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

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

On 2020.03.30 15:16, Ard Biesheuvel wrote:
> On Mon, 30 Mar 2020 at 16:07, Pete Batard <pete@akeo.ie> wrote:
>>
>> 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.
>>
> 
> Well, the reason I asked was because I only want to merge changes that
> are actually need. If there is a valid need, I will happily merge this
> patch.
> 
> But as it turns out, the reason is simply being able to claim that we
> implement the latest ACPI revision, which is actually a bad idea for
> platforms that don't actually rely on any of the new features, since
> it may prevent you from being able to run older OSes

Then we have two very different visions for the platform, especially 
when it comes to the Pi 4, where networking and SD card support is 
pretty much NO_GO for any "older" OS (and troublesome enough as it is in 
modern nones).

This is actually the one place where I'd like to see an actual use-case 
made for "older OSes incompatibility", especially when it comes to 6.0 
MADT vs 6.3 MADT, where the only different is that 6.0 had 3 reserved 
fields and 6.3 started to use 2 of those.

For the record, the MADT blobs we got from Microsoft (for the Pi 3) were 
6.0, and we did downgrade them to 5.1 for convenience (rather than 
compatibility issues), so I really fail to see the issue with bumping 
MADT to 6.3.

Instead, what I'm seeing is folks who need the SPE fields for their 
platform and look at other implementations to see how to set them up 
wasting time having to send a duplicate of the current to edk2 for very 
dubious reasons.

> I disagree. We add what we need when we need it. The patch volume is
> high enough as it is.

Well, considering that different folks will have to send duplicate 
patches, this doesn't entirely come as a surprise...

Sorry, but I really can't see any good reason for this patch to be shot 
down. You guys *will* have to process a similar patch sooner or later.

Regards,

/Pete