1Driver Model 2============ 3 4This README contains high-level information about driver model, a unified 5way of declaring and accessing drivers in U-Boot. The original work was done 6by: 7 8 Marek Vasut <marex@denx.de> 9 Pavel Herrmann <morpheus.ibis@gmail.com> 10 Viktor Křivák <viktor.krivak@gmail.com> 11 Tomas Hlavacek <tmshlvck@gmail.com> 12 13This has been both simplified and extended into the current implementation 14by: 15 16 Simon Glass <sjg@chromium.org> 17 18 19Terminology 20----------- 21 22Uclass - a group of devices which operate in the same way. A uclass provides 23 a way of accessing individual devices within the group, but always 24 using the same interface. For example a GPIO uclass provides 25 operations for get/set value. An I2C uclass may have 10 I2C ports, 26 4 with one driver, and 6 with another. 27 28Driver - some code which talks to a peripheral and presents a higher-level 29 interface to it. 30 31Device - an instance of a driver, tied to a particular port or peripheral. 32 33 34How to try it 35------------- 36 37Build U-Boot sandbox and run it: 38 39 make sandbox_config 40 make 41 ./u-boot 42 43 (type 'reset' to exit U-Boot) 44 45 46There is a uclass called 'demo'. This uclass handles 47saying hello, and reporting its status. There are two drivers in this 48uclass: 49 50 - simple: Just prints a message for hello, doesn't implement status 51 - shape: Prints shapes and reports number of characters printed as status 52 53The demo class is pretty simple, but not trivial. The intention is that it 54can be used for testing, so it will implement all driver model features and 55provide good code coverage of them. It does have multiple drivers, it 56handles parameter data and platdata (data which tells the driver how 57to operate on a particular platform) and it uses private driver data. 58 59To try it, see the example session below: 60 61=>demo hello 1 62Hello '@' from 07981110: red 4 63=>demo status 2 64Status: 0 65=>demo hello 2 66g 67r@ 68e@@ 69e@@@ 70n@@@@ 71g@@@@@ 72=>demo status 2 73Status: 21 74=>demo hello 4 ^ 75 y^^^ 76 e^^^^^ 77l^^^^^^^ 78l^^^^^^^ 79 o^^^^^ 80 w^^^ 81=>demo status 4 82Status: 36 83=> 84 85 86Running the tests 87----------------- 88 89The intent with driver model is that the core portion has 100% test coverage 90in sandbox, and every uclass has its own test. As a move towards this, tests 91are provided in test/dm. To run them, try: 92 93 ./test/dm/test-dm.sh 94 95You should see something like this: 96 97 <...U-Boot banner...> 98 Running 12 driver model tests 99 Test: dm_test_autobind 100 Test: dm_test_autoprobe 101 Test: dm_test_children 102 Test: dm_test_fdt 103 Test: dm_test_gpio 104 sandbox_gpio: sb_gpio_get_value: error: offset 4 not reserved 105 Test: dm_test_leak 106 Warning: Please add '#define DEBUG' to the top of common/dlmalloc.c 107 Warning: Please add '#define DEBUG' to the top of common/dlmalloc.c 108 Test: dm_test_lifecycle 109 Test: dm_test_operations 110 Test: dm_test_ordering 111 Test: dm_test_platdata 112 Test: dm_test_remove 113 Test: dm_test_uclass 114 Failures: 0 115 116(You can add '#define DEBUG' as suggested to check for memory leaks) 117 118 119What is going on? 120----------------- 121 122Let's start at the top. The demo command is in common/cmd_demo.c. It does 123the usual command processing and then: 124 125 struct udevice *demo_dev; 126 127 ret = uclass_get_device(UCLASS_DEMO, devnum, &demo_dev); 128 129UCLASS_DEMO means the class of devices which implement 'demo'. Other 130classes might be MMC, or GPIO, hashing or serial. The idea is that the 131devices in the class all share a particular way of working. The class 132presents a unified view of all these devices to U-Boot. 133 134This function looks up a device for the demo uclass. Given a device 135number we can find the device because all devices have registered with 136the UCLASS_DEMO uclass. 137 138The device is automatically activated ready for use by uclass_get_device(). 139 140Now that we have the device we can do things like: 141 142 return demo_hello(demo_dev, ch); 143 144This function is in the demo uclass. It takes care of calling the 'hello' 145method of the relevant driver. Bearing in mind that there are two drivers, 146this particular device may use one or other of them. 147 148The code for demo_hello() is in drivers/demo/demo-uclass.c: 149 150int demo_hello(struct udevice *dev, int ch) 151{ 152 const struct demo_ops *ops = device_get_ops(dev); 153 154 if (!ops->hello) 155 return -ENOSYS; 156 157 return ops->hello(dev, ch); 158} 159 160As you can see it just calls the relevant driver method. One of these is 161in drivers/demo/demo-simple.c: 162 163static int simple_hello(struct udevice *dev, int ch) 164{ 165 const struct dm_demo_pdata *pdata = dev_get_platdata(dev); 166 167 printf("Hello from %08x: %s %d\n", map_to_sysmem(dev), 168 pdata->colour, pdata->sides); 169 170 return 0; 171} 172 173 174So that is a trip from top (command execution) to bottom (driver action) 175but it leaves a lot of topics to address. 