Re: [PATCH v6 07/12] xen: enable Dom0 to use SVE feature

[Jan Beulich <jbeulich@xxxxxxxx>]

On 24.04.2023 17:34, Luca Fancellu wrote:
>> On 24 Apr 2023, at 16:25, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>> On 24.04.2023 17:18, Luca Fancellu wrote:
>>>> On 24 Apr 2023, at 16:06, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>> On 24.04.2023 16:57, Luca Fancellu wrote:
>>>>>> On 24 Apr 2023, at 15:05, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>>>> On 24.04.2023 16:00, Luca Fancellu wrote:
>>>>>>>> On 24 Apr 2023, at 12:34, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>>>>>> On 24.04.2023 08:02, Luca Fancellu wrote:
>>>>>>>>> @@ -30,9 +37,11 @@ int sve_context_init(struct vcpu *v);
>>>>>>>>> void sve_context_free(struct vcpu *v);
>>>>>>>>> void sve_save_state(struct vcpu *v);
>>>>>>>>> void sve_restore_state(struct vcpu *v);
>>>>>>>>> +bool sve_domctl_vl_param(int val, unsigned int *out);
>>>>>>>>>
>>>>>>>>> #else /* !CONFIG_ARM64_SVE */
>>>>>>>>>
>>>>>>>>> +#define opt_dom0_sve     (0)
>>>>>>>>> #define is_sve_domain(d) (0)
>>>>>>>>>
>>>>>>>>> static inline register_t compute_max_zcr(void)
>>>>>>>>> @@ -59,6 +68,11 @@ static inline void sve_context_free(struct vcpu 
>>>>>>>>> *v) {}
>>>>>>>>> static inline void sve_save_state(struct vcpu *v) {}
>>>>>>>>> static inline void sve_restore_state(struct vcpu *v) {}
>>>>>>>>>
>>>>>>>>> +static inline bool sve_domctl_vl_param(int val, unsigned int *out)
>>>>>>>>> +{
>>>>>>>>> +    return false;
>>>>>>>>> +}
>>>>>>>>
>>>>>>>> Once again I don't see the need for this stub: opt_dom0_sve is 
>>>>>>>> #define-d
>>>>>>>> to plain zero when !ARM64_SVE, so the only call site merely requires a
>>>>>>>> visible declaration, and DCE will take care of eliminating the actual 
>>>>>>>> call.
>>>>>>>
>>>>>>> I’ve tried to do that, I’ve put the declaration outside the ifdef so 
>>>>>>> that it was always included
>>>>>>> and I removed the stub, but I got errors on compilation because of 
>>>>>>> undefined function.
>>>>>>> For that reason  I left that change out.
>>>>>>
>>>>>> Interesting. I don't see where the reference would be coming from.
>>>>>
>>>>> Could it be because the declaration is visible, outside the ifdef, but 
>>>>> the definition is not compiled in? 
>>>>
>>>> Well, yes, likely. But the question isn't that but "Why did the reference
>>>> not get removed, when it's inside an if(0) block?"
>>>
>>> Oh ok, I don’t know, here what I get if for example I build arm32:
>>>
>>> arm-linux-gnueabihf-ld -EL -T arch/arm/xen.lds -N prelink.o \
>>> ./common/symbols-dummy.o -o ./.xen-syms.0
>>> arm-linux-gnueabihf-ld: prelink.o: in function `create_domUs':
>>> (.init.text+0x13464): undefined reference to `sve_domctl_vl_param'
>>
>> In particular with seeing this: What you copied here is a build with the
>> series applied only up to this patch? I ask because the patch here adds a
>> call only out of create_dom0().
> 
> No I’ve do the changes on top of the serie, I’ve tried it now, only to this 
> patch and it builds correctly,
> It was my mistake to don’t read carefully the error output.
> 
> Anyway I guess this change is not applicable because we don’t have a symbol 
> that is plain 0 for domUs
> to be placed inside create_domUs.

Possible, but would you mind first telling me in which other patch(es) the
further reference(s) are being introduced, so I could take a look without
(again) digging through the entire series?

Jan