blob: b2fb2f5e19220e6b59bab121a521af2fe7ce2b54 [file] [log] [blame]
David Gibsonc125a182006-02-01 03:05:22 -08001 Booting the Linux/ppc kernel without Open Firmware
2 --------------------------------------------------
3
David Gibsonc125a182006-02-01 03:05:22 -08004(c) 2005 Benjamin Herrenschmidt <benh at kernel.crashing.org>,
5 IBM Corp.
6(c) 2005 Becky Bruce <becky.bruce at freescale.com>,
7 Freescale Semiconductor, FSL SOC and 32-bit additions
Vitaly Wool28f9ec32006-11-20 16:32:39 +03008(c) 2006 MontaVista Software, Inc.
9 Flash chip node definition
David Gibsonc125a182006-02-01 03:05:22 -080010
Stuart Yoder5e1e9ba2007-06-06 04:29:14 +100011Table of Contents
12=================
13
14 I - Introduction
Grant Likelyede338f2011-04-28 14:27:23 -060015 1) Entry point for arch/arm
16 2) Entry point for arch/powerpc
17 3) Entry point for arch/x86
Stuart Yoder5e1e9ba2007-06-06 04:29:14 +100018
19 II - The DT block format
20 1) Header
21 2) Device tree generalities
22 3) Device tree "structure" block
23 4) Device tree "strings" block
24
25 III - Required content of the device tree
26 1) Note about cells and address representation
27 2) Note about "compatible" properties
28 3) Note about "name" properties
29 4) Note about node and property names and character set
30 5) Required nodes and properties
31 a) The root node
32 b) The /cpus node
33 c) The /cpus/* nodes
34 d) the /memory node(s)
35 e) The /chosen node
36 f) the /soc<SOCname> node
37
38 IV - "dtc", the device tree compiler
39
40 V - Recommendations for a bootloader
41
42 VI - System-on-a-chip devices and nodes
43 1) Defining child nodes of an SOC
44 2) Representing devices without a current OF specification
Stuart Yoder5e1e9ba2007-06-06 04:29:14 +100045
Anton Vorontsovb9e0ba82010-08-11 20:56:03 +040046 VII - Specifying interrupt information for devices
Stuart Yoder5e1e9ba2007-06-06 04:29:14 +100047 1) interrupts property
48 2) interrupt-parent property
49 3) OpenPIC Interrupt Controllers
50 4) ISA Interrupt Controllers
51
Anton Vorontsovb9e0ba82010-08-11 20:56:03 +040052 VIII - Specifying device power management information (sleep property)
Scott Wood2dff4172008-07-11 17:31:15 -050053
Stuart Yoder5e1e9ba2007-06-06 04:29:14 +100054 Appendix A - Sample SOC node for MPC8540
55
56
57Revision Information
58====================
59
David Gibsonc125a182006-02-01 03:05:22 -080060 May 18, 2005: Rev 0.1 - Initial draft, no chapter III yet.
61
62 May 19, 2005: Rev 0.2 - Add chapter III and bits & pieces here or
63 clarifies the fact that a lot of things are
64 optional, the kernel only requires a very
65 small device tree, though it is encouraged
66 to provide an as complete one as possible.
67
68 May 24, 2005: Rev 0.3 - Precise that DT block has to be in RAM
69 - Misc fixes
70 - Define version 3 and new format version 16
71 for the DT block (version 16 needs kernel
72 patches, will be fwd separately).
73 String block now has a size, and full path
74 is replaced by unit name for more
75 compactness.
76 linux,phandle is made optional, only nodes
77 that are referenced by other nodes need it.
78 "name" property is now automatically
79 deduced from the unit name
80
81 June 1, 2005: Rev 0.4 - Correct confusion between OF_DT_END and
82 OF_DT_END_NODE in structure definition.
83 - Change version 16 format to always align
84 property data to 4 bytes. Since tokens are
85 already aligned, that means no specific
Matt LaPlante5d3f0832006-11-30 05:21:10 +010086 required alignment between property size
David Gibsonc125a182006-02-01 03:05:22 -080087 and property data. The old style variable
88 alignment would make it impossible to do
89 "simple" insertion of properties using
Domen Puncer5dd60162007-03-02 21:44:45 +110090 memmove (thanks Milton for
David Gibsonc125a182006-02-01 03:05:22 -080091 noticing). Updated kernel patch as well
Matt LaPlante5d3f0832006-11-30 05:21:10 +010092 - Correct a few more alignment constraints
David Gibsonc125a182006-02-01 03:05:22 -080093 - Add a chapter about the device-tree
94 compiler and the textural representation of
95 the tree that can be "compiled" by dtc.
96
David Gibsonc125a182006-02-01 03:05:22 -080097 November 21, 2005: Rev 0.5
98 - Additions/generalizations for 32-bit
99 - Changed to reflect the new arch/powerpc
100 structure
101 - Added chapter VI
102
103
104 ToDo:
105 - Add some definitions of interrupt tree (simple/complex)
Domen Puncer5dd60162007-03-02 21:44:45 +1100106 - Add some definitions for PCI host bridges
David Gibsonc125a182006-02-01 03:05:22 -0800107 - Add some common address format examples
108 - Add definitions for standard properties and "compatible"
109 names for cells that are not already defined by the existing
110 OF spec.
111 - Compare FSL SOC use of PCI to standard and make sure no new
112 node definition required.
113 - Add more information about node definitions for SOC devices
114 that currently have no standard, like the FSL CPM.
115
116
117I - Introduction
118================
119
Grant Likelycf4e5c62011-01-31 00:12:26 -0700120During the development of the Linux/ppc64 kernel, and more
David Gibsonc125a182006-02-01 03:05:22 -0800121specifically, the addition of new platform types outside of the old
122IBM pSeries/iSeries pair, it was decided to enforce some strict rules
123regarding the kernel entry and bootloader <-> kernel interfaces, in
124order to avoid the degeneration that had become the ppc32 kernel entry
125point and the way a new platform should be added to the kernel. The
126legacy iSeries platform breaks those rules as it predates this scheme,
127but no new board support will be accepted in the main tree that
Lennert Buytenhek475fc7c2010-09-21 23:22:40 +0000128doesn't follow them properly. In addition, since the advent of the
David Gibsonc125a182006-02-01 03:05:22 -0800129arch/powerpc merged architecture for ppc32 and ppc64, new 32-bit
130platforms and 32-bit platforms which move into arch/powerpc will be
131required to use these rules as well.
132
133The main requirement that will be defined in more detail below is
134the presence of a device-tree whose format is defined after Open
135Firmware specification. However, in order to make life easier
136to embedded board vendors, the kernel doesn't require the device-tree
137to represent every device in the system and only requires some nodes
138and properties to be present. This will be described in detail in
139section III, but, for example, the kernel does not require you to
140create a node for every PCI device in the system. It is a requirement
141to have a node for PCI host bridges in order to provide interrupt
Sylvestre Ledruf65e51d2011-04-04 15:04:46 -0700142routing information and memory/IO ranges, among others. It is also
Grant Likelycf4e5c62011-01-31 00:12:26 -0700143recommended to define nodes for on chip devices and other buses that
David Gibsonc125a182006-02-01 03:05:22 -0800144don't specifically fit in an existing OF specification. This creates a
145great flexibility in the way the kernel can then probe those and match
146drivers to device, without having to hard code all sorts of tables. It
147also makes it more flexible for board vendors to do minor hardware
148upgrades without significantly impacting the kernel code or cluttering
149it with special cases.
150
151
Grant Likelyede338f2011-04-28 14:27:23 -06001521) Entry point for arch/arm
153---------------------------
154
155 There is one single entry point to the kernel, at the start
156 of the kernel image. That entry point supports two calling
157 conventions. A summary of the interface is described here. A full
158 description of the boot requirements is documented in
159 Documentation/arm/Booting
160
161 a) ATAGS interface. Minimal information is passed from firmware
162 to the kernel with a tagged list of predefined parameters.
163
164 r0 : 0
165
166 r1 : Machine type number
167
168 r2 : Physical address of tagged list in system RAM
169
170 b) Entry with a flattened device-tree block. Firmware loads the
171 physical address of the flattened device tree block (dtb) into r2,
Masanari Iida40e47122012-03-04 23:16:11 +0900172 r1 is not used, but it is considered good practice to use a valid
Grant Likelyede338f2011-04-28 14:27:23 -0600173 machine number as described in Documentation/arm/Booting.
174
175 r0 : 0
176
177 r1 : Valid machine type number. When using a device tree,
178 a single machine type number will often be assigned to
179 represent a class or family of SoCs.
180
181 r2 : physical pointer to the device-tree block
182 (defined in chapter II) in RAM. Device tree can be located
183 anywhere in system RAM, but it should be aligned on a 64 bit
184 boundary.
185
186 The kernel will differentiate between ATAGS and device tree booting by
187 reading the memory pointed to by r2 and looking for either the flattened
188 device tree block magic value (0xd00dfeed) or the ATAG_CORE value at
189 offset 0x4 from r2 (0x54410001).
