[RFC PATCH] KVM: arm64: Override Microsoft Azure Cobalt 100 MIDR value with ARM Neoverse N2

[Marc Zyngier]

On Tue, 06 Feb 2024 19:58:16 +0000,
Easwar Hariharan <eahariha at linux.microsoft.com> wrote:
> 
> Several workload optimizations and errata depend on validating that the
> optimization or errata are applicable to the particular CPU by checking
> the MIDR_EL1 system register value. With the Microsoft implementer ID
> for Azure Cobalt 100, the value doesn't match and ~20-25% performance
> regression is seen in these workloads. Override the Azure Cobalt 100
> value and replace it with the default ARM Neoverse N2 value that Azure
> Cobalt 100 is based on.

Since you don't disclose *why* this particular value should have any
impact on the behaviour of the kernel, the answer should be "Thanks,
but no, thanks".

Whatever the reason is for doing so, you should make it plain what you
are working around. Blindly overriding ID registers is not an option,
and you should simply add your MIDR value to whatever errata list that
actually matches your implementation.

> 
> Signed-off-by: Easwar Hariharan <eahariha at linux.microsoft.com>
> ---
>  arch/arm64/include/asm/cputype.h   | 3 ++-
>  arch/arm64/include/asm/el2_setup.h | 5 +++++
>  arch/arm64/kvm/sys_regs.c          | 9 ++++++++-
>  3 files changed, 15 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/cputype.h b/arch/arm64/include/asm/cputype.h
> index 7c7493cb571f..0450c6c32377 100644
> --- a/arch/arm64/include/asm/cputype.h
> +++ b/arch/arm64/include/asm/cputype.h
> @@ -262,7 +262,8 @@ is_midr_in_range_list(u32 midr, struct midr_range const *ranges)
>   */
>  static inline u32 __attribute_const__ read_cpuid_id(void)
>  {
> -	return read_cpuid(MIDR_EL1);
> +	return (read_cpuid(MIDR_EL1) == 0x6D0FD490 ? 0x410FD490 :
> +			read_cpuid(MIDR_EL1));
>  }
>  
>  static inline u64 __attribute_const__ read_cpuid_mpidr(void)
> diff --git a/arch/arm64/include/asm/el2_setup.h b/arch/arm64/include/asm/el2_setup.h
> index b7afaa026842..502a14e54a31 100644
> --- a/arch/arm64/include/asm/el2_setup.h
> +++ b/arch/arm64/include/asm/el2_setup.h
> @@ -138,6 +138,11 @@
>  .macro __init_el2_nvhe_idregs
>  	mrs	x0, midr_el1
>  	mrs	x1, mpidr_el1
> +	ldr	x2, =0x6D0FD490
> +	cmp	x0, x2
> +	bne	.Loverride_cobalt100_\@
> +	ldr	x0, =0x410FD490
> +.Loverride_cobalt100_\@:
>  	msr	vpidr_el2, x0
>  	msr	vmpidr_el2, x1
>  .endm
> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> index 30253bd19917..8ea9c7fdabdb 100644
> --- a/arch/arm64/kvm/sys_regs.c
> +++ b/arch/arm64/kvm/sys_regs.c
> @@ -3574,7 +3574,14 @@ id_to_sys_reg_desc(struct kvm_vcpu *vcpu, u64 id,
>  		return ((struct sys_reg_desc *)r)->val;			\
>  	}
>  
> -FUNCTION_INVARIANT(midr_el1)
> +static u64 get_midr_el1(struct kvm_vcpu *v, const struct sys_reg_desc *r)
> +{
> +	((struct sys_reg_desc *)r)->val = read_sysreg(midr_el1);
> +	if (((struct sys_reg_desc *)r)->val == 0x6D0FD490)
> +		((struct sys_reg_desc *)r)->val == 0x410FD490;

As pointed out to me by Joey, this line is really interesting, and
shows that you didn't really test this patch.

Thanks,

	M.

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