/Documentation/powerpc/booting-without-of.txt
Plain Text | 1474 lines | 1197 code | 277 blank | 0 comment | 0 complexity | 1b7eb950c8a73884eab80fa7d74f2ac0 MD5 | raw file
Possible License(s): CC-BY-SA-3.0, GPL-2.0, LGPL-2.0, AGPL-1.0
Large files files are truncated, but you can click here to view the full file
1 Booting the Linux/ppc kernel without Open Firmware
2 --------------------------------------------------
3
4(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
8(c) 2006 MontaVista Software, Inc.
9 Flash chip node definition
10
11Table of Contents
12=================
13
14 I - Introduction
15 1) Entry point for arch/powerpc
16 2) Board support
17
18 II - The DT block format
19 1) Header
20 2) Device tree generalities
21 3) Device tree "structure" block
22 4) Device tree "strings" block
23
24 III - Required content of the device tree
25 1) Note about cells and address representation
26 2) Note about "compatible" properties
27 3) Note about "name" properties
28 4) Note about node and property names and character set
29 5) Required nodes and properties
30 a) The root node
31 b) The /cpus node
32 c) The /cpus/* nodes
33 d) the /memory node(s)
34 e) The /chosen node
35 f) the /soc<SOCname> node
36
37 IV - "dtc", the device tree compiler
38
39 V - Recommendations for a bootloader
40
41 VI - System-on-a-chip devices and nodes
42 1) Defining child nodes of an SOC
43 2) Representing devices without a current OF specification
44 a) PHY nodes
45 b) Interrupt controllers
46 c) 4xx/Axon EMAC ethernet nodes
47 d) Xilinx IP cores
48 e) USB EHCI controllers
49 f) MDIO on GPIOs
50 g) SPI busses
51
52 VII - Marvell Discovery mv64[345]6x System Controller chips
53 1) The /system-controller node
54 2) Child nodes of /system-controller
55 a) Marvell Discovery MDIO bus
56 b) Marvell Discovery ethernet controller
57 c) Marvell Discovery PHY nodes
58 d) Marvell Discovery SDMA nodes
59 e) Marvell Discovery BRG nodes
60 f) Marvell Discovery CUNIT nodes
61 g) Marvell Discovery MPSCROUTING nodes
62 h) Marvell Discovery MPSCINTR nodes
63 i) Marvell Discovery MPSC nodes
64 j) Marvell Discovery Watch Dog Timer nodes
65 k) Marvell Discovery I2C nodes
66 l) Marvell Discovery PIC (Programmable Interrupt Controller) nodes
67 m) Marvell Discovery MPP (Multipurpose Pins) multiplexing nodes
68 n) Marvell Discovery GPP (General Purpose Pins) nodes
69 o) Marvell Discovery PCI host bridge node
70 p) Marvell Discovery CPU Error nodes
71 q) Marvell Discovery SRAM Controller nodes
72 r) Marvell Discovery PCI Error Handler nodes
73 s) Marvell Discovery Memory Controller nodes
74
75 VIII - Specifying interrupt information for devices
76 1) interrupts property
77 2) interrupt-parent property
78 3) OpenPIC Interrupt Controllers
79 4) ISA Interrupt Controllers
80
81 IX - Specifying GPIO information for devices
82 1) gpios property
83 2) gpio-controller nodes
84
85 X - Specifying device power management information (sleep property)
86
87 Appendix A - Sample SOC node for MPC8540
88
89
90Revision Information
91====================
92
93 May 18, 2005: Rev 0.1 - Initial draft, no chapter III yet.
94
95 May 19, 2005: Rev 0.2 - Add chapter III and bits & pieces here or
96 clarifies the fact that a lot of things are
97 optional, the kernel only requires a very
98 small device tree, though it is encouraged
99 to provide an as complete one as possible.
100
101 May 24, 2005: Rev 0.3 - Precise that DT block has to be in RAM
102 - Misc fixes
103 - Define version 3 and new format version 16
104 for the DT block (version 16 needs kernel
105 patches, will be fwd separately).
106 String block now has a size, and full path
107 is replaced by unit name for more
108 compactness.
109 linux,phandle is made optional, only nodes
110 that are referenced by other nodes need it.
111 "name" property is now automatically
112 deduced from the unit name
113
114 June 1, 2005: Rev 0.4 - Correct confusion between OF_DT_END and
115 OF_DT_END_NODE in structure definition.
116 - Change version 16 format to always align
117 property data to 4 bytes. Since tokens are
118 already aligned, that means no specific
119 required alignment between property size
120 and property data. The old style variable
121 alignment would make it impossible to do
122 "simple" insertion of properties using
123 memmove (thanks Milton for
124 noticing). Updated kernel patch as well
125 - Correct a few more alignment constraints
126 - Add a chapter about the device-tree
127 compiler and the textural representation of
128 the tree that can be "compiled" by dtc.
129
130 November 21, 2005: Rev 0.5
131 - Additions/generalizations for 32-bit
132 - Changed to reflect the new arch/powerpc
133 structure
134 - Added chapter VI
135
136
137 ToDo:
138 - Add some definitions of interrupt tree (simple/complex)
139 - Add some definitions for PCI host bridges
140 - Add some common address format examples
141 - Add definitions for standard properties and "compatible"
142 names for cells that are not already defined by the existing
143 OF spec.
144 - Compare FSL SOC use of PCI to standard and make sure no new
145 node definition required.