190
1912) Entry point for arch/powerpc
David Gibsonc125a182006-02-01 03:05:22 -0800192-------------------------------
193
Grant Likelycf4e5c62011-01-31 00:12:26 -0700194 There is one single entry point to the kernel, at the start
David Gibsonc125a182006-02-01 03:05:22 -0800195 of the kernel image. That entry point supports two calling
196 conventions:
197
198 a) Boot from Open Firmware. If your firmware is compatible
199 with Open Firmware (IEEE 1275) or provides an OF compatible
200 client interface API (support for "interpret" callback of
201 forth words isn't required), you can enter the kernel with:
202
203 r5 : OF callback pointer as defined by IEEE 1275
Domen Puncer5dd60162007-03-02 21:44:45 +1100204 bindings to powerpc. Only the 32-bit client interface
David Gibsonc125a182006-02-01 03:05:22 -0800205 is currently supported
206
207 r3, r4 : address & length of an initrd if any or 0
208
209 The MMU is either on or off; the kernel will run the
210 trampoline located in arch/powerpc/kernel/prom_init.c to
211 extract the device-tree and other information from open
212 firmware and build a flattened device-tree as described
213 in b). prom_init() will then re-enter the kernel using
214 the second method. This trampoline code runs in the
215 context of the firmware, which is supposed to handle all
216 exceptions during that time.
217
218 b) Direct entry with a flattened device-tree block. This entry
219 point is called by a) after the OF trampoline and can also be
220 called directly by a bootloader that does not support the Open
221 Firmware client interface. It is also used by "kexec" to
222 implement "hot" booting of a new kernel from a previous
223 running one. This method is what I will describe in more
224 details in this document, as method a) is simply standard Open
225 Firmware, and thus should be implemented according to the
226 various standard documents defining it and its binding to the
227 PowerPC platform. The entry point definition then becomes:
228
229 r3 : physical pointer to the device-tree block
230 (defined in chapter II) in RAM
231
232 r4 : physical pointer to the kernel itself. This is
233 used by the assembly code to properly disable the MMU
234 in case you are entering the kernel with MMU enabled
235 and a non-1:1 mapping.
236
Matt LaPlante2fe0ae72006-10-03 22:50:39 +0200237 r5 : NULL (as to differentiate with method a)
David Gibsonc125a182006-02-01 03:05:22 -0800238
239 Note about SMP entry: Either your firmware puts your other
240 CPUs in some sleep loop or spin loop in ROM where you can get
241 them out via a soft reset or some other means, in which case
242 you don't need to care, or you'll have to enter the kernel
243 with all CPUs. The way to do that with method b) will be
244 described in a later revision of this document.
245
David Gibsonc125a182006-02-01 03:05:22 -0800246 Board supports (platforms) are not exclusive config options. An
247 arbitrary set of board supports can be built in a single kernel
248 image. The kernel will "know" what set of functions to use for a
249 given platform based on the content of the device-tree. Thus, you
250 should:
251
252 a) add your platform support as a _boolean_ option in
253 arch/powerpc/Kconfig, following the example of PPC_PSERIES,
254 PPC_PMAC and PPC_MAPLE. The later is probably a good
255 example of a board support to start from.
256
257 b) create your main platform file as
258 "arch/powerpc/platforms/myplatform/myboard_setup.c" and add it
259 to the Makefile under the condition of your CONFIG_
260 option. This file will define a structure of type "ppc_md"
261 containing the various callbacks that the generic code will
262 use to get to your platform specific code
263
Grant Likelycf4e5c62011-01-31 00:12:26 -0700264 A kernel image may support multiple platforms, but only if the
Domen Puncer5dd60162007-03-02 21:44:45 +1100265 platforms feature the same core architecture. A single kernel build
David Gibsonc125a182006-02-01 03:05:22 -0800266 cannot support both configurations with Book E and configurations
267 with classic Powerpc architectures.
268
Grant Likelyede338f2011-04-28 14:27:23 -06002693) Entry point for arch/x86
Sebastian Andrzej Siewiorda6b7372011-02-22 21:07:37 +0100270-------------------------------
271
272 There is one single 32bit entry point to the kernel at code32_start,
273 the decompressor (the real mode entry point goes to the same 32bit
274 entry point once it switched into protected mode). That entry point
275 supports one calling convention which is documented in
276 Documentation/x86/boot.txt
277 The physical pointer to the device-tree block (defined in chapter II)
278 is passed via setup_data which requires at least boot protocol 2.09.
279 The type filed is defined as
280
281 #define SETUP_DTB 2
282
283 This device-tree is used as an extension to the "boot page". As such it
284 does not parse / consider data which is already covered by the boot
285 page. This includes memory size, reserved ranges, command line arguments
286 or initrd address. It simply holds information which can not be retrieved
287 otherwise like interrupt routing or a list of devices behind an I2C bus.
David Gibsonc125a182006-02-01 03:05:22 -0800288
289II - The DT block format
290========================
291
292
293This chapter defines the actual format of the flattened device-tree
294passed to the kernel. The actual content of it and kernel requirements
295are described later. You can find example of code manipulating that
296format in various places, including arch/powerpc/kernel/prom_init.c
297which will generate a flattened device-tree from the Open Firmware
298representation, or the fs2dt utility which is part of the kexec tools
299which will generate one from a filesystem representation. It is
300expected that a bootloader like uboot provides a bit more support,
301that will be discussed later as well.
302
303Note: The block has to be in main memory. It has to be accessible in
304both real mode and virtual mode with no mapping other than main
305memory. If you are writing a simple flash bootloader, it should copy
306the block to RAM before passing it to the kernel.
307
308
3091) Header
310---------
311
Grant Likelycf4e5c62011-01-31 00:12:26 -0700312 The kernel is passed the physical address pointing to an area of memory
313 that is roughly described in include/linux/of_fdt.h by the structure
David Gibsonc125a182006-02-01 03:05:22 -0800314 boot_param_header:
315
316struct boot_param_header {
317 u32 magic; /* magic word OF_DT_HEADER */
318 u32 totalsize; /* total size of DT block */
319 u32 off_dt_struct; /* offset to structure */
320 u32 off_dt_strings; /* offset to strings */
321 u32 off_mem_rsvmap; /* offset to memory reserve map
Domen Puncer5dd60162007-03-02 21:44:45 +1100322 */
David Gibsonc125a182006-02-01 03:05:22 -0800323 u32 version; /* format version */
324 u32 last_comp_version; /* last compatible version */
325
326 /* version 2 fields below */
327 u32 boot_cpuid_phys; /* Which physical CPU id we're
328 booting on */
329 /* version 3 fields below */
330 u32 size_dt_strings; /* size of the strings block */
David Gibson0e0293c2007-03-14 11:50:40 +1100331
332 /* version 17 fields below */
333 u32 size_dt_struct; /* size of the DT structure block */
David Gibsonc125a182006-02-01 03:05:22 -0800334};
335
336 Along with the constants:
337
338/* Definitions used by the flattened device tree */
339#define OF_DT_HEADER 0xd00dfeed /* 4: version,
340 4: total size */
341#define OF_DT_BEGIN_NODE 0x1 /* Start node: full name
Domen Puncer5dd60162007-03-02 21:44:45 +1100342 */
David Gibsonc125a182006-02-01 03:05:22 -0800343#define OF_DT_END_NODE 0x2 /* End node */
344#define OF_DT_PROP 0x3 /* Property: name off,
345 size, content */
346#define OF_DT_END 0x9
347
348 All values in this header are in big endian format, the various
349 fields in this header are defined more precisely below. All
350 "offset" values are in bytes from the start of the header; that is
Grant Likelycf4e5c62011-01-31 00:12:26 -0700351 from the physical base address of the device tree block.
David Gibsonc125a182006-02-01 03:05:22 -0800352
353 - magic
354
355 This is a magic value that "marks" the beginning of the
356 device-tree block header. It contains the value 0xd00dfeed and is
357 defined by the constant OF_DT_HEADER
358
359 - totalsize
360
361 This is the total size of the DT block including the header. The
362 "DT" block should enclose all data structures defined in this
363 chapter (who are pointed to by offsets in this header). That is,
364 the device-tree structure, strings, and the memory reserve map.
365
366 - off_dt_struct
367
368 This is an offset from the beginning of the header to the start
369 of the "structure" part the device tree. (see 2) device tree)
370
371 - off_dt_strings
372
373 This is an offset from the beginning of the header to the start
374 of the "strings" part of the device-tree
375
376 - off_mem_rsvmap
377
378 This is an offset from the beginning of the header to the start
Domen Puncer5dd60162007-03-02 21:44:45 +1100379 of the reserved memory map. This map is a list of pairs of 64-
David Gibsonc125a182006-02-01 03:05:22 -0800380 bit integers. Each pair is a physical address and a size. The
David Gibsonc125a182006-02-01 03:05:22 -0800381 list is terminated by an entry of size 0. This map provides the
382 kernel with a list of physical memory areas that are "reserved"
383 and thus not to be used for memory allocations, especially during
384 early initialization. The kernel needs to allocate memory during
385 boot for things like un-flattening the device-tree, allocating an
386 MMU hash table, etc... Those allocations must be done in such a
387 way to avoid overriding critical things like, on Open Firmware
388 capable machines, the RTAS instance, or on some pSeries, the TCE
389 tables used for the iommu. Typically, the reserve map should
390 contain _at least_ this DT block itself (header,total_size). If
391 you are passing an initrd to the kernel, you should reserve it as
392 well. You do not need to reserve the kernel image itself. The map
Domen Puncer5dd60162007-03-02 21:44:45 +1100393 should be 64-bit aligned.
