1 Booting the Linux/ppc kernel without Open Firmware
2 --------------------------------------------------
4 (c) 2005 Benjamin Herrenschmidt <benh at kernel.crashing.org>,
6 (c) 2005 Becky Bruce <becky.bruce at freescale.com>,
7 Freescale Semiconductor, FSL SOC and 32-bit additions
8 (c) 2006 MontaVista Software, Inc.
9 Flash chip node definition
15 1) Entry point for arch/powerpc
16 2) Entry point for arch/arm
18 II - The DT block format
20 2) Device tree generalities
21 3) Device tree "structure" block
22 4) Device tree "strings" block
24 III - Required content of the device tree
25 1) Note about cells and address representation
26 2) Note about "compatible" properties
27 3) Note about "name" properties
28 4) Note about node and property names and character set
29 5) Required nodes and properties
33 d) the /memory node(s)
35 f) the /soc<SOCname> node
37 IV - "dtc", the device tree compiler
39 V - Recommendations for a bootloader
41 VI - System-on-a-chip devices and nodes
42 1) Defining child nodes of an SOC
43 2) Representing devices without a current OF specification
45 VII - Specifying interrupt information for devices
46 1) interrupts property
47 2) interrupt-parent property
48 3) OpenPIC Interrupt Controllers
49 4) ISA Interrupt Controllers
51 VIII - Specifying device power management information (sleep property)
53 Appendix A - Sample SOC node for MPC8540
59 May 18, 2005: Rev 0.1 - Initial draft, no chapter III yet.
61 May 19, 2005: Rev 0.2 - Add chapter III and bits & pieces here or
62 clarifies the fact that a lot of things are
63 optional, the kernel only requires a very
64 small device tree, though it is encouraged
65 to provide an as complete one as possible.
67 May 24, 2005: Rev 0.3 - Precise that DT block has to be in RAM
69 - Define version 3 and new format version 16
70 for the DT block (version 16 needs kernel
71 patches, will be fwd separately).
72 String block now has a size, and full path
73 is replaced by unit name for more
75 linux,phandle is made optional, only nodes
76 that are referenced by other nodes need it.
77 "name" property is now automatically
78 deduced from the unit name
80 June 1, 2005: Rev 0.4 - Correct confusion between OF_DT_END and
81 OF_DT_END_NODE in structure definition.
82 - Change version 16 format to always align
83 property data to 4 bytes. Since tokens are
84 already aligned, that means no specific
85 required alignment between property size
86 and property data. The old style variable
87 alignment would make it impossible to do
88 "simple" insertion of properties using
89 memmove (thanks Milton for
90 noticing). Updated kernel patch as well
91 - Correct a few more alignment constraints
92 - Add a chapter about the device-tree
93 compiler and the textural representation of
94 the tree that can be "compiled" by dtc.
96 November 21, 2005: Rev 0.5
97 - Additions/generalizations for 32-bit
98 - Changed to reflect the new arch/powerpc
104 - Add some definitions of interrupt tree (simple/complex)
105 - Add some definitions for PCI host bridges
106 - Add some common address format examples
107 - Add definitions for standard properties and "compatible"
108 names for cells that are not already defined by the existing
110 - Compare FSL SOC use of PCI to standard and make sure no new
111 node definition required.
112 - Add more information about node definitions for SOC devices
113 that currently have no standard, like the FSL CPM.
119 During the development of the Linux/ppc64 kernel, and more
120 specifically, the addition of new platform types outside of the old
121 IBM pSeries/iSeries pair, it was decided to enforce some strict rules
122 regarding the kernel entry and bootloader <-> kernel interfaces, in
123 order to avoid the degeneration that had become the ppc32 kernel entry
124 point and the way a new platform should be added to the kernel. The
125 legacy iSeries platform breaks those rules as it predates this scheme,
126 but no new board support will be accepted in the main tree that
127 doesn't follow them properly. In addition, since the advent of the
128 arch/powerpc merged architecture for ppc32 and ppc64, new 32-bit
129 platforms and 32-bit platforms which move into arch/powerpc will be
130 required to use these rules as well.
132 The main requirement that will be defined in more detail below is
133 the presence of a device-tree whose format is defined after Open
134 Firmware specification. However, in order to make life easier
135 to embedded board vendors, the kernel doesn't require the device-tree
136 to represent every device in the system and only requires some nodes
137 and properties to be present. This will be described in detail in
138 section III, but, for example, the kernel does not require you to
139 create a node for every PCI device in the system. It is a requirement
140 to have a node for PCI host bridges in order to provide interrupt
141 routing informations and memory/IO ranges, among others. It is also
142 recommended to define nodes for on chip devices and other buses that
143 don't specifically fit in an existing OF specification. This creates a
144 great flexibility in the way the kernel can then probe those and match
145 drivers to device, without having to hard code all sorts of tables. It
146 also makes it more flexible for board vendors to do minor hardware
147 upgrades without significantly impacting the kernel code or cluttering
148 it with special cases.
151 1) Entry point for arch/powerpc
152 -------------------------------
154 There is one single entry point to the kernel, at the start
155 of the kernel image. That entry point supports two calling
158 a) Boot from Open Firmware. If your firmware is compatible
159 with Open Firmware (IEEE 1275) or provides an OF compatible
160 client interface API (support for "interpret" callback of
161 forth words isn't required), you can enter the kernel with:
163 r5 : OF callback pointer as defined by IEEE 1275
164 bindings to powerpc. Only the 32-bit client interface
165 is currently supported
167 r3, r4 : address & length of an initrd if any or 0
169 The MMU is either on or off; the kernel will run the
170 trampoline located in arch/powerpc/kernel/prom_init.c to
171 extract the device-tree and other information from open
172 firmware and build a flattened device-tree as described
173 in b). prom_init() will then re-enter the kernel using
174 the second method. This trampoline code runs in the
175 context of the firmware, which is supposed to handle all
176 exceptions during that time.
