RFC: device thermal limits represented in device tree nodes

[Eduardo Valentin]

Hello Grant and Rob,

(Resending, as I got a message saying:
<devicetree-discuss@xxxxxxxxxxxxxxxx>: Recipient address rejected: User
has moved to devicetree at vger.kernel.org)

I am writing this email to you specifically to ask your technical
assessment with respect to representing device thermal limits as device
tree nodes. I am proposing to introduce device tree nodes to describe
these limits as thermal zones, their composition and their relations
with cooling devices and other thermal zones (thermal data).

As you should know, device thermal limits are part of hardware
specification. Considering your board layout, mechanics, power
dissipation and composition of ICs, etc, that will impose thermal
requirements on your system, and infringing these limits can lead to
device damage, device life time reduction or even end user harm. Thus,
the thermal data help to describe the hardware limits and what needs to
be done if those limits are crosses, as part of your board design and
non-functional requirements. Obviously that is very dependent on your
hardware, and not all of them will have these non-functional
requirements. Besides, describing these limits has *nothing* to do with
how you actually find these limits.

In any case, there is a need to properly represent these requirements
and I am proposing to have this representation in device tree. There
were already couple of counter-arguments claiming this is actually about
configuration and performance profile description. But I still stand
against these two readings of this proposal and again state that if one
interprets it as configuration or performance profile, that is a
mis-understanding [0]. Let me state it clear (again [1]), my proposal is
to describe hardware thermal limits, because these limits are part of a
hardware specification; representing in device tree would not infringe
the original purpose of this data structure  ("The Device Tree is a data
structure for describing hardware."[2]).

Before I explain my proposal, I want to highlight also that these data
is represented elsewhere already and it is reused across different OS's.
Thermal data is described using ACPI [3] and operating systems
ACPI-aware do support the interpretation of thermal data. Linux is one
example of such systems (I believe I do not need to enlist here all
systems supporting ACPI). On the other hand, not all systems have ACPI
or are specified to use ACPI.  Thus, here is another reason to represent
properly thermal data, so that we can scale across systems.

In the specific case of Linux, the common thermal concepts between ACPI
systems and non-ACPI systems have been represented in the thermal
framework (CONFIG_THERMAL). Today, on ACPI systems, thermal data is
fetched from bootloader with help from the common ACPI parser. For
non-ACPI systems, the thermal data is actually coded as part of device
drivers.

So, to the point, a brief explanation of my proposal goes as follows:
i   - trip points: a node to describe a point in the temperature domain
in which the system has to take an action. This node describes just the
point, not the action. Properties here are temperature, hysteresis, and
type (critical, hot, passive, active, etc).
ii  - binding parameters: the bind_param node is a node to describe how
actions (cooling devices) get assigned to trip points. Cooling devices
are expected to be loaded in the target system. Properties here are:
cooling device name, weight, trip_mask and limits.
iii - thermal zones: the thermal_zone node is the node containing all
the required info for describing a thermal zone with hardware thermal
limitation, including its bindings with cooling devices. Properties here
are:  type, passive_delay, polling_delay, governor. The thermal_zone
node must contain, apart from its own properties, one node containing
trip nodes and one node containing all the zone bind parameters.

Here is an example (on OMAP4430):
thermal_zone {
	type = "CPU";
	mask = <0x03>; /* trips writability */
	passive_delay = <250>; /* milliseconds */
	polling_delay = <1000>; /* milliseconds */
	governor = "step_wise";
	trips {
		alert@100000{
			temperature = <100000>; /* milliCelsius
			hysteresis = <2000>; /* milliCelsius */
			type = <THERMAL_TRIP_PASSIVE>;
		};
		crit@125000{
			temperature = <125000>; /* milliCelsius
			hysteresis = <2000>; /* milliCelsius */
			type = <THERMAL_TRIP_CRITICAL>;
		};
	};
	bind_params {
		action@0{
			cooling_device = "thermal-cpufreq";
			weight = <100>; /* percentage */
			mask = <0x01>;
			/* no limits, using defaults */
		};
	};
};

In this current proposal, a 'thermal_zone' node would be embedded inside
a temperature sensor node, for simplicity. But other possible builds
could embedded them in the device with thermal limits (CPU nodes, for
instance) or they could be not embedded in any specific node.

A full documented description can be found here [4]. Also a branch
containing:
(a) needed changes in order to have this DT parser;
(b) the DT parser with documentation
(c) examples on how drivers could be changes to use the parser
can be found in my branch here [5]. I wrote the thermal DT parser to
build thermal zones with the thermal framework API. However, if one does
not want to do that, it can simple do not include a CONFIG_THERMAL_OF=y
in her/his build, and the calls will be translated to nops, and the
device tree thermal data can be parsed to somewhere else interested
(other subsystem or even user land). A TODO on this implementation is
that it still lacks the representation of thermal zones composed by
several sensors. However, I believe it is better to take an incremental
approach here.  This series can already be used to improve most of the
existing platform thermal drivers (most are CPU thermal drivers) and to
reuse the existing code of some hwmon sensors to build thermal zones for
board thermal requirements.

I have already posted a patch series with this proposal on [6], that
contains a reference for the original RFC. But looks like my messages
got moderated on device tree mailing list. Obviously, within PM forum,
feedback was quite positive. However, we cannot proceed without proper
assessment of other subsystems. lm-sensors  folks (Guenter) seam to be
strongly against this series, as there is a fear that this may introduce
a mis-usage of DT. I still believe this is needed for  hardware
description, and thus not a infringement on DT purposes.

Please let me know your thoughts on this topic and apologize me if my
previous messages on this topic did not reach you (hope they reach now).

All best,

Eduardo Valentin

[0] - https://lkml.org/lkml/2013/7/17/621
[1] - https://lkml.org/lkml/2013/7/18/279
[2] - www.devicetree.org
[3] - http://www.acpi.info/
[4] -
https://git.kernel.org/cgit/linux/kernel/git/evalenti/linux.git/diff/Documentation/devicetree/bindings/thermal/thermal.txt?h=thermal_work/thermal_core/dt_parser&id=405bf0b51457ed055a082af2653d7ce757bc2e91
[5] -
https://git.kernel.org/cgit/linux/kernel/git/evalenti/linux.git/log/?h=thermal_work/thermal_core/dt_parser
[6] - https://lkml.org/lkml/2013/7/17/923


-- 
You have got to be excited about what you are doing. (L. Lamport)

Eduardo Valentin

Attachment: signature.asc
Description: OpenPGP digital signature