1.. SPDX-License-Identifier: GPL-2.0
   4Linux and the Devicetree
   7The Linux usage model for device tree data
   9:Author: Grant Likely <>
  11This article describes how Linux uses the device tree.  An overview of
  12the device tree data format can be found on the device tree usage page
  13at\ [1]_.
  15.. [1]
  17The "Open Firmware Device Tree", or simply Devicetree (DT), is a data
  18structure and language for describing hardware.  More specifically, it
  19is a description of hardware that is readable by an operating system
  20so that the operating system doesn't need to hard code details of the
  23Structurally, the DT is a tree, or acyclic graph with named nodes, and
  24nodes may have an arbitrary number of named properties encapsulating
  25arbitrary data.  A mechanism also exists to create arbitrary
  26links from one node to another outside of the natural tree structure.
  28Conceptually, a common set of usage conventions, called 'bindings',
  29is defined for how data should appear in the tree to describe typical
  30hardware characteristics including data busses, interrupt lines, GPIO
  31connections, and peripheral devices.
  33As much as possible, hardware is described using existing bindings to
  34maximize use of existing support code, but since property and node
  35names are simply text strings, it is easy to extend existing bindings
  36or create new ones by defining new nodes and properties.  Be wary,
  37however, of creating a new binding without first doing some homework
  38about what already exists.  There are currently two different,
  39incompatible, bindings for i2c busses that came about because the new
  40binding was created without first investigating how i2c devices were
  41already being enumerated in existing systems.
  431. History
  45The DT was originally created by Open Firmware as part of the
  46communication method for passing data from Open Firmware to a client
  47program (like to an operating system).  An operating system used the
  48Device Tree to discover the topology of the hardware at runtime, and
  49thereby support a majority of available hardware without hard coded
  50information (assuming drivers were available for all devices).
  52Since Open Firmware is commonly used on PowerPC and SPARC platforms,
  53the Linux support for those architectures has for a long time used the
  54Device Tree.
  56In 2005, when PowerPC Linux began a major cleanup and to merge 32-bit
  57and 64-bit support, the decision was made to require DT support on all
  58powerpc platforms, regardless of whether or not they used Open
  59Firmware.  To do this, a DT representation called the Flattened Device
  60Tree (FDT) was created which could be passed to the kernel as a binary
  61blob without requiring a real Open Firmware implementation.  U-Boot,
  62kexec, and other bootloaders were modified to support both passing a
  63Device Tree Binary (dtb) and to modify a dtb at boot time.  DT was
  64also added to the PowerPC boot wrapper (``arch/powerpc/boot/*``) so that
  65a dtb could be wrapped up with the kernel image to support booting
  66existing non-DT aware firmware.
  68Some time later, FDT infrastructure was generalized to be usable by
  69all architectures.  At the time of this writing, 6 mainlined
  70architectures (arm, microblaze, mips, powerpc, sparc, and x86) and 1
  71out of mainline (nios) have some level of DT support.
  732. Data Model
  75If you haven't already read the Device Tree Usage\ [1]_ page,
  76then go read it now.  It's okay, I'll wait....
  782.1 High Level View
  80The most important thing to understand is that the DT is simply a data
  81structure that describes the hardware.  There is nothing magical about
  82it, and it doesn't magically make all hardware configuration problems
  83go away.  What it does do is provide a language for decoupling the
  84hardware configuration from the board and device driver support in the
  85Linux kernel (or any other operating system for that matter).  Using
  86it allows board and device support to become data driven; to make
  87setup decisions based on data passed into the kernel instead of on
  88per-machine hard coded selections.
  90Ideally, data driven platform setup should result in less code
  91duplication and make it easier to support a wide range of hardware
  92with a single kernel image.
  94Linux uses DT data for three major purposes:
  961) platform identification,
  972) runtime configuration, and
  983) device population.
 1002.2 Platform Identification
 102First and foremost, the kernel will use data in the DT to identify the
 103specific machine.  In a perfect world, the specific platform shouldn't
 104matter to the kernel because all platform details would be described
 105perfectly by the device tree in a consistent and reliable manner.
 106Hardware is not perfect though, and so the kernel must identify the
 107machine during early boot so that it has the opportunity to run
 108machine-specific fixups.