146 - Add more information about node definitions for SOC devices
147 that currently have no standard, like the FSL CPM.
148
149
150I - Introduction
151================
152
153During the recent development of the Linux/ppc64 kernel, and more
154specifically, the addition of new platform types outside of the old
155IBM pSeries/iSeries pair, it was decided to enforce some strict rules
156regarding the kernel entry and bootloader <-> kernel interfaces, in
157order to avoid the degeneration that had become the ppc32 kernel entry
158point and the way a new platform should be added to the kernel. The
159legacy iSeries platform breaks those rules as it predates this scheme,
160but no new board support will be accepted in the main tree that
161doesn't follows them properly. In addition, since the advent of the
162arch/powerpc merged architecture for ppc32 and ppc64, new 32-bit
163platforms and 32-bit platforms which move into arch/powerpc will be
164required to use these rules as well.
165
166The main requirement that will be defined in more detail below is
167the presence of a device-tree whose format is defined after Open
168Firmware specification. However, in order to make life easier
169to embedded board vendors, the kernel doesn't require the device-tree
170to represent every device in the system and only requires some nodes
171and properties to be present. This will be described in detail in
172section III, but, for example, the kernel does not require you to
173create a node for every PCI device in the system. It is a requirement
174to have a node for PCI host bridges in order to provide interrupt
175routing informations and memory/IO ranges, among others. It is also
176recommended to define nodes for on chip devices and other busses that
177don't specifically fit in an existing OF specification. This creates a
178great flexibility in the way the kernel can then probe those and match
179drivers to device, without having to hard code all sorts of tables. It
180also makes it more flexible for board vendors to do minor hardware
181upgrades without significantly impacting the kernel code or cluttering
182it with special cases.
183
184
1851) Entry point for arch/powerpc
186-------------------------------
187
188 There is one and one single entry point to the kernel, at the start
189 of the kernel image. That entry point supports two calling
190 conventions:
191
192 a) Boot from Open Firmware. If your firmware is compatible
193 with Open Firmware (IEEE 1275) or provides an OF compatible
194 client interface API (support for "interpret" callback of
195 forth words isn't required), you can enter the kernel with:
196
197 r5 : OF callback pointer as defined by IEEE 1275
198 bindings to powerpc. Only the 32-bit client interface
199 is currently supported
200
201 r3, r4 : address & length of an initrd if any or 0
202
203 The MMU is either on or off; the kernel will run the
204 trampoline located in arch/powerpc/kernel/prom_init.c to
205 extract the device-tree and other information from open
206 firmware and build a flattened device-tree as described
207 in b). prom_init() will then re-enter the kernel using
208 the second method. This trampoline code runs in the
209 context of the firmware, which is supposed to handle all
210 exceptions during that time.
211
212 b) Direct entry with a flattened device-tree block. This entry
213 point is called by a) after the OF trampoline and can also be
214 called directly by a bootloader that does not support the Open
215 Firmware client interface. It is also used by "kexec" to
216 implement "hot" booting of a new kernel from a previous
217 running one. This method is what I will describe in more
218 details in this document, as method a) is simply standard Open
219 Firmware, and thus should be implemented according to the
220 various standard documents defining it and its binding to the
221 PowerPC platform. The entry point definition then becomes:
222
223 r3 : physical pointer to the device-tree block
224 (defined in chapter II) in RAM
225
226 r4 : physical pointer to the kernel itself. This is
227 used by the assembly code to properly disable the MMU
228 in case you are entering the kernel with MMU enabled
229 and a non-1:1 mapping.
230
231 r5 : NULL (as to differentiate with method a)
232
233 Note about SMP entry: Either your firmware puts your other
234 CPUs in some sleep loop or spin loop in ROM where you can get
235 them out via a soft reset or some other means, in which case
236 you don't need to care, or you'll have to enter the kernel
237 with all CPUs. The way to do that with method b) will be
238 described in a later revision of this document.
239
240
2412) Board support
242----------------
243
24464-bit kernels:
245
246 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
264 c) Add a reference to your "ppc_md" structure in the
265 "machines" table in arch/powerpc/kernel/setup_64.c if you are
266 a 64-bit platform.
267
268 d) request and get assigned a platform number (see PLATFORM_*
269 constants in arch/powerpc/include/asm/processor.h
270
27132-bit embedded kernels:
272
273 Currently, board support is essentially an exclusive config option.
274 The kernel is configured for a single platform. Part of the reason
275 for this is to keep kernels on embedded systems small and efficient;
276 part of this is due to the fact the code is already that way. In the
277 future, a kernel may support multiple platforms, but only if the
278 platforms feature the same core architecture. A single kernel build
279 cannot support both configurations with Book E and configurations
280 with classic Powerpc architectures.
281
282 32-bit embedded platforms that are moved into arch/powerpc using a
283 flattened device tree should adopt the merged tree practice of
284 setting ppc_md up dynamically, even though the kernel is currently
285 built with support for only a single platform at a time. This allows
286 unification of the setup code, and will make it easier to go to a
287 multiple-platform-support model in the future.
288
289NOTE: I believe the above will be true once Ben's done with the merge
290of the boot sequences.... someone speak up if this is wrong!
291
292 To add a 32-bit embedded platform support, follow the instructions
293 for 64-bit platforms above, with the exception that the Kconfig
294 option should be set up such that the kernel builds exclusively for
295 the platform selected. The processor type for the platform should
296 enable another config option to select the specific board
297 supported.
