1 Booting the Linux/ppc kernel without Open Firmware
2 --------------------------------------------------
5 (c) 2005 Benjamin Herrenschmidt <benh at kernel.crashing.org>,
7 (c) 2005 Becky Bruce <becky.bruce at freescale.com>,
8 Freescale Semiconductor, FSL SOC and 32-bit additions
10 May 18, 2005: Rev 0.1 - Initial draft, no chapter III yet.
12 May 19, 2005: Rev 0.2 - Add chapter III and bits & pieces here or
13 clarifies the fact that a lot of things are
14 optional, the kernel only requires a very
15 small device tree, though it is encouraged
16 to provide an as complete one as possible.
18 May 24, 2005: Rev 0.3 - Precise that DT block has to be in RAM
20 - Define version 3 and new format version 16
21 for the DT block (version 16 needs kernel
22 patches, will be fwd separately).
23 String block now has a size, and full path
24 is replaced by unit name for more
26 linux,phandle is made optional, only nodes
27 that are referenced by other nodes need it.
28 "name" property is now automatically
29 deduced from the unit name
31 June 1, 2005: Rev 0.4 - Correct confusion between OF_DT_END and
32 OF_DT_END_NODE in structure definition.
33 - Change version 16 format to always align
34 property data to 4 bytes. Since tokens are
35 already aligned, that means no specific
36 required alignement between property size
37 and property data. The old style variable
38 alignment would make it impossible to do
39 "simple" insertion of properties using
40 memove (thanks Milton for
41 noticing). Updated kernel patch as well
42 - Correct a few more alignement constraints
43 - Add a chapter about the device-tree
44 compiler and the textural representation of
45 the tree that can be "compiled" by dtc.
48 November 21, 2005: Rev 0.5
49 - Additions/generalizations for 32-bit
50 - Changed to reflect the new arch/powerpc
56 - Add some definitions of interrupt tree (simple/complex)
57 - Add some definitions for pci host bridges
58 - Add some common address format examples
59 - Add definitions for standard properties and "compatible"
60 names for cells that are not already defined by the existing
62 - Compare FSL SOC use of PCI to standard and make sure no new
63 node definition required.
64 - Add more information about node definitions for SOC devices
65 that currently have no standard, like the FSL CPM.
71 During the recent development of the Linux/ppc64 kernel, and more
72 specifically, the addition of new platform types outside of the old
73 IBM pSeries/iSeries pair, it was decided to enforce some strict rules
74 regarding the kernel entry and bootloader <-> kernel interfaces, in
75 order to avoid the degeneration that had become the ppc32 kernel entry
76 point and the way a new platform should be added to the kernel. The
77 legacy iSeries platform breaks those rules as it predates this scheme,
78 but no new board support will be accepted in the main tree that
79 doesn't follows them properly. In addition, since the advent of the
80 arch/powerpc merged architecture for ppc32 and ppc64, new 32-bit
81 platforms and 32-bit platforms which move into arch/powerpc will be
82 required to use these rules as well.
84 The main requirement that will be defined in more detail below is
85 the presence of a device-tree whose format is defined after Open
86 Firmware specification. However, in order to make life easier
87 to embedded board vendors, the kernel doesn't require the device-tree
88 to represent every device in the system and only requires some nodes
89 and properties to be present. This will be described in detail in
90 section III, but, for example, the kernel does not require you to
91 create a node for every PCI device in the system. It is a requirement
92 to have a node for PCI host bridges in order to provide interrupt
93 routing informations and memory/IO ranges, among others. It is also
94 recommended to define nodes for on chip devices and other busses that
95 don't specifically fit in an existing OF specification. This creates a
96 great flexibility in the way the kernel can then probe those and match
97 drivers to device, without having to hard code all sorts of tables. It
98 also makes it more flexible for board vendors to do minor hardware
99 upgrades without significantly impacting the kernel code or cluttering
100 it with special cases.
103 1) Entry point for arch/powerpc
104 -------------------------------
106 There is one and one single entry point to the kernel, at the start
107 of the kernel image. That entry point supports two calling
110 a) Boot from Open Firmware. If your firmware is compatible
111 with Open Firmware (IEEE 1275) or provides an OF compatible
112 client interface API (support for "interpret" callback of
113 forth words isn't required), you can enter the kernel with:
115 r5 : OF callback pointer as defined by IEEE 1275
116 bindings to powerpc. Only the 32 bit client interface
117 is currently supported
119 r3, r4 : address & length of an initrd if any or 0
121 The MMU is either on or off; the kernel will run the
122 trampoline located in arch/powerpc/kernel/prom_init.c to
123 extract the device-tree and other information from open
124 firmware and build a flattened device-tree as described
125 in b). prom_init() will then re-enter the kernel using
126 the second method. This trampoline code runs in the
127 context of the firmware, which is supposed to handle all
128 exceptions during that time.
130 b) Direct entry with a flattened device-tree block. This entry
131 point is called by a) after the OF trampoline and can also be
132 called directly by a bootloader that does not support the Open
133 Firmware client interface. It is also used by "kexec" to
134 implement "hot" booting of a new kernel from a previous
135 running one. This method is what I will describe in more
136 details in this document, as method a) is simply standard Open
137 Firmware, and thus should be implemented according to the
138 various standard documents defining it and its binding to the
139 PowerPC platform. The entry point definition then becomes:
141 r3 : physical pointer to the device-tree block
142 (defined in chapter II) in RAM
144 r4 : physical pointer to the kernel itself. This is
145 used by the assembly code to properly disable the MMU
146 in case you are entering the kernel with MMU enabled
147 and a non-1:1 mapping.
149 r5 : NULL (as to differenciate with method a)
151 Note about SMP entry: Either your firmware puts your other
152 CPUs in some sleep loop or spin loop in ROM where you can get
153 them out via a soft reset or some other means, in which case
154 you don't need to care, or you'll have to enter the kernel
155 with all CPUs. The way to do that with method b) will be
156 described in a later revision of this document.
