u-boot/doc/uImage.FIT/source_file_format.txt

320 lines
12 KiB
Text
Raw Normal View History

U-Boot new uImage source file format (bindings definition)
==========================================================
Author: Marian Balakowicz <m8@semihalf.com>
External data additions, 25/1/16 Simon Glass <sjg@chromium.org>
1) Introduction
---------------
Evolution of the 2.6 Linux kernel for embedded PowerPC systems introduced new
booting method which requires that hardware description is available to the
kernel in the form of Flattened Device Tree.
Booting with a Flattened Device Tree is much more flexible and is intended to
replace direct passing of 'struct bd_info' which was used to boot pre-FDT
kernels.
However, U-Boot needs to support both techniques to provide backward
compatibility for platforms which are not FDT ready. Number of elements
playing role in the booting process has increased and now includes the FDT
blob. Kernel image, FDT blob and possibly ramdisk image - all must be placed
in the system memory and passed to bootm as a arguments. Some of them may be
missing: FDT is not present for legacy platforms, ramdisk is always optional.
Additionally, old uImage format has been extended to support multi sub-images
but the support is limited by simple format of the legacy uImage structure.
Single binary header 'struct legacy_img_hdr' is not flexible enough to cover all
possible scenarios.
All those factors combined clearly show that there is a need for new, more
flexible, multi component uImage format.
2) New uImage format assumptions
--------------------------------
a) Implementation
Libfdt has been selected for the new uImage format implementation as (1) it
provides needed functionality, (2) is actively maintained and developed and
(3) increases code reuse as it is already part of the U-Boot source tree.
b) Terminology
This document defines new uImage structure by providing FDT bindings for new
uImage internals. Bindings are defined from U-Boot perspective, i.e. describe
final form of the uImage at the moment when it reaches U-Boot. User
perspective may be simpler, as some of the properties (like timestamps and
hashes) will need to be filled in automatically by the U-Boot mkimage tool.
To avoid confusion with the kernel FDT the following naming convention is
proposed for the new uImage format related terms:
FIT - Flattened uImage Tree
FIT is formally a flattened device tree (in the libfdt meaning), which
conforms to bindings defined in this document.
.its - image tree source
.itb - flattened image tree blob
c) Image building procedure
The following picture shows how the new uImage is prepared. Input consists of
image source file (.its) and a set of data files. Image is created with the
help of standard U-Boot mkimage tool which in turn uses dtc (device tree
compiler) to produce image tree blob (.itb). Resulting .itb file is the
actual binary of a new uImage.
tqm5200.its
+
vmlinux.bin.gz mkimage + dtc xfer to target
eldk-4.2-ramdisk --------------> tqm5200.itb --------------> bootm
tqm5200.dtb /|\
... |
'new uImage'
- create .its file, automatically filled-in properties are omitted
- call mkimage tool on a .its file
- mkimage calls dtc to create .itb image and assures that
missing properties are added
- .itb (new uImage) is uploaded onto the target and used therein
d) Unique identifiers
To identify FIT sub-nodes representing images, hashes, configurations (which
are defined in the following sections), the "unit name" of the given sub-node
is used as it's identifier as it assures uniqueness without additional
checking required.
3) Root node properties
-----------------------
Root node of the uImage Tree should have the following layout:
/ o image-tree
|- description = "image description"
|- timestamp = <12399321>
|- #address-cells = <1>
|
o images
| |
| o image-1 {...}
| o image-2 {...}
| ...
|
o configurations
|- default = "conf-1"
|
o conf-1 {...}
o conf-2 {...}
...
Optional property:
- description : Textual description of the uImage
Mandatory property:
- timestamp : Last image modification time being counted in seconds since
1970-01-01 00:00:00 - to be automatically calculated by mkimage tool.
