[boot-wrapper PATCH 2/2] aarch64: Enable use of SME by EL2 and below

[Mark Rutland]

Hi Mark,

On Tue, Feb 01, 2022 at 05:21:32PM +0000, broonie at kernel.org wrote:
> From: Mark Brown <broonie at kernel.org>
> 
> Allow lower ELs to use SME when booted on a system that support it. This
> requires us to set two new bits, one in each of SCR_EL3 and CPTR_EL3, set
> the maximum vector length in a similar fashion to SVE and if the optional
> FA64 feature is present then set another feature bit in the new SMCR
> register.

This looks good to me. I'll wait a day or so to allow others to review, but if
no-one shouts I'll apply this (along with patch 1) before the end of the week.

There's one trivial fixup I intend to make when applying (noted below), but
regardless this looks good structurally and value-wise, so no need to respin
unless someone shouts.

I checked the register encodings and bit definitions against developer.arm.com:

* CPTR_EL3
  https://developer.arm.com/documentation/ddi0601/2021-12/AArch64-Registers/CPTR-EL3--Architectural-Feature-Trap-Register--EL3-?lang=en

* ID_AA64PFR1_EL1
  https://developer.arm.com/documentation/ddi0601/2021-12/AArch64-Registers/ID-AA64PFR1-EL1--AArch64-Processor-Feature-Register-1?lang=en

* ID_AA64SMFR0_EL1
  https://developer.arm.com/documentation/ddi0601/2021-12/AArch64-Registers/ID-AA64SMFR0-EL1--SME-Feature-ID-register-0?lang=en

* SCR_EL3
  https://developer.arm.com/documentation/ddi0601/2021-12/AArch64-Registers/SCR-EL3--Secure-Configuration-Register?lang=en

> 
> Signed-off-by: Mark Brown <broonie at kernel.org>
> ---
>  arch/aarch64/include/asm/cpu.h | 10 ++++++++++
>  arch/aarch64/init.c            | 22 ++++++++++++++++++++++
>  2 files changed, 32 insertions(+)
> 
> diff --git a/arch/aarch64/include/asm/cpu.h b/arch/aarch64/include/asm/cpu.h
> index ce80b6e..49f5a71 100644
> --- a/arch/aarch64/include/asm/cpu.h
> +++ b/arch/aarch64/include/asm/cpu.h
> @@ -49,6 +49,7 @@
>  #define SCR_EL3_FGTEN			BIT(27)
>  #define SCR_EL3_ECVEN			BIT(28)
>  #define SCR_EL3_TME			BIT(34)
> +#define SCR_EL3_EnTP2			BIT(41)
>
>  #define HCR_EL2_RES1			BIT(1)
>  
> @@ -69,9 +70,13 @@
>  #define ID_AA64MMFR0_EL1_FGT		BITS(59, 56)
>  #define ID_AA64MMFR0_EL1_ECV		BITS(63, 60)
>  
> +#define ID_AA64PFR1_EL1_SME		BITS(27, 24)
>  #define ID_AA64PFR1_EL1_MTE		BITS(11, 8)
>  #define ID_AA64PFR0_EL1_SVE		BITS(35, 32)
>  
> +#define ID_AA64SMFR0_EL1		s3_0_c0_c4_5
> +#define ID_AA64SMFR0_EL1_FA64		(1UL << 63)

For consistency with the other ID fields, I'm going to make this:

   #define ID_AA64SMFR0_EL1_FA64		BIT(63)

In future I'd like to split the remaining definitions using shifted bits into
separate <register>_<field> and <register>_<field>_<value> definitions (or
something of that rought shape) so that field boundaries are always explicit,
but those can stay as-is for now.

> +
>  /*
>   * Initial register values required for the boot-wrapper to run out-of-reset.
>   */
> @@ -96,6 +101,7 @@
>  #define SPSR_EL2H		(9 << 0)	/* EL2 Handler mode */
>  #define SPSR_HYP		(0x1a << 0)	/* M[3:0] = hyp, M[4] = AArch32 */
>  
> +#define CPTR_EL3_ESM		(1 << 12)
>  #define CPTR_EL3_EZ		(1 << 8)
>  
>  #define ICC_SRE_EL2		S3_4_C12_C9_5
> @@ -107,6 +113,10 @@
>  #define ZCR_EL3			s3_6_c1_c2_0
>  #define ZCR_EL3_LEN_MAX		0xf
>  
> +#define SMCR_EL3		s3_6_c1_c2_6
> +#define SMCR_EL3_FA64		(1 << 31)
> +#define SMCR_EL3_LEN_MAX	0xf
> +
>  #define ID_AA64ISAR2_EL1	s3_0_c0_c6_2
>  
>  #define SCTLR_EL1_CP15BEN	(1 << 5)
> diff --git a/arch/aarch64/init.c b/arch/aarch64/init.c
> index 8bb0524..db73b58 100644
> --- a/arch/aarch64/init.c
> +++ b/arch/aarch64/init.c
> @@ -47,6 +47,7 @@ void cpu_init_el3(void)
>  	unsigned long scr = SCR_EL3_RES1 | SCR_EL3_NS | SCR_EL3_HCE;
>  	unsigned long mdcr = 0;
>  	unsigned long cptr = 0;
> +	unsigned long smcr = 0;
>  
>  	if (cpu_has_pauth())
>  		scr |= SCR_EL3_APK | SCR_EL3_API;
> @@ -95,6 +96,27 @@ void cpu_init_el3(void)
>  		msr(ZCR_EL3, ZCR_EL3_LEN_MAX);
>  	}
>  
> +	if (mrs_field(ID_AA64PFR1_EL1, SME)) {
> +		cptr |= CPTR_EL3_ESM;
> +		msr(CPTR_EL3, cptr);
> +		isb();
> +
> +		scr |= SCR_EL3_EnTP2;
> +		msr(SCR_EL3, scr);
> +		isb();
> +
> +		/*
> +		 * Write the maximum possible vector length, hardware
> +		 * will constrain to the actual limit.
> +		 */
> +		smcr = SMCR_EL3_LEN_MAX;
> +
> +		if (mrs_field(ID_AA64SMFR0_EL1, FA64))
> +			smcr |= SMCR_EL3_FA64;
> +
> +		msr(SMCR_EL3, smcr);
> +	}
> +
>  	msr(CNTFRQ_EL0, COUNTER_FREQ);
>  }
>  
> -- 
> 2.30.2
>