164 Board supports (platforms) are not exclusive config options. An
165 arbitrary set of board supports can be built in a single kernel
166 image. The kernel will "know" what set of functions to use for a
167 given platform based on the content of the device-tree. Thus, you
170 a) add your platform support as a _boolean_ option in
171 arch/powerpc/Kconfig, following the example of PPC_PSERIES,
172 PPC_PMAC and PPC_MAPLE. The later is probably a good
173 example of a board support to start from.
175 b) create your main platform file as
176 "arch/powerpc/platforms/myplatform/myboard_setup.c" and add it
177 to the Makefile under the condition of your CONFIG_
178 option. This file will define a structure of type "ppc_md"
179 containing the various callbacks that the generic code will
180 use to get to your platform specific code
182 c) Add a reference to your "ppc_md" structure in the
183 "machines" table in arch/powerpc/kernel/setup_64.c if you are
186 d) request and get assigned a platform number (see PLATFORM_*
187 constants in include/asm-powerpc/processor.h
189 32-bit embedded kernels:
191 Currently, board support is essentially an exclusive config option.
192 The kernel is configured for a single platform. Part of the reason
193 for this is to keep kernels on embedded systems small and efficient;
194 part of this is due to the fact the code is already that way. In the
195 future, a kernel may support multiple platforms, but only if the
196 platforms feature the same core architectire. A single kernel build
197 cannot support both configurations with Book E and configurations
198 with classic Powerpc architectures.
200 32-bit embedded platforms that are moved into arch/powerpc using a
201 flattened device tree should adopt the merged tree practice of
202 setting ppc_md up dynamically, even though the kernel is currently
203 built with support for only a single platform at a time. This allows
204 unification of the setup code, and will make it easier to go to a
205 multiple-platform-support model in the future.
207 NOTE: I believe the above will be true once Ben's done with the merge
208 of the boot sequences.... someone speak up if this is wrong!
210 To add a 32-bit embedded platform support, follow the instructions
211 for 64-bit platforms above, with the exception that the Kconfig
212 option should be set up such that the kernel builds exclusively for
213 the platform selected. The processor type for the platform should
214 enable another config option to select the specific board
217 NOTE: If ben doesn't merge the setup files, may need to change this to
221 I will describe later the boot process and various callbacks that
222 your platform should implement.
225 II - The DT block format
226 ========================
229 This chapter defines the actual format of the flattened device-tree
230 passed to the kernel. The actual content of it and kernel requirements
231 are described later. You can find example of code manipulating that
232 format in various places, including arch/powerpc/kernel/prom_init.c
233 which will generate a flattened device-tree from the Open Firmware
234 representation, or the fs2dt utility which is part of the kexec tools
235 which will generate one from a filesystem representation. It is
236 expected that a bootloader like uboot provides a bit more support,
237 that will be discussed later as well.
239 Note: The block has to be in main memory. It has to be accessible in
240 both real mode and virtual mode with no mapping other than main
241 memory. If you are writing a simple flash bootloader, it should copy
242 the block to RAM before passing it to the kernel.
248 The kernel is entered with r3 pointing to an area of memory that is
249 roughtly described in include/asm-powerpc/prom.h by the structure
252 struct boot_param_header {
253 u32 magic; /* magic word OF_DT_HEADER */
254 u32 totalsize; /* total size of DT block */
255 u32 off_dt_struct; /* offset to structure */
256 u32 off_dt_strings; /* offset to strings */
257 u32 off_mem_rsvmap; /* offset to memory reserve map
259 u32 version; /* format version */
260 u32 last_comp_version; /* last compatible version */
262 /* version 2 fields below */
263 u32 boot_cpuid_phys; /* Which physical CPU id we're
265 /* version 3 fields below */
266 u32 size_dt_strings; /* size of the strings block */
269 Along with the constants:
271 /* Definitions used by the flattened device tree */
272 #define OF_DT_HEADER 0xd00dfeed /* 4: version,
274 #define OF_DT_BEGIN_NODE 0x1 /* Start node: full name
276 #define OF_DT_END_NODE 0x2 /* End node */
277 #define OF_DT_PROP 0x3 /* Property: name off,
279 #define OF_DT_END 0x9
281 All values in this header are in big endian format, the various
282 fields in this header are defined more precisely below. All
283 "offset" values are in bytes from the start of the header; that is
284 from the value of r3.
288 This is a magic value that "marks" the beginning of the
289 device-tree block header. It contains the value 0xd00dfeed and is
290 defined by the constant OF_DT_HEADER
294 This is the total size of the DT block including the header. The
295 "DT" block should enclose all data structures defined in this
296 chapter (who are pointed to by offsets in this header). That is,
297 the device-tree structure, strings, and the memory reserve map.
301 This is an offset from the beginning of the header to the start
302 of the "structure" part the device tree. (see 2) device tree)
306 This is an offset from the beginning of the header to the start
307 of the "strings" part of the device-tree
311 This is an offset from the beginning of the header to the start
312 of the reserved memory map. This map is a list of pairs of 64
313 bit integers. Each pair is a physical address and a size. The
315 list is terminated by an entry of size 0. This map provides the
316 kernel with a list of physical memory areas that are "reserved"
317 and thus not to be used for memory allocations, especially during
318 early initialization. The kernel needs to allocate memory during
319 boot for things like un-flattening the device-tree, allocating an
320 MMU hash table, etc... Those allocations must be done in such a
321 way to avoid overriding critical things like, on Open Firmware
322 capable machines, the RTAS instance, or on some pSeries, the TCE
323 tables used for the iommu. Typically, the reserve map should
324 contain _at least_ this DT block itself (header,total_size). If
325 you are passing an initrd to the kernel, you should reserve it as
326 well. You do not need to reserve the kernel image itself. The map
327 should be 64 bit aligned.
331 This is the version of this structure. Version 1 stops
332 here. Version 2 adds an additional field boot_cpuid_phys.
