking of ITS version for KVM

[Marc Zyngier]

On Fri, 11 Jun 2021 06:52:26 +0100,
Shaokun Zhang <zhangshaokun@xxxxxxxxxxxxx> wrote:
> 
> From: Wudi Wang <wangwudi@xxxxxxxxxxxxx>
> 
> The version of GIC used by KVM is provided by gic_v3_kvm_info.
> The KVM that supports GICv4 or GICv4.1 only checks GIC
> version. Actually, the GIC and ITS need to work together.
> So we add the checking of ITS version for KVM: If and only if
> both GIC & ITS support GICv4, gic_kvm_info.has_v4 is true.
> If and only if both GIC & ITS support GICv4.1,
> gic_kvm_info.has_v4_1 is true.
> 
> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: Marc Zyngier <maz@xxxxxxxxxx>
> Signed-off-by: Wudi Wang <wangwudi@xxxxxxxxxxxxx>
> Signed-off-by: Shaokun Zhang <zhangshaokun@xxxxxxxxxxxxx>
> ---
>  drivers/irqchip/irq-gic-common.h |  2 ++
>  drivers/irqchip/irq-gic-v3-its.c |  3 +++
>  drivers/irqchip/irq-gic-v3.c     | 10 +++++-----
>  3 files changed, 10 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/irqchip/irq-gic-common.h b/drivers/irqchip/irq-gic-common.h
> index ccba8b0fe0f5..e5d44998445a 100644
> --- a/drivers/irqchip/irq-gic-common.h
> +++ b/drivers/irqchip/irq-gic-common.h
> @@ -10,6 +10,8 @@
>  #include <linux/irqdomain.h>
>  #include <linux/irqchip/arm-gic-common.h>
>  
> +extern struct gic_kvm_info gic_v3_kvm_info;
> +
>  struct gic_quirk {
>  	const char *desc;
>  	const char *compatible;
> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> index 2e6923c2c8a8..45d6163c14d5 100644
> --- a/drivers/irqchip/irq-gic-v3-its.c
> +++ b/drivers/irqchip/irq-gic-v3-its.c
> @@ -5419,6 +5419,9 @@ int __init its_init(struct fwnode_handle *handle, struct rdists *rdists,
>  		has_v4_1 |= is_v4_1(its);
>  	}
>  
> +	gic_v3_kvm_info.has_v4 = has_v4;
> +	gic_v3_kvm_info.has_v4_1 = has_v4_1;

If you are going down that road: what if you have multiple ITSs,
implementing a variety of v3, v4, v4.1? We currently support this to
some extent, but this is breaking it.

What case are you exactly trying to fix?

> +
>  	/* Don't bother with inconsistent systems */
>  	if (WARN_ON(!has_v4_1 && rdists->has_rvpeid))
>  		rdists->has_rvpeid = false;
> diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
> index 37a23aa6de37..7454f99bf580 100644
> --- a/drivers/irqchip/irq-gic-v3.c
> +++ b/drivers/irqchip/irq-gic-v3.c
> @@ -103,7 +103,7 @@ EXPORT_SYMBOL(gic_nonsecure_priorities);
>  /* ppi_nmi_refs[n] == number of cpus having ppi[n + 16] set as NMI */
>  static refcount_t *ppi_nmi_refs;
>  
> -static struct gic_kvm_info gic_v3_kvm_info;
> +struct gic_kvm_info gic_v3_kvm_info;

This will conflict with the rework that is in -next, and maybe cause
some lifetime issue (see how the structure is now tagged __initdata).

>  static DEFINE_PER_CPU(bool, has_rss);
>  
>  #define MPIDR_RS(mpidr)			(((mpidr) & 0xF0UL) >> 4)
> @@ -1850,8 +1850,8 @@ static void __init gic_of_setup_kvm_info(struct device_node *node)
>  	if (!ret)
>  		gic_v3_kvm_info.vcpu = r;
>  
> -	gic_v3_kvm_info.has_v4 = gic_data.rdists.has_vlpis;
> -	gic_v3_kvm_info.has_v4_1 = gic_data.rdists.has_rvpeid;
> +	gic_v3_kvm_info.has_v4 &= gic_data.rdists.has_vlpis;
> +	gic_v3_kvm_info.has_v4_1 &= gic_data.rdists.has_rvpeid;
>  	gic_set_kvm_info(&gic_v3_kvm_info);
>  }
>  
> @@ -2166,8 +2166,8 @@ static void __init gic_acpi_setup_kvm_info(void)
>  		vcpu->end = vcpu->start + ACPI_GICV2_VCPU_MEM_SIZE - 1;
>  	}
>  
> -	gic_v3_kvm_info.has_v4 = gic_data.rdists.has_vlpis;
> -	gic_v3_kvm_info.has_v4_1 = gic_data.rdists.has_rvpeid;
> +	gic_v3_kvm_info.has_v4 &= gic_data.rdists.has_vlpis;
> +	gic_v3_kvm_info.has_v4_1 &= gic_data.rdists.has_rvpeid;
>  	gic_set_kvm_info(&gic_v3_kvm_info);
>  }

Thanks,

	M.

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