178 b) Direct entry with a flattened device-tree block. This entry
179 point is called by a) after the OF trampoline and can also be
180 called directly by a bootloader that does not support the Open
181 Firmware client interface. It is also used by "kexec" to
182 implement "hot" booting of a new kernel from a previous
183 running one. This method is what I will describe in more
184 details in this document, as method a) is simply standard Open
185 Firmware, and thus should be implemented according to the
186 various standard documents defining it and its binding to the
187 PowerPC platform. The entry point definition then becomes:
189 r3 : physical pointer to the device-tree block
190 (defined in chapter II) in RAM
192 r4 : physical pointer to the kernel itself. This is
193 used by the assembly code to properly disable the MMU
194 in case you are entering the kernel with MMU enabled
195 and a non-1:1 mapping.
197 r5 : NULL (as to differentiate with method a)
199 Note about SMP entry: Either your firmware puts your other
200 CPUs in some sleep loop or spin loop in ROM where you can get
201 them out via a soft reset or some other means, in which case
202 you don't need to care, or you'll have to enter the kernel
203 with all CPUs. The way to do that with method b) will be
204 described in a later revision of this document.
206 Board supports (platforms) are not exclusive config options. An
207 arbitrary set of board supports can be built in a single kernel
208 image. The kernel will "know" what set of functions to use for a
209 given platform based on the content of the device-tree. Thus, you
212 a) add your platform support as a _boolean_ option in
213 arch/powerpc/Kconfig, following the example of PPC_PSERIES,
214 PPC_PMAC and PPC_MAPLE. The later is probably a good
215 example of a board support to start from.
217 b) create your main platform file as
218 "arch/powerpc/platforms/myplatform/myboard_setup.c" and add it
219 to the Makefile under the condition of your CONFIG_
220 option. This file will define a structure of type "ppc_md"
221 containing the various callbacks that the generic code will
222 use to get to your platform specific code
224 A kernel image may support multiple platforms, but only if the
225 platforms feature the same core architecture. A single kernel build
226 cannot support both configurations with Book E and configurations
227 with classic Powerpc architectures.
229 2) Entry point for arch/arm
230 ---------------------------
232 There is one single entry point to the kernel, at the start
233 of the kernel image. That entry point supports two calling
234 conventions. A summary of the interface is described here. A full
235 description of the boot requirements is documented in
236 Documentation/arm/Booting
238 a) ATAGS interface. Minimal information is passed from firmware
239 to the kernel with a tagged list of predefined parameters.
243 r1 : Machine type number
245 r2 : Physical address of tagged list in system RAM
247 b) Entry with a flattened device-tree block. Firmware loads the
248 physical address of the flattened device tree block (dtb) into r2,
249 r1 is not used, but it is considered good practise to use a valid
250 machine number as described in Documentation/arm/Booting.
254 r1 : Valid machine type number. When using a device tree,
255 a single machine type number will often be assigned to
256 represent a class or family of SoCs.
258 r2 : physical pointer to the device-tree block
259 (defined in chapter II) in RAM. Device tree can be located
260 anywhere in system RAM, but it should be aligned on a 32 bit
263 The kernel will differentiate between ATAGS and device tree booting by
264 reading the memory pointed to by r1 and looking for either the flattened
265 device tree block magic value (0xd00dfeed) or the ATAG_CORE value at
266 offset 0x4 from r2 (0x54410001).
269 II - The DT block format
270 ========================
273 This chapter defines the actual format of the flattened device-tree
274 passed to the kernel. The actual content of it and kernel requirements
275 are described later. You can find example of code manipulating that
276 format in various places, including arch/powerpc/kernel/prom_init.c
277 which will generate a flattened device-tree from the Open Firmware
278 representation, or the fs2dt utility which is part of the kexec tools
279 which will generate one from a filesystem representation. It is
280 expected that a bootloader like uboot provides a bit more support,
281 that will be discussed later as well.
283 Note: The block has to be in main memory. It has to be accessible in
284 both real mode and virtual mode with no mapping other than main
285 memory. If you are writing a simple flash bootloader, it should copy
286 the block to RAM before passing it to the kernel.
292 The kernel is passed the physical address pointing to an area of memory
293 that is roughly described in include/linux/of_fdt.h by the structure
296 struct boot_param_header {
297 u32 magic; /* magic word OF_DT_HEADER */
298 u32 totalsize; /* total size of DT block */
299 u32 off_dt_struct; /* offset to structure */
300 u32 off_dt_strings; /* offset to strings */
301 u32 off_mem_rsvmap; /* offset to memory reserve map
303 u32 version; /* format version */
304 u32 last_comp_version; /* last compatible version */
306 /* version 2 fields below */
307 u32 boot_cpuid_phys; /* Which physical CPU id we're
309 /* version 3 fields below */
310 u32 size_dt_strings; /* size of the strings block */
312 /* version 17 fields below */
313 u32 size_dt_struct; /* size of the DT structure block */
316 Along with the constants:
318 /* Definitions used by the flattened device tree */
319 #define OF_DT_HEADER 0xd00dfeed /* 4: version,
321 #define OF_DT_BEGIN_NODE 0x1 /* Start node: full name
323 #define OF_DT_END_NODE 0x2 /* End node */
324 #define OF_DT_PROP 0x3 /* Property: name off,
326 #define OF_DT_END 0x9
328 All values in this header are in big endian format, the various
329 fields in this header are defined more precisely below. All
330 "offset" values are in bytes from the start of the header; that is
331 from the physical base address of the device tree block.
335 This is a magic value that "marks" the beginning of the
336 device-tree block header. It contains the value 0xd00dfeed and is
337 defined by the constant OF_DT_HEADER
341 This is the total size of the DT block including the header. The
342 "DT" block should enclose all data structures defined in this
343 chapter (who are pointed to by offsets in this header). That is,
344 the device-tree structure, strings, and the memory reserve map.
348 This is an offset from the beginning of the header to the start
349 of the "structure" part the device tree. (see 2) device tree)
353 This is an offset from the beginning of the header to the start
354 of the "strings" part of the device-tree
358 This is an offset from the beginning of the header to the start
359 of the reserved memory map. This map is a list of pairs of 64-
360 bit integers. Each pair is a physical address and a size. The
361 list is terminated by an entry of size 0. This map provides the
362 kernel with a list of physical memory areas that are "reserved"
363 and thus not to be used for memory allocations, especially during
364 early initialization. The kernel needs to allocate memory during
365 boot for things like un-flattening the device-tree, allocating an
366 MMU hash table, etc... Those allocations must be done in such a
367 way to avoid overriding critical things like, on Open Firmware
368 capable machines, the RTAS instance, or on some pSeries, the TCE
369 tables used for the iommu. Typically, the reserve map should
370 contain _at least_ this DT block itself (header,total_size). If
371 you are passing an initrd to the kernel, you should reserve it as
372 well. You do not need to reserve the kernel image itself. The map
373 should be 64-bit aligned.