298
299NOTE: If Ben doesn't merge the setup files, may need to change this to
300point to setup_32.c
301
302
303 I will describe later the boot process and various callbacks that
304 your platform should implement.
305
306
307II - The DT block format
308========================
309
310
311This chapter defines the actual format of the flattened device-tree
312passed to the kernel. The actual content of it and kernel requirements
313are described later. You can find example of code manipulating that
314format in various places, including arch/powerpc/kernel/prom_init.c
315which will generate a flattened device-tree from the Open Firmware
316representation, or the fs2dt utility which is part of the kexec tools
317which will generate one from a filesystem representation. It is
318expected that a bootloader like uboot provides a bit more support,
319that will be discussed later as well.
320
321Note: The block has to be in main memory. It has to be accessible in
322both real mode and virtual mode with no mapping other than main
323memory. If you are writing a simple flash bootloader, it should copy
324the block to RAM before passing it to the kernel.
325
326
3271) Header
328---------
329
330 The kernel is entered with r3 pointing to an area of memory that is
331 roughly described in arch/powerpc/include/asm/prom.h by the structure
332 boot_param_header:
333
334struct boot_param_header {
335 u32 magic; /* magic word OF_DT_HEADER */
336 u32 totalsize; /* total size of DT block */
337 u32 off_dt_struct; /* offset to structure */
338 u32 off_dt_strings; /* offset to strings */
339 u32 off_mem_rsvmap; /* offset to memory reserve map
340 */
341 u32 version; /* format version */
342 u32 last_comp_version; /* last compatible version */
343
344 /* version 2 fields below */
345 u32 boot_cpuid_phys; /* Which physical CPU id we're
346 booting on */
347 /* version 3 fields below */
348 u32 size_dt_strings; /* size of the strings block */
349
350 /* version 17 fields below */
351 u32 size_dt_struct; /* size of the DT structure block */
352};
353
354 Along with the constants:
355
356/* Definitions used by the flattened device tree */
357#define OF_DT_HEADER 0xd00dfeed /* 4: version,
358 4: total size */
359#define OF_DT_BEGIN_NODE 0x1 /* Start node: full name
360 */
361#define OF_DT_END_NODE 0x2 /* End node */
362#define OF_DT_PROP 0x3 /* Property: name off,
363 size, content */
364#define OF_DT_END 0x9
365
366 All values in this header are in big endian format, the various
367 fields in this header are defined more precisely below. All
368 "offset" values are in bytes from the start of the header; that is
369 from the value of r3.
370
371 - magic
372
373 This is a magic value that "marks" the beginning of the
374 device-tree block header. It contains the value 0xd00dfeed and is
375 defined by the constant OF_DT_HEADER
376
377 - totalsize
378
379 This is the total size of the DT block including the header. The
380 "DT" block should enclose all data structures defined in this
381 chapter (who are pointed to by offsets in this header). That is,
382 the device-tree structure, strings, and the memory reserve map.
383
384 - off_dt_struct
385
386 This is an offset from the beginning of the header to the start
387 of the "structure" part the device tree. (see 2) device tree)
388
389 - off_dt_strings
390
391 This is an offset from the beginning of the header to the start
392 of the "strings" part of the device-tree
393
394 - off_mem_rsvmap
395
396 This is an offset from the beginning of the header to the start
397 of the reserved memory map. This map is a list of pairs of 64-
398 bit integers. Each pair is a physical address and a size. The
399 list is terminated by an entry of size 0. This map provides the
400 kernel with a list of physical memory areas that are "reserved"
401 and thus not to be used for memory allocations, especially during
402 early initialization. The kernel needs to allocate memory during
403 boot for things like un-flattening the device-tree, allocating an
404 MMU hash table, etc... Those allocations must be done in such a
405 way to avoid overriding critical things like, on Open Firmware
406 capable machines, the RTAS instance, or on some pSeries, the TCE
407 tables used for the iommu. Typically, the reserve map should
408 contain _at least_ this DT block itself (header,total_size). If
409 you are passing an initrd to the kernel, you should reserve it as
410 well. You do not need to reserve the kernel image itself. The map
411 should be 64-bit aligned.
412
413 - version
414
415 This is the version of this structure. Version 1 stops
416 here. Version 2 adds an additional field boot_cpuid_phys.
417 Version 3 adds the size of the strings block, allowing the kernel
418 to reallocate it easily at boot and free up the unused flattened
419 structure after expansion. Version 16 introduces a new more
420 "compact" format for the tree itself that is however not backward
421 compatible. Version 17 adds an additional field, size_dt_struct,
422 allowing it to be reallocated or moved more easily (this is
423 particularly useful for bootloaders which need to make
424 adjustments to a device tree based on probed information). You
425 should always generate a structure of the highest version defined
426 at the time of your implementation. Currently that is version 17,
427 unless you explicitly aim at being backward compatible.
428
429 - last_comp_version
430
431 Last compatible version. This indicates down to what version of
432 the DT block you are backward compatible. For example, version 2
433 is backward compatible with version 1 (that is, a kernel build
434 for version 1 will be able to boot with a version 2 format). You
435 should put a 1 in this field if you generate a device tree of
436 version 1 to 3, or 16 if you generate a tree of version 16 or 17
437 using the new unit name format.