176 177 178Declaring Drivers 179----------------- 180 181A driver declaration looks something like this (see 182drivers/demo/demo-shape.c): 183 184static const struct demo_ops shape_ops = { 185 .hello = shape_hello, 186 .status = shape_status, 187}; 188 189U_BOOT_DRIVER(demo_shape_drv) = { 190 .name = "demo_shape_drv", 191 .id = UCLASS_DEMO, 192 .ops = &shape_ops, 193 .priv_data_size = sizeof(struct shape_data), 194}; 195 196 197This driver has two methods (hello and status) and requires a bit of 198private data (accessible through dev_get_priv(dev) once the driver has 199been probed). It is a member of UCLASS_DEMO so will register itself 200there. 201 202In U_BOOT_DRIVER it is also possible to specify special methods for bind 203and unbind, and these are called at appropriate times. For many drivers 204it is hoped that only 'probe' and 'remove' will be needed. 205 206The U_BOOT_DRIVER macro creates a data structure accessible from C, 207so driver model can find the drivers that are available. 208 209The methods a device can provide are documented in the device.h header. 210Briefly, they are: 211 212 bind - make the driver model aware of a device (bind it to its driver) 213 unbind - make the driver model forget the device 214 ofdata_to_platdata - convert device tree data to platdata - see later 215 probe - make a device ready for use 216 remove - remove a device so it cannot be used until probed again 217 218The sequence to get a device to work is bind, ofdata_to_platdata (if using 219device tree) and probe. 220 221 222Platform Data 223------------- 224 225Platform data is like Linux platform data, if you are familiar with that. 226It provides the board-specific information to start up a device. 227 228Why is this information not just stored in the device driver itself? The 229idea is that the device driver is generic, and can in principle operate on 230any board that has that type of device. For example, with modern 231highly-complex SoCs it is common for the IP to come from an IP vendor, and 232therefore (for example) the MMC controller may be the same on chips from 233different vendors. It makes no sense to write independent drivers for the 234MMC controller on each vendor's SoC, when they are all almost the same. 235Similarly, we may have 6 UARTs in an SoC, all of which are mostly the same, 236but lie at different addresses in the address space. 237 238Using the UART example, we have a single driver and it is instantiated 6 239times by supplying 6 lots of platform data. Each lot of platform data 240gives the driver name and a pointer to a structure containing information 241about this instance - e.g. the address of the register space. It may be that 242one of the UARTS supports RS-485 operation - this can be added as a flag in 243the platform data, which is set for this one port and clear for the rest. 244 245Think of your driver as a generic piece of code which knows how to talk to 246a device, but needs to know where it is, any variant/option information and 247so on. Platform data provides this link between the generic piece of code 248and the specific way it is bound on a particular board. 249 250Examples of platform data include: 251 252 - The base address of the IP block's register space 253 - Configuration options, like: 254 - the SPI polarity and maximum speed for a SPI controller 255 - the I2C speed to use for an I2C device 256 - the number of GPIOs available in a GPIO device 257 258Where does the platform data come from? It is either held in a structure 259which is compiled into U-Boot, or it can be parsed from the Device Tree 260(see 'Device Tree' below). 261 262For an example of how it can be compiled in, see demo-pdata.c which 263sets up a table of driver names and their associated platform data. 264The data can be interpreted by the drivers however they like - it is 265basically a communication scheme between the board-specific code and 266the generic drivers, which are intended to work on any board. 267 268Drivers can access their data via dev->info->platdata. Here is 269the declaration for the platform data, which would normally appear 270in the board file. 271 272 static const struct dm_demo_cdata red_square = { 273 .colour = "red", 274 .sides = 4. 