377 This is the version of this structure. Version 1 stops
378 here. Version 2 adds an additional field boot_cpuid_phys.
379 Version 3 adds the size of the strings block, allowing the kernel
380 to reallocate it easily at boot and free up the unused flattened
381 structure after expansion. Version 16 introduces a new more
382 "compact" format for the tree itself that is however not backward
383 compatible. Version 17 adds an additional field, size_dt_struct,
384 allowing it to be reallocated or moved more easily (this is
385 particularly useful for bootloaders which need to make
386 adjustments to a device tree based on probed information). You
387 should always generate a structure of the highest version defined
388 at the time of your implementation. Currently that is version 17,
389 unless you explicitly aim at being backward compatible.
393 Last compatible version. This indicates down to what version of
394 the DT block you are backward compatible. For example, version 2
395 is backward compatible with version 1 (that is, a kernel build
396 for version 1 will be able to boot with a version 2 format). You
397 should put a 1 in this field if you generate a device tree of
398 version 1 to 3, or 16 if you generate a tree of version 16 or 17
399 using the new unit name format.
403 This field only exist on version 2 headers. It indicate which
404 physical CPU ID is calling the kernel entry point. This is used,
405 among others, by kexec. If you are on an SMP system, this value
406 should match the content of the "reg" property of the CPU node in
407 the device-tree corresponding to the CPU calling the kernel entry
408 point (see further chapters for more informations on the required
409 device-tree contents)
413 This field only exists on version 3 and later headers. It
414 gives the size of the "strings" section of the device tree (which
415 starts at the offset given by off_dt_strings).
419 This field only exists on version 17 and later headers. It gives
420 the size of the "structure" section of the device tree (which
421 starts at the offset given by off_dt_struct).
423 So the typical layout of a DT block (though the various parts don't
424 need to be in that order) looks like this (addresses go from top to
428 ------------------------------
429 base -> | struct boot_param_header |
430 ------------------------------
431 | (alignment gap) (*) |
432 ------------------------------
433 | memory reserve map |
434 ------------------------------
436 ------------------------------
438 | device-tree structure |
440 ------------------------------
442 ------------------------------
444 | device-tree strings |
446 -----> ------------------------------
449 --- (base + totalsize)
451 (*) The alignment gaps are not necessarily present; their presence
452 and size are dependent on the various alignment requirements of
453 the individual data blocks.
456 2) Device tree generalities
457 ---------------------------
459 This device-tree itself is separated in two different blocks, a
460 structure block and a strings block. Both need to be aligned to a 4
463 First, let's quickly describe the device-tree concept before detailing
464 the storage format. This chapter does _not_ describe the detail of the
465 required types of nodes & properties for the kernel, this is done
466 later in chapter III.
468 The device-tree layout is strongly inherited from the definition of
469 the Open Firmware IEEE 1275 device-tree. It's basically a tree of
470 nodes, each node having two or more named properties. A property can
473 It is a tree, so each node has one and only one parent except for the
474 root node who has no parent.
476 A node has 2 names. The actual node name is generally contained in a
477 property of type "name" in the node property list whose value is a
478 zero terminated string and is mandatory for version 1 to 3 of the
479 format definition (as it is in Open Firmware). Version 16 makes it
480 optional as it can generate it from the unit name defined below.
482 There is also a "unit name" that is used to differentiate nodes with
483 the same name at the same level, it is usually made of the node
484 names, the "@" sign, and a "unit address", which definition is
485 specific to the bus type the node sits on.
487 The unit name doesn't exist as a property per-se but is included in
488 the device-tree structure. It is typically used to represent "path" in
489 the device-tree. More details about the actual format of these will be
492 The kernel generic code does not make any formal use of the
493 unit address (though some board support code may do) so the only real
494 requirement here for the unit address is to ensure uniqueness of
495 the node unit name at a given level of the tree. Nodes with no notion
496 of address and no possible sibling of the same name (like /memory or
497 /cpus) may omit the unit address in the context of this specification,
498 or use the "@0" default unit address. The unit name is used to define
499 a node "full path", which is the concatenation of all parent node
500 unit names separated with "/".
502 The root node doesn't have a defined name, and isn't required to have
503 a name property either if you are using version 3 or earlier of the
504 format. It also has no unit address (no @ symbol followed by a unit
505 address). The root node unit name is thus an empty string. The full
506 path to the root node is "/".
508 Every node which actually represents an actual device (that is, a node
509 which isn't only a virtual "container" for more nodes, like "/cpus"
510 is) is also required to have a "compatible" property indicating the
511 specific hardware and an optional list of devices it is fully
512 backwards compatible with.
514 Finally, every node that can be referenced from a property in another
515 node is required to have either a "phandle" or a "linux,phandle"
516 property. Real Open Firmware implementations provide a unique
517 "phandle" value for every node that the "prom_init()" trampoline code
518 turns into "linux,phandle" properties. However, this is made optional
519 if the flattened device tree is used directly. An example of a node
520 referencing another node via "phandle" is when laying out the
521 interrupt tree which will be described in a further version of this
524 The "phandle" property is a 32-bit value that uniquely
525 identifies a node. You are free to use whatever values or system of
526 values, internal pointers, or whatever to generate these, the only
527 requirement is that every node for which you provide that property has
528 a unique value for it.
530 Here is an example of a simple device-tree. In this example, an "o"
531 designates a node followed by the node unit name. Properties are
532 presented with their name followed by their content. "content"
533 represents an ASCII string (zero terminated) value, while <content>
534 represents a 32-bit hexadecimal value. The various nodes in this
535 example will be discussed in a later chapter. At this point, it is
536 only meant to give you a idea of what a device-tree looks like. I have
537 purposefully kept the "name" and "linux,phandle" properties which
538 aren't necessary in order to give you a better idea of what the tree
539 looks like in practice.