Conditionally mandatory property:
- #address-cells : Number of 32bit cells required to represent entry and
load addresses supplied within sub-image nodes. May be omitted when no
entry or load addresses are used.
doc: FIT image: Clarify format and simplify syntax ** Introduction There are currently four ways to load an OS image with u-boot 1. SPL -> u-boot -> bootm 2. SPL blue falcon mode 3. "Basic" FIT image (CONFIG_LOAD_FIT) 4. "Full-featured" FIT image (CONFIG_LOAD_FIT_FULL) These four code paths were developed independently, and share very little code. (3) and (4), behave very differently, are littered with special cases. They even have different DTS syntax and properties. The cause of this divergence is that the FIT format specification leaves a number of things open to interpretation. The purpose of this change is to enable the reduction of code size, duplication, and complexity by updating and streamlining the FIT format. We are only marginally concerned with backwards compatibility, because we don't have inter-compatibility. For example, CONFIG_LOAD_FIT is able to load images that CONFIG_LOAD_FIT_FULL won't. This is a direct result of the incompatible syntax between the two implementations. Ideally, these changes would enable "simple" FIT to be a subset of the "full" fit implementation, and share most code. These changes should also eliminate the need for falcon mode (although we are not advocating for the removal of falcon mode at this time). ** Description of changes * The "configurations" node is now mandatory Guessing how to load components based on their "os" and "type" invites confusion and superfluous heuristics. Instead, require each FIT image to be explicit on how components should be loaded. * Eliminate "ramdisk", "setup", "standalone", and "fpga" properties Having too many special purpose properties requires special-casing FIT loading code. When a special property can be handled by another property, it is redundant. - A "ramdisk" is identical to a loadable. Thus ramdisk images should be placed under "loadables". - A "setup" node can be achieved by using a "kernel" or "firmware" property instead. - "standalone" is used for u-boot nodes. The correct property to use in this case is "firmware". - "fpga" is a loadable * Prioritize control between "firmware" and "kernel" "firmware" and "kernel" are special nodes in that control is passed to the "entry-point" of the image. Both can be present, for example, an OP-TEE firmware with a linux kernel. When both are present, control is passed to the "firmware" image. ** Further generalizations (not included herein) The "firmware" and "kernel" properties could be generalized as a "next-boot-stage", or similar name. This "next" stage would be special in that it is both executable, and is the stage that is passed control. For example, "next-stage" could be an op-tee image, with linux as a loadable, or a u-boot image. Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
2020-12-15 19:15:43 +00:00
Mandatory nodes:
- images : This node contains a set of sub-nodes, each of them representing
single component sub-image (like kernel, ramdisk, etc.). At least one
sub-image is required.
- configurations : Contains a set of available configuration nodes and
defines a default configuration.
4) '/images' node
-----------------
This node is a container node for component sub-image nodes. Each sub-node of
the '/images' node should have the following layout:
o image-1
|- description = "component sub-image description"
|- data = /incbin/("path/to/data/file.bin")
|- type = "sub-image type name"
|- arch = "ARCH name"
|- os = "OS name"
|- compression = "compression name"
|- load = <00000000>
|- entry = <00000000>
|
o hash-1 {...}
o hash-2 {...}
...
Mandatory properties:
- description : Textual description of the component sub-image
- type : Name of component sub-image type, supported types are:
"standalone", "kernel", "kernel_noload", "ramdisk", "firmware", "script",
"filesystem", "flat_dt" and others (see uimage_type in common/image.c).
- data : Path to the external file which contains this node's binary data.
- compression : Compression used by included data. Supported compressions
are "gzip" and "bzip2". If no compression is used compression property
should be set to "none". If the data is compressed but it should not be
uncompressed by U-Boot (e.g. compressed ramdisk), this should also be set
to "none".
