Re: [PATCH v4 2/3] arm/efi: Use dom0less configuration when using EFI boot

[Jan Beulich <jbeulich@xxxxxxxx>]

On 08.10.2021 15:38, Luca Fancellu wrote:
>> On 7 Oct 2021, at 08:15, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>> On 01.10.2021 17:13, Luca Fancellu wrote:
>>>> On 1 Oct 2021, at 15:22, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>> On 01.10.2021 15:55, Luca Fancellu wrote:
>>>>>> On 1 Oct 2021, at 12:02, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>>>> On 30.09.2021 16:28, Luca Fancellu wrote:
>>>>>>> @@ -1361,12 +1361,30 @@ efi_start(EFI_HANDLE ImageHandle, 
>>>>>>> EFI_SYSTEM_TABLE *SystemTable)
>>>>>>>       efi_bs->FreePages(cfg.addr, PFN_UP(cfg.size));
>>>>>>>       cfg.addr = 0;
>>>>>>>
>>>>>>> -        dir_handle->Close(dir_handle);
>>>>>>> -
>>>>>>>       if ( gop && !base_video )
>>>>>>>           gop_mode = efi_find_gop_mode(gop, cols, rows, depth);
>>>>>>>   }
>>>>>>>
>>>>>>> +#ifdef CONFIG_HAS_DEVICE_TREE
>>>>>>> +    /* Get the number of boot modules specified on the DT or an error 
>>>>>>> (<0) */
>>>>>>> +    dt_modules_found = efi_arch_check_dt_boot(dir_handle);
>>>>>>> +#endif
>>>>>>
>>>>>> So I had asked to add a stub enclosed in such an #ifdef, to avoid the
>>>>>> #ifdef here. I may be willing to let you keep things as you have them
>>>>>> now, but I'd like to understand why you've picked that different
>>>>>> approach despite the prior discussion.
>>>>>
>>>>> There must be a misunderstanding, your message in the v3 was:
>>>>>
>>>>> "Every time I see this addition I'm getting puzzled. As a result I'm
>>>>> afraid I now need to finally ask you to do something about this (and
>>>>> I'm sorry for doing so only now). There would better be no notion of
>>>>> DT in x86 code, and there would better also not be a need for
>>>>> architectures not supporting DT to each supply such a stub. Instead
>>>>> I think you want to put this stub in xen/common/efi/boot.c, inside a
>>>>> suitable #ifdef.”
>>>>>
>>>>> So I thought you wanted me to remove the stub in x86 (since it doesn’t 
>>>>> support DT)
>>>>> and put this call under #ifdef so it won’t be compiled for arch not 
>>>>> supporting DT.
>>>>
>>>> So FTAOD I'll repeat the crucial part: "I think you want to put this
>>>> stub in xen/common/efi/boot.c". There was nothing about removing the
>>>> stub altogether.
>>>
>>> Oh ok, now I see, so in your opinion this is a better idea:
>>>
>>> #ifndef CONFIG_HAS_DEVICE_TREE
>>> static inline int __init efi_arch_check_dt_boot(EFI_FILE_HANDLE dir_handle)
>>> {
>>>    return 0;
>>> }
>>> #endif
>>>
>>> But I would like to understand the advantage respect of my approach, could 
>>> you
>>> explain me?
>>
>> Well, to a degree it's a matter of taste. Your approach may lead to a long
>> series of various #ifdef sections in a single function, harming readability.
>> Having stubs instead (usually placed in headers, albeit not in this case)
>> allows the main bodies of code to remain more tidy.
> 
> Yes right, in this case I did in another way because declaring the stub in 
> the .c file
> was (in my opinion) not the right thing to do, since also the name 
> (efi_arch_*) recalls
> something arch oriented and so not to be put in the common code.

Feel free to drop "arch" from the hook name.

> In this way any future architecture supporting DT, can just provide the 
> function (or a 
> stub) and we don’t have stubs in architectures that won’t ever support DT.
> 
> In your opinion that solution could be acceptable?

Yes, but not preferable.

Jan