438
439 - boot_cpuid_phys
440
441 This field only exist on version 2 headers. It indicate which
442 physical CPU ID is calling the kernel entry point. This is used,
443 among others, by kexec. If you are on an SMP system, this value
444 should match the content of the "reg" property of the CPU node in
445 the device-tree corresponding to the CPU calling the kernel entry
446 point (see further chapters for more informations on the required
447 device-tree contents)
448
449 - size_dt_strings
450
451 This field only exists on version 3 and later headers. It
452 gives the size of the "strings" section of the device tree (which
453 starts at the offset given by off_dt_strings).
454
455 - size_dt_struct
456
457 This field only exists on version 17 and later headers. It gives
458 the size of the "structure" section of the device tree (which
459 starts at the offset given by off_dt_struct).
460
461 So the typical layout of a DT block (though the various parts don't
462 need to be in that order) looks like this (addresses go from top to
463 bottom):
464
465
466 ------------------------------
467 r3 -> | struct boot_param_header |
468 ------------------------------
469 | (alignment gap) (*) |
470 ------------------------------
471 | memory reserve map |
472 ------------------------------
473 | (alignment gap) |
474 ------------------------------
475 | |
476 | device-tree structure |
477 | |
478 ------------------------------
479 | (alignment gap) |
480 ------------------------------
481 | |
482 | device-tree strings |
483 | |
484 -----> ------------------------------
485 |
486 |
487 --- (r3 + totalsize)
488
489 (*) The alignment gaps are not necessarily present; their presence
490 and size are dependent on the various alignment requirements of
491 the individual data blocks.
492
493
4942) Device tree generalities
495---------------------------
496
497This device-tree itself is separated in two different blocks, a
498structure block and a strings block. Both need to be aligned to a 4
499byte boundary.
500
501First, let's quickly describe the device-tree concept before detailing
502the storage format. This chapter does _not_ describe the detail of the
503required types of nodes & properties for the kernel, this is done
504later in chapter III.
505
506The device-tree layout is strongly inherited from the definition of
507the Open Firmware IEEE 1275 device-tree. It's basically a tree of
508nodes, each node having two or more named properties. A property can
509have a value or not.
510
511It is a tree, so each node has one and only one parent except for the
512root node who has no parent.
513
514A node has 2 names. The actual node name is generally contained in a
515property of type "name" in the node property list whose value is a
516zero terminated string and is mandatory for version 1 to 3 of the
517format definition (as it is in Open Firmware). Version 16 makes it
518optional as it can generate it from the unit name defined below.
519
520There is also a "unit name" that is used to differentiate nodes with
521the same name at the same level, it is usually made of the node
522names, the "@" sign, and a "unit address", which definition is
523specific to the bus type the node sits on.
524
525The unit name doesn't exist as a property per-se but is included in
526the device-tree structure. It is typically used to represent "path" in
527the device-tree. More details about the actual format of these will be
528below.
529
530The kernel powerpc generic code does not make any formal use of the
531unit address (though some board support code may do) so the only real
532requirement here for the unit address is to ensure uniqueness of
533the node unit name at a given level of the tree. Nodes with no notion
534of address and no possible sibling of the same name (like /memory or
535/cpus) may omit the unit address in the context of this specification,
536or use the "@0" default unit address. The unit name is used to define
537a node "full path", which is the concatenation of all parent node
538unit names separated with "/".
539
540The root node doesn't have a defined name, and isn't required to have
541a name property either if you are using version 3 or earlier of the
542format. It also has no unit address (no @ symbol followed by a unit
543address). The root node unit name is thus an empty string. The full
544path to the root node is "/".
545
546Every node which actually represents an actual device (that is, a node
547which isn't only a virtual "container" for more nodes, like "/cpus"
548is) is also required to have a "device_type" property indicating the
549type of node .
550
551Finally, every node that can be referenced from a property in another
552node is required to have a "linux,phandle" property. Real open
553firmware implementations provide a unique "phandle" value for every
554node that the "prom_init()" trampoline code turns into
555"linux,phandle" properties. However, this is made optional if the
556flattened device tree is used directly. An example of a node
557referencing another node via "phandle" is when laying out the
558interrupt tree which will be described in a further version of this
559document.
560
561This "linux, phandle" property is a 32-bit value that uniquely
562identifies a node. You are free to use whatever values or system of
563values, internal pointers, or whatever to generate these, the only
564requirement is that every node for which you provide that property has
565a unique value for it.
566
567Here is an example of a simple device-tree. In this example, an "o"
568designates a node followed by the node unit name. Properties are
569presented with their name followed by their content. "content"
570represents an ASCII string (zero terminated) value, while <content>
571represents a 32-bit hexadecimal value. The various nodes in this
572example will be discussed in a later chapter. At this point, it is
573only meant to give you a idea of what a device-tree looks like. I have
574purposefully kept the "name" and "linux,phandle" properties which
575aren't necessary in order to give you a better idea of what the tree
576looks like in practice.
