[RFC] ACPI on arm64 TODO List

[Grant Likely]

On Wed, Dec 17, 2014 at 10:26 PM, Grant Likely <grant.likely at linaro.org> wrote:
> On Tue, Dec 16, 2014 at 11:27 AM, Arnd Bergmann <arnd at arndb.de> wrote:
>> On Monday 15 December 2014 19:18:16 Al Stone wrote:
>>> 7. Why is ACPI required?
>>>    * Problem:
>>>      * arm64 maintainers still haven't been convinced that ACPI is
>>>        necessary.
>>>      * Why do hardware and OS vendors say ACPI is required?
>>>    * Status: Al & Grant collecting statements from OEMs to be posted
>>>      publicly early in the new year; firmware summit for broader
>>>      discussion planned.
>>
>> I was particularly hoping to see better progress on this item. It
>> really shouldn't be that hard to explain why someone wants this feature.
>
> I've written something up in as a reply on the firmware summit thread.
> I'm going to rework it to be a standalone document and post it
> publicly. I hope that should resolve this issue.

I've posted an article on my blog, but I'm reposting it here because
the mailing list is more conducive to discussion...

http://www.secretlab.ca/archives/151

Why ACPI on ARM?
----------------

Why are we doing ACPI on ARM? That question has been asked many times,
but we haven't yet had a good summary of the most important reasons
for wanting ACPI on ARM. This article is an attempt to state the
rationale clearly.

During an email conversation late last year, Catalin Marinas asked for
a summary of exactly why we want ACPI on ARM, Dong Wei replied with
the following list:
> 1. Support multiple OSes, including Linux and Windows
> 2. Support device configurations
> 3. Support dynamic device configurations (hot add/removal)
> 4. Support hardware abstraction through control methods
> 5. Support power management
> 6. Support thermal management
> 7. Support RAS interfaces

The above list is certainly true in that all of them need to be
supported. However, that list doesn't give the rationale for choosing
ACPI. We already have DT mechanisms for doing most of the above, and
can certainly create new bindings for anything that is missing. So, if
it isn't an issue of functionality, then how does ACPI differ from DT
and why is ACPI a better fit for general purpose ARM servers?

The difference is in the support model. To explain what I mean, I'm
first going to expand on each of the items above and discuss the
similarities and differences between ACPI and DT. Then, with that as
the groundwork, I'll discuss how ACPI is a better fit for the general
purpose hardware support model.

Device Configurations
---------------------
2. Support device configurations
3. Support dynamic device configurations (hot add/removal)

>From day one, DT was about device configurations. There isn't any
significant difference between ACPI & DT here. In fact, the majority
of ACPI tables are completely analogous to DT descriptions. With the
exception of the DSDT and SSDT tables, most ACPI tables are merely
flat data used to describe hardware.

DT platforms have also supported dynamic configuration and hotplug for
years. There isn't a lot here that differentiates between ACPI and DT.
The biggest difference is that dynamic changes to the ACPI namespace
can be triggered by ACPI methods, whereas for DT changes are received
as messages from firmware and have been very much platform specific
(e.g. IBM pSeries does this)

Power Management Model
----------------------
4. Support hardware abstraction through control methods
5. Support power management
6. Support thermal management

Power, thermal, and clock management can all be dealt with as a group.
ACPI defines a power management model (OSPM) that both the platform
and the OS conform to. The OS implements the OSPM state machine, but
the platform can provide state change behaviour in the form of
bytecode methods. Methods can access hardware directly or hand off PM
operations to a coprocessor. The OS really doesn't have to care about
the details as long as the platform obeys the rules of the OSPM model.

With DT, the kernel has device drivers for each and every component in
the platform, and configures them using DT data. DT itself doesn't
have a PM model. Rather the PM model is an implementation detail of
the kernel. Device drivers use DT data to decide how to handle PM
state changes. We have clock, pinctrl, and regulator frameworks in the
kernel for working out runtime PM. However, this only works when all
the drivers and support code have been merged into the kernel. When
the kernel's PM model doesn't work for new hardware, then we change
the model. This works very well for mobile/embedded because the vendor
controls the kernel. We can change things when we need to, but we also
struggle with getting board support mainlined.

This difference has a big impact when it comes to OS support.
Engineers from hardware vendors, Microsoft, and most vocally Red Hat
have all told me bluntly that rebuilding the kernel doesn't work for
enterprise OS support. Their model is based around a fixed OS release
that ideally boots out-of-the-box. It may still need additional device
drivers for specific peripherals/features, but from a system view, the
OS works. When additional drivers are provided separately, those
drivers fit within the existing OSPM model for power management. This
is where ACPI has a technical advantage over DT. The ACPI OSPM model
and it's bytecode gives the HW vendors a level of abstraction under
their control, not the kernel's. When the hardware behaves differently
from what the OS expects, the vendor is able to change the behaviour
without changing the HW or patching the OS.

At this point you'd be right to point out that it is harder to get the
whole system working correctly when behaviour is split between the
kernel and the platform. The OS must trust that the platform doesn't
violate the OSPM model. All manner of bad things happen if it does.
That is exactly why the DT model doesn't encode behaviour: It is
easier to make changes and fix bugs when everything is within the same
code base. We don't need a platform/kernel split when we can modify
the kernel.

However, the enterprise folks don't have that luxury. The
platform/kernel split isn't a design choice. It is a characteristic of
the market. Hardware and OS vendors each have their own product
timetables, and they don't line up. The timeline for getting patches
into the kernel and flowing through into OS releases puts OS support
far downstream from the actual release of hardware. Hardware vendors
simply cannot wait for OS support to come online to be able to release
their products. They need to be able to work with available releases,
and make their hardware behave in the way the OS expects. The
advantage of ACPI OSPM is that it defines behaviour and limits what
the hardware is allowed to do without involving the kernel.

