[PATCH 2/2] arm64: don't load 32-bit binaries if platform has no aarch32_el0

[mark.rutland@arm.com (Mark Rutland)]

On Tue, Sep 01, 2015 at 03:41:12PM +0100, Yury Norov wrote:
> Kernel option COMPAT defines the ability of executing aarch32 binaries.
> Some platforms does not support aarch32 mode, and so cannot execute that
> binaries. But we cannot just disable COMPAT for them because the same
> kernel binary may be used by multiple platforms.
> 
> In this patch, system_supports_aarch32_el0() is introduced to detect
> aarch32 support at run-time.
> 
> Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
> ---
>  arch/arm64/include/asm/cpufeature.h | 1 +
>  arch/arm64/include/asm/elf.h        | 6 ++++--
>  arch/arm64/kernel/cpuinfo.c         | 9 +++++++++
>  3 files changed, 14 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/cpufeature.h b/arch/arm64/include/asm/cpufeature.h
> index 20cdc26..d24ea15 100644
> --- a/arch/arm64/include/asm/cpufeature.h
> +++ b/arch/arm64/include/asm/cpufeature.h
> @@ -81,6 +81,7 @@ void check_local_cpu_errata(void);
>  void check_local_cpu_features(void);
>  bool cpu_supports_mixed_endian_el0(void);
>  bool system_supports_mixed_endian_el0(void);
> +bool system_supports_aarch32_el0(void);
>  
>  #endif /* __ASSEMBLY__ */
>  
> diff --git a/arch/arm64/include/asm/elf.h b/arch/arm64/include/asm/elf.h
> index faad6df..461897b 100644
> --- a/arch/arm64/include/asm/elf.h
> +++ b/arch/arm64/include/asm/elf.h
> @@ -21,6 +21,7 @@
>  /*
>   * ELF register definitions..
>   */
> +#include <asm/cpufeature.h>
>  #include <asm/ptrace.h>
>  #include <asm/user.h>
>  
> @@ -173,8 +174,9 @@ typedef compat_elf_greg_t		compat_elf_gregset_t[COMPAT_ELF_NGREG];
>  
>  /* AArch32 EABI. */
>  #define EF_ARM_EABI_MASK		0xff000000
> -#define compat_elf_check_arch(x)	(((x)->e_machine == EM_ARM) && \
> -					 ((x)->e_flags & EF_ARM_EABI_MASK))
> +#define compat_elf_check_arch(x)	(system_supports_aarch32_el0()	\
> +					&& ((x)->e_machine == EM_ARM)	\
> +					&& ((x)->e_flags & EF_ARM_EABI_MASK))
>  
>  #define compat_start_thread		compat_start_thread
>  #define COMPAT_SET_PERSONALITY(ex)	set_thread_flag(TIF_32BIT);
> diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c
> index 75d5a86..95d953f 100644
> --- a/arch/arm64/kernel/cpuinfo.c
> +++ b/arch/arm64/kernel/cpuinfo.c
> @@ -79,6 +79,15 @@ bool system_supports_mixed_endian_el0(void)
>  	return mixed_endian_el0;
>  }
>  
> +#define AARCH64		1
> +#define AARCH32_64	2

These should be better namespaced. Perhaps ID_AA64PFR0_EL1_EL0_64 and
ID_AA64PFR0_EL1_EL0_6432 ?

> +bool system_supports_aarch32_el0(void)
> +{
> +	struct cpuinfo_arm64 *info = this_cpu_ptr(&cpu_data);
> +	u64 arm64_el0 = info->reg_id_aa64pfr0 & 0xf;
> +	return arm64_el0 == AARCH32_64;
> +}

We should handle this the same way as we do for endianness support and
check that all CPUs support AArch32, and set a global flag, rather than
assuming that all CPUs are symmetric. Likewise for any other feature we
have to dynamically detect.

Thanks,
Mark.