577
578 / o device-tree
579 |- name = "device-tree"
580 |- model = "MyBoardName"
581 |- compatible = "MyBoardFamilyName"
582 |- #address-cells = <2>
583 |- #size-cells = <2>
584 |- linux,phandle = <0>
585 |
586 o cpus
587 | | - name = "cpus"
588 | | - linux,phandle = <1>
589 | | - #address-cells = <1>
590 | | - #size-cells = <0>
591 | |
592 | o PowerPC,970@0
593 | |- name = "PowerPC,970"
594 | |- device_type = "cpu"
595 | |- reg = <0>
596 | |- clock-frequency = <5f5e1000>
597 | |- 64-bit
598 | |- linux,phandle = <2>
599 |
600 o memory@0
601 | |- name = "memory"
602 | |- device_type = "memory"
603 | |- reg = <00000000 00000000 00000000 20000000>
604 | |- linux,phandle = <3>
605 |
606 o chosen
607 |- name = "chosen"
608 |- bootargs = "root=/dev/sda2"
609 |- linux,phandle = <4>
610
611This tree is almost a minimal tree. It pretty much contains the
612minimal set of required nodes and properties to boot a linux kernel;
613that is, some basic model informations at the root, the CPUs, and the
614physical memory layout. It also includes misc information passed
615through /chosen, like in this example, the platform type (mandatory)
616and the kernel command line arguments (optional).
617
618The /cpus/PowerPC,970@0/64-bit property is an example of a
619property without a value. All other properties have a value. The
620significance of the #address-cells and #size-cells properties will be
621explained in chapter IV which defines precisely the required nodes and
622properties and their content.
623
624
6253) Device tree "structure" block
626
627The structure of the device tree is a linearized tree structure. The
628"OF_DT_BEGIN_NODE" token starts a new node, and the "OF_DT_END_NODE"
629ends that node definition. Child nodes are simply defined before
630"OF_DT_END_NODE" (that is nodes within the node). A 'token' is a 32
631bit value. The tree has to be "finished" with a OF_DT_END token
632
633Here's the basic structure of a single node:
634
635 * token OF_DT_BEGIN_NODE (that is 0x00000001)
636 * for version 1 to 3, this is the node full path as a zero
637 terminated string, starting with "/". For version 16 and later,
638 this is the node unit name only (or an empty string for the
639 root node)
640 * [align gap to next 4 bytes boundary]
641 * for each property:
642 * token OF_DT_PROP (that is 0x00000003)
643 * 32-bit value of property value size in bytes (or 0 if no
644 value)
645 * 32-bit value of offset in string block of property name
646 * property value data if any
647 * [align gap to next 4 bytes boundary]
648 * [child nodes if any]
649 * token OF_DT_END_NODE (that is 0x00000002)
650
651So the node content can be summarized as a start token, a full path,
652a list of properties, a list of child nodes, and an end token. Every
653child node is a full node structure itself as defined above.
654
655NOTE: The above definition requires that all property definitions for
656a particular node MUST precede any subnode definitions for that node.
657Although the structure would not be ambiguous if properties and
658subnodes were intermingled, the kernel parser requires that the
659properties come first (up until at least 2.6.22). Any tools
660manipulating a flattened tree must take care to preserve this
661constraint.
662
6634) Device tree "strings" block
664
665In order to save space, property names, which are generally redundant,
666are stored separately in the "strings" block. This block is simply the
667whole bunch of zero terminated strings for all property names
668concatenated together. The device-tree property definitions in the
669structure block will contain offset values from the beginning of the
670strings block.
671
672
673III - Required content of the device tree
674=========================================
675
676WARNING: All "linux,*" properties defined in this document apply only
677to a flattened device-tree. If your platform uses a real
678implementation of Open Firmware or an implementation compatible with
679the Open Firmware client interface, those properties will be created
680by the trampoline code in the kernel's prom_init() file. For example,
681that's where you'll have to add code to detect your board model and
682set the platform number. However, when using the flattened device-tree
683entry point, there is no prom_init() pass, and thus you have to
684provide those properties yourself.
685
686
6871) Note about cells and address representation
688----------------------------------------------
689
690The general rule is documented in the various Open Firmware
691documentations. If you choose to describe a bus with the device-tree
692and there exist an OF bus binding, then you should follow the
693specification. However, the kernel does not require every single
694device or bus to be described by the device tree.
695
696In general, the format of an address for a device is defined by the
697parent bus type, based on the #address-cells and #size-cells
698properties. Note that the parent's parent definitions of #address-cells
699and #size-cells are not inherited so every node with children must specify
700them. The kernel requires the root node to have those properties defining
701addresses format for devices directly mapped on the processor bus.
702
703Those 2 properties define 'cells' for representing an address and a
704size. A "cell" is a 32-bit number. For example, if both contain 2
705like the example tree given above, then an address and a size are both
706composed of 2 cells, and each is a 64-bit number (cells are
707concatenated and expected to be in big endian format). Another example
708is the way Apple firmware defines them, with 2 cells for an address
709and one cell for a size. Most 32-bit implementations should define
710#address-cells and #size-cells to 1, which represents a 32-bit value.
711Some 32-bit processors allow for physical addresses greater than 32
712bits; these processors should define #address-cells as 2.
713
714"reg" properties are always a tuple of the type "address size" where
715the number of cells of address and size is specified by the bus
716#address-cells and #size-cells. When a bus supports various address
717spaces and other flags relative to a given address allocation (like
718prefetchable, etc...) those flags are usually added to the top level
719bits of the physical address. For example, a PCI physical address is
720made of 3 cells, the bottom two containing the actual address itself
721while the top cell contains address space indication, flags, and pci
722bus & device numbers.
723
724For busses that support dynamic allocation, it's the accepted practice
725to then not provide the address in "reg" (keep it 0) though while
726providing a flag indicating the address is dynamically allocated, and
727then, to provide a separate "assigned-addresses" property that
728contains the fully allocated addresses. See the PCI OF bindings for
729details.