542 |- name = "device-tree"
543 |- model = "MyBoardName"
544 |- compatible = "MyBoardFamilyName"
545 |- #address-cells = <2>
547 |- linux,phandle = <0>
551 | | - linux,phandle = <1>
552 | | - #address-cells = <1>
553 | | - #size-cells = <0>
556 | |- name = "PowerPC,970"
557 | |- device_type = "cpu"
559 | |- clock-frequency = <5f5e1000>
561 | |- linux,phandle = <2>
565 | |- device_type = "memory"
566 | |- reg = <00000000 00000000 00000000 20000000>
567 | |- linux,phandle = <3>
571 |- bootargs = "root=/dev/sda2"
572 |- linux,phandle = <4>
574 This tree is almost a minimal tree. It pretty much contains the
575 minimal set of required nodes and properties to boot a linux kernel;
576 that is, some basic model informations at the root, the CPUs, and the
577 physical memory layout. It also includes misc information passed
578 through /chosen, like in this example, the platform type (mandatory)
579 and the kernel command line arguments (optional).
581 The /cpus/PowerPC,970@0/64-bit property is an example of a
582 property without a value. All other properties have a value. The
583 significance of the #address-cells and #size-cells properties will be
584 explained in chapter IV which defines precisely the required nodes and
585 properties and their content.
588 3) Device tree "structure" block
590 The structure of the device tree is a linearized tree structure. The
591 "OF_DT_BEGIN_NODE" token starts a new node, and the "OF_DT_END_NODE"
592 ends that node definition. Child nodes are simply defined before
593 "OF_DT_END_NODE" (that is nodes within the node). A 'token' is a 32
594 bit value. The tree has to be "finished" with a OF_DT_END token
596 Here's the basic structure of a single node:
598 * token OF_DT_BEGIN_NODE (that is 0x00000001)
599 * for version 1 to 3, this is the node full path as a zero
600 terminated string, starting with "/". For version 16 and later,
601 this is the node unit name only (or an empty string for the
603 * [align gap to next 4 bytes boundary]
605 * token OF_DT_PROP (that is 0x00000003)
606 * 32-bit value of property value size in bytes (or 0 if no
608 * 32-bit value of offset in string block of property name
609 * property value data if any
610 * [align gap to next 4 bytes boundary]
611 * [child nodes if any]
612 * token OF_DT_END_NODE (that is 0x00000002)
614 So the node content can be summarized as a start token, a full path,
615 a list of properties, a list of child nodes, and an end token. Every
616 child node is a full node structure itself as defined above.
618 NOTE: The above definition requires that all property definitions for
619 a particular node MUST precede any subnode definitions for that node.
620 Although the structure would not be ambiguous if properties and
621 subnodes were intermingled, the kernel parser requires that the
622 properties come first (up until at least 2.6.22). Any tools
623 manipulating a flattened tree must take care to preserve this
626 4) Device tree "strings" block
628 In order to save space, property names, which are generally redundant,
629 are stored separately in the "strings" block. This block is simply the
630 whole bunch of zero terminated strings for all property names
631 concatenated together. The device-tree property definitions in the
632 structure block will contain offset values from the beginning of the
636 III - Required content of the device tree
637 =========================================
639 WARNING: All "linux,*" properties defined in this document apply only
640 to a flattened device-tree. If your platform uses a real
641 implementation of Open Firmware or an implementation compatible with
642 the Open Firmware client interface, those properties will be created
643 by the trampoline code in the kernel's prom_init() file. For example,
644 that's where you'll have to add code to detect your board model and
645 set the platform number. However, when using the flattened device-tree
646 entry point, there is no prom_init() pass, and thus you have to
647 provide those properties yourself.
650 1) Note about cells and address representation
651 ----------------------------------------------
653 The general rule is documented in the various Open Firmware
654 documentations. If you choose to describe a bus with the device-tree
655 and there exist an OF bus binding, then you should follow the
656 specification. However, the kernel does not require every single
657 device or bus to be described by the device tree.
659 In general, the format of an address for a device is defined by the
660 parent bus type, based on the #address-cells and #size-cells
661 properties. Note that the parent's parent definitions of #address-cells
662 and #size-cells are not inherited so every node with children must specify
663 them. The kernel requires the root node to have those properties defining
664 addresses format for devices directly mapped on the processor bus.
666 Those 2 properties define 'cells' for representing an address and a
667 size. A "cell" is a 32-bit number. For example, if both contain 2
668 like the example tree given above, then an address and a size are both
669 composed of 2 cells, and each is a 64-bit number (cells are
670 concatenated and expected to be in big endian format). Another example
671 is the way Apple firmware defines them, with 2 cells for an address
672 and one cell for a size. Most 32-bit implementations should define
673 #address-cells and #size-cells to 1, which represents a 32-bit value.
674 Some 32-bit processors allow for physical addresses greater than 32
675 bits; these processors should define #address-cells as 2.
677 "reg" properties are always a tuple of the type "address size" where
678 the number of cells of address and size is specified by the bus
679 #address-cells and #size-cells. When a bus supports various address
680 spaces and other flags relative to a given address allocation (like
681 prefetchable, etc...) those flags are usually added to the top level
682 bits of the physical address. For example, a PCI physical address is
683 made of 3 cells, the bottom two containing the actual address itself
684 while the top cell contains address space indication, flags, and pci
685 bus & device numbers.
687 For buses that support dynamic allocation, it's the accepted practice
688 to then not provide the address in "reg" (keep it 0) though while
689 providing a flag indicating the address is dynamically allocated, and
690 then, to provide a separate "assigned-addresses" property that
691 contains the fully allocated addresses. See the PCI OF bindings for
694 In general, a simple bus with no address space bits and no dynamic
695 allocation is preferred if it reflects your hardware, as the existing
696 kernel address parsing functions will work out of the box. If you
697 define a bus type with a more complex address format, including things
698 like address space bits, you'll have to add a bus translator to the
699 prom_parse.c file of the recent kernels for your bus type.
