Re: [PATCH 1/1] EmbeddedPkg/AcpiLib: add GICC table init macro for ACPI 6.3

["Pete Batard" <pete@akeo.ie>]

On 2020.03.30 15:01, Ard Biesheuvel wrote:
> On Mon, 30 Mar 2020 at 15:56, Pete Batard <pete@akeo.ie> wrote:
>>
>> On 2020.03.30 14:20, Ard Biesheuvel wrote:
>>> On Mon, 30 Mar 2020 at 15:12, Ard Biesheuvel <ard.biesheuvel@linaro.org> wrote:
>>>>
>>>> On Mon, 30 Mar 2020 at 15:09, Pete Batard <pete@akeo.ie> wrote:
>>>>>
>>>>> On 2020.03.30 14:06, Ard Biesheuvel wrote:
>>>>>> On Fri, 27 Mar 2020 at 14:06, Pete Batard <pete@akeo.ie> wrote:
>>>>>>>
>>>>>>> Incidentally, this is not an [edk2-platform] patch, as the subject line
>>>>>>> from previous mail seemed to indicate, but an [edk2] patch.
>>>>>>>
>>>>>>
>>>>>> Do we have a user for this?
>>>>>
>>>>> Yes we do. I have a pachset lined up that updates the Raspberry Pi ACPI
>>>>> to 6.3, that has a dependency on this.
>>>>>
>>>>
>>>> But does the RPi have SPE and the associated overflow interrupt?
>>
>> No, but it doesn't matter since the specs indicate that SPE values can
>> be set to zero if unused/non-applicable.
>>
>>>> ACPI
>>>> is designed to be backward compatible, so it is perfectly acceptable
>>>> to use the 6.2 macros in the context of a firmware implementation that
>>>> complies with 6.3.
>>
>> This is what happens if you try to use EFI_ACPI_6_0_GICC_STRUCTURE_INIT
>> in a 6.3 context:
>>
>> /usr/src/edk2/MdePkg/Include/IndustryStandard/Acpi10.h:297:33: error:
>> excess elements in scalar initializer [-Werror]
>>    #define EFI_ACPI_RESERVED_BYTE  0x00
>>                                    ^~~~
>> Building ...
>> /usr/src/edk2/MdePkg/Library/DxeCoreHobLib/DxeCoreHobLib.inf [AARCH64]
>> /usr/src/edk2/EmbeddedPkg/Include/Library/AcpiLib.h:64:30: note: in
>> expansion of macro ‘EFI_ACPI_RESERVED_BYTE’
>>        {EFI_ACPI_RESERVED_BYTE, EFI_ACPI_RESERVED_BYTE,
>> EFI_ACPI_RESERVED_BYTE}         \
>>                                 ^~~~~~~~~~~~~~~~~~~~~~
>> /usr/src/edk2-platforms/Platform/RaspberryPi/AcpiTables/Madt.aslc:64:5:
>> note: in expansion of macro ‘EFI_ACPI_6_0_GICC_STRUCTURE_INIT’
>>        EFI_ACPI_6_0_GICC_STRUCTURE_INIT (
>>        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>
> 
> What do you mean exactly by 'in a 6.3 context': are you trying to
> statically initialize a 63 struct with the 60 macro?

Yes. I am trying to upgrade all of our ACPI tables to 6.3, on account 
that (part of a commit message from the edk2-platform I have lined up):

-----------------------------------------------------------------------
Because of its widespread availability and low price, we expect the
Raspberry Pi source to be used by platform developers as a starting
point to create their own platform implementation.

As such, it makes a lot of sense to want to use the most up to date
underlying standards, even if the pay off is limited in this case,
as it may help others benefit from the latest improvements and
features brought by modern ACPI.
-----------------------------------------------------------------------

>> The only reason I'm sending an EDK2 patch, which I'd always rather avoid
>> so that edk2-platforms patches can be applied faster, is that I haven't
>> been able to find a way to make the existing 6.0 macros work in a 6.3
>> context, and I expect that this will be the case for others.
>>
> 
> By why do we need the 6.3 context? If 6.0 can describe our platform
> fully, it is actually better for compatibility to stick with it rather
> than upgrade to 6.3.

See above.

I have to say that I'm a bit taken aback by the idea that, even though 
we can anticipate that there will be a need for a 6.3 macro that does 
initialise the SPE field, there seems to be strong reluctance to add 
that macro before someone makes the case for it.

Ideally, we should update our macros the minute the specs are released, 
regardless of whether we believe there's a use-case for it or not.

Regards,

/Pete

> 
>>>> Or is there another reason you want to update the
>>>> MADT to 6.3?
>>
>> I just want to fix the compilation error, as well as make sure that
>> folks who need the ACPI 6.3 SPE init can get it without having to spend
>> time waiting for an EDK2 patch to be applied.
>>
>>> BTW, this patch sets the size of the GICC entry to 'sizeof
>>> (EFI_ACPI_6_0_GIC_STRUCTURE)' so it is likely that the parser will
>>> choke on it.
>>
>> The structure size hasn't changed. There were 3 reserved bytes, and now
>> there's one reserve byte and one 16-bit word for the SPE.
>>
>> Since I actually need this patch as part of a platform update, I did
>> validate compilation, so, no, the parser doesn't choke on it.
>>
> 
> I didn't mean to imply that you are being sloppy. I'd just like to
> understand what the point is of all of this, given that ACPI 6.3 is
> not needed to describe the RPi4
>