On 03/02/2018 21:44, Richard Henderson wrote:
> On 02/03/2018 07:39 AM, Paolo Bonzini wrote:
>> +/* This function gives an error if an invalid, non-NULL pointer type is
>> passed
>> + * to QEMU_MAKE_LOCKABLE. For optimized builds, we can rely on dead-code
>> elimination
>> + * from the compiler, and give the errors already at link time.
>> + */
>> +#ifdef __OPTIMIZE__
>> +void unknown_lock_type(void *);
>> +#else
>> +static inline void unknown_lock_type(void *unused)
>> +{
>> + abort();
>> +}
>
> Since you're using __builtin_choose_expr, I'm surprised you would need to test
> __OPTIMZE__. The nature of the builtin is such that it *must* eliminate the
> other branch; otherwise the types don't even match up.
__builtin_choose_expr works fine. However we also have
return x ? __builtin_choose_expr(..., unknown_lock_type) : NULL;
and if "x" is NULL then its type is a void*, and the LHS will refer to
unknown_lock_type. I was expecting __always_inline__ to be enough to
avoid this, but apparently this is not the case.
Paolo
> It might be worth using __attribute__((error("message"))) on the function too.
>
> Otherwise,
>
> Reviewed-by: Richard Henderson <richard.hender...@linaro.org>
>
>
> r~
>
