Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 1 | .. SPDX-License-Identifier: GPL-2.0+ |
| 2 | .. sectionauthor:: Simon Glass <sjg@chromium.org> |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 3 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 4 | Live Device Tree |
| 5 | ================ |
| 6 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 7 | |
| 8 | Introduction |
| 9 | ------------ |
| 10 | |
| 11 | Traditionally U-Boot has used a 'flat' device tree. This means that it |
| 12 | reads directly from the device tree binary structure. It is called a flat |
| 13 | device tree because nodes are listed one after the other, with the |
| 14 | hierarchy detected by tags in the format. |
| 15 | |
| 16 | This document describes U-Boot's support for a 'live' device tree, meaning |
| 17 | that the tree is loaded into a hierarchical data structure within U-Boot. |
| 18 | |
| 19 | |
| 20 | Motivation |
| 21 | ---------- |
| 22 | |
| 23 | The flat device tree has several advantages: |
| 24 | |
| 25 | - it is the format produced by the device tree compiler, so no translation |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 26 | is needed |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 27 | |
| 28 | - it is fairly compact (e.g. there is no need for pointers) |
| 29 | |
| 30 | - it is accessed by the libfdt library, which is well tested and stable |
| 31 | |
| 32 | |
| 33 | However the flat device tree does have some limitations. Adding new |
| 34 | properties can involve copying large amounts of data around to make room. |
| 35 | The overall tree has a fixed maximum size so sometimes the tree must be |
| 36 | rebuilt in a new location to create more space. Even if not adding new |
| 37 | properties or nodes, scanning the tree can be slow. For example, finding |
| 38 | the parent of a node is a slow process. Reading from nodes involves a |
| 39 | small amount parsing which takes a little time. |
| 40 | |
| 41 | Driver model scans the entire device tree sequentially on start-up which |
| 42 | avoids the worst of the flat tree's limitations. But if the tree is to be |
| 43 | modified at run-time, a live tree is much faster. Even if no modification |
| 44 | is necessary, parsing the tree once and using a live tree from then on |
| 45 | seems to save a little time. |
| 46 | |
| 47 | |
| 48 | Implementation |
| 49 | -------------- |
| 50 | |
| 51 | In U-Boot a live device tree ('livetree') is currently supported only |
| 52 | after relocation. Therefore we need a mechanism to specify a device |
| 53 | tree node regardless of whether it is in the flat tree or livetree. |
| 54 | |
| 55 | The 'ofnode' type provides this. An ofnode can point to either a flat tree |
| 56 | node (when the live tree node is not yet set up) or a livetree node. The |
| 57 | caller of an ofnode function does not need to worry about these details. |
| 58 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 59 | The main users of the information in a device tree are drivers. These have |
| 60 | a 'struct udevice \*' which is attached to a device tree node. Therefore it |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 61 | makes sense to be able to read device tree properties using the |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 62 | 'struct udevice \*', rather than having to obtain the ofnode first. |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 63 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 64 | The 'dev_read\_...()' interface provides this. It allows properties to be |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 65 | easily read from the device tree using only a device pointer. Under the |
| 66 | hood it uses ofnode so it works with both flat and live device trees. |
| 67 | |
| 68 | |
| 69 | Enabling livetree |
| 70 | ----------------- |
| 71 | |
| 72 | CONFIG_OF_LIVE enables livetree. When this option is enabled, the flat |
| 73 | tree will be used in SPL and before relocation in U-Boot proper. Just |
| 74 | before relocation a livetree is built, and this is used for U-Boot proper |
| 75 | after relocation. |
| 76 | |
| 77 | Most checks for livetree use CONFIG_IS_ENABLED(OF_LIVE). This means that |
| 78 | for SPL, the CONFIG_SPL_OF_LIVE option is checked. At present this does |
| 79 | not exist, since SPL does not support livetree. |
| 80 | |
| 81 | |
| 82 | Porting drivers |
| 83 | --------------- |
| 84 | |
| 85 | Many existing drivers use the fdtdec interface to read device tree |