Conditionally mandatory property:
doc: FIT image: Clarify format and simplify syntax ** Introduction There are currently four ways to load an OS image with u-boot 1. SPL -> u-boot -> bootm 2. SPL blue falcon mode 3. "Basic" FIT image (CONFIG_LOAD_FIT) 4. "Full-featured" FIT image (CONFIG_LOAD_FIT_FULL) These four code paths were developed independently, and share very little code. (3) and (4), behave very differently, are littered with special cases. They even have different DTS syntax and properties. The cause of this divergence is that the FIT format specification leaves a number of things open to interpretation. The purpose of this change is to enable the reduction of code size, duplication, and complexity by updating and streamlining the FIT format. We are only marginally concerned with backwards compatibility, because we don't have inter-compatibility. For example, CONFIG_LOAD_FIT is able to load images that CONFIG_LOAD_FIT_FULL won't. This is a direct result of the incompatible syntax between the two implementations. Ideally, these changes would enable "simple" FIT to be a subset of the "full" fit implementation, and share most code. These changes should also eliminate the need for falcon mode (although we are not advocating for the removal of falcon mode at this time). ** Description of changes * The "configurations" node is now mandatory Guessing how to load components based on their "os" and "type" invites confusion and superfluous heuristics. Instead, require each FIT image to be explicit on how components should be loaded. * Eliminate "ramdisk", "setup", "standalone", and "fpga" properties Having too many special purpose properties requires special-casing FIT loading code. When a special property can be handled by another property, it is redundant. - A "ramdisk" is identical to a loadable. Thus ramdisk images should be placed under "loadables". - A "setup" node can be achieved by using a "kernel" or "firmware" property instead. - "standalone" is used for u-boot nodes. The correct property to use in this case is "firmware". - "fpga" is a loadable * Prioritize control between "firmware" and "kernel" "firmware" and "kernel" are special nodes in that control is passed to the "entry-point" of the image. Both can be present, for example, an OP-TEE firmware with a linux kernel. When both are present, control is passed to the "firmware" image. ** Further generalizations (not included herein) The "firmware" and "kernel" properties could be generalized as a "next-boot-stage", or similar name. This "next" stage would be special in that it is both executable, and is the stage that is passed control. For example, "next-stage" could be an op-tee image, with linux as a loadable, or a u-boot image. Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
2020-12-15 19:15:43 +00:00
- os : OS name, mandatory for types "kernel". Valid OS names are:
"openbsd", "netbsd", "freebsd", "4_4bsd", "linux", "svr4", "esix",
"solaris", "irix", "sco", "dell", "ncr", "lynxos", "vxworks", "psos", "qnx",
"u-boot", "rtems", "unity", "integrity".
- arch : Architecture name, mandatory for types: "standalone", "kernel",
"firmware", "ramdisk" and "fdt". Valid architecture names are: "alpha",
"arm", "i386", "ia64", "mips", "mips64", "ppc", "s390", "sh", "sparc",
"sparc64", "m68k", "microblaze", "nios2", "blackfin", "avr32", "st200",
"sandbox".
- entry : entry point address, address size is determined by
doc: FIT image: Clarify format and simplify syntax ** Introduction There are currently four ways to load an OS image with u-boot 1. SPL -> u-boot -> bootm 2. SPL blue falcon mode 3. "Basic" FIT image (CONFIG_LOAD_FIT) 4. "Full-featured" FIT image (CONFIG_LOAD_FIT_FULL) These four code paths were developed independently, and share very little code. (3) and (4), behave very differently, are littered with special cases. They even have different DTS syntax and properties. The cause of this divergence is that the FIT format specification leaves a number of things open to interpretation. The purpose of this change is to enable the reduction of code size, duplication, and complexity by updating and streamlining the FIT format. We are only marginally concerned with backwards compatibility, because we don't have inter-compatibility. For example, CONFIG_LOAD_FIT is able to load images that CONFIG_LOAD_FIT_FULL won't. This is a direct result of the incompatible syntax between the two implementations. Ideally, these changes would enable "simple" FIT to be a subset of the "full" fit implementation, and share most code. These changes should also eliminate the need for falcon mode (although we are not advocating for the removal of falcon mode at this time). ** Description of changes * The "configurations" node is now mandatory Guessing how to load components based on their "os" and "type" invites confusion and superfluous heuristics. Instead, require each FIT image to be explicit on how components should be loaded. * Eliminate "ramdisk", "setup", "standalone", and "fpga" properties Having too many special purpose properties requires special-casing FIT loading code. When a special property can be handled by another property, it is redundant. - A "ramdisk" is identical to a loadable. Thus ramdisk images should be placed under "loadables". - A "setup" node can be achieved by using a "kernel" or "firmware" property instead. - "standalone" is used for u-boot nodes. The correct property to use in this case is "firmware". - "fpga" is a loadable * Prioritize control between "firmware" and "kernel" "firmware" and "kernel" are special nodes in that control is passed to the "entry-point" of the image. Both can be present, for example, an OP-TEE firmware with a linux kernel. When both are present, control is passed to the "firmware" image. ** Further generalizations (not included herein) The "firmware" and "kernel" properties could be generalized as a "next-boot-stage", or similar name. This "next" stage would be special in that it is both executable, and is the stage that is passed control. For example, "next-stage" could be an op-tee image, with linux as a loadable, or a u-boot image. Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
2020-12-15 19:15:43 +00:00
'#address-cells' property of the root node.