701 The "reg" property only defines addresses and sizes (if #size-cells is
702 non-0) within a given bus. In order to translate addresses upward
703 (that is into parent bus addresses, and possibly into CPU physical
704 addresses), all buses must contain a "ranges" property. If the
705 "ranges" property is missing at a given level, it's assumed that
706 translation isn't possible, i.e., the registers are not visible on the
707 parent bus. The format of the "ranges" property for a bus is a list
710 bus address, parent bus address, size
712 "bus address" is in the format of the bus this bus node is defining,
713 that is, for a PCI bridge, it would be a PCI address. Thus, (bus
714 address, size) defines a range of addresses for child devices. "parent
715 bus address" is in the format of the parent bus of this bus. For
716 example, for a PCI host controller, that would be a CPU address. For a
717 PCI<->ISA bridge, that would be a PCI address. It defines the base
718 address in the parent bus where the beginning of that range is mapped.
720 For new 64-bit board support, I recommend either the 2/2 format or
721 Apple's 2/1 format which is slightly more compact since sizes usually
722 fit in a single 32-bit word. New 32-bit board support should use a
723 1/1 format, unless the processor supports physical addresses greater
724 than 32-bits, in which case a 2/1 format is recommended.
726 Alternatively, the "ranges" property may be empty, indicating that the
727 registers are visible on the parent bus using an identity mapping
728 translation. In other words, the parent bus address space is the same
729 as the child bus address space.
731 2) Note about "compatible" properties
732 -------------------------------------
734 These properties are optional, but recommended in devices and the root
735 node. The format of a "compatible" property is a list of concatenated
736 zero terminated strings. They allow a device to express its
737 compatibility with a family of similar devices, in some cases,
738 allowing a single driver to match against several devices regardless
739 of their actual names.
741 3) Note about "name" properties
742 -------------------------------
744 While earlier users of Open Firmware like OldWorld macintoshes tended
745 to use the actual device name for the "name" property, it's nowadays
746 considered a good practice to use a name that is closer to the device
747 class (often equal to device_type). For example, nowadays, Ethernet
748 controllers are named "ethernet", an additional "model" property
749 defining precisely the chip type/model, and "compatible" property
750 defining the family in case a single driver can driver more than one
751 of these chips. However, the kernel doesn't generally put any
752 restriction on the "name" property; it is simply considered good
753 practice to follow the standard and its evolutions as closely as
756 Note also that the new format version 16 makes the "name" property
757 optional. If it's absent for a node, then the node's unit name is then
758 used to reconstruct the name. That is, the part of the unit name
759 before the "@" sign is used (or the entire unit name if no "@" sign
762 4) Note about node and property names and character set
763 -------------------------------------------------------
765 While Open Firmware provides more flexible usage of 8859-1, this
766 specification enforces more strict rules. Nodes and properties should
767 be comprised only of ASCII characters 'a' to 'z', '0' to
768 '9', ',', '.', '_', '+', '#', '?', and '-'. Node names additionally
769 allow uppercase characters 'A' to 'Z' (property names should be
770 lowercase. The fact that vendors like Apple don't respect this rule is
771 irrelevant here). Additionally, node and property names should always
772 begin with a character in the range 'a' to 'z' (or 'A' to 'Z' for node
775 The maximum number of characters for both nodes and property names
776 is 31. In the case of node names, this is only the leftmost part of
777 a unit name (the pure "name" property), it doesn't include the unit
778 address which can extend beyond that limit.
781 5) Required nodes and properties
782 --------------------------------
783 These are all that are currently required. However, it is strongly
784 recommended that you expose PCI host bridges as documented in the
785 PCI binding to Open Firmware, and your interrupt tree as documented
786 in OF interrupt tree specification.
790 The root node requires some properties to be present:
792 - model : this is your board name/model
793 - #address-cells : address representation for "root" devices
794 - #size-cells: the size representation for "root" devices
795 - compatible : the board "family" generally finds its way here,
796 for example, if you have 2 board models with a similar layout,
797 that typically get driven by the same platform code in the
798 kernel, you would specify the exact board model in the
799 compatible property followed by an entry that represents the SoC
802 The root node is also generally where you add additional properties
803 specific to your board like the serial number if any, that sort of
804 thing. It is recommended that if you add any "custom" property whose
805 name may clash with standard defined ones, you prefix them with your
806 vendor name and a comma.
810 This node is the parent of all individual CPU nodes. It doesn't
811 have any specific requirements, though it's generally good practice
814 #address-cells = <00000001>
815 #size-cells = <00000000>
817 This defines that the "address" for a CPU is a single cell, and has
818 no meaningful size. This is not necessary but the kernel will assume
819 that format when reading the "reg" properties of a CPU node, see
824 So under /cpus, you are supposed to create a node for every CPU on
825 the machine. There is no specific restriction on the name of the
826 CPU, though it's common to call it <architecture>,<core>. For
827 example, Apple uses PowerPC,G5 while IBM uses PowerPC,970FX.
828 However, the Generic Names convention suggests that it would be
829 better to simply use 'cpu' for each cpu node and use the compatible
830 property to identify the specific cpu core.
834 - device_type : has to be "cpu"
835 - reg : This is the physical CPU number, it's a single 32-bit cell
836 and is also used as-is as the unit number for constructing the
837 unit name in the full path. For example, with 2 CPUs, you would
839 /cpus/PowerPC,970FX@0
840 /cpus/PowerPC,970FX@1
841 (unit addresses do not require leading zeroes)
842 - d-cache-block-size : one cell, L1 data cache block size in bytes (*)
843 - i-cache-block-size : one cell, L1 instruction cache block size in
845 - d-cache-size : one cell, size of L1 data cache in bytes
846 - i-cache-size : one cell, size of L1 instruction cache in bytes
848 (*) The cache "block" size is the size on which the cache management
849 instructions operate. Historically, this document used the cache
850 "line" size here which is incorrect. The kernel will prefer the cache
851 block size and will fallback to cache line size for backward
854 Recommended properties:
856 - timebase-frequency : a cell indicating the frequency of the
857 timebase in Hz. This is not directly used by the generic code,
858 but you are welcome to copy/paste the pSeries code for setting
859 the kernel timebase/decrementer calibration based on this
861 - clock-frequency : a cell indicating the CPU core clock frequency
862 in Hz. A new property will be defined for 64-bit values, but if
863 your frequency is < 4Ghz, one cell is enough. Here as well as
864 for the above, the common code doesn't use that property, but
865 you are welcome to re-use the pSeries or Maple one. A future
866 kernel version might provide a common function for this.