| 86 | properties. This only works with a flat device tree. The drivers should be |
| 87 | converted to use the dev_read_() interface. |
| 88 | |
| 89 | For example, the old code may be like this: |
| 90 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 91 | .. code-block:: c |
| 92 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 93 | struct udevice *bus; |
| 94 | const void *blob = gd->fdt_blob; |
| 95 | int node = dev_of_offset(bus); |
| 96 | |
| 97 | i2c_bus->regs = (struct i2c_ctlr *)devfdt_get_addr(dev); |
| 98 | plat->frequency = fdtdec_get_int(blob, node, "spi-max-frequency", 500000); |
| 99 | |
| 100 | The new code is: |
| 101 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 102 | .. code-block:: c |
| 103 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 104 | struct udevice *bus; |
| 105 | |
| 106 | i2c_bus->regs = (struct i2c_ctlr *)dev_read_addr(dev); |
| 107 | plat->frequency = dev_read_u32_default(bus, "spi-max-frequency", 500000); |
| 108 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 109 | The dev_read\_...() interface is more convenient and works with both the |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 110 | flat and live device trees. See include/dm/read.h for a list of functions. |
| 111 | |
| 112 | Where properties must be read from sub-nodes or other nodes, you must fall |
| 113 | back to using ofnode. For example, for old code like this: |
| 114 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 115 | .. code-block:: c |
| 116 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 117 | const void *blob = gd->fdt_blob; |
| 118 | int subnode; |
| 119 | |
| 120 | fdt_for_each_subnode(subnode, blob, dev_of_offset(dev)) { |
| 121 | freq = fdtdec_get_int(blob, node, "spi-max-frequency", 500000); |
| 122 | ... |
| 123 | } |
| 124 | |
| 125 | you should use: |
| 126 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 127 | .. code-block:: c |
| 128 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 129 | ofnode subnode; |
| 130 | |
| 131 | ofnode_for_each_subnode(subnode, dev_ofnode(dev)) { |
| 132 | freq = ofnode_read_u32(node, "spi-max-frequency", 500000); |
| 133 | ... |
| 134 | } |
| 135 | |
| 136 | |
| 137 | Useful ofnode functions |
| 138 | ----------------------- |
| 139 | |
| 140 | The internal data structures of the livetree are defined in include/dm/of.h : |
| 141 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 142 | :struct device_node: holds information about a device tree node |
| 143 | :struct property: holds information about a property within a node |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 144 | |
| 145 | Nodes have pointers to their first property, their parent, their first child |
| 146 | and their sibling. This allows nodes to be linked together in a hierarchical |
| 147 | tree. |
| 148 | |
| 149 | Properties have pointers to the next property. This allows all properties of |
| 150 | a node to be linked together in a chain. |
| 151 | |
| 152 | It should not be necessary to use these data structures in normal code. In |
| 153 | particular, you should refrain from using functions which access the livetree |
| 154 | directly, such as of_read_u32(). Use ofnode functions instead, to allow your |
| 155 | code to work with a flat tree also. |
| 156 | |
| 157 | Some conversion functions are used internally. Generally these are not needed |
| 158 | for driver code. Note that they will not work if called in the wrong context. |
| 159 | For example it is invalid to call ofnode_to_no() when a flat tree is being |
| 160 | used. Similarly it is not possible to call ofnode_to_offset() on a livetree |
| 161 | node. |
| 162 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 163 | ofnode_to_np(): |
| 164 | converts ofnode to struct device_node * |
| 165 | ofnode_to_offset(): |
| 166 | converts ofnode to offset |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 167 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 168 | no_to_ofnode(): |
| 169 | converts node pointer to ofnode |
| 170 | offset_to_ofnode(): |
| 171 | converts offset to ofnode |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 172 | |
| 173 | |
| 174 | Other useful functions: |
| 175 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 176 | of_live_active(): |
| 177 | returns true if livetree is in use, false if flat tree |
| 178 | ofnode_valid(): |
| 179 | return true if a given node is valid |
| 180 | ofnode_is_np(): |
| 181 | returns true if a given node is a livetree node |
| 182 | ofnode_equal(): |