275 }; 276 static const struct driver_info info[] = { 277 { 278 .name = "demo_shape_drv", 279 .platdata = &red_square, 280 }, 281 }; 282 283 demo1 = driver_bind(root, &info[0]); 284 285 286Device Tree 287----------- 288 289While platdata is useful, a more flexible way of providing device data is 290by using device tree. With device tree we replace the above code with the 291following device tree fragment: 292 293 red-square { 294 compatible = "demo-shape"; 295 colour = "red"; 296 sides = <4>; 297 }; 298 299This means that instead of having lots of U_BOOT_DEVICE() declarations in 300the board file, we put these in the device tree. This approach allows a lot 301more generality, since the same board file can support many types of boards 302(e,g. with the same SoC) just by using different device trees. An added 303benefit is that the Linux device tree can be used, thus further simplifying 304the task of board-bring up either for U-Boot or Linux devs (whoever gets to 305the board first!). 306 307The easiest way to make this work it to add a few members to the driver: 308 309 .platdata_auto_alloc_size = sizeof(struct dm_test_pdata), 310 .ofdata_to_platdata = testfdt_ofdata_to_platdata, 311 312The 'auto_alloc' feature allowed space for the platdata to be allocated 313and zeroed before the driver's ofdata_to_platdata() method is called. The 314ofdata_to_platdata() method, which the driver write supplies, should parse 315the device tree node for this device and place it in dev->platdata. Thus 316when the probe method is called later (to set up the device ready for use) 317the platform data will be present. 318 319Note that both methods are optional. If you provide an ofdata_to_platdata 320method then it will be called first (during activation). If you provide a 321probe method it will be called next. See Driver Lifecycle below for more 322details. 323 324If you don't want to have the platdata automatically allocated then you 325can leave out platdata_auto_alloc_size. In this case you can use malloc 326in your ofdata_to_platdata (or probe) method to allocate the required memory, 327and you should free it in the remove method. 328 329 330Declaring Uclasses 331------------------ 332 333The demo uclass is declared like this: 334 335U_BOOT_CLASS(demo) = { 336 .id = UCLASS_DEMO, 337}; 338 339It is also possible to specify special methods for probe, etc. The uclass 340numbering comes from include/dm/uclass.h. To add a new uclass, add to the 341end of the enum there, then declare your uclass as above. 342 343 344Driver Lifecycle 345---------------- 346 347Here are the stages that a device goes through in driver model. Note that all 348methods mentioned here are optional - e.g. if there is no probe() method for 349a device then it will not be called. A simple device may have very few 350methods actually defined. 351 3521. Bind stage 353 354A device and its driver are bound using one of these two methods: 355 356 - Scan the U_BOOT_DEVICE() definitions. U-Boot It looks up the 357name specified by each, to find the appropriate driver. It then calls 358device_bind() to create a new device and bind' it to its driver. This will 359call the device's bind() method. 360 361 - Scan through the device tree definitions. U-Boot looks at top-level 362nodes in the the device tree. It looks at the compatible string in each node 363and uses the of_match part of the U_BOOT_DRIVER() structure to find the 364right driver for each node. It then calls device_bind() to bind the 365newly-created device to its driver (thereby creating a device structure). 366This will also call the device's bind() method. 367 368At this point all the devices are known, and bound to their drivers. There 369is a 'struct udevice' allocated for all devices. However, nothing has been 370activated (except for the root device). Each bound device that was created 371from a U_BOOT_DEVICE() declaration will hold the platdata pointer specified 372in that declaration. For a bound device created from the device tree, 373platdata will be NULL, but of_offset will be the offset of the device tree 374node that caused the device to be created. The uclass is set correctly for 375the device. 376 377The device's bind() method is permitted to perform simple actions, but 378should not scan the device tree node, not initialise hardware, nor set up 379structures or allocate memory. All of these tasks should be left for 380the probe() method. 381 382Note that compared to Linux, U-Boot's driver model has a separate step of 383probe/remove which is independent of bind/unbind. This is partly because in 384U-Boot it may be expensive to probe devices and we don't want to do it until 385they are needed, or perhaps until after relocation. 