730
731In general, a simple bus with no address space bits and no dynamic
732allocation is preferred if it reflects your hardware, as the existing
733kernel address parsing functions will work out of the box. If you
734define a bus type with a more complex address format, including things
735like address space bits, you'll have to add a bus translator to the
736prom_parse.c file of the recent kernels for your bus type.
737
738The "reg" property only defines addresses and sizes (if #size-cells is
739non-0) within a given bus. In order to translate addresses upward
740(that is into parent bus addresses, and possibly into CPU physical
741addresses), all busses must contain a "ranges" property. If the
742"ranges" property is missing at a given level, it's assumed that
743translation isn't possible, i.e., the registers are not visible on the
744parent bus. The format of the "ranges" property for a bus is a list
745of:
746
747 bus address, parent bus address, size
748
749"bus address" is in the format of the bus this bus node is defining,
750that is, for a PCI bridge, it would be a PCI address. Thus, (bus
751address, size) defines a range of addresses for child devices. "parent
752bus address" is in the format of the parent bus of this bus. For
753example, for a PCI host controller, that would be a CPU address. For a
754PCI<->ISA bridge, that would be a PCI address. It defines the base
755address in the parent bus where the beginning of that range is mapped.
756
757For a new 64-bit powerpc board, I recommend either the 2/2 format or
758Apple's 2/1 format which is slightly more compact since sizes usually
759fit in a single 32-bit word. New 32-bit powerpc boards should use a
7601/1 format, unless the processor supports physical addresses greater
761than 32-bits, in which case a 2/1 format is recommended.
762
763Alternatively, the "ranges" property may be empty, indicating that the
764registers are visible on the parent bus using an identity mapping
765translation. In other words, the parent bus address space is the same
766as the child bus address space.
767
7682) Note about "compatible" properties
769-------------------------------------
770
771These properties are optional, but recommended in devices and the root
772node. The format of a "compatible" property is a list of concatenated
773zero terminated strings. They allow a device to express its
774compatibility with a family of similar devices, in some cases,
775allowing a single driver to match against several devices regardless
776of their actual names.
777
7783) Note about "name" properties
779-------------------------------
780
781While earlier users of Open Firmware like OldWorld macintoshes tended
782to use the actual device name for the "name" property, it's nowadays
783considered a good practice to use a name that is closer to the device
784class (often equal to device_type). For example, nowadays, ethernet
785controllers are named "ethernet", an additional "model" property
786defining precisely the chip type/model, and "compatible" property
787defining the family in case a single driver can driver more than one
788of these chips. However, the kernel doesn't generally put any
789restriction on the "name" property; it is simply considered good
790practice to follow the standard and its evolutions as closely as
791possible.
792
793Note also that the new format version 16 makes the "name" property
794optional. If it's absent for a node, then the node's unit name is then
795used to reconstruct the name. That is, the part of the unit name
796before the "@" sign is used (or the entire unit name if no "@" sign
797is present).
798
7994) Note about node and property names and character set
800-------------------------------------------------------
801
802While open firmware provides more flexible usage of 8859-1, this
803specification enforces more strict rules. Nodes and properties should
804be comprised only of ASCII characters 'a' to 'z', '0' to
805'9', ',', '.', '_', '+', '#', '?', and '-'. Node names additionally
806allow uppercase characters 'A' to 'Z' (property names should be
807lowercase. The fact that vendors like Apple don't respect this rule is
808irrelevant here). Additionally, node and property names should always
809begin with a character in the range 'a' to 'z' (or 'A' to 'Z' for node
810names).
811
812The maximum number of characters for both nodes and property names
813is 31. In the case of node names, this is only the leftmost part of
814a unit name (the pure "name" property), it doesn't include the unit
815address which can extend beyond that limit.
816
817
8185) Required nodes and properties
819--------------------------------
820 These are all that are currently required. However, it is strongly
821 recommended that you expose PCI host bridges as documented in the
822 PCI binding to open firmware, and your interrupt tree as documented
823 in OF interrupt tree specification.
824
825 a) The root node
826
827 The root node requires some properties to be present:
828
829 - model : this is your board name/model
830 - #address-cells : address representation for "root" devices
831 - #size-cells: the size representation for "root" devices
832 - device_type : This property shouldn't be necessary. However, if
833 you decide to create a device_type for your root node, make sure it
834 is _not_ "chrp" unless your platform is a pSeries or PAPR compliant
835 one for 64-bit, or a CHRP-type machine for 32-bit as this will
836 matched by the kernel this way.
837
838 Additionally, some recommended properties are:
839
840 - compatible : the board "family" generally finds its way here,
841 for example, if you have 2 board models with a similar layout,
842 that typically get driven by the same platform code in the
843 kernel, you would use a different "model" property but put a
844 value in "compatible". The kernel doesn't directly use that
845 value but it is generally useful.
846
847 The root node is also generally where you add additional properties
848 specific to your board like the serial number if any, that sort of
849 thing. It is recommended that if you add any "custom" property whose
850 name may clash with standard defined ones, you prefix them with your
851 vendor name and a comma.