David Gibsonc125a182006-02-01 03:05:22 -0800394
395 - version
396
397 This is the version of this structure. Version 1 stops
398 here. Version 2 adds an additional field boot_cpuid_phys.
399 Version 3 adds the size of the strings block, allowing the kernel
400 to reallocate it easily at boot and free up the unused flattened
401 structure after expansion. Version 16 introduces a new more
402 "compact" format for the tree itself that is however not backward
David Gibson0e0293c2007-03-14 11:50:40 +1100403 compatible. Version 17 adds an additional field, size_dt_struct,
404 allowing it to be reallocated or moved more easily (this is
405 particularly useful for bootloaders which need to make
406 adjustments to a device tree based on probed information). You
407 should always generate a structure of the highest version defined
408 at the time of your implementation. Currently that is version 17,
409 unless you explicitly aim at being backward compatible.
David Gibsonc125a182006-02-01 03:05:22 -0800410
411 - last_comp_version
412
413 Last compatible version. This indicates down to what version of
414 the DT block you are backward compatible. For example, version 2
415 is backward compatible with version 1 (that is, a kernel build
416 for version 1 will be able to boot with a version 2 format). You
417 should put a 1 in this field if you generate a device tree of
David Gibson0e0293c2007-03-14 11:50:40 +1100418 version 1 to 3, or 16 if you generate a tree of version 16 or 17
David Gibsonc125a182006-02-01 03:05:22 -0800419 using the new unit name format.
420
421 - boot_cpuid_phys
422
423 This field only exist on version 2 headers. It indicate which
424 physical CPU ID is calling the kernel entry point. This is used,
425 among others, by kexec. If you are on an SMP system, this value
426 should match the content of the "reg" property of the CPU node in
427 the device-tree corresponding to the CPU calling the kernel entry
Sylvestre Ledruf65e51d2011-04-04 15:04:46 -0700428 point (see further chapters for more information on the required
David Gibsonc125a182006-02-01 03:05:22 -0800429 device-tree contents)
430
David Gibson0e0293c2007-03-14 11:50:40 +1100431 - size_dt_strings
432
433 This field only exists on version 3 and later headers. It
434 gives the size of the "strings" section of the device tree (which
435 starts at the offset given by off_dt_strings).
436
437 - size_dt_struct
438
439 This field only exists on version 17 and later headers. It gives
440 the size of the "structure" section of the device tree (which
441 starts at the offset given by off_dt_struct).
David Gibsonc125a182006-02-01 03:05:22 -0800442
443 So the typical layout of a DT block (though the various parts don't
444 need to be in that order) looks like this (addresses go from top to
445 bottom):
446
447
448 ------------------------------
Grant Likelycf4e5c62011-01-31 00:12:26 -0700449 base -> | struct boot_param_header |
David Gibsonc125a182006-02-01 03:05:22 -0800450 ------------------------------
451 | (alignment gap) (*) |
452 ------------------------------
453 | memory reserve map |
454 ------------------------------
455 | (alignment gap) |
456 ------------------------------
457 | |
458 | device-tree structure |
459 | |
460 ------------------------------
461 | (alignment gap) |
462 ------------------------------
463 | |
464 | device-tree strings |
465 | |
466 -----> ------------------------------
467 |
468 |
Grant Likelycf4e5c62011-01-31 00:12:26 -0700469 --- (base + totalsize)
David Gibsonc125a182006-02-01 03:05:22 -0800470
471 (*) The alignment gaps are not necessarily present; their presence
472 and size are dependent on the various alignment requirements of
473 the individual data blocks.
474
475
4762) Device tree generalities
477---------------------------
478
479This device-tree itself is separated in two different blocks, a
480structure block and a strings block. Both need to be aligned to a 4
481byte boundary.
482
483First, let's quickly describe the device-tree concept before detailing
484the storage format. This chapter does _not_ describe the detail of the
485required types of nodes & properties for the kernel, this is done
486later in chapter III.
487
488The device-tree layout is strongly inherited from the definition of
489the Open Firmware IEEE 1275 device-tree. It's basically a tree of
490nodes, each node having two or more named properties. A property can
491have a value or not.
492
493It is a tree, so each node has one and only one parent except for the
494root node who has no parent.
495
496A node has 2 names. The actual node name is generally contained in a
497property of type "name" in the node property list whose value is a
498zero terminated string and is mandatory for version 1 to 3 of the
David Gibson0e0293c2007-03-14 11:50:40 +1100499format definition (as it is in Open Firmware). Version 16 makes it
David Gibsonc125a182006-02-01 03:05:22 -0800500optional as it can generate it from the unit name defined below.
501
Matt LaPlante2fe0ae72006-10-03 22:50:39 +0200502There is also a "unit name" that is used to differentiate nodes with
David Gibsonc125a182006-02-01 03:05:22 -0800503the same name at the same level, it is usually made of the node
Matt LaPlante2fe0ae72006-10-03 22:50:39 +0200504names, the "@" sign, and a "unit address", which definition is
David Gibsonc125a182006-02-01 03:05:22 -0800505specific to the bus type the node sits on.
506
507The unit name doesn't exist as a property per-se but is included in
508the device-tree structure. It is typically used to represent "path" in
509the device-tree. More details about the actual format of these will be
510below.
511
Grant Likelycf4e5c62011-01-31 00:12:26 -0700512The kernel generic code does not make any formal use of the
David Gibsonc125a182006-02-01 03:05:22 -0800513unit address (though some board support code may do) so the only real
514requirement here for the unit address is to ensure uniqueness of
515the node unit name at a given level of the tree. Nodes with no notion
516of address and no possible sibling of the same name (like /memory or
517/cpus) may omit the unit address in the context of this specification,
518or use the "@0" default unit address. The unit name is used to define
519a node "full path", which is the concatenation of all parent node
520unit names separated with "/".
521
522The root node doesn't have a defined name, and isn't required to have
523a name property either if you are using version 3 or earlier of the
524format. It also has no unit address (no @ symbol followed by a unit
525address). The root node unit name is thus an empty string. The full
526path to the root node is "/".
527
528Every node which actually represents an actual device (that is, a node
529which isn't only a virtual "container" for more nodes, like "/cpus"
Grant Likelycf4e5c62011-01-31 00:12:26 -0700530is) is also required to have a "compatible" property indicating the
531specific hardware and an optional list of devices it is fully
532backwards compatible with.
David Gibsonc125a182006-02-01 03:05:22 -0800533
534Finally, every node that can be referenced from a property in another
Grant Likelycf4e5c62011-01-31 00:12:26 -0700535node is required to have either a "phandle" or a "linux,phandle"
536property. Real Open Firmware implementations provide a unique
537"phandle" value for every node that the "prom_init()" trampoline code
538turns into "linux,phandle" properties. However, this is made optional
539if the flattened device tree is used directly. An example of a node
David Gibsonc125a182006-02-01 03:05:22 -0800540referencing another node via "phandle" is when laying out the
541interrupt tree which will be described in a further version of this
542document.
543
Grant Likelycf4e5c62011-01-31 00:12:26 -0700544The "phandle" property is a 32-bit value that uniquely
David Gibsonc125a182006-02-01 03:05:22 -0800545identifies a node. You are free to use whatever values or system of
546values, internal pointers, or whatever to generate these, the only
547requirement is that every node for which you provide that property has
548a unique value for it.
549
550Here is an example of a simple device-tree. In this example, an "o"
551designates a node followed by the node unit name. Properties are
552presented with their name followed by their content. "content"
553represents an ASCII string (zero terminated) value, while <content>
Roland Stigge36793622012-05-16 22:33:55 +0200554represents a 32-bit value, specified in decimal or hexadecimal (the
555latter prefixed 0x). The various nodes in this example will be
556discussed in a later chapter. At this point, it is only meant to give
557you a idea of what a device-tree looks like. I have purposefully kept
558the "name" and "linux,phandle" properties which aren't necessary in
559order to give you a better idea of what the tree looks like in
560practice.
David Gibsonc125a182006-02-01 03:05:22 -0800561
562 / o device-tree
563 |- name = "device-tree"
564 |- model = "MyBoardName"
565 |- compatible = "MyBoardFamilyName"
566 |- #address-cells = <2>
567 |- #size-cells = <2>
568 |- linux,phandle = <0>
569 |
570 o cpus
571 | | - name = "cpus"
572 | | - linux,phandle = <1>
573 | | - #address-cells = <1>
574 | | - #size-cells = <0>
575 | |
576 | o PowerPC,970@0
577 | |- name = "PowerPC,970"
578 | |- device_type = "cpu"
579 | |- reg = <0>
Roland Stigge36793622012-05-16 22:33:55 +0200580 | |- clock-frequency = <0x5f5e1000>
Timur Tabi32aed2a2007-02-14 15:29:07 -0600581 | |- 64-bit
David Gibsonc125a182006-02-01 03:05:22 -0800582 | |- linux,phandle = <2>
583 |
584 o memory@0
585 | |- name = "memory"
586 | |- device_type = "memory"
Roland Stigge36793622012-05-16 22:33:55 +0200587 | |- reg = <0x00000000 0x00000000 0x00000000 0x20000000>
David Gibsonc125a182006-02-01 03:05:22 -0800588 | |- linux,phandle = <3>
589 |
590 o chosen
591 |- name = "chosen"
592 |- bootargs = "root=/dev/sda2"
David Gibsonc125a182006-02-01 03:05:22 -0800593 |- linux,phandle = <4>
594
595This tree is almost a minimal tree. It pretty much contains the
596minimal set of required nodes and properties to boot a linux kernel;
Sylvestre Ledruf65e51d2011-04-04 15:04:46 -0700597that is, some basic model information at the root, the CPUs, and the
David Gibsonc125a182006-02-01 03:05:22 -0800598physical memory layout. It also includes misc information passed
599through /chosen, like in this example, the platform type (mandatory)
600and the kernel command line arguments (optional).