 110In the majority of cases, the machine identity is irrelevant, and the
 111kernel will instead select setup code based on the machine's core
 112CPU or SoC.  On ARM for example, setup_arch() in
 113arch/arm/kernel/setup.c will call setup_machine_fdt() in
 114arch/arm/kernel/devtree.c which searches through the machine_desc
 115table and selects the machine_desc which best matches the device tree
 116data.  It determines the best match by looking at the 'compatible'
 117property in the root device tree node, and comparing it with the
 118dt_compat list in struct machine_desc (which is defined in
 119arch/arm/include/asm/mach/arch.h if you're curious).
 121The 'compatible' property contains a sorted list of strings starting
 122with the exact name of the machine, followed by an optional list of
 123boards it is compatible with sorted from most compatible to least.  For
 124example, the root compatible properties for the TI BeagleBoard and its
 125successor, the BeagleBoard xM board might look like, respectively::
 127        compatible = "ti,omap3-beagleboard", "ti,omap3450", "ti,omap3";
 128        compatible = "ti,omap3-beagleboard-xm", "ti,omap3450", "ti,omap3";
 130Where "ti,omap3-beagleboard-xm" specifies the exact model, it also
 131claims that it compatible with the OMAP 3450 SoC, and the omap3 family
 132of SoCs in general.  You'll notice that the list is sorted from most
 133specific (exact board) to least specific (SoC family).
 135Astute readers might point out that the Beagle xM could also claim
 136compatibility with the original Beagle board.  However, one should be
 137cautioned about doing so at the board level since there is typically a
 138high level of change from one board to another, even within the same
 139product line, and it is hard to nail down exactly what is meant when one
 140board claims to be compatible with another.  For the top level, it is
 141better to err on the side of caution and not claim one board is
 142compatible with another.  The notable exception would be when one
 143board is a carrier for another, such as a CPU module attached to a
 144carrier board.
 146One more note on compatible values.  Any string used in a compatible
 147property must be documented as to what it indicates.  Add
 148documentation for compatible strings in Documentation/devicetree/bindings.
 150Again on ARM, for each machine_desc, the kernel looks to see if
 151any of the dt_compat list entries appear in the compatible property.
 152If one does, then that machine_desc is a candidate for driving the
 153machine.  After searching the entire table of machine_descs,
 154setup_machine_fdt() returns the 'most compatible' machine_desc based
 155on which entry in the compatible property each machine_desc matches
 156against.  If no matching machine_desc is found, then it returns NULL.
 158The reasoning behind this scheme is the observation that in the majority
 159of cases, a single machine_desc can support a large number of boards
 160if they all use the same SoC, or same family of SoCs.  However,
 161invariably there will be some exceptions where a specific board will
 162require special setup code that is not useful in the generic case.
 163Special cases could be handled by explicitly checking for the
 164troublesome board(s) in generic setup code, but doing so very quickly
 165becomes ugly and/or unmaintainable if it is more than just a couple of
 168Instead, the compatible list allows a generic machine_desc to provide
 169support for a wide common set of boards by specifying "less
 170compatible" values in the dt_compat list.  In the example above,
 171generic board support can claim compatibility with "ti,omap3" or
 172"ti,omap3450".  If a bug was discovered on the original beagleboard
 173that required special workaround code during early boot, then a new
 174machine_desc could be added which implements the workarounds and only
 175matches on "ti,omap3-beagleboard".
 177PowerPC uses a slightly different scheme where it calls the .probe()
 178hook from each machine_desc, and the first one returning TRUE is used.
 179However, this approach does not take into account the priority of the
 180compatible list, and probably should be avoided for new architecture
 1832.3 Runtime configuration
 185In most cases, a DT will be the sole method of communicating data from
 186firmware to the kernel, so also gets used to pass in runtime and
 187configuration data like the kernel parameters string and the location
 188of an initrd image.
 190Most of this data is contained in the /chosen node, and when booting
 191Linux it will look something like this::
 193        chosen {
 194                bootargs = "console=ttyS0,115200 loglevel=8";
 195                initrd-start = <0xc8000000>;
 196                initrd-end = <0xc8200000>;
 197        };
 199The bootargs property contains the kernel arguments, and the initrd-*
 200properties define the address and size of an initrd blob.  Note that
 201initrd-end is the first address after the initrd image, so this doesn't
 202match the usual semantic of struct resource.  The chosen node may also
 203optionally contain an arbitrary number of additional properties for
 204platform-specific configuration data.
 206During early boot, the architecture setup code calls of_scan_flat_dt()
 207several times with different helper callbacks to parse device tree
 208data before paging is setup.  The of_scan_flat_dt() code scans through
 209the device tree and uses the helpers to extract information required
 210during early boot.  Typically the early_init_dt_scan_chosen() helper
 211is used to parse the chosen node including kernel parameters,
 212early_init_dt_scan_root() to initialize the DT address space model,
 213and early_init_dt_scan_memory() to determine the size and
 214location of usable RAM.