What remains is sorting out how we make sure everything works. How do
we make sure there is enough cross platform testing to ensure new
hardware doesn't ship broken and that new OS releases don't break on
old hardware? Those are the reasons why a UEFI/ACPI firmware summit is
being organized, it's why the UEFI forum holds plugfests 3 times a
year, and it is why we're working on FWTS and LuvOS.

Reliability, Availability & Serviceability (RAS)
------------------------------------------------
7. Support RAS interfaces

This isn't a question of whether or not DT can support RAS. Of course
it can. Rather it is a matter of RAS bindings already existing for
ACPI, including a usage model. We've barely begun to explore this on
DT. This item doesn't make ACPI technically superior to DT, but it
certainly makes it more mature.

Multiplatform support
---------------------
1. Support multiple OSes, including Linux and Windows

I'm tackling this item last because I think it is the most contentious
for those of us in the Linux world. I wanted to get the other issues
out of the way before addressing it.

The separation between hardware vendors and OS vendors in the server
market is new for ARM. For the first time ARM hardware and OS release
cycles are completely decoupled from each other, and neither are
expected to have specific knowledge of the other (ie. the hardware
vendor doesn't control the choice of OS). ARM and their partners want
to create an ecosystem of independent OSes and hardware platforms that
don't explicitly require the former to be ported to the latter.

Now, one could argue that Linux is driving the potential market for
ARM servers, and therefore Linux is the only thing that matters, but
hardware vendors don't see it that way. For hardware vendors it is in
their best interest to support as wide a choice of OSes as possible in
order to catch the widest potential customer base. Even if the
majority choose Linux, some will choose BSD, some will choose Windows,
and some will choose something else. Whether or not we think this is
foolish is beside the point; it isn't something we have influence
over.

During early ARM server planning meetings between ARM, its partners
and other industry representatives (myself included) we discussed this
exact point. Before us were two options, DT and ACPI. As one of the
Linux people in the room, I advised that ACPI's closed governance
model was a show stopper for Linux and that DT is the working
interface. Microsoft on the other hand made it abundantly clear that
ACPI was the only interface that they would support. For their part,
the hardware vendors stated the platform abstraction behaviour of ACPI
is a hard requirement for their support model and that they would not
close the door on either Linux or Windows.

However, the one thing that all of us could agree on was that
supporting multiple interfaces doesn't help anyone: It would require
twice as much effort on defining bindings (once for Linux-DT and once
for Windows-ACPI) and it would require firmware to describe everything
twice. Eventually we reached the compromise to use ACPI, but on the
condition of opening the governance process to give Linux engineers
equal influence over the specification. The fact that we now have a
much better seat at the ACPI table, for both ARM and x86, is a direct
result of these early ARM server negotiations. We are no longer second
class citizens in the ACPI world and are actually driving much of the
recent development.

I know that this line of thought is more about market forces rather
than a hard technical argument between ACPI and DT, but it is an
equally significant one. Agreeing on a single way of doing things is
important. The ARM server ecosystem is better for the agreement to use
the same interface for all operating systems. This is what is meant by
standards compliant. The standard is a codification of the mutually
agreed interface. It provides confidence that all vendors are using
the same rules for interoperability.

Summary
-------
To summarize, here is the short form rationale for ACPI on ARM:

 - ACPI's bytecode allows the platform to encode behaviour. DT
explicitly does not support this. For hardware vendors, being able to
encode behaviour is an important tool for supporting operating system
releases on new hardware.
 - ACPI's OSPM defines a power management model that constrains what
the platform is allowed into a specific model while still having
flexibility in hardware design.
 - For enterprise use-cases, ACPI has extablished bindings, such as
for RAS, which are used in production. DT does not. Yes, we can define
those bindings but doing so means ARM and x86 will use completely
different code paths in both firmware and the kernel.
 - Choosing a single interface for platform/OS abstraction is
important. It is not reasonable to require vendors to implement both
DT and ACPI if they want to support multiple operating systems.
Agreeing on a single interface instead of being fragmented into per-OS
interfaces makes for better interoperability overall.
 - The ACPI governance process works well and we're at the same table
as HW vendors and other OS vendors. In fact, there is no longer any
reason to feel that ACPI is a Windows thing or that we are playing
second fiddle to Microsoft. The move of ACPI governance into the UEFI
forum has significantly opened up the processes, and currently, a
large portion of the changes being made to ACPI is being driven by
Linux.

At the beginning of this article I made the statement that the
difference is in the support model. For servers, responsibility for
hardware behaviour cannot be purely the domain of the kernel, but
rather is split between the platform and the kernel. ACPI frees the OS
from needing to understand all the minute details of the hardware so
that the OS doesn't need to be ported to each and every device
individually. It allows the hardware vendors to take responsibility
for PM behaviour without depending on an OS release cycle which it is
not under their control.

ACPI is also important because hardware and OS vendors have already
worked out how to use it to support the general purpose ecosystem. The
infrastructure is in place, the bindings are in place, and the process
is in place. DT does exactly what we need it to when working with
vertically integrated devices, but we don't have good processes for
supporting what the server vendors need. We could potentially get
there with DT, but doing so doesn't buy us anything. ACPI already does
what the hardware vendors need, Microsoft won't collaborate with us on
DT, and the hardware vendors would still need to provide two
completely separate firmware interface; one for Linux and one for
Windows.