333 Version 3 adds the size of the strings block, allowing the kernel
334 to reallocate it easily at boot and free up the unused flattened
335 structure after expansion. Version 16 introduces a new more
336 "compact" format for the tree itself that is however not backward
337 compatible. You should always generate a structure of the highest
338 version defined at the time of your implementation. Currently
339 that is version 16, unless you explicitely aim at being backward
344 Last compatible version. This indicates down to what version of
345 the DT block you are backward compatible. For example, version 2
346 is backward compatible with version 1 (that is, a kernel build
347 for version 1 will be able to boot with a version 2 format). You
348 should put a 1 in this field if you generate a device tree of
349 version 1 to 3, or 0x10 if you generate a tree of version 0x10
350 using the new unit name format.
354 This field only exist on version 2 headers. It indicate which
355 physical CPU ID is calling the kernel entry point. This is used,
356 among others, by kexec. If you are on an SMP system, this value
357 should match the content of the "reg" property of the CPU node in
358 the device-tree corresponding to the CPU calling the kernel entry
359 point (see further chapters for more informations on the required
360 device-tree contents)
363 So the typical layout of a DT block (though the various parts don't
364 need to be in that order) looks like this (addresses go from top to
368 ------------------------------
369 r3 -> | struct boot_param_header |
370 ------------------------------
371 | (alignment gap) (*) |
372 ------------------------------
373 | memory reserve map |
374 ------------------------------
376 ------------------------------
378 | device-tree structure |
380 ------------------------------
382 ------------------------------
384 | device-tree strings |
386 -----> ------------------------------
391 (*) The alignment gaps are not necessarily present; their presence
392 and size are dependent on the various alignment requirements of
393 the individual data blocks.
396 2) Device tree generalities
397 ---------------------------
399 This device-tree itself is separated in two different blocks, a
400 structure block and a strings block. Both need to be aligned to a 4
403 First, let's quickly describe the device-tree concept before detailing
404 the storage format. This chapter does _not_ describe the detail of the
405 required types of nodes & properties for the kernel, this is done
406 later in chapter III.
408 The device-tree layout is strongly inherited from the definition of
409 the Open Firmware IEEE 1275 device-tree. It's basically a tree of
410 nodes, each node having two or more named properties. A property can
413 It is a tree, so each node has one and only one parent except for the
414 root node who has no parent.
416 A node has 2 names. The actual node name is generally contained in a
417 property of type "name" in the node property list whose value is a
418 zero terminated string and is mandatory for version 1 to 3 of the
419 format definition (as it is in Open Firmware). Version 0x10 makes it
420 optional as it can generate it from the unit name defined below.
422 There is also a "unit name" that is used to differenciate nodes with
423 the same name at the same level, it is usually made of the node
424 name's, the "@" sign, and a "unit address", which definition is
425 specific to the bus type the node sits on.
427 The unit name doesn't exist as a property per-se but is included in
428 the device-tree structure. It is typically used to represent "path" in
429 the device-tree. More details about the actual format of these will be
432 The kernel powerpc generic code does not make any formal use of the
433 unit address (though some board support code may do) so the only real
434 requirement here for the unit address is to ensure uniqueness of
435 the node unit name at a given level of the tree. Nodes with no notion
436 of address and no possible sibling of the same name (like /memory or
437 /cpus) may omit the unit address in the context of this specification,
438 or use the "@0" default unit address. The unit name is used to define
439 a node "full path", which is the concatenation of all parent node
440 unit names separated with "/".
442 The root node doesn't have a defined name, and isn't required to have
443 a name property either if you are using version 3 or earlier of the
444 format. It also has no unit address (no @ symbol followed by a unit
445 address). The root node unit name is thus an empty string. The full
446 path to the root node is "/".
448 Every node which actually represents an actual device (that is, a node
449 which isn't only a virtual "container" for more nodes, like "/cpus"
450 is) is also required to have a "device_type" property indicating the
453 Finally, every node that can be referenced from a property in another
454 node is required to have a "linux,phandle" property. Real open
455 firmware implementations provide a unique "phandle" value for every
456 node that the "prom_init()" trampoline code turns into
457 "linux,phandle" properties. However, this is made optional if the
458 flattened device tree is used directly. An example of a node
459 referencing another node via "phandle" is when laying out the
460 interrupt tree which will be described in a further version of this
463 This "linux, phandle" property is a 32 bit value that uniquely
464 identifies a node. You are free to use whatever values or system of
465 values, internal pointers, or whatever to generate these, the only
466 requirement is that every node for which you provide that property has
467 a unique value for it.
469 Here is an example of a simple device-tree. In this example, an "o"
470 designates a node followed by the node unit name. Properties are
471 presented with their name followed by their content. "content"
472 represents an ASCII string (zero terminated) value, while <content>
473 represents a 32 bit hexadecimal value. The various nodes in this
474 example will be discussed in a later chapter. At this point, it is
475 only meant to give you a idea of what a device-tree looks like. I have
476 purposefully kept the "name" and "linux,phandle" properties which
477 aren't necessary in order to give you a better idea of what the tree
478 looks like in practice.
481 |- name = "device-tree"
482 |- model = "MyBoardName"
483 |- compatible = "MyBoardFamilyName"
484 |- #address-cells = <2>
486 |- linux,phandle = <0>
490 | | - linux,phandle = <1>
491 | | - #address-cells = <1>
492 | | - #size-cells = <0>
495 | |- name = "PowerPC,970"
496 | |- device_type = "cpu"
498 | |- clock-frequency = <5f5e1000>
500 | |- linux,phandle = <2>
504 | |- device_type = "memory"
505 | |- reg = <00000000 00000000 00000000 20000000>
506 | |- linux,phandle = <3>
510 |- bootargs = "root=/dev/sda2"
511 |- linux,platform = <00000600>
512 |- linux,phandle = <4>
514 This tree is almost a minimal tree. It pretty much contains the
515 minimal set of required nodes and properties to boot a linux kernel;
516 that is, some basic model informations at the root, the CPUs, and the
517 physical memory layout. It also includes misc information passed
518 through /chosen, like in this example, the platform type (mandatory)
519 and the kernel command line arguments (optional).