 216On ARM, the function setup_machine_fdt() is responsible for early
 217scanning of the device tree after selecting the correct machine_desc
 218that supports the board.
 2202.4 Device population
 222After the board has been identified, and after the early configuration data
 223has been parsed, then kernel initialization can proceed in the normal
 224way.  At some point in this process, unflatten_device_tree() is called
 225to convert the data into a more efficient runtime representation.
 226This is also when machine-specific setup hooks will get called, like
 227the machine_desc .init_early(), .init_irq() and .init_machine() hooks
 228on ARM.  The remainder of this section uses examples from the ARM
 229implementation, but all architectures will do pretty much the same
 230thing when using a DT.
 232As can be guessed by the names, .init_early() is used for any machine-
 233specific setup that needs to be executed early in the boot process,
 234and .init_irq() is used to set up interrupt handling.  Using a DT
 235doesn't materially change the behaviour of either of these functions.
 236If a DT is provided, then both .init_early() and .init_irq() are able
 237to call any of the DT query functions (of_* in include/linux/of*.h) to
 238get additional data about the platform.
 240The most interesting hook in the DT context is .init_machine() which
 241is primarily responsible for populating the Linux device model with
 242data about the platform.  Historically this has been implemented on
 243embedded platforms by defining a set of static clock structures,
 244platform_devices, and other data in the board support .c file, and
 245registering it en-masse in .init_machine().  When DT is used, then
 246instead of hard coding static devices for each platform, the list of
 247devices can be obtained by parsing the DT, and allocating device
 248structures dynamically.
 250The simplest case is when .init_machine() is only responsible for
 251registering a block of platform_devices.  A platform_device is a concept
 252used by Linux for memory or I/O mapped devices which cannot be detected
 253by hardware, and for 'composite' or 'virtual' devices (more on those
 254later).  While there is no 'platform device' terminology for the DT,
 255platform devices roughly correspond to device nodes at the root of the
 256tree and children of simple memory mapped bus nodes.
 258About now is a good time to lay out an example.  Here is part of the
 259device tree for the NVIDIA Tegra board::
 261  /{
 262        compatible = "nvidia,harmony", "nvidia,tegra20";
 263        #address-cells = <1>;
 264        #size-cells = <1>;
 265        interrupt-parent = <&intc>;
 267        chosen { };
 268        aliases { };
 270        memory {
 271                device_type = "memory";
 272                reg = <0x00000000 0x40000000>;
 273        };
 275        soc {
 276                compatible = "nvidia,tegra20-soc", "simple-bus";
 277                #address-cells = <1>;
 278                #size-cells = <1>;
 279                ranges;
 281                intc: interrupt-controller@50041000 {
 282                        compatible = "nvidia,tegra20-gic";
 283                        interrupt-controller;
 284                        #interrupt-cells = <1>;
 285                        reg = <0x50041000 0x1000>, < 0x50040100 0x0100 >;
 286                };
 288                serial@70006300 {
 289                        compatible = "nvidia,tegra20-uart";
 290                        reg = <0x70006300 0x100>;
 291                        interrupts = <122>;
 292                };
 294                i2s1: i2s@70002800 {
 295                        compatible = "nvidia,tegra20-i2s";
 296                        reg = <0x70002800 0x100>;
 297                        interrupts = <77>;
 298                        codec = <&wm8903>;
 299                };
 301                i2c@7000c000 {
 302                        compatible = "nvidia,tegra20-i2c";
 303                        #address-cells = <1>;
 304                        #size-cells = <0>;
 305                        reg = <0x7000c000 0x100>;
 306                        interrupts = <70>;
 308                        wm8903: codec@1a {
 309                                compatible = "wlf,wm8903";
 310                                reg = <0x1a>;
 311                                interrupts = <347>;
 312                        };
 313                };
 314        };
 316        sound {
 317                compatible = "nvidia,harmony-sound";
 318                i2s-controller = <&i2s1>;
 319                i2s-codec = <&wm8903>;
 320        };
 321  };
 323At .init_machine() time, Tegra board support code will need to look at
 324this DT and decide which nodes to create platform_devices for.