386 3872. Activation/probe 388 389When a device needs to be used, U-Boot activates it, by following these 390steps (see device_probe()): 391 392 a. If priv_auto_alloc_size is non-zero, then the device-private space 393 is allocated for the device and zeroed. It will be accessible as 394 dev->priv. The driver can put anything it likes in there, but should use 395 it for run-time information, not platform data (which should be static 396 and known before the device is probed). 397 398 b. If platdata_auto_alloc_size is non-zero, then the platform data space 399 is allocated. This is only useful for device tree operation, since 400 otherwise you would have to specific the platform data in the 401 U_BOOT_DEVICE() declaration. The space is allocated for the device and 402 zeroed. It will be accessible as dev->platdata. 403 404 c. If the device's uclass specifies a non-zero per_device_auto_alloc_size, 405 then this space is allocated and zeroed also. It is allocated for and 406 stored in the device, but it is uclass data. owned by the uclass driver. 407 It is possible for the device to access it. 408 409 d. All parent devices are probed. It is not possible to activate a device 410 unless its predecessors (all the way up to the root device) are activated. 411 This means (for example) that an I2C driver will require that its bus 412 be activated. 413 414 e. If the driver provides an ofdata_to_platdata() method, then this is 415 called to convert the device tree data into platform data. This should 416 do various calls like fdtdec_get_int(gd->fdt_blob, dev->of_offset, ...) 417 to access the node and store the resulting information into dev->platdata. 418 After this point, the device works the same way whether it was bound 419 using a device tree node or U_BOOT_DEVICE() structure. In either case, 420 the platform data is now stored in the platdata structure. Typically you 421 will use the platdata_auto_alloc_size feature to specify the size of the 422 platform data structure, and U-Boot will automatically allocate and zero 423 it for you before entry to ofdata_to_platdata(). But if not, you can 424 allocate it yourself in ofdata_to_platdata(). Note that it is preferable 425 to do all the device tree decoding in ofdata_to_platdata() rather than 426 in probe(). (Apart from the ugliness of mixing configuration and run-time 427 data, one day it is possible that U-Boot will cache platformat data for 428 devices which are regularly de/activated). 429 430 f. The device's probe() method is called. This should do anything that 431 is required by the device to get it going. This could include checking 432 that the hardware is actually present, setting up clocks for the 433 hardware and setting up hardware registers to initial values. The code 434 in probe() can access: 435 436 - platform data in dev->platdata (for configuration) 437 - private data in dev->priv (for run-time state) 438 - uclass data in dev->uclass_priv (for things the uclass stores 439 about this device) 440 441 Note: If you don't use priv_auto_alloc_size then you will need to 442 allocate the priv space here yourself. The same applies also to 443 platdata_auto_alloc_size. Remember to free them in the remove() method. 444 445 g. The device is marked 'activated' 446 447 h. The uclass's post_probe() method is called, if one exists. This may 448 cause the uclass to do some housekeeping to record the device as 449 activated and 'known' by the uclass. 450 4513. Running stage 452 453The device is now activated and can be used. From now until it is removed 454all of the above structures are accessible. The device appears in the 455uclass's list of devices (so if the device is in UCLASS_GPIO it will appear 456as a device in the GPIO uclass). This is the 'running' state of the device. 457 4584. Removal stage 459 460When the device is no-longer required, you can call device_remove() to 461remove it. This performs the probe steps in reverse: 462 463 a. The uclass's pre_remove() method is called, if one exists. This may 464 cause the uclass to do some housekeeping to record the device as 465 deactivated and no-longer 'known' by the uclass. 466 467 b. All the device's children are removed. It is not permitted to have 468 an active child device with a non-active parent. This means that 469 device_remove() is called for all the children recursively at this point. 470 471 c. The device's remove() method is called. At this stage nothing has been 472 deallocated so platform data, private data and the uclass data will all 473 still be present. This is where the hardware can be shut down. It is 474 intended that the device be completely inactive at this point, For U-Boot 475 to be sure that no hardware is running, it should be enough to remove 476 all devices. 