601
Timur Tabi32aed2a2007-02-14 15:29:07 -0600602The /cpus/PowerPC,970@0/64-bit property is an example of a
David Gibsonc125a182006-02-01 03:05:22 -0800603property without a value. All other properties have a value. The
604significance of the #address-cells and #size-cells properties will be
605explained in chapter IV which defines precisely the required nodes and
606properties and their content.
607
608
6093) Device tree "structure" block
610
611The structure of the device tree is a linearized tree structure. The
612"OF_DT_BEGIN_NODE" token starts a new node, and the "OF_DT_END_NODE"
613ends that node definition. Child nodes are simply defined before
614"OF_DT_END_NODE" (that is nodes within the node). A 'token' is a 32
615bit value. The tree has to be "finished" with a OF_DT_END token
616
617Here's the basic structure of a single node:
618
619 * token OF_DT_BEGIN_NODE (that is 0x00000001)
620 * for version 1 to 3, this is the node full path as a zero
621 terminated string, starting with "/". For version 16 and later,
622 this is the node unit name only (or an empty string for the
623 root node)
624 * [align gap to next 4 bytes boundary]
625 * for each property:
626 * token OF_DT_PROP (that is 0x00000003)
Domen Puncer5dd60162007-03-02 21:44:45 +1100627 * 32-bit value of property value size in bytes (or 0 if no
628 value)
629 * 32-bit value of offset in string block of property name
David Gibsonc125a182006-02-01 03:05:22 -0800630 * property value data if any
631 * [align gap to next 4 bytes boundary]
632 * [child nodes if any]
633 * token OF_DT_END_NODE (that is 0x00000002)
634
Domen Puncer5dd60162007-03-02 21:44:45 +1100635So the node content can be summarized as a start token, a full path,
Matt LaPlante53cb4722006-10-03 22:55:17 +0200636a list of properties, a list of child nodes, and an end token. Every
David Gibsonc125a182006-02-01 03:05:22 -0800637child node is a full node structure itself as defined above.
638
David Gibsoneff2ebd2007-06-28 15:56:26 +1000639NOTE: The above definition requires that all property definitions for
640a particular node MUST precede any subnode definitions for that node.
641Although the structure would not be ambiguous if properties and
642subnodes were intermingled, the kernel parser requires that the
643properties come first (up until at least 2.6.22). Any tools
644manipulating a flattened tree must take care to preserve this
645constraint.
646
Matt LaPlante53cb4722006-10-03 22:55:17 +02006474) Device tree "strings" block
David Gibsonc125a182006-02-01 03:05:22 -0800648
649In order to save space, property names, which are generally redundant,
650are stored separately in the "strings" block. This block is simply the
651whole bunch of zero terminated strings for all property names
652concatenated together. The device-tree property definitions in the
653structure block will contain offset values from the beginning of the
654strings block.
655
656
657III - Required content of the device tree
658=========================================
659
660WARNING: All "linux,*" properties defined in this document apply only
661to a flattened device-tree. If your platform uses a real
662implementation of Open Firmware or an implementation compatible with
663the Open Firmware client interface, those properties will be created
664by the trampoline code in the kernel's prom_init() file. For example,
665that's where you'll have to add code to detect your board model and
Matt LaPlantea2ffd272006-10-03 22:49:15 +0200666set the platform number. However, when using the flattened device-tree
David Gibsonc125a182006-02-01 03:05:22 -0800667entry point, there is no prom_init() pass, and thus you have to
668provide those properties yourself.
669
670
6711) Note about cells and address representation
672----------------------------------------------
673
674The general rule is documented in the various Open Firmware
Domen Puncer5dd60162007-03-02 21:44:45 +1100675documentations. If you choose to describe a bus with the device-tree
David Gibsonc125a182006-02-01 03:05:22 -0800676and there exist an OF bus binding, then you should follow the
677specification. However, the kernel does not require every single
678device or bus to be described by the device tree.
679
680In general, the format of an address for a device is defined by the
681parent bus type, based on the #address-cells and #size-cells
Mark A. Greer5b14e5f2008-01-04 02:40:47 +1100682properties. Note that the parent's parent definitions of #address-cells
Matt LaPlanted9195882008-07-25 19:45:33 -0700683and #size-cells are not inherited so every node with children must specify
Mark A. Greer5b14e5f2008-01-04 02:40:47 +1100684them. The kernel requires the root node to have those properties defining
685addresses format for devices directly mapped on the processor bus.
David Gibsonc125a182006-02-01 03:05:22 -0800686
687Those 2 properties define 'cells' for representing an address and a
Domen Puncer5dd60162007-03-02 21:44:45 +1100688size. A "cell" is a 32-bit number. For example, if both contain 2
David Gibsonc125a182006-02-01 03:05:22 -0800689like the example tree given above, then an address and a size are both
Domen Puncer5dd60162007-03-02 21:44:45 +1100690composed of 2 cells, and each is a 64-bit number (cells are
David Gibsonc125a182006-02-01 03:05:22 -0800691concatenated and expected to be in big endian format). Another example
692is the way Apple firmware defines them, with 2 cells for an address
693and one cell for a size. Most 32-bit implementations should define
694#address-cells and #size-cells to 1, which represents a 32-bit value.
695Some 32-bit processors allow for physical addresses greater than 32
696bits; these processors should define #address-cells as 2.
697
698"reg" properties are always a tuple of the type "address size" where
699the number of cells of address and size is specified by the bus
700#address-cells and #size-cells. When a bus supports various address
701spaces and other flags relative to a given address allocation (like
702prefetchable, etc...) those flags are usually added to the top level
703bits of the physical address. For example, a PCI physical address is
704made of 3 cells, the bottom two containing the actual address itself
705while the top cell contains address space indication, flags, and pci
706bus & device numbers.
707
Grant Likelycf4e5c62011-01-31 00:12:26 -0700708For buses that support dynamic allocation, it's the accepted practice
David Gibsonc125a182006-02-01 03:05:22 -0800709to then not provide the address in "reg" (keep it 0) though while
710providing a flag indicating the address is dynamically allocated, and
711then, to provide a separate "assigned-addresses" property that
712contains the fully allocated addresses. See the PCI OF bindings for
713details.
714
715In general, a simple bus with no address space bits and no dynamic
716allocation is preferred if it reflects your hardware, as the existing
717kernel address parsing functions will work out of the box. If you
718define a bus type with a more complex address format, including things
719like address space bits, you'll have to add a bus translator to the
720prom_parse.c file of the recent kernels for your bus type.
721
Stephen Neuendorffere1fd1862007-12-04 12:08:57 +1100722The "reg" property only defines addresses and sizes (if #size-cells is
723non-0) within a given bus. In order to translate addresses upward
Domen Puncer5dd60162007-03-02 21:44:45 +1100724(that is into parent bus addresses, and possibly into CPU physical
Grant Likelycf4e5c62011-01-31 00:12:26 -0700725addresses), all buses must contain a "ranges" property. If the
David Gibsonc125a182006-02-01 03:05:22 -0800726"ranges" property is missing at a given level, it's assumed that
Stephen Neuendorffere1fd1862007-12-04 12:08:57 +1100727translation isn't possible, i.e., the registers are not visible on the
728parent bus. The format of the "ranges" property for a bus is a list
729of:
David Gibsonc125a182006-02-01 03:05:22 -0800730
731 bus address, parent bus address, size
732
733"bus address" is in the format of the bus this bus node is defining,
734that is, for a PCI bridge, it would be a PCI address. Thus, (bus
735address, size) defines a range of addresses for child devices. "parent
736bus address" is in the format of the parent bus of this bus. For
737example, for a PCI host controller, that would be a CPU address. For a
738PCI<->ISA bridge, that would be a PCI address. It defines the base
739address in the parent bus where the beginning of that range is mapped.
740
Grant Likelycf4e5c62011-01-31 00:12:26 -0700741For new 64-bit board support, I recommend either the 2/2 format or
David Gibsonc125a182006-02-01 03:05:22 -0800742Apple's 2/1 format which is slightly more compact since sizes usually
Grant Likelycf4e5c62011-01-31 00:12:26 -0700743fit in a single 32-bit word. New 32-bit board support should use a
David Gibsonc125a182006-02-01 03:05:22 -08007441/1 format, unless the processor supports physical addresses greater
745than 32-bits, in which case a 2/1 format is recommended.
