Re: [PATCH v3 3/7] xen/arm: create a cpuinfo structure for guest

[Julien Grall <julien@xxxxxxx>]

Hi Bertand,

On 09/12/2020 16:30, Bertrand Marquis wrote:
> Create a cpuinfo structure for guest and mask into it the features that
> we do not support in Xen or that we do not want to publish to guests.
>
> Modify some values in the cpuinfo structure for guests to mask some
> features which we do not want to allow to guests (like AMU) or we do not
> support (like SVE).
>
> The code is trying to group together registers modifications for the
> same feature to be able in the long term to easily enable/disable a
> feature depending on user parameters or add other registers modification
> in the same place (like enabling/disabling HCR bits).
>
> Signed-off-by: Bertrand Marquis <bertrand.marquis@xxxxxxx>
> ---
> Changes in V2: Rebase
> Changes in V3:
>    Use current_cpu_data info instead of recalling identify_cpu
>
> ---
>   xen/arch/arm/cpufeature.c        | 51 ++++++++++++++++++++++++++++++++
>   xen/include/asm-arm/cpufeature.h |  2 ++
>   2 files changed, 53 insertions(+)
>
> diff --git a/xen/arch/arm/cpufeature.c b/xen/arch/arm/cpufeature.c
> index bc7ee5ac95..7255383504 100644
> --- a/xen/arch/arm/cpufeature.c
> +++ b/xen/arch/arm/cpufeature.c
> @@ -24,6 +24,8 @@
>   
>   DECLARE_BITMAP(cpu_hwcaps, ARM_NCAPS);
>   
> +struct cpuinfo_arm __read_mostly guest_cpuinfo;
> +
>   void update_cpu_capabilities(const struct arm_cpu_capabilities *caps,
>                                const char *info)
>   {
> @@ -157,6 +159,55 @@ void identify_cpu(struct cpuinfo_arm *c)
>   #endif
>   }
>   
> +/*
> + * This function is creating a cpuinfo structure with values modified to mask
> + * all cpu features that should not be published to guest.
> + * The created structure is then used to provide ID registers values to guests.
> + */
> +static int __init create_guest_cpuinfo(void)
> +{
> +    /*
> +     * TODO: The code is currently using only the features detected on the boot
> +     * core. In the long term we should try to compute values containing only
> +     * features supported by all cores.
> +     */
> +    guest_cpuinfo = current_cpu_data;
It would be more logical to use boot_cpu_data as this would be easier to 
match with your comment.
> +
> +#ifdef CONFIG_ARM_64
> +    /* Disable MPAM as xen does not support it */
> +    guest_cpuinfo.pfr64.mpam = 0;
> +    guest_cpuinfo.pfr64.mpam_frac = 0;
> +
> +    /* Disable SVE as Xen does not support it */
> +    guest_cpuinfo.pfr64.sve = 0;
> +    guest_cpuinfo.zfr64.bits[0] = 0;
> +
> +    /* Disable MTE as Xen does not support it */
> +    guest_cpuinfo.pfr64.mte = 0;
> +#endif
> +
> +    /* Disable AMU */
> +#ifdef CONFIG_ARM_64
> +    guest_cpuinfo.pfr64.amu = 0;
> +#endif
> +    guest_cpuinfo.pfr32.amu = 0;
> +
> +    /* Disable RAS as Xen does not support it */
> +#ifdef CONFIG_ARM_64
> +    guest_cpuinfo.pfr64.ras = 0;
> +    guest_cpuinfo.pfr64.ras_frac = 0;
> +#endif
> +    guest_cpuinfo.pfr32.ras = 0;
> +    guest_cpuinfo.pfr32.ras_frac = 0;
How about all the fields that are currently marked as RES0/RES1? 
Shouldn't we make sure they will stay like that even if newer 
architecture use them?
> +
> +    return 0;
> +}
> +/*
> + * This function needs to be run after all smp are started to have
> + * cpuinfo structures for all cores.
> + */
> +__initcall(create_guest_cpuinfo);
> +
>   /*
>    * Local variables:
>    * mode: C
> diff --git a/xen/include/asm-arm/cpufeature.h b/xen/include/asm-arm/cpufeature.h
> index 6cf83d775b..10b62bd324 100644
> --- a/xen/include/asm-arm/cpufeature.h
> +++ b/xen/include/asm-arm/cpufeature.h
> @@ -283,6 +283,8 @@ extern void identify_cpu(struct cpuinfo_arm *);
>   extern struct cpuinfo_arm cpu_data[];
>   #define current_cpu_data cpu_data[smp_processor_id()]
>   
> +extern struct cpuinfo_arm guest_cpuinfo;
> +
>   #endif /* __ASSEMBLY__ */
>   
>   #endif
--
Julien Grall