Mandatory for types: "firmware", and "kernel".
- load : load address, address size is determined by '#address-cells'
doc: FIT image: Clarify format and simplify syntax ** Introduction There are currently four ways to load an OS image with u-boot 1. SPL -> u-boot -> bootm 2. SPL blue falcon mode 3. "Basic" FIT image (CONFIG_LOAD_FIT) 4. "Full-featured" FIT image (CONFIG_LOAD_FIT_FULL) These four code paths were developed independently, and share very little code. (3) and (4), behave very differently, are littered with special cases. They even have different DTS syntax and properties. The cause of this divergence is that the FIT format specification leaves a number of things open to interpretation. The purpose of this change is to enable the reduction of code size, duplication, and complexity by updating and streamlining the FIT format. We are only marginally concerned with backwards compatibility, because we don't have inter-compatibility. For example, CONFIG_LOAD_FIT is able to load images that CONFIG_LOAD_FIT_FULL won't. This is a direct result of the incompatible syntax between the two implementations. Ideally, these changes would enable "simple" FIT to be a subset of the "full" fit implementation, and share most code. These changes should also eliminate the need for falcon mode (although we are not advocating for the removal of falcon mode at this time). ** Description of changes * The "configurations" node is now mandatory Guessing how to load components based on their "os" and "type" invites confusion and superfluous heuristics. Instead, require each FIT image to be explicit on how components should be loaded. * Eliminate "ramdisk", "setup", "standalone", and "fpga" properties Having too many special purpose properties requires special-casing FIT loading code. When a special property can be handled by another property, it is redundant. - A "ramdisk" is identical to a loadable. Thus ramdisk images should be placed under "loadables". - A "setup" node can be achieved by using a "kernel" or "firmware" property instead. - "standalone" is used for u-boot nodes. The correct property to use in this case is "firmware". - "fpga" is a loadable * Prioritize control between "firmware" and "kernel" "firmware" and "kernel" are special nodes in that control is passed to the "entry-point" of the image. Both can be present, for example, an OP-TEE firmware with a linux kernel. When both are present, control is passed to the "firmware" image. ** Further generalizations (not included herein) The "firmware" and "kernel" properties could be generalized as a "next-boot-stage", or similar name. This "next" stage would be special in that it is both executable, and is the stage that is passed control. For example, "next-stage" could be an op-tee image, with linux as a loadable, or a u-boot image. Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
2020-12-15 19:15:43 +00:00
property of the root node.
Mandatory for types: "firmware", and "kernel".
- compatible : compatible method for loading image.
Mandatory for types: "fpga", and images that do not specify a load address.
Supported compatible methods:
"u-boot,fpga-legacy" - the generic fpga loading routine.
"u-boot,zynqmp-fpga-ddrauth" - signed non-encrypted FPGA bitstream for
Xilinx Zynq UltraScale+ (ZymqMP) device.
"u-boot,zynqmp-fpga-enc" - encrypted FPGA bitstream for Xilinx Zynq
UltraScale+ (ZynqMP) device.
- phase : U-Boot phase for which the image is intended.
"spl" - image is an SPL image
"u-boot" - image is a U-Boot image
Optional nodes:
- hash-1 : Each hash sub-node represents separate hash or checksum
calculated for node's data according to specified algorithm.
5) Hash nodes
-------------
o hash-1
|- algo = "hash or checksum algorithm name"
|- value = [hash or checksum value]
Mandatory properties:
- algo : Algorithm name, supported are "crc32", "md5" and "sha1".
- value : Actual checksum or hash value, correspondingly 4, 16 or 20 bytes
long.