746
Stephen Neuendorffere1fd1862007-12-04 12:08:57 +1100747Alternatively, the "ranges" property may be empty, indicating that the
748registers are visible on the parent bus using an identity mapping
749translation. In other words, the parent bus address space is the same
750as the child bus address space.
David Gibsonc125a182006-02-01 03:05:22 -0800751
7522) Note about "compatible" properties
753-------------------------------------
754
755These properties are optional, but recommended in devices and the root
756node. The format of a "compatible" property is a list of concatenated
757zero terminated strings. They allow a device to express its
758compatibility with a family of similar devices, in some cases,
759allowing a single driver to match against several devices regardless
760of their actual names.
761
7623) Note about "name" properties
763-------------------------------
764
765While earlier users of Open Firmware like OldWorld macintoshes tended
766to use the actual device name for the "name" property, it's nowadays
767considered a good practice to use a name that is closer to the device
Grant Likelycf4e5c62011-01-31 00:12:26 -0700768class (often equal to device_type). For example, nowadays, Ethernet
David Gibsonc125a182006-02-01 03:05:22 -0800769controllers are named "ethernet", an additional "model" property
770defining precisely the chip type/model, and "compatible" property
771defining the family in case a single driver can driver more than one
772of these chips. However, the kernel doesn't generally put any
773restriction on the "name" property; it is simply considered good
774practice to follow the standard and its evolutions as closely as
775possible.
776
777Note also that the new format version 16 makes the "name" property
778optional. If it's absent for a node, then the node's unit name is then
779used to reconstruct the name. That is, the part of the unit name
780before the "@" sign is used (or the entire unit name if no "@" sign
781is present).
782
7834) Note about node and property names and character set
784-------------------------------------------------------
785
Grant Likelycf4e5c62011-01-31 00:12:26 -0700786While Open Firmware provides more flexible usage of 8859-1, this
David Gibsonc125a182006-02-01 03:05:22 -0800787specification enforces more strict rules. Nodes and properties should
788be comprised only of ASCII characters 'a' to 'z', '0' to
789'9', ',', '.', '_', '+', '#', '?', and '-'. Node names additionally
790allow uppercase characters 'A' to 'Z' (property names should be
791lowercase. The fact that vendors like Apple don't respect this rule is
792irrelevant here). Additionally, node and property names should always
793begin with a character in the range 'a' to 'z' (or 'A' to 'Z' for node
794names).
795
796The maximum number of characters for both nodes and property names
797is 31. In the case of node names, this is only the leftmost part of
798a unit name (the pure "name" property), it doesn't include the unit
799address which can extend beyond that limit.
800
801
8025) Required nodes and properties
803--------------------------------
804 These are all that are currently required. However, it is strongly
805 recommended that you expose PCI host bridges as documented in the
Grant Likelycf4e5c62011-01-31 00:12:26 -0700806 PCI binding to Open Firmware, and your interrupt tree as documented
David Gibsonc125a182006-02-01 03:05:22 -0800807 in OF interrupt tree specification.
808
809 a) The root node
810
811 The root node requires some properties to be present:
812
813 - model : this is your board name/model
814 - #address-cells : address representation for "root" devices
815 - #size-cells: the size representation for "root" devices
David Gibsonc125a182006-02-01 03:05:22 -0800816 - compatible : the board "family" generally finds its way here,
817 for example, if you have 2 board models with a similar layout,
818 that typically get driven by the same platform code in the
Grant Likelycf4e5c62011-01-31 00:12:26 -0700819 kernel, you would specify the exact board model in the
820 compatible property followed by an entry that represents the SoC
821 model.
David Gibsonc125a182006-02-01 03:05:22 -0800822
823 The root node is also generally where you add additional properties
824 specific to your board like the serial number if any, that sort of
Matt LaPlante6c28f2c2006-10-03 22:46:31 +0200825 thing. It is recommended that if you add any "custom" property whose
David Gibsonc125a182006-02-01 03:05:22 -0800826 name may clash with standard defined ones, you prefix them with your
827 vendor name and a comma.
828
829 b) The /cpus node
830
831 This node is the parent of all individual CPU nodes. It doesn't
832 have any specific requirements, though it's generally good practice
833 to have at least:
834
835 #address-cells = <00000001>
836 #size-cells = <00000000>
837
838 This defines that the "address" for a CPU is a single cell, and has
839 no meaningful size. This is not necessary but the kernel will assume
840 that format when reading the "reg" properties of a CPU node, see
841 below
842
843 c) The /cpus/* nodes
844
845 So under /cpus, you are supposed to create a node for every CPU on
846 the machine. There is no specific restriction on the name of the
Grant Likelycf4e5c62011-01-31 00:12:26 -0700847 CPU, though it's common to call it <architecture>,<core>. For
David Gibsonc125a182006-02-01 03:05:22 -0800848 example, Apple uses PowerPC,G5 while IBM uses PowerPC,970FX.
Grant Likelycf4e5c62011-01-31 00:12:26 -0700849 However, the Generic Names convention suggests that it would be
850 better to simply use 'cpu' for each cpu node and use the compatible
851 property to identify the specific cpu core.
David Gibsonc125a182006-02-01 03:05:22 -0800852
853 Required properties:
854
855 - device_type : has to be "cpu"
Domen Puncer5dd60162007-03-02 21:44:45 +1100856 - reg : This is the physical CPU number, it's a single 32-bit cell
David Gibsonc125a182006-02-01 03:05:22 -0800857 and is also used as-is as the unit number for constructing the
858 unit name in the full path. For example, with 2 CPUs, you would
859 have the full path:
860 /cpus/PowerPC,970FX@0
861 /cpus/PowerPC,970FX@1
862 (unit addresses do not require leading zeroes)
Benjamin Herrenschmidt20474ab2007-10-28 08:49:28 +1100863 - d-cache-block-size : one cell, L1 data cache block size in bytes (*)
864 - i-cache-block-size : one cell, L1 instruction cache block size in
David Gibsonc125a182006-02-01 03:05:22 -0800865 bytes
866 - d-cache-size : one cell, size of L1 data cache in bytes
867 - i-cache-size : one cell, size of L1 instruction cache in bytes
David Gibsonc125a182006-02-01 03:05:22 -0800868
Benjamin Herrenschmidt20474ab2007-10-28 08:49:28 +1100869(*) The cache "block" size is the size on which the cache management
870instructions operate. Historically, this document used the cache
871"line" size here which is incorrect. The kernel will prefer the cache
872block size and will fallback to cache line size for backward
873compatibility.
874
David Gibsonc125a182006-02-01 03:05:22 -0800875 Recommended properties:
876
877 - timebase-frequency : a cell indicating the frequency of the
878 timebase in Hz. This is not directly used by the generic code,
879 but you are welcome to copy/paste the pSeries code for setting
880 the kernel timebase/decrementer calibration based on this
881 value.
882 - clock-frequency : a cell indicating the CPU core clock frequency
Domen Puncer5dd60162007-03-02 21:44:45 +1100883 in Hz. A new property will be defined for 64-bit values, but if
David Gibsonc125a182006-02-01 03:05:22 -0800884 your frequency is < 4Ghz, one cell is enough. Here as well as
885 for the above, the common code doesn't use that property, but
886 you are welcome to re-use the pSeries or Maple one. A future
887 kernel version might provide a common function for this.
Benjamin Herrenschmidt20474ab2007-10-28 08:49:28 +1100888 - d-cache-line-size : one cell, L1 data cache line size in bytes
889 if different from the block size
890 - i-cache-line-size : one cell, L1 instruction cache line size in
891 bytes if different from the block size
David Gibsonc125a182006-02-01 03:05:22 -0800892
893 You are welcome to add any property you find relevant to your board,
894 like some information about the mechanism used to soft-reset the
895 CPUs. For example, Apple puts the GPIO number for CPU soft reset
896 lines in there as a "soft-reset" property since they start secondary
897 CPUs by soft-resetting them.
898
899
900 d) the /memory node(s)
901
902 To define the physical memory layout of your board, you should
903 create one or more memory node(s). You can either create a single
904 node with all memory ranges in its reg property, or you can create
905 several nodes, as you wish. The unit address (@ part) used for the
906 full path is the address of the first range of memory defined by a
907 given node. If you use a single memory node, this will typically be
908 @0.
909
910 Required properties:
911
912 - device_type : has to be "memory"
913 - reg : This property contains all the physical memory ranges of
914 your board. It's a list of addresses/sizes concatenated
915 together, with the number of cells of each defined by the
916 #address-cells and #size-cells of the root node. For example,
Matt LaPlante6c28f2c2006-10-03 22:46:31 +0200917 with both of these properties being 2 like in the example given
David Gibsonc125a182006-02-01 03:05:22 -0800918 earlier, a 970 based machine with 6Gb of RAM could typically
919 have a "reg" property here that looks like:
920
921 00000000 00000000 00000000 80000000
922 00000001 00000000 00000001 00000000
923
924 That is a range starting at 0 of 0x80000000 bytes and a range
925 starting at 0x100000000 and of 0x100000000 bytes. You can see
926 that there is no memory covering the IO hole between 2Gb and
927 4Gb. Some vendors prefer splitting those ranges into smaller
928 segments, but the kernel doesn't care.
