Re: [PATCH v2 2/3] riscv: cpufeature: Add ISA extension parsing for Supm

[Samuel Holland <samuel.holland@sifive.com>]

Hi Guodong,

On 2026-01-24 7:36 PM, Guodong Xu wrote:
> Supm has been ratified in the RISC-V Pointer Masking specification
> (Version 1.0, 10/2024) and is mandated in RVA23 Profiles (Version 1.0,
> 2024-10-17) for RVA23U64. Supm indicates userspace pointer masking
> support.
> 
> Remove the previous macro aliasing of Supm to Ssnpm/Smnpm in hwcap.h,
> treating Supm as a distinct RISC-V ISA extension ID.
> 
> Add ISA parsing logic for Supm, and implement a validator to ensure
> that Supm is only reported as available if Kconfig allows it and the
> underlying Ssnpm (for supervisor mode) or Smnpm (for machine mode)
> extension is present. Supm relies on Ssnpm or Smnpm to provide the
> underlying hardware implementation.
> 
> With this change, "supm" will be reported (when available) in
> /proc/cpuinfo as part of the "isa" and "hart isa" string.
> 
> Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
> Link: https://lore.kernel.org/lkml/20260101-legume-engraved-0fae8282cfbe@spud/#r [1]
> Link: https://lore.kernel.org/all/4ebbe14b-2579-4ba6-808d-d50c24641d04@sifive.com/#r [2]
> Signed-off-by: Guodong Xu <guodong@riscstar.com>
> ---
> v2: Add Reviewed-by from Conor.
>     Update RISCV_ISA_EXT_SUPM id to solve rebase conflict.
> ---
>  arch/riscv/include/asm/hwcap.h |  3 +--
>  arch/riscv/kernel/cpufeature.c | 35 +++++++++++++++++++++++++++++++++--
>  2 files changed, 34 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/riscv/include/asm/hwcap.h b/arch/riscv/include/asm/hwcap.h
> index 7ef8e5f55c8d..aa2af21f3bd3 100644
> --- a/arch/riscv/include/asm/hwcap.h
> +++ b/arch/riscv/include/asm/hwcap.h
> @@ -112,6 +112,7 @@
>  #define RISCV_ISA_EXT_ZCLSD		103
>  #define RISCV_ISA_EXT_ZICFILP		104
>  #define RISCV_ISA_EXT_ZICFISS		105
> +#define RISCV_ISA_EXT_SUPM		106
>  
>  #define RISCV_ISA_EXT_XLINUXENVCFG	127
>  
> @@ -120,10 +121,8 @@
>  
>  #ifdef CONFIG_RISCV_M_MODE
>  #define RISCV_ISA_EXT_SxAIA		RISCV_ISA_EXT_SMAIA
> -#define RISCV_ISA_EXT_SUPM		RISCV_ISA_EXT_SMNPM
>  #else
>  #define RISCV_ISA_EXT_SxAIA		RISCV_ISA_EXT_SSAIA
> -#define RISCV_ISA_EXT_SUPM		RISCV_ISA_EXT_SSNPM
>  #endif
>  
>  #endif /* _ASM_RISCV_HWCAP_H */
> diff --git a/arch/riscv/kernel/cpufeature.c b/arch/riscv/kernel/cpufeature.c
> index 1734f9a4c2fd..e1f7ad882289 100644
> --- a/arch/riscv/kernel/cpufeature.c
> +++ b/arch/riscv/kernel/cpufeature.c
> @@ -317,6 +317,27 @@ static int riscv_cfiss_validate(const struct riscv_isa_ext_data *data,
>  	return 0;
>  }
>  
> +static int riscv_ext_supm_validate(const struct riscv_isa_ext_data *data,
> +				   const unsigned long *isa_bitmap)
> +{
> +	if (!IS_ENABLED(CONFIG_RISCV_ISA_SUPM))
> +		return -EINVAL;
> +
> +	/*
> +	 * Supm requires Ssnpm for S-mode or Smnpm for M-mode to provide
> +	 * pointer masking for the U-mode execution environment.
> +	 */
> +	if (IS_ENABLED(CONFIG_RISCV_M_MODE)) {
> +		if (__riscv_isa_extension_available(isa_bitmap, RISCV_ISA_EXT_SMNPM))
> +			return 0;
> +	} else {
> +		if (__riscv_isa_extension_available(isa_bitmap, RISCV_ISA_EXT_SSNPM))
> +			return 0;
> +	}
> +
> +	return -EPROBE_DEFER;
> +}
> +
>  static const unsigned int riscv_a_exts[] = {
>  	RISCV_ISA_EXT_ZAAMO,
>  	RISCV_ISA_EXT_ZALRSC,
> @@ -450,6 +471,15 @@ static const unsigned int riscv_c_exts[] = {
>  	RISCV_ISA_EXT_ZCD,
>  };
>  
> +/*
> + * Smnpm and Ssnpm provide pointer masking for the next lower privilege mode
> + * (U-mode), thus enabling Supm. Both extensions imply the same subset.