6) '/configurations' node
-------------------------
doc: FIT image: Clarify format and simplify syntax ** Introduction There are currently four ways to load an OS image with u-boot 1. SPL -> u-boot -> bootm 2. SPL blue falcon mode 3. "Basic" FIT image (CONFIG_LOAD_FIT) 4. "Full-featured" FIT image (CONFIG_LOAD_FIT_FULL) These four code paths were developed independently, and share very little code. (3) and (4), behave very differently, are littered with special cases. They even have different DTS syntax and properties. The cause of this divergence is that the FIT format specification leaves a number of things open to interpretation. The purpose of this change is to enable the reduction of code size, duplication, and complexity by updating and streamlining the FIT format. We are only marginally concerned with backwards compatibility, because we don't have inter-compatibility. For example, CONFIG_LOAD_FIT is able to load images that CONFIG_LOAD_FIT_FULL won't. This is a direct result of the incompatible syntax between the two implementations. Ideally, these changes would enable "simple" FIT to be a subset of the "full" fit implementation, and share most code. These changes should also eliminate the need for falcon mode (although we are not advocating for the removal of falcon mode at this time). ** Description of changes * The "configurations" node is now mandatory Guessing how to load components based on their "os" and "type" invites confusion and superfluous heuristics. Instead, require each FIT image to be explicit on how components should be loaded. * Eliminate "ramdisk", "setup", "standalone", and "fpga" properties Having too many special purpose properties requires special-casing FIT loading code. When a special property can be handled by another property, it is redundant. - A "ramdisk" is identical to a loadable. Thus ramdisk images should be placed under "loadables". - A "setup" node can be achieved by using a "kernel" or "firmware" property instead. - "standalone" is used for u-boot nodes. The correct property to use in this case is "firmware". - "fpga" is a loadable * Prioritize control between "firmware" and "kernel" "firmware" and "kernel" are special nodes in that control is passed to the "entry-point" of the image. Both can be present, for example, an OP-TEE firmware with a linux kernel. When both are present, control is passed to the "firmware" image. ** Further generalizations (not included herein) The "firmware" and "kernel" properties could be generalized as a "next-boot-stage", or similar name. This "next" stage would be special in that it is both executable, and is the stage that is passed control. For example, "next-stage" could be an op-tee image, with linux as a loadable, or a u-boot image. Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
2020-12-15 19:15:43 +00:00
The 'configurations' node creates convenient, labeled boot configurations,
which combine together kernel images with their ramdisks and fdt blobs.
The 'configurations' node has has the following structure:
o configurations
|- default = "default configuration sub-node unit name"
|
o config-1 {...}
o config-2 {...}
...
Optional property:
- default : Selects one of the configuration sub-nodes as a default
configuration.
Mandatory nodes:
- configuration-sub-node-unit-name : At least one of the configuration
sub-nodes is required.
7) Configuration nodes
----------------------
Each configuration has the following structure:
o config-1
|- description = "configuration description"
|- kernel = "kernel sub-node unit name"
|- fdt = "fdt sub-node unit-name" [, "fdt overlay sub-node unit-name", ...]
|- loadables = "loadables sub-node unit-name"
|- compatible = "vendor,board-style device tree compatible string"
Mandatory properties:
- description : Textual configuration description.