521 The /cpus/PowerPC,970@0/linux,boot-cpu property is an example of a
522 property without a value. All other properties have a value. The
523 significance of the #address-cells and #size-cells properties will be
524 explained in chapter IV which defines precisely the required nodes and
525 properties and their content.
528 3) Device tree "structure" block
530 The structure of the device tree is a linearized tree structure. The
531 "OF_DT_BEGIN_NODE" token starts a new node, and the "OF_DT_END_NODE"
532 ends that node definition. Child nodes are simply defined before
533 "OF_DT_END_NODE" (that is nodes within the node). A 'token' is a 32
534 bit value. The tree has to be "finished" with a OF_DT_END token
536 Here's the basic structure of a single node:
538 * token OF_DT_BEGIN_NODE (that is 0x00000001)
539 * for version 1 to 3, this is the node full path as a zero
540 terminated string, starting with "/". For version 16 and later,
541 this is the node unit name only (or an empty string for the
543 * [align gap to next 4 bytes boundary]
545 * token OF_DT_PROP (that is 0x00000003)
546 * 32 bit value of property value size in bytes (or 0 of no
548 * 32 bit value of offset in string block of property name
549 * property value data if any
550 * [align gap to next 4 bytes boundary]
551 * [child nodes if any]
552 * token OF_DT_END_NODE (that is 0x00000002)
554 So the node content can be summmarised as a start token, a full path,
555 a list of properties, a list of child node and an end token. Every
556 child node is a full node structure itself as defined above.
558 4) Device tree 'strings" block
560 In order to save space, property names, which are generally redundant,
561 are stored separately in the "strings" block. This block is simply the
562 whole bunch of zero terminated strings for all property names
563 concatenated together. The device-tree property definitions in the
564 structure block will contain offset values from the beginning of the
568 III - Required content of the device tree
569 =========================================
571 WARNING: All "linux,*" properties defined in this document apply only
572 to a flattened device-tree. If your platform uses a real
573 implementation of Open Firmware or an implementation compatible with
574 the Open Firmware client interface, those properties will be created
575 by the trampoline code in the kernel's prom_init() file. For example,
576 that's where you'll have to add code to detect your board model and
577 set the platform number. However, when using the flatenned device-tree
578 entry point, there is no prom_init() pass, and thus you have to
579 provide those properties yourself.
582 1) Note about cells and address representation
583 ----------------------------------------------
585 The general rule is documented in the various Open Firmware
586 documentations. If you chose to describe a bus with the device-tree
587 and there exist an OF bus binding, then you should follow the
588 specification. However, the kernel does not require every single
589 device or bus to be described by the device tree.
591 In general, the format of an address for a device is defined by the
592 parent bus type, based on the #address-cells and #size-cells
593 property. In the absence of such a property, the parent's parent
594 values are used, etc... The kernel requires the root node to have
595 those properties defining addresses format for devices directly mapped
596 on the processor bus.
598 Those 2 properties define 'cells' for representing an address and a
599 size. A "cell" is a 32 bit number. For example, if both contain 2
600 like the example tree given above, then an address and a size are both
601 composed of 2 cells, and each is a 64 bit number (cells are
602 concatenated and expected to be in big endian format). Another example
603 is the way Apple firmware defines them, with 2 cells for an address
604 and one cell for a size. Most 32-bit implementations should define
605 #address-cells and #size-cells to 1, which represents a 32-bit value.
606 Some 32-bit processors allow for physical addresses greater than 32
607 bits; these processors should define #address-cells as 2.
609 "reg" properties are always a tuple of the type "address size" where
610 the number of cells of address and size is specified by the bus
611 #address-cells and #size-cells. When a bus supports various address
612 spaces and other flags relative to a given address allocation (like
613 prefetchable, etc...) those flags are usually added to the top level
614 bits of the physical address. For example, a PCI physical address is
615 made of 3 cells, the bottom two containing the actual address itself
616 while the top cell contains address space indication, flags, and pci
617 bus & device numbers.
619 For busses that support dynamic allocation, it's the accepted practice
620 to then not provide the address in "reg" (keep it 0) though while
621 providing a flag indicating the address is dynamically allocated, and
622 then, to provide a separate "assigned-addresses" property that
623 contains the fully allocated addresses. See the PCI OF bindings for
626 In general, a simple bus with no address space bits and no dynamic
627 allocation is preferred if it reflects your hardware, as the existing
628 kernel address parsing functions will work out of the box. If you
629 define a bus type with a more complex address format, including things
630 like address space bits, you'll have to add a bus translator to the
631 prom_parse.c file of the recent kernels for your bus type.
633 The "reg" property only defines addresses and sizes (if #size-cells
635 non-0) within a given bus. In order to translate addresses upward
636 (that is into parent bus addresses, and possibly into cpu physical
637 addresses), all busses must contain a "ranges" property. If the
638 "ranges" property is missing at a given level, it's assumed that
639 translation isn't possible. The format of the "ranges" proprety for a
642 bus address, parent bus address, size
644 "bus address" is in the format of the bus this bus node is defining,
645 that is, for a PCI bridge, it would be a PCI address. Thus, (bus
646 address, size) defines a range of addresses for child devices. "parent
647 bus address" is in the format of the parent bus of this bus. For
648 example, for a PCI host controller, that would be a CPU address. For a
649 PCI<->ISA bridge, that would be a PCI address. It defines the base
650 address in the parent bus where the beginning of that range is mapped.
652 For a new 64 bit powerpc board, I recommend either the 2/2 format or
653 Apple's 2/1 format which is slightly more compact since sizes usually
654 fit in a single 32 bit word. New 32 bit powerpc boards should use a
655 1/1 format, unless the processor supports physical addresses greater
656 than 32-bits, in which case a 2/1 format is recommended.