929
930 e) The /chosen node
931
Grant Likelycf4e5c62011-01-31 00:12:26 -0700932 This node is a bit "special". Normally, that's where Open Firmware
David Gibsonc125a182006-02-01 03:05:22 -0800933 puts some variable environment information, like the arguments, or
Stuart Yoderd1bff9e2007-02-19 11:25:05 -0600934 the default input/output devices.
David Gibsonc125a182006-02-01 03:05:22 -0800935
936 This specification makes a few of these mandatory, but also defines
937 some linux-specific properties that would be normally constructed by
938 the prom_init() trampoline when booting with an OF client interface,
939 but that you have to provide yourself when using the flattened format.
940
David Gibsonc125a182006-02-01 03:05:22 -0800941 Recommended properties:
942
943 - bootargs : This zero-terminated string is passed as the kernel
944 command line
945 - linux,stdout-path : This is the full path to your standard
946 console device if any. Typically, if you have serial devices on
947 your board, you may want to put the full path to the one set as
948 the default console in the firmware here, for the kernel to pick
Grant Likelycf4e5c62011-01-31 00:12:26 -0700949 it up as its own default console.
David Gibsonc125a182006-02-01 03:05:22 -0800950
951 Note that u-boot creates and fills in the chosen node for platforms
952 that use it.
953
Stuart Yoderd1bff9e2007-02-19 11:25:05 -0600954 (Note: a practice that is now obsolete was to include a property
955 under /chosen called interrupt-controller which had a phandle value
956 that pointed to the main interrupt controller)
957
David Gibsonc125a182006-02-01 03:05:22 -0800958 f) the /soc<SOCname> node
959
Grant Likelycf4e5c62011-01-31 00:12:26 -0700960 This node is used to represent a system-on-a-chip (SoC) and must be
961 present if the processor is a SoC. The top-level soc node contains
962 information that is global to all devices on the SoC. The node name
963 should contain a unit address for the SoC, which is the base address
964 of the memory-mapped register set for the SoC. The name of an SoC
David Gibsonc125a182006-02-01 03:05:22 -0800965 node should start with "soc", and the remainder of the name should
966 represent the part number for the soc. For example, the MPC8540's
967 soc node would be called "soc8540".
968
969 Required properties:
970
David Gibsonc125a182006-02-01 03:05:22 -0800971 - ranges : Should be defined as specified in 1) to describe the
Grant Likelycf4e5c62011-01-31 00:12:26 -0700972 translation of SoC addresses for memory mapped SoC registers.
973 - bus-frequency: Contains the bus frequency for the SoC node.
Becky Bruce7d4b95a2006-02-06 14:26:31 -0600974 Typically, the value of this field is filled in by the boot
Stefan Roeseefcc2da2009-04-16 15:11:54 -0600975 loader.
Grant Likelycf4e5c62011-01-31 00:12:26 -0700976 - compatible : Exact model of the SoC
Becky Bruce7d4b95a2006-02-06 14:26:31 -0600977
David Gibsonc125a182006-02-01 03:05:22 -0800978
979 Recommended properties:
980
981 - reg : This property defines the address and size of the
982 memory-mapped registers that are used for the SOC node itself.
983 It does not include the child device registers - these will be
984 defined inside each child node. The address specified in the
985 "reg" property should match the unit address of the SOC node.
986 - #address-cells : Address representation for "soc" devices. The
987 format of this field may vary depending on whether or not the
988 device registers are memory mapped. For memory mapped
989 registers, this field represents the number of cells needed to
990 represent the address of the registers. For SOCs that do not
991 use MMIO, a special address format should be defined that
992 contains enough cells to represent the required information.
993 See 1) above for more details on defining #address-cells.
994 - #size-cells : Size representation for "soc" devices
995 - #interrupt-cells : Defines the width of cells used to represent
996 interrupts. Typically this value is <2>, which includes a
997 32-bit number that represents the interrupt number, and a
998 32-bit number that represents the interrupt sense and level.
999 This field is only needed if the SOC contains an interrupt
1000 controller.
1001
1002 The SOC node may contain child nodes for each SOC device that the
1003 platform uses. Nodes should not be created for devices which exist
1004 on the SOC but are not used by a particular platform. See chapter VI
Domen Puncer5dd60162007-03-02 21:44:45 +11001005 for more information on how to specify devices that are part of a SOC.
David Gibsonc125a182006-02-01 03:05:22 -08001006
1007 Example SOC node for the MPC8540:
1008
1009 soc8540@e0000000 {
1010 #address-cells = <1>;
1011 #size-cells = <1>;
1012 #interrupt-cells = <2>;
1013 device_type = "soc";
Roland Stigge36793622012-05-16 22:33:55 +02001014 ranges = <0x00000000 0xe0000000 0x00100000>
1015 reg = <0xe0000000 0x00003000>;
Becky Bruce7d4b95a2006-02-06 14:26:31 -06001016 bus-frequency = <0>;
David Gibsonc125a182006-02-01 03:05:22 -08001017 }
1018
1019
1020
1021IV - "dtc", the device tree compiler
1022====================================
1023
1024
1025dtc source code can be found at
Justin P. Mattock0ea6e612010-07-23 20:51:24 -07001026<http://git.jdl.com/gitweb/?p=dtc.git>
David Gibsonc125a182006-02-01 03:05:22 -08001027
1028WARNING: This version is still in early development stage; the
1029resulting device-tree "blobs" have not yet been validated with the
Lennert Buytenhek475fc7c2010-09-21 23:22:40 +00001030kernel. The current generated block lacks a useful reserve map (it will
David Gibsonc125a182006-02-01 03:05:22 -08001031be fixed to generate an empty one, it's up to the bootloader to fill
1032it up) among others. The error handling needs work, bugs are lurking,
1033etc...
1034
1035dtc basically takes a device-tree in a given format and outputs a
1036device-tree in another format. The currently supported formats are:
1037
1038 Input formats:
1039 -------------
1040
1041 - "dtb": "blob" format, that is a flattened device-tree block
1042 with
1043 header all in a binary blob.
1044 - "dts": "source" format. This is a text file containing a
1045 "source" for a device-tree. The format is defined later in this
1046 chapter.
1047 - "fs" format. This is a representation equivalent to the
1048 output of /proc/device-tree, that is nodes are directories and
1049 properties are files
1050
1051 Output formats:
1052 ---------------
1053
1054 - "dtb": "blob" format
1055 - "dts": "source" format
1056 - "asm": assembly language file. This is a file that can be
1057 sourced by gas to generate a device-tree "blob". That file can
1058 then simply be added to your Makefile. Additionally, the
Matt LaPlante6c28f2c2006-10-03 22:46:31 +02001059 assembly file exports some symbols that can be used.
David Gibsonc125a182006-02-01 03:05:22 -08001060
1061
1062The syntax of the dtc tool is
1063
1064 dtc [-I <input-format>] [-O <output-format>]
1065 [-o output-filename] [-V output_version] input_filename
1066
1067
Domen Puncer5dd60162007-03-02 21:44:45 +11001068The "output_version" defines what version of the "blob" format will be
David Gibsonc125a182006-02-01 03:05:22 -08001069generated. Supported versions are 1,2,3 and 16. The default is
1070currently version 3 but that may change in the future to version 16.
1071
1072Additionally, dtc performs various sanity checks on the tree, like the
Matt LaPlante6c28f2c2006-10-03 22:46:31 +02001073uniqueness of linux, phandle properties, validity of strings, etc...
David Gibsonc125a182006-02-01 03:05:22 -08001074
1075The format of the .dts "source" file is "C" like, supports C and C++
Matt LaPlante6c28f2c2006-10-03 22:46:31 +02001076style comments.
David Gibsonc125a182006-02-01 03:05:22 -08001077
1078/ {
1079}
1080
1081The above is the "device-tree" definition. It's the only statement
1082supported currently at the toplevel.
1083
1084/ {
1085 property1 = "string_value"; /* define a property containing a 0
1086 * terminated string
1087 */
1088
Roland Stigge36793622012-05-16 22:33:55 +02001089 property2 = <0x1234abcd>; /* define a property containing a
Domen Puncer5dd60162007-03-02 21:44:45 +11001090 * numerical 32-bit value (hexadecimal)
David Gibsonc125a182006-02-01 03:05:22 -08001091 */
1092
Roland Stigge36793622012-05-16 22:33:55 +02001093 property3 = <0x12345678 0x12345678 0xdeadbeef>;
David Gibsonc125a182006-02-01 03:05:22 -08001094 /* define a property containing 3
Domen Puncer5dd60162007-03-02 21:44:45 +11001095 * numerical 32-bit values (cells) in
David Gibsonc125a182006-02-01 03:05:22 -08001096 * hexadecimal
1097 */
Roland Stigge36793622012-05-16 22:33:55 +02001098 property4 = [0x0a 0x0b 0x0c 0x0d 0xde 0xea 0xad 0xbe 0xef];
David Gibsonc125a182006-02-01 03:05:22 -08001099 /* define a property whose content is
1100 * an arbitrary array of bytes
1101 */
1102
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04001103 childnode@address { /* define a child node named "childnode"
David Gibsonc125a182006-02-01 03:05:22 -08001104 * whose unit name is "childnode at
1105 * address"
1106 */
1107
1108 childprop = "hello\n"; /* define a property "childprop" of
1109 * childnode (in this case, a string)
1110 */
1111 };
1112};
1113
1114Nodes can contain other nodes etc... thus defining the hierarchical
1115structure of the tree.