| 183 | compares two ofnodes |
| 184 | ofnode_null(): |
| 185 | returns a null ofnode (for which ofnode_valid() returns false) |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 186 | |
| 187 | |
| 188 | Phandles |
| 189 | -------- |
| 190 | |
| 191 | There is full phandle support for live tree. All functions make use of |
| 192 | struct ofnode_phandle_args, which has an ofnode within it. This supports both |
| 193 | livetree and flat tree transparently. See for example |
| 194 | ofnode_parse_phandle_with_args(). |
| 195 | |
| 196 | |
| 197 | Reading addresses |
| 198 | ----------------- |
| 199 | |
| 200 | You should use dev_read_addr() and friends to read addresses from device-tree |
| 201 | nodes. |
| 202 | |
| 203 | |
| 204 | fdtdec |
| 205 | ------ |
| 206 | |
| 207 | The existing fdtdec interface will eventually be retired. Please try to avoid |
| 208 | using it in new code. |
| 209 | |
| 210 | |
| 211 | Modifying the livetree |
| 212 | ---------------------- |
| 213 | |
Simon Glass | 1817f71 | 2022-07-30 15:52:07 -0600 | [diff] [blame] | 214 | This is supported in a limited way, with ofnode_write_prop() and related |
| 215 | functions. |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 216 | |
Simon Glass | 1817f71 | 2022-07-30 15:52:07 -0600 | [diff] [blame] | 217 | The unflattening algorithm results in a single block of memory being |
| 218 | allocated for the whole tree. When writing new properties, these are |
| 219 | allocated new memory outside that block. When the block is freed, the |
| 220 | allocated properties remain. This can result in a memory leak. |
| 221 | |
| 222 | The solution to this leak would be to add a flag for properties (and nodes when |
| 223 | support is provided for adding those) that indicates that they should be |
| 224 | freed. Then the tree can be scanned for these 'separately allocated' nodes and |
| 225 | properties before freeing the memory block. |
| 226 | |
Simon Glass | 2f0cb8f | 2022-07-30 15:52:12 -0600 | [diff] [blame] | 227 | The ofnode_write\_...() functions also support writing to the flat tree. Care |
| 228 | should be taken however, since this can change the position of node names and |
| 229 | properties in the flat tree, thus affecting the live tree. Generally this does |
| 230 | not matter, since when we fire up the live tree we don't ever use the flat tree |
| 231 | again. But in the case of tests, this can cause a problem. |
| 232 | |
| 233 | The sandbox tests typically run with OF_LIVE enabled but with the actual live |
| 234 | tree either present or absent. This is to make sure that the flat tree functions |
| 235 | work correctly even with OF_LIVE is enabled. But if a test modifies the flat |
| 236 | device tree, then the live tree can become invalid. Any live tree tests that run |
| 237 | after that point will use a corrupted tree, e.g. with an incorrect property name |
Simon Glass | 3ba7675 | 2022-09-06 20:27:05 -0600 | [diff] [blame] | 238 | or worse. To deal with this we take a copy of the device tree and restore it |
| 239 | after any test that modifies it. Note that this copy is not made on other |
| 240 | boards, only sandbox. |
Simon Glass | 2f0cb8f | 2022-07-30 15:52:12 -0600 | [diff] [blame] | 241 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 242 | |
Simon Glass | ef75c59 | 2022-07-30 15:52:08 -0600 | [diff] [blame] | 243 | Multiple livetrees |
| 244 | ------------------ |
| 245 | |
| 246 | The livetree implementation was originally designed for use with the control |
| 247 | FDT. This means that the FDT fix-ups (ft_board_setup() and the like, must use |
| 248 | a flat tree. |
| 249 | |
| 250 | It would be helpful to use livetree for fixups, since adding a lot of nodes and |
| 251 | properties would involve less memory copying and be more efficient. As a step |
| 252 | towards this, an `oftree` type has been introduced. It is normally set to |
Simon Glass | cb13a1b | 2022-09-06 20:27:26 -0600 | [diff] [blame] | 253 | oftree_default() but can be set to other values using oftree_from_fdt(). |
| 254 | So long as OF_LIVE is disabled, it is possible to do fixups using the ofnode |
| 255 | interface. The OF_LIVE support required addition of the flattening step at the |
| 256 | end. |
Simon Glass | ef75c59 | 2022-07-30 15:52:08 -0600 | [diff] [blame] | 257 | |
Simon Glass | e6a211c | 2022-10-11 09:47:19 -0600 | [diff] [blame] | 258 | See dm_test_ofnode_root() for some examples. The oftree_from_fdt() function |