659 2) Note about "compatible" properties
660 -------------------------------------
662 These properties are optional, but recommended in devices and the root
663 node. The format of a "compatible" property is a list of concatenated
664 zero terminated strings. They allow a device to express its
665 compatibility with a family of similar devices, in some cases,
666 allowing a single driver to match against several devices regardless
667 of their actual names.
669 3) Note about "name" properties
670 -------------------------------
672 While earlier users of Open Firmware like OldWorld macintoshes tended
673 to use the actual device name for the "name" property, it's nowadays
674 considered a good practice to use a name that is closer to the device
675 class (often equal to device_type). For example, nowadays, ethernet
676 controllers are named "ethernet", an additional "model" property
677 defining precisely the chip type/model, and "compatible" property
678 defining the family in case a single driver can driver more than one
679 of these chips. However, the kernel doesn't generally put any
680 restriction on the "name" property; it is simply considered good
681 practice to follow the standard and its evolutions as closely as
684 Note also that the new format version 16 makes the "name" property
685 optional. If it's absent for a node, then the node's unit name is then
686 used to reconstruct the name. That is, the part of the unit name
687 before the "@" sign is used (or the entire unit name if no "@" sign
690 4) Note about node and property names and character set
691 -------------------------------------------------------
693 While open firmware provides more flexibe usage of 8859-1, this
694 specification enforces more strict rules. Nodes and properties should
695 be comprised only of ASCII characters 'a' to 'z', '0' to
696 '9', ',', '.', '_', '+', '#', '?', and '-'. Node names additionally
697 allow uppercase characters 'A' to 'Z' (property names should be
698 lowercase. The fact that vendors like Apple don't respect this rule is
699 irrelevant here). Additionally, node and property names should always
700 begin with a character in the range 'a' to 'z' (or 'A' to 'Z' for node
703 The maximum number of characters for both nodes and property names
704 is 31. In the case of node names, this is only the leftmost part of
705 a unit name (the pure "name" property), it doesn't include the unit
706 address which can extend beyond that limit.
709 5) Required nodes and properties
710 --------------------------------
711 These are all that are currently required. However, it is strongly
712 recommended that you expose PCI host bridges as documented in the
713 PCI binding to open firmware, and your interrupt tree as documented
714 in OF interrupt tree specification.
718 The root node requires some properties to be present:
720 - model : this is your board name/model
721 - #address-cells : address representation for "root" devices
722 - #size-cells: the size representation for "root" devices
724 Additionally, some recommended properties are:
726 - compatible : the board "family" generally finds its way here,
727 for example, if you have 2 board models with a similar layout,
728 that typically get driven by the same platform code in the
729 kernel, you would use a different "model" property but put a
730 value in "compatible". The kernel doesn't directly use that
731 value (see /chosen/linux,platform for how the kernel choses a
732 platform type) but it is generally useful.
734 The root node is also generally where you add additional properties
735 specific to your board like the serial number if any, that sort of
736 thing. it is recommended that if you add any "custom" property whose
737 name may clash with standard defined ones, you prefix them with your
738 vendor name and a comma.
742 This node is the parent of all individual CPU nodes. It doesn't
743 have any specific requirements, though it's generally good practice
746 #address-cells = <00000001>
747 #size-cells = <00000000>
749 This defines that the "address" for a CPU is a single cell, and has
750 no meaningful size. This is not necessary but the kernel will assume
751 that format when reading the "reg" properties of a CPU node, see
756 So under /cpus, you are supposed to create a node for every CPU on
757 the machine. There is no specific restriction on the name of the
758 CPU, though It's common practice to call it PowerPC,<name>. For
759 example, Apple uses PowerPC,G5 while IBM uses PowerPC,970FX.
763 - device_type : has to be "cpu"
764 - reg : This is the physical cpu number, it's a single 32 bit cell
765 and is also used as-is as the unit number for constructing the
766 unit name in the full path. For example, with 2 CPUs, you would
768 /cpus/PowerPC,970FX@0
769 /cpus/PowerPC,970FX@1
770 (unit addresses do not require leading zeroes)
771 - d-cache-line-size : one cell, L1 data cache line size in bytes
772 - i-cache-line-size : one cell, L1 instruction cache line size in
774 - d-cache-size : one cell, size of L1 data cache in bytes
775 - i-cache-size : one cell, size of L1 instruction cache in bytes
776 - linux, boot-cpu : Should be defined if this cpu is the boot cpu.
778 Recommended properties:
780 - timebase-frequency : a cell indicating the frequency of the
781 timebase in Hz. This is not directly used by the generic code,
782 but you are welcome to copy/paste the pSeries code for setting
783 the kernel timebase/decrementer calibration based on this
785 - clock-frequency : a cell indicating the CPU core clock frequency
786 in Hz. A new property will be defined for 64 bit values, but if
787 your frequency is < 4Ghz, one cell is enough. Here as well as
788 for the above, the common code doesn't use that property, but
789 you are welcome to re-use the pSeries or Maple one. A future
790 kernel version might provide a common function for this.
792 You are welcome to add any property you find relevant to your board,
793 like some information about the mechanism used to soft-reset the
794 CPUs. For example, Apple puts the GPIO number for CPU soft reset
795 lines in there as a "soft-reset" property since they start secondary
796 CPUs by soft-resetting them.
799 d) the /memory node(s)
801 To define the physical memory layout of your board, you should
802 create one or more memory node(s). You can either create a single
803 node with all memory ranges in its reg property, or you can create
804 several nodes, as you wish. The unit address (@ part) used for the
805 full path is the address of the first range of memory defined by a
806 given node. If you use a single memory node, this will typically be
811 - device_type : has to be "memory"
812 - reg : This property contains all the physical memory ranges of
813 your board. It's a list of addresses/sizes concatenated
814 together, with the number of cells of each defined by the
815 #address-cells and #size-cells of the root node. For example,
816 with both of these properties beeing 2 like in the example given
817 earlier, a 970 based machine with 6Gb of RAM could typically
818 have a "reg" property here that looks like:
820 00000000 00000000 00000000 80000000
821 00000001 00000000 00000001 00000000
823 That is a range starting at 0 of 0x80000000 bytes and a range
824 starting at 0x100000000 and of 0x100000000 bytes. You can see
825 that there is no memory covering the IO hole between 2Gb and
826 4Gb. Some vendors prefer splitting those ranges into smaller
827 segments, but the kernel doesn't care.