477 478 d. The device memory is freed (platform data, private data, uclass data). 479 480 Note: Because the platform data for a U_BOOT_DEVICE() is defined with a 481 static pointer, it is not de-allocated during the remove() method. For 482 a device instantiated using the device tree data, the platform data will 483 be dynamically allocated, and thus needs to be deallocated during the 484 remove() method, either: 485 486 1. if the platdata_auto_alloc_size is non-zero, the deallocation 487 happens automatically within the driver model core; or 488 489 2. when platdata_auto_alloc_size is 0, both the allocation (in probe() 490 or preferably ofdata_to_platdata()) and the deallocation in remove() 491 are the responsibility of the driver author. 492 493 e. The device is marked inactive. Note that it is still bound, so the 494 device structure itself is not freed at this point. Should the device be 495 activated again, then the cycle starts again at step 2 above. 496 4975. Unbind stage 498 499The device is unbound. This is the step that actually destroys the device. 500If a parent has children these will be destroyed first. After this point 501the device does not exist and its memory has be deallocated. 502 503 504Data Structures 505--------------- 506 507Driver model uses a doubly-linked list as the basic data structure. Some 508nodes have several lists running through them. Creating a more efficient 509data structure might be worthwhile in some rare cases, once we understand 510what the bottlenecks are. 511 512 513Changes since v1 514---------------- 515 516For the record, this implementation uses a very similar approach to the 517original patches, but makes at least the following changes: 518 519- Tried to aggressively remove boilerplate, so that for most drivers there 520is little or no 'driver model' code to write. 521- Moved some data from code into data structure - e.g. store a pointer to 522the driver operations structure in the driver, rather than passing it 523to the driver bind function. 524- Rename some structures to make them more similar to Linux (struct udevice 525instead of struct instance, struct platdata, etc.) 526- Change the name 'core' to 'uclass', meaning U-Boot class. It seems that 527this concept relates to a class of drivers (or a subsystem). We shouldn't 528use 'class' since it is a C++ reserved word, so U-Boot class (uclass) seems 529better than 'core'. 530- Remove 'struct driver_instance' and just use a single 'struct udevice'. 531This removes a level of indirection that doesn't seem necessary. 532- Built in device tree support, to avoid the need for platdata 533- Removed the concept of driver relocation, and just make it possible for 534the new driver (created after relocation) to access the old driver data. 535I feel that relocation is a very special case and will only apply to a few 536drivers, many of which can/will just re-init anyway. So the overhead of 537dealing with this might not be worth it. 538- Implemented a GPIO system, trying to keep it simple 539 540 541Things to punt for later 542------------------------ 543 544- SPL support - this will have to be present before many drivers can be 545converted, but it seems like we can add it once we are happy with the 546core implementation. 547- Pre-relocation support - similar story 548 549That is not to say that no thinking has gone into these - in fact there 550is quite a lot there. However, getting these right is non-trivial and 551there is a high cost associated with going down the wrong path. 552 553For SPL, it may be possible to fit in a simplified driver model with only 554bind and probe methods, to reduce size. 555 556For pre-relocation we can simply call the driver model init function. Then 557post relocation we throw that away and re-init driver model again. For drivers 558which require some sort of continuity between pre- and post-relocation 559devices, we can provide access to the pre-relocation device pointers. 560 561Uclasses are statically numbered at compile time. It would be possible to 562change this to dynamic numbering, but then we would require some sort of 563lookup service, perhaps searching by name. This is slightly less efficient 564so has been left out for now. One small advantage of dynamic numbering might 565be fewer merge conflicts in uclass-id.h. 566 567 568Simon Glass 569sjg@chromium.org 570April 2013 571Updated 7-May-13 572Updated 14-Jun-13 573Updated 18-Oct-13 574Updated 5-Nov-13 575