[PATCH 19/19] Documentation: ACPI for ARM64

[graeme.gregory@linaro.org (Graeme Gregory)]

On 28/07/2014 10:07, Arnd Bergmann wrote:
> On Saturday 26 July 2014 19:34:48 Olof Johansson wrote:
>> On Thu, Jul 24, 2014 at 6:00 AM, Hanjun Guo <hanjun.guo@linaro.org> wrote:
>>> +Relationship with Device Tree
>>> +-----------------------------
>>> +
>>> +ACPI support in drivers and subsystems for ARMv8 should never be mutually
>>> +exclusive with DT support at compile time.
>>> +
>>> +At boot time the kernel will only use one description method depending on
>>> +parameters passed from the bootloader.
>> Possibly overriden by kernel bootargs. And as debated for quite a
>> while earlier this year, acpi should still default to off -- if a DT
>> and ACPI are both passed in, DT should at this time be given priority.
> I think this would be harder to do with the way that ACPI is passed in
> to the kernel. IIRC, you always have a minimal DT information based on
> the ARM64 boot protocol, but in the case of ACPI, this contains pointers
> to the ACPI tables, which are then used for populating the Linux platform
> devices (unless acpi=disabled is set), while the other contents of the
> DTB may be present but we skip the of_platform_populate state.
>
> If this is correct, then replacing the firmware-generated dtb with a
> user-provided on would implicitly remove the ACPI tables from visibility,
> which is exactly what we want.
>
> It's possible that I'm misremembering it though, and it should be
> documented better.
>
>>> +Regardless of whether DT or ACPI is used, the kernel must always be capable
>>> +of booting with either scheme.
>> It should always be possible to compile out ACPI. There will be plenty
>> of platforms that will not implement it, so disabling CONFIG_ACPI
>> needs to be possible.
> Right.
>
>>> +Clocks
>>> +------
>>> +
>>> +Like clocks that are part of the power resources there is no standard way
>>> +to represent a clock tree in ACPI 5.1 in a similar manner to how it is
>>> +described in DT.
>>> +
>>> +Devices affected by this include things like UARTs, SoC driven LCD displays,
>>> +etc.
>>> +
>>> +The firmware for example UEFI should initialise these clocks to fixed working
>>> +values before the kernel is executed. If a driver requires to know rates of
>>> +clocks set by firmware then they can be passed to kernel using _DSD.
>>> +
>>> +example :-
>>> +
>>> +Device (CLK0) {
>>> +       ...
>>> +
>>> +       Name (_DSD, Package() {
>>> +               ToUUID("XXXXX"),
>>> +               Package() {
>>> +                       Package(2) {"#clock-cells", 0},
>> Clock-cells? What do they mean here? Is this specified in the ACPI
>> standards? I had to register to get access to it, and didn't feel like
>> doing that right now. I guess it's not _all_ that open a spec. :(
> ...
>>> +                       Package(2) {"clock-frequency", "10000"}
>>> +               }
>>> +       })
>>> +
>>> +       ...
>>> +}
>>> +
>>> +Device (USR1) {
>>> +       ...
>>> +
>>> +       Name (_DSD, Package() {
>>> +               ToUUID("XXXXX"),
>>> +               Package() {
>>> +                       Package(2) {"clocks", Package() {1, ^CLK0}}},
>> A clock is a device in the ACPI model? Why not just provide the rate
>> as data into the device here? You said you're not trying to model the
>> clock tree, so why reference an external node for it?
> Exactly. I think what is going on here is a conflict of interests between
> Intel's embedded ACPI uses and the ARM64 server requirements. The above
> closely resembles what we do in DT, and that makes perfect sense for
> Intel's machines so they can reuse a lot of the infrastructure we put
> in place for DT. I also suspect it will take a few more years before
> this actually gets accepted into both an ACPI specification and the
> common operating systems (no point doing it if only Linux is going to
> adopt it).
>
> For the servers, I don't see how it makes any sense at all, independent
> of the architecture, and relying on a feature like this would only serve
> to delay the adoption of ACPI (whether that is a good or bad thing
> may be a matter of perspective).
>
> Maybe Graeme or others can comment on where this is coming from. What kind
> of driver would actually need to find out the clock rate of a device on
> an arm64 server? The examples above list "UARTs, SoC driven LCD displays,
> etc.". For all I know, the UART is required to be PL01x (without DMA)
> compatible, which should be fully described in ACPI, and I don't see why
> a server would come with an LCD.
>
>
The PL011 UART is the use-case I keep hitting, that IP block has a 
variable input clock on pretty much everything I have seen in the wild. 
I really hope that this use does not spread beyond a few essential 
devices like the UART. IMO all real hardware should be the other side of 
a PCIe bridge.

Graeme