If Linux is running in S-mode, then Smnpm does _not_ imply Supm. So this list
cannot be shared between Ssnpm and Smnpm. (When running Linux in M-mode, I think
we assume S-mode isn't supported, so the opposite case isn't possible. If you do
run M-mode Linux on M/S/U hardware, I think there are other things that explode
spectacularly, since we will touch for example the wrong envcfg register.)

> + */
> +static const unsigned int riscv_supm_exts[] = {
> +	RISCV_ISA_EXT_XLINUXENVCFG,
> +	RISCV_ISA_EXT_SUPM
> +};
> +
>  /*
>   * The canonical order of ISA extension names in the ISA string is defined in
>   * chapter 27 of the unprivileged specification.
> @@ -577,12 +607,13 @@ const struct riscv_isa_ext_data riscv_isa_ext[] = {
>  	__RISCV_ISA_EXT_DATA_VALIDATE(zvkt, RISCV_ISA_EXT_ZVKT, riscv_ext_vector_crypto_validate),
>  	__RISCV_ISA_EXT_DATA(smaia, RISCV_ISA_EXT_SMAIA),
>  	__RISCV_ISA_EXT_DATA(smmpm, RISCV_ISA_EXT_SMMPM),
> -	__RISCV_ISA_EXT_SUPERSET(smnpm, RISCV_ISA_EXT_SMNPM, riscv_xlinuxenvcfg_exts),
> +	__RISCV_ISA_EXT_SUPERSET(smnpm, RISCV_ISA_EXT_SMNPM, riscv_supm_exts),
>  	__RISCV_ISA_EXT_DATA(smstateen, RISCV_ISA_EXT_SMSTATEEN),
>  	__RISCV_ISA_EXT_DATA(ssaia, RISCV_ISA_EXT_SSAIA),
>  	__RISCV_ISA_EXT_DATA(sscofpmf, RISCV_ISA_EXT_SSCOFPMF),
> -	__RISCV_ISA_EXT_SUPERSET(ssnpm, RISCV_ISA_EXT_SSNPM, riscv_xlinuxenvcfg_exts),
> +	__RISCV_ISA_EXT_SUPERSET(ssnpm, RISCV_ISA_EXT_SSNPM, riscv_supm_exts),
>  	__RISCV_ISA_EXT_DATA(sstc, RISCV_ISA_EXT_SSTC),
> +	__RISCV_ISA_EXT_DATA_VALIDATE(supm, RISCV_ISA_EXT_SUPM, riscv_ext_supm_validate),

I don't think this quite matches what we want either. We don't want to accept
Supm from the devicetree at all, so it shouldn't be in this array (just like
Xlinuxenvcfg is not in this array). You'll need new code in riscv_resolve_isa()
(or its callers) to set the RISCV_ISA_EXT_SUPM bit under the right conditions.

Regards,
Samuel

>  	__RISCV_ISA_EXT_DATA(svade, RISCV_ISA_EXT_SVADE),
>  	__RISCV_ISA_EXT_DATA_VALIDATE(svadu, RISCV_ISA_EXT_SVADU, riscv_ext_svadu_validate),
>  	__RISCV_ISA_EXT_DATA(svinval, RISCV_ISA_EXT_SVINVAL),
>