Re: [PATCH v1 4/8] target/riscv: Remove the hardcoded MSTATUS_SD macro

[Richard Henderson]

On 4/2/21 1:02 PM, Alistair Francis wrote:
> @@ -369,6 +369,9 @@ static void gen_jal(DisasContext *ctx, int rd, target_ulong 
> imm)
>   static void mark_fs_dirty(DisasContext *ctx)
>   {
>       TCGv tmp;
> +    CPUState *cpu = ctx->cs;
> +    CPURISCVState *env = cpu->env_ptr;
> +
>       if (ctx->mstatus_fs == MSTATUS_FS) {
>           return;
>       }
> @@ -377,12 +380,24 @@ static void mark_fs_dirty(DisasContext *ctx)
>   
>       tmp = tcg_temp_new();
>       tcg_gen_ld_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus));
> -    tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | MSTATUS_SD);
> +    if (riscv_cpu_is_32bit(env)) {

This is less than ideal, and will be incorrect long term.
You should check ctx->misa instead.

Eventually you'll need to change riscv_tr_init_disas_context to not just copy 
ctx->misa from env.  At present we flush all translation blocks when misa 
changes, which works.  But you won't want to do that when the hypervisor is 
64-bit and the guest is 32-bit.

Anyway, I think it would be a good idea to create a helper local to translate, 
akin to has_ext().

> +        tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | MSTATUS32_SD);
> +    } else {
> +#if defined(TARGET_RISCV64)
> +        tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | MSTATUS64_SD);
> +#endif

The ifdefs are ugly.  I presume there's some sort of compiler warning here? 
Does it go away if you cast to target_ulong?

How about

    target_ulong sd = is_32bit(ctx) ? MSTATUS32_SD : MSTATUS64_SD;
    tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | sd);


r~