867 - d-cache-line-size : one cell, L1 data cache line size in bytes
868 if different from the block size
869 - i-cache-line-size : one cell, L1 instruction cache line size in
870 bytes if different from the block size
872 You are welcome to add any property you find relevant to your board,
873 like some information about the mechanism used to soft-reset the
874 CPUs. For example, Apple puts the GPIO number for CPU soft reset
875 lines in there as a "soft-reset" property since they start secondary
876 CPUs by soft-resetting them.
879 d) the /memory node(s)
881 To define the physical memory layout of your board, you should
882 create one or more memory node(s). You can either create a single
883 node with all memory ranges in its reg property, or you can create
884 several nodes, as you wish. The unit address (@ part) used for the
885 full path is the address of the first range of memory defined by a
886 given node. If you use a single memory node, this will typically be
891 - device_type : has to be "memory"
892 - reg : This property contains all the physical memory ranges of
893 your board. It's a list of addresses/sizes concatenated
894 together, with the number of cells of each defined by the
895 #address-cells and #size-cells of the root node. For example,
896 with both of these properties being 2 like in the example given
897 earlier, a 970 based machine with 6Gb of RAM could typically
898 have a "reg" property here that looks like:
900 00000000 00000000 00000000 80000000
901 00000001 00000000 00000001 00000000
903 That is a range starting at 0 of 0x80000000 bytes and a range
904 starting at 0x100000000 and of 0x100000000 bytes. You can see
905 that there is no memory covering the IO hole between 2Gb and
906 4Gb. Some vendors prefer splitting those ranges into smaller
907 segments, but the kernel doesn't care.
911 This node is a bit "special". Normally, that's where Open Firmware
912 puts some variable environment information, like the arguments, or
913 the default input/output devices.
915 This specification makes a few of these mandatory, but also defines
916 some linux-specific properties that would be normally constructed by
917 the prom_init() trampoline when booting with an OF client interface,
918 but that you have to provide yourself when using the flattened format.
920 Recommended properties:
922 - bootargs : This zero-terminated string is passed as the kernel
924 - linux,stdout-path : This is the full path to your standard
925 console device if any. Typically, if you have serial devices on
926 your board, you may want to put the full path to the one set as
927 the default console in the firmware here, for the kernel to pick
928 it up as its own default console.
930 Note that u-boot creates and fills in the chosen node for platforms
933 (Note: a practice that is now obsolete was to include a property
934 under /chosen called interrupt-controller which had a phandle value
935 that pointed to the main interrupt controller)
937 f) the /soc<SOCname> node
939 This node is used to represent a system-on-a-chip (SoC) and must be
940 present if the processor is a SoC. The top-level soc node contains
941 information that is global to all devices on the SoC. The node name
942 should contain a unit address for the SoC, which is the base address
943 of the memory-mapped register set for the SoC. The name of an SoC
944 node should start with "soc", and the remainder of the name should
945 represent the part number for the soc. For example, the MPC8540's
946 soc node would be called "soc8540".
950 - ranges : Should be defined as specified in 1) to describe the
951 translation of SoC addresses for memory mapped SoC registers.
952 - bus-frequency: Contains the bus frequency for the SoC node.
953 Typically, the value of this field is filled in by the boot
955 - compatible : Exact model of the SoC
958 Recommended properties:
960 - reg : This property defines the address and size of the
961 memory-mapped registers that are used for the SOC node itself.
962 It does not include the child device registers - these will be
963 defined inside each child node. The address specified in the
964 "reg" property should match the unit address of the SOC node.
965 - #address-cells : Address representation for "soc" devices. The
966 format of this field may vary depending on whether or not the
967 device registers are memory mapped. For memory mapped
968 registers, this field represents the number of cells needed to
969 represent the address of the registers. For SOCs that do not
970 use MMIO, a special address format should be defined that
971 contains enough cells to represent the required information.
972 See 1) above for more details on defining #address-cells.
973 - #size-cells : Size representation for "soc" devices
974 - #interrupt-cells : Defines the width of cells used to represent
975 interrupts. Typically this value is <2>, which includes a
976 32-bit number that represents the interrupt number, and a
977 32-bit number that represents the interrupt sense and level.
978 This field is only needed if the SOC contains an interrupt
981 The SOC node may contain child nodes for each SOC device that the
982 platform uses. Nodes should not be created for devices which exist
983 on the SOC but are not used by a particular platform. See chapter VI
984 for more information on how to specify devices that are part of a SOC.
986 Example SOC node for the MPC8540:
989 #address-cells = <1>;
991 #interrupt-cells = <2>;
993 ranges = <00000000 e0000000 00100000>
994 reg = <e0000000 00003000>;
1000 IV - "dtc", the device tree compiler
1001 ====================================
1004 dtc source code can be found at
1005 <http://git.jdl.com/gitweb/?p=dtc.git>
1007 WARNING: This version is still in early development stage; the
1008 resulting device-tree "blobs" have not yet been validated with the
1009 kernel. The current generated block lacks a useful reserve map (it will
1010 be fixed to generate an empty one, it's up to the bootloader to fill
1011 it up) among others. The error handling needs work, bugs are lurking,
1014 dtc basically takes a device-tree in a given format and outputs a
1015 device-tree in another format. The currently supported formats are:
1020 - "dtb": "blob" format, that is a flattened device-tree block
1022 header all in a binary blob.
1023 - "dts": "source" format. This is a text file containing a
1024 "source" for a device-tree. The format is defined later in this
1026 - "fs" format. This is a representation equivalent to the
1027 output of /proc/device-tree, that is nodes are directories and
1028 properties are files
1033 - "dtb": "blob" format
1034 - "dts": "source" format
1035 - "asm": assembly language file. This is a file that can be
1036 sourced by gas to generate a device-tree "blob". That file can
1037 then simply be added to your Makefile. Additionally, the
1038 assembly file exports some symbols that can be used.
1041 The syntax of the dtc tool is
1043 dtc [-I <input-format>] [-O <output-format>]
1044 [-o output-filename] [-V output_version] input_filename
1047 The "output_version" defines what version of the "blob" format will be
1048 generated. Supported versions are 1,2,3 and 16. The default is
1049 currently version 3 but that may change in the future to version 16.