1116
1117Strings support common escape sequences from C: "\n", "\t", "\r",
1118"\(octal value)", "\x(hex value)".
1119
1120It is also suggested that you pipe your source file through cpp (gcc
1121preprocessor) so you can use #include's, #define for constants, etc...
1122
1123Finally, various options are planned but not yet implemented, like
1124automatic generation of phandles, labels (exported to the asm file so
1125you can point to a property content and change it easily from whatever
1126you link the device-tree with), label or path instead of numeric value
1127in some cells to "point" to a node (replaced by a phandle at compile
1128time), export of reserve map address to the asm file, ability to
1129specify reserve map content at compile time, etc...
1130
1131We may provide a .h include file with common definitions of that
1132proves useful for some properties (like building PCI properties or
1133interrupt maps) though it may be better to add a notion of struct
1134definitions to the compiler...
1135
1136
1137V - Recommendations for a bootloader
1138====================================
1139
1140
1141Here are some various ideas/recommendations that have been proposed
1142while all this has been defined and implemented.
1143
1144 - The bootloader may want to be able to use the device-tree itself
1145 and may want to manipulate it (to add/edit some properties,
1146 like physical memory size or kernel arguments). At this point, 2
1147 choices can be made. Either the bootloader works directly on the
1148 flattened format, or the bootloader has its own internal tree
1149 representation with pointers (similar to the kernel one) and
1150 re-flattens the tree when booting the kernel. The former is a bit
1151 more difficult to edit/modify, the later requires probably a bit
1152 more code to handle the tree structure. Note that the structure
1153 format has been designed so it's relatively easy to "insert"
1154 properties or nodes or delete them by just memmoving things
1155 around. It contains no internal offsets or pointers for this
1156 purpose.
1157
Matt LaPlanted6bc8ac2006-10-03 22:54:15 +02001158 - An example of code for iterating nodes & retrieving properties
David Gibsonc125a182006-02-01 03:05:22 -08001159 directly from the flattened tree format can be found in the kernel
Grant Likelycf4e5c62011-01-31 00:12:26 -07001160 file drivers/of/fdt.c. Look at the of_scan_flat_dt() function,
Matt LaPlanted6bc8ac2006-10-03 22:54:15 +02001161 its usage in early_init_devtree(), and the corresponding various
David Gibsonc125a182006-02-01 03:05:22 -08001162 early_init_dt_scan_*() callbacks. That code can be re-used in a
1163 GPL bootloader, and as the author of that code, I would be happy
Domen Puncer5dd60162007-03-02 21:44:45 +11001164 to discuss possible free licensing to any vendor who wishes to
David Gibsonc125a182006-02-01 03:05:22 -08001165 integrate all or part of this code into a non-GPL bootloader.
Grant Likelycf4e5c62011-01-31 00:12:26 -07001166 (reference needed; who is 'I' here? ---gcl Jan 31, 2011)
David Gibsonc125a182006-02-01 03:05:22 -08001167
1168
1169
1170VI - System-on-a-chip devices and nodes
1171=======================================
1172
1173Many companies are now starting to develop system-on-a-chip
Domen Puncer5dd60162007-03-02 21:44:45 +11001174processors, where the processor core (CPU) and many peripheral devices
David Gibsonc125a182006-02-01 03:05:22 -08001175exist on a single piece of silicon. For these SOCs, an SOC node
1176should be used that defines child nodes for the devices that make
1177up the SOC. While platforms are not required to use this model in
1178order to boot the kernel, it is highly encouraged that all SOC
1179implementations define as complete a flat-device-tree as possible to
1180describe the devices on the SOC. This will allow for the
1181genericization of much of the kernel code.
1182
1183
11841) Defining child nodes of an SOC
1185---------------------------------
1186
1187Each device that is part of an SOC may have its own node entry inside
1188the SOC node. For each device that is included in the SOC, the unit
1189address property represents the address offset for this device's
1190memory-mapped registers in the parent's address space. The parent's
1191address space is defined by the "ranges" property in the top-level soc
1192node. The "reg" property for each node that exists directly under the
1193SOC node should contain the address mapping from the child address space
1194to the parent SOC address space and the size of the device's
1195memory-mapped register file.
1196
1197For many devices that may exist inside an SOC, there are predefined
1198specifications for the format of the device tree node. All SOC child
1199nodes should follow these specifications, except where noted in this
1200document.
1201
1202See appendix A for an example partial SOC node definition for the
1203MPC8540.
1204
1205
Stuart Yoder27565902007-03-02 13:42:33 -060012062) Representing devices without a current OF specification
David Gibsonc125a182006-02-01 03:05:22 -08001207----------------------------------------------------------
1208
Grant Likelycf4e5c62011-01-31 00:12:26 -07001209Currently, there are many devices on SoCs that do not have a standard
1210representation defined as part of the Open Firmware specifications,
1211mainly because the boards that contain these SoCs are not currently
1212booted using Open Firmware. Binding documentation for new devices
1213should be added to the Documentation/devicetree/bindings directory.
1214That directory will expand as device tree support is added to more and
1215more SoCs.
1216
David Gibsonc125a182006-02-01 03:05:22 -08001217
Kumar Galab053dc52009-06-19 08:31:05 -05001218VII - Specifying interrupt information for devices
Stuart Yoder27565902007-03-02 13:42:33 -06001219===================================================
1220
Grant Likelycf4e5c62011-01-31 00:12:26 -07001221The device tree represents the buses and devices of a hardware
Stuart Yoder27565902007-03-02 13:42:33 -06001222system in a form similar to the physical bus topology of the
1223hardware.
1224
1225In addition, a logical 'interrupt tree' exists which represents the
1226hierarchy and routing of interrupts in the hardware.
1227
1228The interrupt tree model is fully described in the
1229document "Open Firmware Recommended Practice: Interrupt
1230Mapping Version 0.9". The document is available at:
Christian Kujau242260f2013-02-21 16:43:05 -08001231<http://www.openfirmware.org/ofwg/practice/>
Stuart Yoder27565902007-03-02 13:42:33 -06001232
12331) interrupts property
1234----------------------
1235
1236Devices that generate interrupts to a single interrupt controller
1237should use the conventional OF representation described in the
1238OF interrupt mapping documentation.
1239
1240Each device which generates interrupts must have an 'interrupt'
1241property. The interrupt property value is an arbitrary number of
1242of 'interrupt specifier' values which describe the interrupt or
1243interrupts for the device.
1244
1245The encoding of an interrupt specifier is determined by the
1246interrupt domain in which the device is located in the
1247interrupt tree. The root of an interrupt domain specifies in
1248its #interrupt-cells property the number of 32-bit cells
1249required to encode an interrupt specifier. See the OF interrupt
1250mapping documentation for a detailed description of domains.
1251
1252For example, the binding for the OpenPIC interrupt controller
1253specifies an #interrupt-cells value of 2 to encode the interrupt
1254number and level/sense information. All interrupt children in an
1255OpenPIC interrupt domain use 2 cells per interrupt in their interrupts
1256property.
1257
1258The PCI bus binding specifies a #interrupt-cell value of 1 to encode
1259which interrupt pin (INTA,INTB,INTC,INTD) is used.
1260
12612) interrupt-parent property
1262----------------------------
1263
1264The interrupt-parent property is specified to define an explicit
1265link between a device node and its interrupt parent in
1266the interrupt tree. The value of interrupt-parent is the
1267phandle of the parent node.
1268
Francis Galieguea33f3222010-04-23 00:08:02 +02001269If the interrupt-parent property is not defined for a node, its
Stuart Yoder27565902007-03-02 13:42:33 -06001270interrupt parent is assumed to be an ancestor in the node's
1271_device tree_ hierarchy.
1272
12733) OpenPIC Interrupt Controllers
1274--------------------------------
1275
1276OpenPIC interrupt controllers require 2 cells to encode
1277interrupt information. The first cell defines the interrupt
1278number. The second cell defines the sense and level
1279information.
1280
1281Sense and level information should be encoded as follows:
1282
1283 0 = low to high edge sensitive type enabled
1284 1 = active low level sensitive type enabled
1285 2 = active high level sensitive type enabled
1286 3 = high to low edge sensitive type enabled
1287
12884) ISA Interrupt Controllers
1289----------------------------
1290
1291ISA PIC interrupt controllers require 2 cells to encode
1292interrupt information. The first cell defines the interrupt
1293number. The second cell defines the sense and level
1294information.
1295
1296ISA PIC interrupt controllers should adhere to the ISA PIC
1297encodings listed below:
1298
1299 0 = active low level sensitive type enabled
1300 1 = active high level sensitive type enabled
1301 2 = high to low edge sensitive type enabled
1302 3 = low to high edge sensitive type enabled
1303
Kumar Galab053dc52009-06-19 08:31:05 -05001304VIII - Specifying Device Power Management Information (sleep property)
Scott Wood2dff4172008-07-11 17:31:15 -05001305===================================================================
1306
1307Devices on SOCs often have mechanisms for placing devices into low-power
1308states that are decoupled from the devices' own register blocks. Sometimes,
1309this information is more complicated than a cell-index property can
1310reasonably describe. Thus, each device controlled in such a manner
1311may contain a "sleep" property which describes these connections.