831 This node is a bit "special". Normally, that's where open firmware
832 puts some variable environment information, like the arguments, or
833 phandle pointers to nodes like the main interrupt controller, or the
834 default input/output devices.
836 This specification makes a few of these mandatory, but also defines
837 some linux-specific properties that would be normally constructed by
838 the prom_init() trampoline when booting with an OF client interface,
839 but that you have to provide yourself when using the flattened format.
843 - linux,platform : This is your platform number as assigned by the
844 architecture maintainers
846 Recommended properties:
848 - bootargs : This zero-terminated string is passed as the kernel
850 - linux,stdout-path : This is the full path to your standard
851 console device if any. Typically, if you have serial devices on
852 your board, you may want to put the full path to the one set as
853 the default console in the firmware here, for the kernel to pick
854 it up as it's own default console. If you look at the funciton
855 set_preferred_console() in arch/ppc64/kernel/setup.c, you'll see
856 that the kernel tries to find out the default console and has
857 knowledge of various types like 8250 serial ports. You may want
858 to extend this function to add your own.
859 - interrupt-controller : This is one cell containing a phandle
860 value that matches the "linux,phandle" property of your main
861 interrupt controller node. May be used for interrupt routing.
864 Note that u-boot creates and fills in the chosen node for platforms
867 f) the /soc<SOCname> node
869 This node is used to represent a system-on-a-chip (SOC) and must be
870 present if the processor is a SOC. The top-level soc node contains
871 information that is global to all devices on the SOC. The node name
872 should contain a unit address for the SOC, which is the base address
873 of the memory-mapped register set for the SOC. The name of an soc
874 node should start with "soc", and the remainder of the name should
875 represent the part number for the soc. For example, the MPC8540's
876 soc node would be called "soc8540".
880 - device_type : Should be "soc"
881 - ranges : Should be defined as specified in 1) to describe the
882 translation of SOC addresses for memory mapped SOC registers.
883 - bus-frequency: Contains the bus frequency for the SOC node.
884 Typically, the value of this field is filled in by the boot
888 Recommended properties:
890 - reg : This property defines the address and size of the
891 memory-mapped registers that are used for the SOC node itself.
892 It does not include the child device registers - these will be
893 defined inside each child node. The address specified in the
894 "reg" property should match the unit address of the SOC node.
895 - #address-cells : Address representation for "soc" devices. The
896 format of this field may vary depending on whether or not the
897 device registers are memory mapped. For memory mapped
898 registers, this field represents the number of cells needed to
899 represent the address of the registers. For SOCs that do not
900 use MMIO, a special address format should be defined that
901 contains enough cells to represent the required information.
902 See 1) above for more details on defining #address-cells.
903 - #size-cells : Size representation for "soc" devices
904 - #interrupt-cells : Defines the width of cells used to represent
905 interrupts. Typically this value is <2>, which includes a
906 32-bit number that represents the interrupt number, and a
907 32-bit number that represents the interrupt sense and level.
908 This field is only needed if the SOC contains an interrupt
911 The SOC node may contain child nodes for each SOC device that the
912 platform uses. Nodes should not be created for devices which exist
913 on the SOC but are not used by a particular platform. See chapter VI
914 for more information on how to specify devices that are part of an
917 Example SOC node for the MPC8540:
920 #address-cells = <1>;
922 #interrupt-cells = <2>;
924 ranges = <00000000 e0000000 00100000>
925 reg = <e0000000 00003000>;
931 IV - "dtc", the device tree compiler
932 ====================================
935 dtc source code can be found at
936 <http://ozlabs.org/~dgibson/dtc/dtc.tar.gz>
938 WARNING: This version is still in early development stage; the
939 resulting device-tree "blobs" have not yet been validated with the
940 kernel. The current generated bloc lacks a useful reserve map (it will
941 be fixed to generate an empty one, it's up to the bootloader to fill
942 it up) among others. The error handling needs work, bugs are lurking,
945 dtc basically takes a device-tree in a given format and outputs a
946 device-tree in another format. The currently supported formats are:
951 - "dtb": "blob" format, that is a flattened device-tree block
953 header all in a binary blob.
954 - "dts": "source" format. This is a text file containing a
955 "source" for a device-tree. The format is defined later in this
957 - "fs" format. This is a representation equivalent to the
958 output of /proc/device-tree, that is nodes are directories and
964 - "dtb": "blob" format
965 - "dts": "source" format
966 - "asm": assembly language file. This is a file that can be
967 sourced by gas to generate a device-tree "blob". That file can
968 then simply be added to your Makefile. Additionally, the
969 assembly file exports some symbols that can be use
972 The syntax of the dtc tool is
974 dtc [-I <input-format>] [-O <output-format>]
975 [-o output-filename] [-V output_version] input_filename
978 The "output_version" defines what versio of the "blob" format will be
979 generated. Supported versions are 1,2,3 and 16. The default is
980 currently version 3 but that may change in the future to version 16.
982 Additionally, dtc performs various sanity checks on the tree, like the
983 uniqueness of linux,phandle properties, validity of strings, etc...
985 The format of the .dts "source" file is "C" like, supports C and C++
991 The above is the "device-tree" definition. It's the only statement
992 supported currently at the toplevel.