1051 Additionally, dtc performs various sanity checks on the tree, like the
1052 uniqueness of linux, phandle properties, validity of strings, etc...
1054 The format of the .dts "source" file is "C" like, supports C and C++
1060 The above is the "device-tree" definition. It's the only statement
1061 supported currently at the toplevel.
1064 property1 = "string_value"; /* define a property containing a 0
1068 property2 = <1234abcd>; /* define a property containing a
1069 * numerical 32-bit value (hexadecimal)
1072 property3 = <12345678 12345678 deadbeef>;
1073 /* define a property containing 3
1074 * numerical 32-bit values (cells) in
1077 property4 = [0a 0b 0c 0d de ea ad be ef];
1078 /* define a property whose content is
1079 * an arbitrary array of bytes
1082 childnode@address { /* define a child node named "childnode"
1083 * whose unit name is "childnode at
1087 childprop = "hello\n"; /* define a property "childprop" of
1088 * childnode (in this case, a string)
1093 Nodes can contain other nodes etc... thus defining the hierarchical
1094 structure of the tree.
1096 Strings support common escape sequences from C: "\n", "\t", "\r",
1097 "\(octal value)", "\x(hex value)".
1099 It is also suggested that you pipe your source file through cpp (gcc
1100 preprocessor) so you can use #include's, #define for constants, etc...
1102 Finally, various options are planned but not yet implemented, like
1103 automatic generation of phandles, labels (exported to the asm file so
1104 you can point to a property content and change it easily from whatever
1105 you link the device-tree with), label or path instead of numeric value
1106 in some cells to "point" to a node (replaced by a phandle at compile
1107 time), export of reserve map address to the asm file, ability to
1108 specify reserve map content at compile time, etc...
1110 We may provide a .h include file with common definitions of that
1111 proves useful for some properties (like building PCI properties or
1112 interrupt maps) though it may be better to add a notion of struct
1113 definitions to the compiler...
1116 V - Recommendations for a bootloader
1117 ====================================
1120 Here are some various ideas/recommendations that have been proposed
1121 while all this has been defined and implemented.
1123 - The bootloader may want to be able to use the device-tree itself
1124 and may want to manipulate it (to add/edit some properties,
1125 like physical memory size or kernel arguments). At this point, 2
1126 choices can be made. Either the bootloader works directly on the
1127 flattened format, or the bootloader has its own internal tree
1128 representation with pointers (similar to the kernel one) and
1129 re-flattens the tree when booting the kernel. The former is a bit
1130 more difficult to edit/modify, the later requires probably a bit
1131 more code to handle the tree structure. Note that the structure
1132 format has been designed so it's relatively easy to "insert"
1133 properties or nodes or delete them by just memmoving things
1134 around. It contains no internal offsets or pointers for this
1137 - An example of code for iterating nodes & retrieving properties
1138 directly from the flattened tree format can be found in the kernel
1139 file drivers/of/fdt.c. Look at the of_scan_flat_dt() function,
1140 its usage in early_init_devtree(), and the corresponding various
1141 early_init_dt_scan_*() callbacks. That code can be re-used in a
1142 GPL bootloader, and as the author of that code, I would be happy
1143 to discuss possible free licensing to any vendor who wishes to
1144 integrate all or part of this code into a non-GPL bootloader.
1145 (reference needed; who is 'I' here? ---gcl Jan 31, 2011)
1149 VI - System-on-a-chip devices and nodes
1150 =======================================
1152 Many companies are now starting to develop system-on-a-chip
1153 processors, where the processor core (CPU) and many peripheral devices
1154 exist on a single piece of silicon. For these SOCs, an SOC node
1155 should be used that defines child nodes for the devices that make
1156 up the SOC. While platforms are not required to use this model in
1157 order to boot the kernel, it is highly encouraged that all SOC
1158 implementations define as complete a flat-device-tree as possible to
1159 describe the devices on the SOC. This will allow for the
1160 genericization of much of the kernel code.
1163 1) Defining child nodes of an SOC
1164 ---------------------------------
1166 Each device that is part of an SOC may have its own node entry inside
1167 the SOC node. For each device that is included in the SOC, the unit
1168 address property represents the address offset for this device's
1169 memory-mapped registers in the parent's address space. The parent's
1170 address space is defined by the "ranges" property in the top-level soc
1171 node. The "reg" property for each node that exists directly under the
1172 SOC node should contain the address mapping from the child address space
1173 to the parent SOC address space and the size of the device's
1174 memory-mapped register file.
1176 For many devices that may exist inside an SOC, there are predefined
1177 specifications for the format of the device tree node. All SOC child
1178 nodes should follow these specifications, except where noted in this
1181 See appendix A for an example partial SOC node definition for the
1185 2) Representing devices without a current OF specification
1186 ----------------------------------------------------------
1188 Currently, there are many devices on SoCs that do not have a standard
1189 representation defined as part of the Open Firmware specifications,
1190 mainly because the boards that contain these SoCs are not currently
1191 booted using Open Firmware. Binding documentation for new devices
1192 should be added to the Documentation/devicetree/bindings directory.
1193 That directory will expand as device tree support is added to more and
1197 VII - Specifying interrupt information for devices
1198 ===================================================
1200 The device tree represents the buses and devices of a hardware
1201 system in a form similar to the physical bus topology of the
1204 In addition, a logical 'interrupt tree' exists which represents the
1205 hierarchy and routing of interrupts in the hardware.
1207 The interrupt tree model is fully described in the
1208 document "Open Firmware Recommended Practice: Interrupt
1209 Mapping Version 0.9". The document is available at:
1210 <http://playground.sun.com/1275/practice>.
1212 1) interrupts property
1213 ----------------------
1215 Devices that generate interrupts to a single interrupt controller
1216 should use the conventional OF representation described in the
1217 OF interrupt mapping documentation.
1219 Each device which generates interrupts must have an 'interrupt'
1220 property. The interrupt property value is an arbitrary number of
1221 of 'interrupt specifier' values which describe the interrupt or
1222 interrupts for the device.