852
853 b) The /cpus node
854
855 This node is the parent of all individual CPU nodes. It doesn't
856 have any specific requirements, though it's generally good practice
857 to have at least:
858
859 #address-cells = <00000001>
860 #size-cells = <00000000>
861
862 This defines that the "address" for a CPU is a single cell, and has
863 no meaningful size. This is not necessary but the kernel will assume
864 that format when reading the "reg" properties of a CPU node, see
865 below
866
867 c) The /cpus/* nodes
868
869 So under /cpus, you are supposed to create a node for every CPU on
870 the machine. There is no specific restriction on the name of the
871 CPU, though It's common practice to call it PowerPC,<name>. For
872 example, Apple uses PowerPC,G5 while IBM uses PowerPC,970FX.
873
874 Required properties:
875
876 - device_type : has to be "cpu"
877 - reg : This is the physical CPU number, it's a single 32-bit cell
878 and is also used as-is as the unit number for constructing the
879 unit name in the full path. For example, with 2 CPUs, you would
880 have the full path:
881 /cpus/PowerPC,970FX@0
882 /cpus/PowerPC,970FX@1
883 (unit addresses do not require leading zeroes)
884 - d-cache-block-size : one cell, L1 data cache block size in bytes (*)
885 - i-cache-block-size : one cell, L1 instruction cache block size in
886 bytes
887 - d-cache-size : one cell, size of L1 data cache in bytes
888 - i-cache-size : one cell, size of L1 instruction cache in bytes
889
890(*) The cache "block" size is the size on which the cache management
891instructions operate. Historically, this document used the cache
892"line" size here which is incorrect. The kernel will prefer the cache
893block size and will fallback to cache line size for backward
894compatibility.
895
896 Recommended properties:
897
898 - timebase-frequency : a cell indicating the frequency of the
899 timebase in Hz. This is not directly used by the generic code,
900 but you are welcome to copy/paste the pSeries code for setting
901 the kernel timebase/decrementer calibration based on this
902 value.
903 - clock-frequency : a cell indicating the CPU core clock frequency
904 in Hz. A new property will be defined for 64-bit values, but if
905 your frequency is < 4Ghz, one cell is enough. Here as well as
906 for the above, the common code doesn't use that property, but
907 you are welcome to re-use the pSeries or Maple one. A future
908 kernel version might provide a common function for this.
909 - d-cache-line-size : one cell, L1 data cache line size in bytes
910 if different from the block size
911 - i-cache-line-size : one cell, L1 instruction cache line size in
912 bytes if different from the block size
913
914 You are welcome to add any property you find relevant to your board,
915 like some information about the mechanism used to soft-reset the
916 CPUs. For example, Apple puts the GPIO number for CPU soft reset
917 lines in there as a "soft-reset" property since they start secondary
918 CPUs by soft-resetting them.
919
920
921 d) the /memory node(s)
922
923 To define the physical memory layout of your board, you should
924 create one or more memory node(s). You can either create a single
925 node with all memory ranges in its reg property, or you can create
926 several nodes, as you wish. The unit address (@ part) used for the
927 full path is the address of the first range of memory defined by a
928 given node. If you use a single memory node, this will typically be
929 @0.
930
931 Required properties:
932
933 - device_type : has to be "memory"
934 - reg : This property contains all the physical memory ranges of
935 your board. It's a list of addresses/sizes concatenated
936 together, with the number of cells of each defined by the
937 #address-cells and #size-cells of the root node. For example,
938 with both of these properties being 2 like in the example given
939 earlier, a 970 based machine with 6Gb of RAM could typically
940 have a "reg" property here that looks like:
941
942 00000000 00000000 00000000 80000000
943 00000001 00000000 00000001 00000000
944
945 That is a range starting at 0 of 0x80000000 bytes and a range
946 starting at 0x100000000 and of 0x100000000 bytes. You can see
947 that there is no memory covering the IO hole between 2Gb and
948 4Gb. Some vendors prefer splitting those ranges into smaller
949 segments, but the kernel doesn't care.
950
951 e) The /chosen node
952
953 This node is a bit "special". Normally, that's where open firmware
954 puts some variable environment information, like the arguments, or
955 the default input/output devices.
956
957 This specification makes a few of these mandatory, but also defines
958 some linux-specific properties that would be normally constructed by
959 the prom_init() trampoline when booting with an OF client interface,
960 but that you have to provide yourself when using the flattened format.
961
962 Recommended properties:
963
964 - bootargs : This zero-terminated string is passed as the kernel
965 command line
966 - linux,stdout-path : This is the full path to your standard
967 console device if any. Typically, if you have serial devices on
968 your board, you may want to put the full path to the one set as
969 the default console in the firmware here, for the kernel to pick
970 it up as its own default console. If you look at the function
971 set_preferred_console() in arch/ppc64/kernel/setup.c, you'll see
972 that the kernel tries to find out the default console and has
973 knowledge of various types like 8250 serial ports. You may want
974 to extend this function to add your own.
975
976 Note that u-boot creates and fills in the chosen node for platforms
977 that use it.
978
979 (Note: a practice that is now obsolete was to include a property
980 under /chosen called interrupt-controller which had a phandle value
981 that pointed to the main interrupt controller)
982
983 f) the /soc<SOCname> node
984
985 This node is used to represent a system-on-a-chip (SOC) and must be
986 present if the processor is a SOC. The top-level soc node contains
987 information that is global to all devices on the SOC. The node name
988 should contain a unit address for the SOC, which is the base address
989 of the memory-mapped register set for the SOC. The name of an soc
990 node should start with "soc", and the remainder of the name should
991 represent the part number for the soc. For example, the MPC8540's
992 soc node would be called "soc8540".