 325However, looking at the tree, it is not immediately obvious what kind
 326of device each node represents, or even if a node represents a device
 327at all.  The /chosen, /aliases, and /memory nodes are informational
 328nodes that don't describe devices (although arguably memory could be
 329considered a device).  The children of the /soc node are memory mapped
 330devices, but the codec@1a is an i2c device, and the sound node
 331represents not a device, but rather how other devices are connected
 332together to create the audio subsystem.  I know what each device is
 333because I'm familiar with the board design, but how does the kernel
 334know what to do with each node?
 336The trick is that the kernel starts at the root of the tree and looks
 337for nodes that have a 'compatible' property.  First, it is generally
 338assumed that any node with a 'compatible' property represents a device
 339of some kind, and second, it can be assumed that any node at the root
 340of the tree is either directly attached to the processor bus, or is a
 341miscellaneous system device that cannot be described any other way.
 342For each of these nodes, Linux allocates and registers a
 343platform_device, which in turn may get bound to a platform_driver.
 345Why is using a platform_device for these nodes a safe assumption?
 346Well, for the way that Linux models devices, just about all bus_types
 347assume that its devices are children of a bus controller.  For
 348example, each i2c_client is a child of an i2c_master.  Each spi_device
 349is a child of an SPI bus.  Similarly for USB, PCI, MDIO, etc.  The
 350same hierarchy is also found in the DT, where I2C device nodes only
 351ever appear as children of an I2C bus node.  Ditto for SPI, MDIO, USB,
 352etc.  The only devices which do not require a specific type of parent
 353device are platform_devices (and amba_devices, but more on that
 354later), which will happily live at the base of the Linux /sys/devices
 355tree.  Therefore, if a DT node is at the root of the tree, then it
 356really probably is best registered as a platform_device.
 358Linux board support code calls of_platform_populate(NULL, NULL, NULL, NULL)
 359to kick off discovery of devices at the root of the tree.  The
 360parameters are all NULL because when starting from the root of the
 361tree, there is no need to provide a starting node (the first NULL), a
 362parent struct device (the last NULL), and we're not using a match
 363table (yet).  For a board that only needs to register devices,
 364.init_machine() can be completely empty except for the
 365of_platform_populate() call.
 367In the Tegra example, this accounts for the /soc and /sound nodes, but
 368what about the children of the SoC node?  Shouldn't they be registered
 369as platform devices too?  For Linux DT support, the generic behaviour
 370is for child devices to be registered by the parent's device driver at
 371driver .probe() time.  So, an i2c bus device driver will register a
 372i2c_client for each child node, an SPI bus driver will register
 373its spi_device children, and similarly for other bus_types.
 374According to that model, a driver could be written that binds to the
 375SoC node and simply registers platform_devices for each of its
 376children.  The board support code would allocate and register an SoC
 377device, a (theoretical) SoC device driver could bind to the SoC device,
 378and register platform_devices for /soc/interrupt-controller, /soc/serial,
 379/soc/i2s, and /soc/i2c in its .probe() hook.  Easy, right?
 381Actually, it turns out that registering children of some
 382platform_devices as more platform_devices is a common pattern, and the
 383device tree support code reflects that and makes the above example
 384simpler.  The second argument to of_platform_populate() is an
 385of_device_id table, and any node that matches an entry in that table
 386will also get its child nodes registered.  In the Tegra case, the code
 387can look something like this::
 389  static void __init harmony_init_machine(void)
 390  {
 391        /* ... */
 392        of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
 393  }
 395"simple-bus" is defined in the Devicetree Specification as a property
 396meaning a simple memory mapped bus, so the of_platform_populate() code
 397could be written to just assume simple-bus compatible nodes will
 398always be traversed.  However, we pass it in as an argument so that
 399board support code can always override the default behaviour.
 401[Need to add discussion of adding i2c/spi/etc child devices]
 403Appendix A: AMBA devices
 406ARM Primecells are a certain kind of device attached to the ARM AMBA
 407bus which include some support for hardware detection and power
 408management.  In Linux, struct amba_device and the amba_bus_type is
 409used to represent Primecell devices.  However, the fiddly bit is that
 410not all devices on an AMBA bus are Primecells, and for Linux it is
 411typical for both amba_device and platform_device instances to be
 412siblings of the same bus segment.
 414When using the DT, this creates problems for of_platform_populate()
 415because it must decide whether to register each node as either a
 416platform_device or an amba_device.  This unfortunately complicates the
 417device creation model a little bit, but the solution turns out not to
 418be too invasive.  If a node is compatible with "arm,amba-primecell", then
 419of_platform_populate() will register it as an amba_device instead of a