1224 The encoding of an interrupt specifier is determined by the
1225 interrupt domain in which the device is located in the
1226 interrupt tree. The root of an interrupt domain specifies in
1227 its #interrupt-cells property the number of 32-bit cells
1228 required to encode an interrupt specifier. See the OF interrupt
1229 mapping documentation for a detailed description of domains.
1231 For example, the binding for the OpenPIC interrupt controller
1232 specifies an #interrupt-cells value of 2 to encode the interrupt
1233 number and level/sense information. All interrupt children in an
1234 OpenPIC interrupt domain use 2 cells per interrupt in their interrupts
1237 The PCI bus binding specifies a #interrupt-cell value of 1 to encode
1238 which interrupt pin (INTA,INTB,INTC,INTD) is used.
1240 2) interrupt-parent property
1241 ----------------------------
1243 The interrupt-parent property is specified to define an explicit
1244 link between a device node and its interrupt parent in
1245 the interrupt tree. The value of interrupt-parent is the
1246 phandle of the parent node.
1248 If the interrupt-parent property is not defined for a node, its
1249 interrupt parent is assumed to be an ancestor in the node's
1250 _device tree_ hierarchy.
1252 3) OpenPIC Interrupt Controllers
1253 --------------------------------
1255 OpenPIC interrupt controllers require 2 cells to encode
1256 interrupt information. The first cell defines the interrupt
1257 number. The second cell defines the sense and level
1260 Sense and level information should be encoded as follows:
1262 0 = low to high edge sensitive type enabled
1263 1 = active low level sensitive type enabled
1264 2 = active high level sensitive type enabled
1265 3 = high to low edge sensitive type enabled
1267 4) ISA Interrupt Controllers
1268 ----------------------------
1270 ISA PIC interrupt controllers require 2 cells to encode
1271 interrupt information. The first cell defines the interrupt
1272 number. The second cell defines the sense and level
1275 ISA PIC interrupt controllers should adhere to the ISA PIC
1276 encodings listed below:
1278 0 = active low level sensitive type enabled
1279 1 = active high level sensitive type enabled
1280 2 = high to low edge sensitive type enabled
1281 3 = low to high edge sensitive type enabled
1283 VIII - Specifying Device Power Management Information (sleep property)
1284 ===================================================================
1286 Devices on SOCs often have mechanisms for placing devices into low-power
1287 states that are decoupled from the devices' own register blocks. Sometimes,
1288 this information is more complicated than a cell-index property can
1289 reasonably describe. Thus, each device controlled in such a manner
1290 may contain a "sleep" property which describes these connections.
1292 The sleep property consists of one or more sleep resources, each of
1293 which consists of a phandle to a sleep controller, followed by a
1294 controller-specific sleep specifier of zero or more cells.
1296 The semantics of what type of low power modes are possible are defined
1297 by the sleep controller. Some examples of the types of low power modes
1298 that may be supported are:
1300 - Dynamic: The device may be disabled or enabled at any time.
1301 - System Suspend: The device may request to be disabled or remain
1302 awake during system suspend, but will not be disabled until then.
1303 - Permanent: The device is disabled permanently (until the next hard
1306 Some devices may share a clock domain with each other, such that they should
1307 only be suspended when none of the devices are in use. Where reasonable,
1308 such nodes should be placed on a virtual bus, where the bus has the sleep
1309 property. If the clock domain is shared among devices that cannot be
1310 reasonably grouped in this manner, then create a virtual sleep controller
1311 (similar to an interrupt nexus, except that defining a standardized
1312 sleep-map should wait until its necessity is demonstrated).
1314 Appendix A - Sample SOC node for MPC8540
1315 ========================================
1318 #address-cells = <1>;
1320 compatible = "fsl,mpc8540-ccsr", "simple-bus";
1321 device_type = "soc";
1322 ranges = <0x00000000 0xe0000000 0x00100000>
1323 bus-frequency = <0>;
1324 interrupt-parent = <&pic>;
1327 #address-cells = <1>;
1329 device_type = "network";
1331 compatible = "gianfar", "simple-bus";
1332 reg = <0x24000 0x1000>;
1333 local-mac-address = [ 00 E0 0C 00 73 00 ];
1334 interrupts = <29 2 30 2 34 2>;
1335 phy-handle = <&phy0>;
1336 sleep = <&pmc 00000080>;
1340 reg = <0x24520 0x20>;
1341 compatible = "fsl,gianfar-mdio";
1343 phy0: ethernet-phy@0 {
1346 device_type = "ethernet-phy";
1349 phy1: ethernet-phy@1 {
1352 device_type = "ethernet-phy";
1355 phy3: ethernet-phy@3 {
1358 device_type = "ethernet-phy";
1364 device_type = "network";
1366 compatible = "gianfar";
1367 reg = <0x25000 0x1000>;
1368 local-mac-address = [ 00 E0 0C 00 73 01 ];
1369 interrupts = <13 2 14 2 18 2>;
1370 phy-handle = <&phy1>;
1371 sleep = <&pmc 00000040>;
1375 device_type = "network";
1377 compatible = "gianfar";
1378 reg = <0x26000 0x1000>;
1379 local-mac-address = [ 00 E0 0C 00 73 02 ];
1380 interrupts = <41 2>;
1381 phy-handle = <&phy3>;
1382 sleep = <&pmc 00000020>;
1386 #address-cells = <1>;
1388 compatible = "fsl,mpc8540-duart", "simple-bus";
1389 sleep = <&pmc 00000002>;
1393 device_type = "serial";
1394 compatible = "ns16550";
1395 reg = <0x4500 0x100>;
1396 clock-frequency = <0>;
1397 interrupts = <42 2>;
1401 device_type = "serial";
1402 compatible = "ns16550";
1403 reg = <0x4600 0x100>;
1404 clock-frequency = <0>;
1405 interrupts = <42 2>;
1410 interrupt-controller;
1411 #address-cells = <0>;
1412 #interrupt-cells = <2>;
1413 reg = <0x40000 0x40000>;
1414 compatible = "chrp,open-pic";
1415 device_type = "open-pic";
1419 interrupts = <43 2>;
1420 reg = <0x3000 0x100>;
1421 compatible = "fsl-i2c";
1423 sleep = <&pmc 00000004>;
1427 compatible = "fsl,mpc8540-pmc", "fsl,mpc8548-pmc";
1428 reg = <0xe0070 0x20>;