993
994 Required properties:
995
996 - device_type : Should be "soc"
997 - ranges : Should be defined as specified in 1) to describe the
998 translation of SOC addresses for memory mapped SOC registers.
999 - bus-frequency: Contains the bus frequency for the SOC node.
1000 Typically, the value of this field is filled in by the boot
1001 loader.
1002
1003
1004 Recommended properties:
1005
1006 - reg : This property defines the address and size of the
1007 memory-mapped registers that are used for the SOC node itself.
1008 It does not include the child device registers - these will be
1009 defined inside each child node. The address specified in the
1010 "reg" property should match the unit address of the SOC node.
1011 - #address-cells : Address representation for "soc" devices. The
1012 format of this field may vary depending on whether or not the
1013 device registers are memory mapped. For memory mapped
1014 registers, this field represents the number of cells needed to
1015 represent the address of the registers. For SOCs that do not
1016 use MMIO, a special address format should be defined that
1017 contains enough cells to represent the required information.
1018 See 1) above for more details on defining #address-cells.
1019 - #size-cells : Size representation for "soc" devices
1020 - #interrupt-cells : Defines the width of cells used to represent
1021 interrupts. Typically this value is <2>, which includes a
1022 32-bit number that represents the interrupt number, and a
1023 32-bit number that represents the interrupt sense and level.
1024 This field is only needed if the SOC contains an interrupt
1025 controller.
1026
1027 The SOC node may contain child nodes for each SOC device that the
1028 platform uses. Nodes should not be created for devices which exist
1029 on the SOC but are not used by a particular platform. See chapter VI
1030 for more information on how to specify devices that are part of a SOC.
1031
1032 Example SOC node for the MPC8540:
1033
1034 soc8540@e0000000 {
1035 #address-cells = <1>;
1036 #size-cells = <1>;
1037 #interrupt-cells = <2>;
1038 device_type = "soc";
1039 ranges = <00000000 e0000000 00100000>
1040 reg = <e0000000 00003000>;
1041 bus-frequency = <0>;
1042 }
1043
1044
1045
1046IV - "dtc", the device tree compiler
1047====================================
1048
1049
1050dtc source code can be found at
1051<http://ozlabs.org/~dgibson/dtc/dtc.tar.gz>
1052
1053WARNING: This version is still in early development stage; the
1054resulting device-tree "blobs" have not yet been validated with the
1055kernel. The current generated bloc lacks a useful reserve map (it will
1056be fixed to generate an empty one, it's up to the bootloader to fill
1057it up) among others. The error handling needs work, bugs are lurking,
1058etc...
1059
1060dtc basically takes a device-tree in a given format and outputs a
1061device-tree in another format. The currently supported formats are:
1062
1063 Input formats:
1064 -------------
1065
1066 - "dtb": "blob" format, that is a flattened device-tree block
1067 with
1068 header all in a binary blob.
1069 - "dts": "source" format. This is a text file containing a
1070 "source" for a device-tree. The format is defined later in this
1071 chapter.
1072 - "fs" format. This is a representation equivalent to the
1073 output of /proc/device-tree, that is nodes are directories and
1074 properties are files
1075
1076 Output formats:
1077 ---------------
1078
1079 - "dtb": "blob" format
1080 - "dts": "source" format
1081 - "asm": assembly language file. This is a file that can be
1082 sourced by gas to generate a device-tree "blob". That file can
1083 then simply be added to your Makefile. Additionally, the
1084 assembly file exports some symbols that can be used.
1085
1086
1087The syntax of the dtc tool is
1088
1089 dtc [-I <input-format>] [-O <output-format>]
1090 [-o output-filename] [-V output_version] input_filename
1091
1092
1093The "output_version" defines what version of the "blob" format will be
1094generated. Supported versions are 1,2,3 and 16. The default is
1095currently version 3 but that may change in the future to version 16.
1096
1097Additionally, dtc performs various sanity checks on the tree, like the
1098uniqueness of linux, phandle properties, validity of strings, etc...
1099
1100The format of the .dts "source" file is "C" like, supports C and C++
1101style comments.
1102
1103/ {
1104}
1105
1106The above is the "device-tree" definition. It's the only statement
1107supported currently at the toplevel.
1108
1109/ {
1110 property1 = "string_value"; /* define a property containing a 0
1111 * terminated string
1112 */
1113
1114 property2 = <1234abcd>; /* define a property containing a
1115 * numerical 32-bit value (hexadecimal)
1116 */
1117
1118 property3 = <12345678 12345678 deadbeef>;
1119 /* define a property containing 3
1120 * numerical 32-bit values (cells) in
1121 * hexadecimal
1122 */
1123 property4 = [0a 0b 0c 0d de ea ad be ef];
1124 /* define a property whose content is
1125 * an arbitrary array of bytes
1126 */
1127
1128 childnode@addresss { /* define a child node named "childnode"
1129 * whose unit name is "childnode at
1130 * address"
1131 */
1132
1133 childprop = "hello\n"; /* define a property "childprop" of
1134 * childnode (in this case, a string)
1135 */
1136 };
1137};
1138
1139Nodes can contain other nodes etc... thus defining the hierarchical
1140structure of the tree.
1141
1142Strings support common escape sequences from C: "\n", "\t", "\r",
1143"\(octal value)", "\x(hex value)".
1144
1145It is also suggested that you pipe your source file through cpp (gcc
1146preprocessor) so you can use #include's, #define for constants, …Large files files are truncated, but you can click here to view the full file