995 property1 = "string_value"; /* define a property containing a 0
999 property2 = <1234abcd>; /* define a property containing a
1000 * numerical 32 bits value (hexadecimal)
1003 property3 = <12345678 12345678 deadbeef>;
1004 /* define a property containing 3
1005 * numerical 32 bits values (cells) in
1008 property4 = [0a 0b 0c 0d de ea ad be ef];
1009 /* define a property whose content is
1010 * an arbitrary array of bytes
1013 childnode@addresss { /* define a child node named "childnode"
1014 * whose unit name is "childnode at
1018 childprop = "hello\n"; /* define a property "childprop" of
1019 * childnode (in this case, a string)
1024 Nodes can contain other nodes etc... thus defining the hierarchical
1025 structure of the tree.
1027 Strings support common escape sequences from C: "\n", "\t", "\r",
1028 "\(octal value)", "\x(hex value)".
1030 It is also suggested that you pipe your source file through cpp (gcc
1031 preprocessor) so you can use #include's, #define for constants, etc...
1033 Finally, various options are planned but not yet implemented, like
1034 automatic generation of phandles, labels (exported to the asm file so
1035 you can point to a property content and change it easily from whatever
1036 you link the device-tree with), label or path instead of numeric value
1037 in some cells to "point" to a node (replaced by a phandle at compile
1038 time), export of reserve map address to the asm file, ability to
1039 specify reserve map content at compile time, etc...
1041 We may provide a .h include file with common definitions of that
1042 proves useful for some properties (like building PCI properties or
1043 interrupt maps) though it may be better to add a notion of struct
1044 definitions to the compiler...
1047 V - Recommendations for a bootloader
1048 ====================================
1051 Here are some various ideas/recommendations that have been proposed
1052 while all this has been defined and implemented.
1054 - The bootloader may want to be able to use the device-tree itself
1055 and may want to manipulate it (to add/edit some properties,
1056 like physical memory size or kernel arguments). At this point, 2
1057 choices can be made. Either the bootloader works directly on the
1058 flattened format, or the bootloader has its own internal tree
1059 representation with pointers (similar to the kernel one) and
1060 re-flattens the tree when booting the kernel. The former is a bit
1061 more difficult to edit/modify, the later requires probably a bit
1062 more code to handle the tree structure. Note that the structure
1063 format has been designed so it's relatively easy to "insert"
1064 properties or nodes or delete them by just memmoving things
1065 around. It contains no internal offsets or pointers for this
1068 - An example of code for iterating nodes & retreiving properties
1069 directly from the flattened tree format can be found in the kernel
1070 file arch/ppc64/kernel/prom.c, look at scan_flat_dt() function,
1071 it's usage in early_init_devtree(), and the corresponding various
1072 early_init_dt_scan_*() callbacks. That code can be re-used in a
1073 GPL bootloader, and as the author of that code, I would be happy
1074 do discuss possible free licencing to any vendor who wishes to
1075 integrate all or part of this code into a non-GPL bootloader.
1079 VI - System-on-a-chip devices and nodes
1080 =======================================
1082 Many companies are now starting to develop system-on-a-chip
1083 processors, where the processor core (cpu) and many peripheral devices
1084 exist on a single piece of silicon. For these SOCs, an SOC node
1085 should be used that defines child nodes for the devices that make
1086 up the SOC. While platforms are not required to use this model in
1087 order to boot the kernel, it is highly encouraged that all SOC
1088 implementations define as complete a flat-device-tree as possible to
1089 describe the devices on the SOC. This will allow for the
1090 genericization of much of the kernel code.
1093 1) Defining child nodes of an SOC
1094 ---------------------------------
1096 Each device that is part of an SOC may have its own node entry inside
1097 the SOC node. For each device that is included in the SOC, the unit
1098 address property represents the address offset for this device's
1099 memory-mapped registers in the parent's address space. The parent's
1100 address space is defined by the "ranges" property in the top-level soc
1101 node. The "reg" property for each node that exists directly under the
1102 SOC node should contain the address mapping from the child address space
1103 to the parent SOC address space and the size of the device's
1104 memory-mapped register file.
1106 For many devices that may exist inside an SOC, there are predefined
1107 specifications for the format of the device tree node. All SOC child
1108 nodes should follow these specifications, except where noted in this
1111 See appendix A for an example partial SOC node definition for the
1115 2) Specifying interrupt information for SOC devices
1116 ---------------------------------------------------
1118 Each device that is part of an SOC and which generates interrupts
1119 should have the following properties:
1121 - interrupt-parent : contains the phandle of the interrupt
1122 controller which handles interrupts for this device
1123 - interrupts : a list of tuples representing the interrupt
1124 number and the interrupt sense and level for each interupt
1127 This information is used by the kernel to build the interrupt table
1128 for the interrupt controllers in the system.
1130 Sense and level information should be encoded as follows:
1132 Devices connected to openPIC-compatible controllers should encode
1133 sense and polarity as follows:
1135 0 = high to low edge sensitive type enabled
1136 1 = active low level sensitive type enabled
1137 2 = low to high edge sensitive type enabled
1138 3 = active high level sensitive type enabled
1140 ISA PIC interrupt controllers should adhere to the ISA PIC
1141 encodings listed below:
1143 0 = active low level sensitive type enabled
1144 1 = active high level sensitive type enabled
1145 2 = high to low edge sensitive type enabled
1146 3 = low to high edge sensitive type enabled
1150 3) Representing devices without a current OF specification
1151 ----------------------------------------------------------
1153 Currently, there are many devices on SOCs that do not have a standard
1154 representation pre-defined as part of the open firmware
1155 specifications, mainly because the boards that contain these SOCs are
1156 not currently booted using open firmware. This section contains
1157 descriptions for the SOC devices for which new nodes have been
1158 defined; this list will expand as more and more SOC-containing
1159 platforms are moved over to use the flattened-device-tree model.