doc: FIT image: Clarify format and simplify syntax ** Introduction There are currently four ways to load an OS image with u-boot 1. SPL -> u-boot -> bootm 2. SPL blue falcon mode 3. "Basic" FIT image (CONFIG_LOAD_FIT) 4. "Full-featured" FIT image (CONFIG_LOAD_FIT_FULL) These four code paths were developed independently, and share very little code. (3) and (4), behave very differently, are littered with special cases. They even have different DTS syntax and properties. The cause of this divergence is that the FIT format specification leaves a number of things open to interpretation. The purpose of this change is to enable the reduction of code size, duplication, and complexity by updating and streamlining the FIT format. We are only marginally concerned with backwards compatibility, because we don't have inter-compatibility. For example, CONFIG_LOAD_FIT is able to load images that CONFIG_LOAD_FIT_FULL won't. This is a direct result of the incompatible syntax between the two implementations. Ideally, these changes would enable "simple" FIT to be a subset of the "full" fit implementation, and share most code. These changes should also eliminate the need for falcon mode (although we are not advocating for the removal of falcon mode at this time). ** Description of changes * The "configurations" node is now mandatory Guessing how to load components based on their "os" and "type" invites confusion and superfluous heuristics. Instead, require each FIT image to be explicit on how components should be loaded. * Eliminate "ramdisk", "setup", "standalone", and "fpga" properties Having too many special purpose properties requires special-casing FIT loading code. When a special property can be handled by another property, it is redundant. - A "ramdisk" is identical to a loadable. Thus ramdisk images should be placed under "loadables". - A "setup" node can be achieved by using a "kernel" or "firmware" property instead. - "standalone" is used for u-boot nodes. The correct property to use in this case is "firmware". - "fpga" is a loadable * Prioritize control between "firmware" and "kernel" "firmware" and "kernel" are special nodes in that control is passed to the "entry-point" of the image. Both can be present, for example, an OP-TEE firmware with a linux kernel. When both are present, control is passed to the "firmware" image. ** Further generalizations (not included herein) The "firmware" and "kernel" properties could be generalized as a "next-boot-stage", or similar name. This "next" stage would be special in that it is both executable, and is the stage that is passed control. For example, "next-stage" could be an op-tee image, with linux as a loadable, or a u-boot image. Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
2020-12-15 19:15:43 +00:00
- kernel or firmware: Unit name of the corresponding kernel or firmware
(u-boot, op-tee, etc) image. If both "kernel" and "firmware" are specified,
control is passed to the firmware image.
Optional properties:
- fdt : Unit name of the corresponding fdt blob (component image node of a
"fdt type"). Additional fdt overlay nodes can be supplied which signify
that the resulting device tree blob is generated by the first base fdt
blob with all subsequent overlays applied.
- fpga : Unit name of the corresponding fpga bitstream blob
(component image node of a "fpga type").
- loadables : Unit name containing a list of additional binaries to be
loaded at their given locations. "loadables" is a comma-separated list
of strings. U-Boot will load each binary at its given start-address and
may optionally invoke additional post-processing steps on this binary based
on its component image node type.
- compatible : The root compatible string of the U-Boot device tree that
this configuration shall automatically match when CONFIG_FIT_BEST_MATCH is
enabled. If this property is not provided, the compatible string will be
extracted from the fdt blob instead. This is only possible if the fdt is
not compressed, so images with compressed fdts that want to use compatible
string matching must always provide this property.
The FDT blob is required to properly boot FDT based kernel, so the minimal
configuration for 2.6 FDT kernel is (kernel, fdt) pair.
Older, 2.4 kernel and 2.6 non-FDT kernel do not use FDT blob, in such cases
'struct bd_info' must be passed instead of FDT blob, thus fdt property *must
not* be specified in a configuration node.
8) External data
----------------
The above format shows a 'data' property which holds the data for each image.
It is also possible for this data to reside outside the FIT itself. This
allows the FIT to be quite small, so that it can be loaded and scanned
without loading a large amount of data. Then when an image is needed it can
be loaded from an external source.
In this case the 'data' property is omitted. Instead you can use:
- data-offset : offset of the data in a separate image store. The image
store is placed immediately after the last byte of the device tree binary,
aligned to a 4-byte boundary.
- data-size : size of the data in bytes
The 'data-offset' property can be substituted with 'data-position', which
defines an absolute position or address as the offset. This is helpful when
booting U-Boot proper before performing relocation. Pass '-p [offset]' to
mkimage to enable 'data-position'.
Normal kernel FIT image has data embedded within FIT structure. U-Boot image
for SPL boot has external data. Existence of 'data-offset' can be used to
identify which format is used.
For FIT image with external data, it would be better to align each blob of data
to block(512 byte) for block device, so that we don't need to do the copy when
read the image data in SPL. Pass '-B 0x200' to mkimage to align the FIT
structure and data to 512 byte, other values available for other align size.
9) Examples
-----------
Please see doc/uImage.FIT/*.its for actual image source files.