Re: [PATCH RESEND v8 15/20] target/riscv/cpu.c: introduce cpu_cfg_ext_auto_update()

[Andrew Jones <ajones@ventanamicro.com>]

On Thu, Aug 24, 2023 at 07:14:35PM -0300, Daniel Henrique Barboza wrote:
> During realize() time we're activating a lot of extensions based on some
> criteria, e.g.:
> 
>     if (cpu->cfg.ext_zk) {
>         cpu->cfg.ext_zkn = true;
>         cpu->cfg.ext_zkr = true;
>         cpu->cfg.ext_zkt = true;
>     }
> 
> This practice resulted in at least one case where we ended up enabling
> something we shouldn't: RVC enabling zca/zcd/zcf when using a CPU that
> has priv_spec older than 1.12.0.
> 
> We're also not considering user choice. There's no way of doing it now
> but this is about to change in the next few patches.
> 
> cpu_cfg_ext_auto_update() will check for priv version mismatches before
> enabling extensions. If we have a mismatch between the current priv
> version and the extension we want to enable, do not enable it. In the
> near future, this same function will also consider user choice when
> deciding if we're going to enable/disable an extension or not.
> 
> For now let's use it to handle zca/zcd/zcf enablement if RVC is enabled.
> 
> Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
> Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
> ---
>  target/riscv/cpu.c | 44 +++++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 41 insertions(+), 3 deletions(-)
> 
> diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
> index ddbf82f859..c56abf8395 100644
> --- a/target/riscv/cpu.c
> +++ b/target/riscv/cpu.c
> @@ -177,6 +177,44 @@ static void isa_ext_update_enabled(RISCVCPU *cpu, uint32_t ext_offset,
>      *ext_enabled = en;
>  }
>  
> +static int cpu_cfg_ext_get_min_version(uint32_t ext_offset)
> +{
> +    int i;
> +
> +    for (i = 0; i < ARRAY_SIZE(isa_edata_arr); i++) {
> +        if (isa_edata_arr[i].ext_enable_offset != ext_offset) {
> +            continue;
> +        }
> +
> +        return isa_edata_arr[i].min_version;
> +    }
> +
> +    /* Default to oldest priv spec if no match found */
> +    return PRIV_VERSION_1_10_0;

This seems backwards to me. If we don't know for how long an extension
has been ratified, then it seems like the default should be "Well, I
guess it's new". Or, we could assert here to ensure we don't have any
extensions having their minimum version checked which are missing version
information. An assert also makes sense for the case that an invalid
ext_offset was passed to this function.

> +}
> +
> +static void cpu_cfg_ext_auto_update(RISCVCPU *cpu, uint32_t ext_offset,
> +                                    bool value)
> +{
> +    CPURISCVState *env = &cpu->env;
> +    bool prev_val = isa_ext_is_enabled(cpu, ext_offset);
> +    int min_version;
> +
> +    if (prev_val == value) {
> +        return;
> +    }
> +
> +    if (value && env->priv_ver != PRIV_VERSION_LATEST) {
> +        /* Do not enable it if priv_ver is older than min_version */
> +        min_version = cpu_cfg_ext_get_min_version(ext_offset);
> +        if (env->priv_ver < min_version) {
> +            return;
> +        }
> +    }
> +
> +    isa_ext_update_enabled(cpu, ext_offset, value);
> +}
> +
>  const char * const riscv_int_regnames[] = {
>      "x0/zero", "x1/ra",  "x2/sp",  "x3/gp",  "x4/tp",  "x5/t0",   "x6/t1",
>      "x7/t2",   "x8/s0",  "x9/s1",  "x10/a0", "x11/a1", "x12/a2",  "x13/a3",
> @@ -1268,12 +1306,12 @@ void riscv_cpu_validate_set_extensions(RISCVCPU *cpu, Error **errp)
>  
>      /* zca, zcd and zcf has a PRIV 1.12.0 restriction */
>      if (riscv_has_ext(env, RVC) && env->priv_ver >= PRIV_VERSION_1_12_0) {
> -        cpu->cfg.ext_zca = true;
> +        cpu_cfg_ext_auto_update(cpu, CPU_CFG_OFFSET(ext_zca), true);
>          if (riscv_has_ext(env, RVF) && env->misa_mxl_max == MXL_RV32) {
> -            cpu->cfg.ext_zcf = true;
> +            cpu_cfg_ext_auto_update(cpu, CPU_CFG_OFFSET(ext_zcf), true);
>          }
>          if (riscv_has_ext(env, RVD)) {
> -            cpu->cfg.ext_zcd = true;
> +            cpu_cfg_ext_auto_update(cpu, CPU_CFG_OFFSET(ext_zcd), true);
>          }
>      }
>  
> -- 
> 2.41.0
> 
>

Thanks,
drew