1163 The MDIO is a bus to which the PHY devices are connected. For each
1164 device that exists on this bus, a child node should be created. See
1165 the definition of the PHY node below for an example of how to define
1168 Required properties:
1169 - reg : Offset and length of the register set for the device
1170 - device_type : Should be "mdio"
1171 - compatible : Should define the compatible device type for the
1172 mdio. Currently, this is most likely to be "gianfar"
1178 device_type = "mdio";
1179 compatible = "gianfar";
1187 b) Gianfar-compatible ethernet nodes
1189 Required properties:
1191 - device_type : Should be "network"
1192 - model : Model of the device. Can be "TSEC", "eTSEC", or "FEC"
1193 - compatible : Should be "gianfar"
1194 - reg : Offset and length of the register set for the device
1195 - address : List of bytes representing the ethernet address of
1197 - interrupts : <a b> where a is the interrupt number and b is a
1198 field that represents an encoding of the sense and level
1199 information for the interrupt. This should be encoded based on
1200 the information in section 2) depending on the type of interrupt
1201 controller you have.
1202 - interrupt-parent : the phandle for the interrupt controller that
1203 services interrupts for this device.
1204 - phy-handle : The phandle for the PHY connected to this ethernet
1211 device_type = "network";
1213 compatible = "gianfar";
1215 address = [ 00 E0 0C 00 73 00 ];
1216 interrupts = <d 3 e 3 12 3>;
1217 interrupt-parent = <40000>;
1218 phy-handle = <2452000>
1225 Required properties:
1227 - device_type : Should be "ethernet-phy"
1228 - interrupts : <a b> where a is the interrupt number and b is a
1229 field that represents an encoding of the sense and level
1230 information for the interrupt. This should be encoded based on
1231 the information in section 2) depending on the type of interrupt
1232 controller you have.
1233 - interrupt-parent : the phandle for the interrupt controller that
1234 services interrupts for this device.
1235 - reg : The ID number for the phy, usually a small integer
1236 - linux,phandle : phandle for this node; likely referenced by an
1237 ethernet controller node.
1243 linux,phandle = <2452000>
1244 interrupt-parent = <40000>;
1245 interrupts = <35 1>;
1247 device_type = "ethernet-phy";
1251 d) Interrupt controllers
1253 Some SOC devices contain interrupt controllers that are different
1254 from the standard Open PIC specification. The SOC device nodes for
1255 these types of controllers should be specified just like a standard
1256 OpenPIC controller. Sense and level information should be encoded
1257 as specified in section 2) of this chapter for each device that
1258 specifies an interrupt.
1263 linux,phandle = <40000>;
1264 clock-frequency = <0>;
1265 interrupt-controller;
1266 #address-cells = <0>;
1267 reg = <40000 40000>;
1269 compatible = "chrp,open-pic";
1270 device_type = "open-pic";
1277 Required properties :
1279 - device_type : Should be "i2c"
1280 - reg : Offset and length of the register set for the device
1282 Recommended properties :
1284 - compatible : Should be "fsl-i2c" for parts compatible with
1285 Freescale I2C specifications.
1286 - interrupts : <a b> where a is the interrupt number and b is a
1287 field that represents an encoding of the sense and level
1288 information for the interrupt. This should be encoded based on
1289 the information in section 2) depending on the type of interrupt
1290 controller you have.
1291 - interrupt-parent : the phandle for the interrupt controller that
1292 services interrupts for this device.
1293 - dfsrr : boolean; if defined, indicates that this I2C device has
1294 a digital filter sampling rate register
1295 - fsl5200-clocking : boolean; if defined, indicated that this device
1296 uses the FSL 5200 clocking mechanism.
1301 interrupt-parent = <40000>;
1302 interrupts = <1b 3>;
1304 device_type = "i2c";
1305 compatible = "fsl-i2c";
1310 More devices will be defined as this spec matures.
1313 Appendix A - Sample SOC node for MPC8540
1314 ========================================
1316 Note that the #address-cells and #size-cells for the SoC node
1317 in this example have been explicitly listed; these are likely
1318 not necessary as they are usually the same as the root node.
1321 #address-cells = <1>;
1323 #interrupt-cells = <2>;
1324 device_type = "soc";
1325 ranges = <00000000 e0000000 00100000>
1326 reg = <e0000000 00003000>;
1327 bus-frequency = <0>;
1331 device_type = "mdio";
1332 compatible = "gianfar";
1335 linux,phandle = <2452000>
1336 interrupt-parent = <40000>;
1337 interrupts = <35 1>;
1339 device_type = "ethernet-phy";
1343 linux,phandle = <2452001>
1344 interrupt-parent = <40000>;
1345 interrupts = <35 1>;
1347 device_type = "ethernet-phy";
1351 linux,phandle = <2452002>
1352 interrupt-parent = <40000>;
1353 interrupts = <35 1>;
1355 device_type = "ethernet-phy";
1362 device_type = "network";
1364 compatible = "gianfar";
1366 address = [ 00 E0 0C 00 73 00 ];
1367 interrupts = <d 3 e 3 12 3>;
1368 interrupt-parent = <40000>;
1369 phy-handle = <2452000>;
1373 #address-cells = <1>;
1375 device_type = "network";
1377 compatible = "gianfar";
1379 address = [ 00 E0 0C 00 73 01 ];
1380 interrupts = <13 3 14 3 18 3>;
1381 interrupt-parent = <40000>;
1382 phy-handle = <2452001>;
1386 #address-cells = <1>;
1388 device_type = "network";
1390 compatible = "gianfar";
1392 address = [ 00 E0 0C 00 73 02 ];
1393 interrupts = <19 3>;
1394 interrupt-parent = <40000>;
1395 phy-handle = <2452002>;
1399 device_type = "serial";
1400 compatible = "ns16550";
1402 clock-frequency = <0>;
1403 interrupts = <1a 3>;
1404 interrupt-parent = <40000>;
1408 linux,phandle = <40000>;
1409 clock-frequency = <0>;
1410 interrupt-controller;
1411 #address-cells = <0>;
1412 reg = <40000 40000>;
1414 compatible = "chrp,open-pic";
1415 device_type = "open-pic";
1420 interrupt-parent = <40000>;
1421 interrupts = <1b 3>;
1423 device_type = "i2c";
1424 compatible = "fsl-i2c";