Re: [PATCH-for-10.1 v3 19/19] tcg: Convert TARGET_SUPPORTS_MTTCG to TCGCPUOps::mttcg_supported field

[Richard Henderson <richard.henderson@linaro.org>]

On 4/3/25 15:04, Philippe Mathieu-Daudé wrote:
> Instead of having a compile-time TARGET_SUPPORTS_MTTCG definition,
> have each target set the 'mttcg_supported' field in the TCGCPUOps
> structure.
> 
> Since so far we only emulate one target architecture at a time,
> tcg_init_machine() gets whether MTTCG is supported via the
> current CPU class (CPU_RESOLVING_TYPE).
> 
> Since TARGET_SUPPORTS_MTTCG isn't available anymore,
> instead of emiting a warning when the 'thread' property
> is set in tcg_set_thread(), emit it in tcg_init_machine()
> where it is consumed.
> 
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> Reviewed-by: Anton Johansson <anjo@rev.ng>
> Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
> Reviewed-by: Richard Henderson <richard.henderson@linaro.org>

This doesn't work for i386 and riscv, which have both been "updated" to initialize 
cc->tcg_ops via ${arch}_tcg_cpu_accel_type_info ...

>   static void mttcg_init(TCGState *s)
>   {
> -#ifdef TARGET_SUPPORTS_MTTCG
> -    bool mttcg_supported = true;
> -#else
> -    bool mttcg_supported = false;
> -#endif
> +    CPUClass *cc = CPU_CLASS(object_class_by_name(CPU_RESOLVING_TYPE));
> +    bool mttcg_supported = cc->tcg_ops->mttcg_supported;

... which has not yet been called, so NULL pointer dereference.

I'm not sure why the init of tcg_ops was moved from ${arch}_cpu_class_init.
Probably it looked nice to remove the #ifdef CONFIG_TCG.

There's a definite ordering problem here, and I'm not sure how to fix it.


r~