1312
1313The sleep property consists of one or more sleep resources, each of
1314which consists of a phandle to a sleep controller, followed by a
1315controller-specific sleep specifier of zero or more cells.
1316
1317The semantics of what type of low power modes are possible are defined
1318by the sleep controller. Some examples of the types of low power modes
1319that may be supported are:
1320
1321 - Dynamic: The device may be disabled or enabled at any time.
1322 - System Suspend: The device may request to be disabled or remain
1323 awake during system suspend, but will not be disabled until then.
1324 - Permanent: The device is disabled permanently (until the next hard
1325 reset).
1326
1327Some devices may share a clock domain with each other, such that they should
1328only be suspended when none of the devices are in use. Where reasonable,
1329such nodes should be placed on a virtual bus, where the bus has the sleep
1330property. If the clock domain is shared among devices that cannot be
1331reasonably grouped in this manner, then create a virtual sleep controller
1332(similar to an interrupt nexus, except that defining a standardized
1333sleep-map should wait until its necessity is demonstrated).
1334
David Gibsonc125a182006-02-01 03:05:22 -08001335Appendix A - Sample SOC node for MPC8540
1336========================================
1337
Scott Wood7e720632008-06-25 12:07:39 -05001338 soc@e0000000 {
David Gibsonc125a182006-02-01 03:05:22 -08001339 #address-cells = <1>;
1340 #size-cells = <1>;
Scott Wood7e720632008-06-25 12:07:39 -05001341 compatible = "fsl,mpc8540-ccsr", "simple-bus";
David Gibsonc125a182006-02-01 03:05:22 -08001342 device_type = "soc";
Scott Wood7e720632008-06-25 12:07:39 -05001343 ranges = <0x00000000 0xe0000000 0x00100000>
Becky Bruce7d4b95a2006-02-06 14:26:31 -06001344 bus-frequency = <0>;
Scott Wood7e720632008-06-25 12:07:39 -05001345 interrupt-parent = <&pic>;
David Gibsonc125a182006-02-01 03:05:22 -08001346
David Gibsonc125a182006-02-01 03:05:22 -08001347 ethernet@24000 {
Scott Wood2dff4172008-07-11 17:31:15 -05001348 #address-cells = <1>;
1349 #size-cells = <1>;
David Gibsonc125a182006-02-01 03:05:22 -08001350 device_type = "network";
1351 model = "TSEC";
Scott Wood2dff4172008-07-11 17:31:15 -05001352 compatible = "gianfar", "simple-bus";
Scott Wood7e720632008-06-25 12:07:39 -05001353 reg = <0x24000 0x1000>;
Roland Stigge36793622012-05-16 22:33:55 +02001354 local-mac-address = [ 0x00 0xE0 0x0C 0x00 0x73 0x00 ];
1355 interrupts = <0x29 2 0x30 2 0x34 2>;
Scott Wood7e720632008-06-25 12:07:39 -05001356 phy-handle = <&phy0>;
Roland Stigge36793622012-05-16 22:33:55 +02001357 sleep = <&pmc 0x00000080>;
Scott Wood2dff4172008-07-11 17:31:15 -05001358 ranges;
1359
1360 mdio@24520 {
Scott Wood7e720632008-06-25 12:07:39 -05001361 reg = <0x24520 0x20>;
Scott Wood2dff4172008-07-11 17:31:15 -05001362 compatible = "fsl,gianfar-mdio";
1363
Scott Wood7e720632008-06-25 12:07:39 -05001364 phy0: ethernet-phy@0 {
1365 interrupts = <5 1>;
Scott Wood2dff4172008-07-11 17:31:15 -05001366 reg = <0>;
1367 device_type = "ethernet-phy";
1368 };
1369
Scott Wood7e720632008-06-25 12:07:39 -05001370 phy1: ethernet-phy@1 {
1371 interrupts = <5 1>;
Scott Wood2dff4172008-07-11 17:31:15 -05001372 reg = <1>;
1373 device_type = "ethernet-phy";
1374 };
1375
Scott Wood7e720632008-06-25 12:07:39 -05001376 phy3: ethernet-phy@3 {
1377 interrupts = <7 1>;
Scott Wood2dff4172008-07-11 17:31:15 -05001378 reg = <3>;
1379 device_type = "ethernet-phy";
1380 };
1381 };
David Gibsonc125a182006-02-01 03:05:22 -08001382 };
1383
1384 ethernet@25000 {
David Gibsonc125a182006-02-01 03:05:22 -08001385 device_type = "network";
1386 model = "TSEC";
1387 compatible = "gianfar";
Scott Wood7e720632008-06-25 12:07:39 -05001388 reg = <0x25000 0x1000>;
Roland Stigge36793622012-05-16 22:33:55 +02001389 local-mac-address = [ 0x00 0xE0 0x0C 0x00 0x73 0x01 ];
1390 interrupts = <0x13 2 0x14 2 0x18 2>;
Scott Wood7e720632008-06-25 12:07:39 -05001391 phy-handle = <&phy1>;
Roland Stigge36793622012-05-16 22:33:55 +02001392 sleep = <&pmc 0x00000040>;
David Gibsonc125a182006-02-01 03:05:22 -08001393 };
1394
1395 ethernet@26000 {
David Gibsonc125a182006-02-01 03:05:22 -08001396 device_type = "network";
1397 model = "FEC";
1398 compatible = "gianfar";
Scott Wood7e720632008-06-25 12:07:39 -05001399 reg = <0x26000 0x1000>;
Roland Stigge36793622012-05-16 22:33:55 +02001400 local-mac-address = [ 0x00 0xE0 0x0C 0x00 0x73 0x02 ];
1401 interrupts = <0x41 2>;
Scott Wood7e720632008-06-25 12:07:39 -05001402 phy-handle = <&phy3>;
Roland Stigge36793622012-05-16 22:33:55 +02001403 sleep = <&pmc 0x00000020>;
David Gibsonc125a182006-02-01 03:05:22 -08001404 };
1405
1406 serial@4500 {
Scott Wood2dff4172008-07-11 17:31:15 -05001407 #address-cells = <1>;
1408 #size-cells = <1>;
1409 compatible = "fsl,mpc8540-duart", "simple-bus";
Roland Stigge36793622012-05-16 22:33:55 +02001410 sleep = <&pmc 0x00000002>;
Scott Wood2dff4172008-07-11 17:31:15 -05001411 ranges;
1412
1413 serial@4500 {
1414 device_type = "serial";
1415 compatible = "ns16550";
Scott Wood7e720632008-06-25 12:07:39 -05001416 reg = <0x4500 0x100>;
Scott Wood2dff4172008-07-11 17:31:15 -05001417 clock-frequency = <0>;
Roland Stigge36793622012-05-16 22:33:55 +02001418 interrupts = <0x42 2>;
Scott Wood2dff4172008-07-11 17:31:15 -05001419 };
1420
1421 serial@4600 {
1422 device_type = "serial";
1423 compatible = "ns16550";
Scott Wood7e720632008-06-25 12:07:39 -05001424 reg = <0x4600 0x100>;
Scott Wood2dff4172008-07-11 17:31:15 -05001425 clock-frequency = <0>;
Roland Stigge36793622012-05-16 22:33:55 +02001426 interrupts = <0x42 2>;
Scott Wood2dff4172008-07-11 17:31:15 -05001427 };
David Gibsonc125a182006-02-01 03:05:22 -08001428 };
1429
Scott Wood7e720632008-06-25 12:07:39 -05001430 pic: pic@40000 {
David Gibsonc125a182006-02-01 03:05:22 -08001431 interrupt-controller;
1432 #address-cells = <0>;
Scott Wood7e720632008-06-25 12:07:39 -05001433 #interrupt-cells = <2>;
1434 reg = <0x40000 0x40000>;
David Gibsonc125a182006-02-01 03:05:22 -08001435 compatible = "chrp,open-pic";
1436 device_type = "open-pic";
David Gibsonc125a182006-02-01 03:05:22 -08001437 };
1438
1439 i2c@3000 {
Roland Stigge36793622012-05-16 22:33:55 +02001440 interrupts = <0x43 2>;
Scott Wood7e720632008-06-25 12:07:39 -05001441 reg = <0x3000 0x100>;
David Gibsonc125a182006-02-01 03:05:22 -08001442 compatible = "fsl-i2c";
1443 dfsrr;
Roland Stigge36793622012-05-16 22:33:55 +02001444 sleep = <&pmc 0x00000004>;
David Gibsonc125a182006-02-01 03:05:22 -08001445 };
1446
Scott Wood2dff4172008-07-11 17:31:15 -05001447 pmc: power@e0070 {
1448 compatible = "fsl,mpc8540-pmc", "fsl,mpc8548-pmc";
Scott Wood7e720632008-06-25 12:07:39 -05001449 reg = <0xe0070 0x20>;
Scott Wood2dff4172008-07-11 17:31:15 -05001450 };
David Gibsonc125a182006-02-01 03:05:22 -08001451 };