Simon Glass | cb13a1b | 2022-09-06 20:27:26 -0600 | [diff] [blame] | 259 | causes a flat device tree to be 'registered' such that it can be used by the |
| 260 | ofnode interface. |
Simon Glass | ef75c59 | 2022-07-30 15:52:08 -0600 | [diff] [blame] | 261 | |
| 262 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 263 | Internal implementation |
| 264 | ----------------------- |
| 265 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 266 | The dev_read\_...() functions have two implementations. When |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 267 | CONFIG_DM_DEV_READ_INLINE is enabled, these functions simply call the ofnode |
| 268 | functions directly. This is useful when livetree is not enabled. The ofnode |
| 269 | functions call ofnode_is_np(node) which will always return false if livetree |
| 270 | is disabled, just falling back to flat tree code. |
| 271 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 272 | This optimisation means that without livetree enabled, the dev_read\_...() and |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 273 | ofnode interfaces do not noticeably add to code size. |
| 274 | |
| 275 | The CONFIG_DM_DEV_READ_INLINE option defaults to enabled when livetree is |
| 276 | disabled. |
| 277 | |
| 278 | Most livetree code comes directly from Linux and is modified as little as |
| 279 | possible. This is deliberate since this code is fairly stable and does what |
| 280 | we want. Some features (such as get/put) are not supported. Internal macros |
| 281 | take care of removing these features silently. |
| 282 | |
| 283 | Within the of_access.c file there are pointers to the alias node, the chosen |
| 284 | node and the stdout-path alias. |
| 285 | |
| 286 | |
| 287 | Errors |
| 288 | ------ |
| 289 | |
| 290 | With a flat device tree, libfdt errors are returned (e.g. -FDT_ERR_NOTFOUND). |
| 291 | For livetree normal 'errno' errors are returned (e.g. -ENOTFOUND). At present |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 292 | the ofnode and dev_read\_...() functions return either one or other type of |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 293 | error. This is clearly not desirable. Once tests are added for all the |
| 294 | functions this can be tidied up. |
| 295 | |
| 296 | |
| 297 | Adding new access functions |
| 298 | --------------------------- |
| 299 | |
| 300 | Adding a new function for device-tree access involves the following steps: |
| 301 | |
| 302 | - Add two dev_read() functions: |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 303 | - inline version in the read.h header file, which calls an ofnode function |
| 304 | - standard version in the read.c file (or perhaps another file), which |
| 305 | also calls an ofnode function |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 306 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 307 | The implementations of these functions can be the same. The purpose |
| 308 | of the inline version is purely to reduce code size impact. |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 309 | |
| 310 | - Add an ofnode function. This should call ofnode_is_np() to work out |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 311 | whether a livetree or flat tree is used. For the livetree it should |
| 312 | call an of\_...() function. For the flat tree it should call an |
| 313 | fdt\_...() function. The livetree version will be optimised out at |
| 314 | compile time if livetree is not enabled. |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 315 | |
Bin Meng | 5a59347 | 2019-07-18 00:33:54 -0700 | [diff] [blame] | 316 | - Add an of\_...() function for the livetree implementation. If a similar |
| 317 | function is available in Linux, the implementation should be taken |
| 318 | from there and modified as little as possible (generally not at all). |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 319 | |
| 320 | |
| 321 | Future work |
| 322 | ----------- |
| 323 | |
Simon Glass | 9e7a42a | 2022-09-06 20:27:30 -0600 | [diff] [blame] | 324 | Live tree support was introduced in U-Boot 2017.07. Some possible enhancements |
| 325 | are: |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 326 | |
Simon Glass | a7e7e08 | 2017-08-31 05:59:55 -0600 | [diff] [blame] | 327 | - support for livetree in SPL and before relocation (if desired) |
Simon Glass | 17abed0 | 2022-09-06 20:27:32 -0600 | [diff] [blame] | 328 | - freeing leaked memory caused by writing new nodes / property values to the |
| 329 | livetree (ofnode_write_prop()) |