On 6/21/21 11:15 AM, Philippe Mathieu-Daudé wrote:
> On 6/21/21 10:44 AM, Thomas Huth wrote:
>> On 17/06/2021 17.48, Philippe Mathieu-Daudé wrote:
>> [...]
>>> This works, but I'd rather use:
>>>
>>> if (sev_enabled()) {
>>> sev_kernel_loader_calc_cmdline_hash(&sev_loader_context,
>>> kernel_cmdline);
>>> }
>>>
>>> And have sev_enabled() defined as:
>>>
>>> #ifdef CONFIG_SEV
>>> bool sev_enabled(void);
>>> #else
>>> #define sev_enabled() false
>>> #endif
>>>
>>> So the compiler could elide the statement if SEV is disabled,
>>> and stub is not necessary.
>>>
>>> But that means we'd need to add "#include CONFIG_DEVICES" in
>>> a sysemu/ header, which looks like an anti-pattern.
>>>
>>> Thomas / Paolo, what do you think?
>>
>> I'd only do that if you are very, very sure that the header file is
>> only included from target-specific files. Otherwise this will of course
>> cause more trouble than benefit.
Back to Paolo, I think the problem is we started to use target-specific
features in Kconfig, which was designed for devices (not
target-specific). We have the same problem (another thread) with the
semihosting architectural feature.
> Hmm it could be clearer to rearrange the target-specific sysemu/
> headers. For this example, eventually sysemu/i386/sev.h?
>
> Phil.
>
