Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | |
| 2 | Device Drivers |
| 3 | |
| 4 | struct device_driver { |
| 5 | char * name; |
| 6 | struct bus_type * bus; |
| 7 | |
| 8 | rwlock_t lock; |
| 9 | atomic_t refcount; |
| 10 | |
| 11 | list_t bus_list; |
| 12 | list_t devices; |
| 13 | |
| 14 | struct driver_dir_entry dir; |
| 15 | |
| 16 | int (*probe) (struct device * dev); |
| 17 | int (*remove) (struct device * dev); |
| 18 | |
Pavel Machek | 438510f | 2005-04-16 15:25:24 -0700 | [diff] [blame^] | 19 | int (*suspend) (struct device * dev, pm_message_t state, u32 level); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | int (*resume) (struct device * dev, u32 level); |
| 21 | |
| 22 | void (*release) (struct device_driver * drv); |
| 23 | }; |
| 24 | |
| 25 | |
| 26 | |
| 27 | Allocation |
| 28 | ~~~~~~~~~~ |
| 29 | |
| 30 | Device drivers are statically allocated structures. Though there may |
| 31 | be multiple devices in a system that a driver supports, struct |
| 32 | device_driver represents the driver as a whole (not a particular |
| 33 | device instance). |
| 34 | |
| 35 | Initialization |
| 36 | ~~~~~~~~~~~~~~ |
| 37 | |
| 38 | The driver must initialize at least the name and bus fields. It should |
| 39 | also initialize the devclass field (when it arrives), so it may obtain |
| 40 | the proper linkage internally. It should also initialize as many of |
| 41 | the callbacks as possible, though each is optional. |
| 42 | |
| 43 | Declaration |
| 44 | ~~~~~~~~~~~ |
| 45 | |
| 46 | As stated above, struct device_driver objects are statically |
| 47 | allocated. Below is an example declaration of the eepro100 |
| 48 | driver. This declaration is hypothetical only; it relies on the driver |
| 49 | being converted completely to the new model. |
| 50 | |
| 51 | static struct device_driver eepro100_driver = { |
| 52 | .name = "eepro100", |
| 53 | .bus = &pci_bus_type, |
| 54 | .devclass = ðernet_devclass, /* when it's implemented */ |
| 55 | |
| 56 | .probe = eepro100_probe, |
| 57 | .remove = eepro100_remove, |
| 58 | .suspend = eepro100_suspend, |
| 59 | .resume = eepro100_resume, |
| 60 | }; |
| 61 | |
| 62 | Most drivers will not be able to be converted completely to the new |
| 63 | model because the bus they belong to has a bus-specific structure with |
| 64 | bus-specific fields that cannot be generalized. |
| 65 | |
| 66 | The most common example of this are device ID structures. A driver |
| 67 | typically defines an array of device IDs that it supports. The format |
| 68 | of these structures and the semantics for comparing device IDs are |
| 69 | completely bus-specific. Defining them as bus-specific entities would |
| 70 | sacrifice type-safety, so we keep bus-specific structures around. |
| 71 | |
| 72 | Bus-specific drivers should include a generic struct device_driver in |
| 73 | the definition of the bus-specific driver. Like this: |
| 74 | |
| 75 | struct pci_driver { |
| 76 | const struct pci_device_id *id_table; |
| 77 | struct device_driver driver; |
| 78 | }; |
| 79 | |
| 80 | A definition that included bus-specific fields would look like |
| 81 | (using the eepro100 driver again): |
| 82 | |
| 83 | static struct pci_driver eepro100_driver = { |
| 84 | .id_table = eepro100_pci_tbl, |
| 85 | .driver = { |
| 86 | .name = "eepro100", |
| 87 | .bus = &pci_bus_type, |
| 88 | .devclass = ðernet_devclass, /* when it's implemented */ |
| 89 | .probe = eepro100_probe, |
| 90 | .remove = eepro100_remove, |
| 91 | .suspend = eepro100_suspend, |
| 92 | .resume = eepro100_resume, |
| 93 | }, |
| 94 | }; |
| 95 | |
| 96 | Some may find the syntax of embedded struct initialization awkward or |
| 97 | even a bit ugly. So far, it's the best way we've found to do what we want... |
| 98 | |
| 99 | Registration |
| 100 | ~~~~~~~~~~~~ |
| 101 | |
| 102 | int driver_register(struct device_driver * drv); |
| 103 | |
| 104 | The driver registers the structure on startup. For drivers that have |
| 105 | no bus-specific fields (i.e. don't have a bus-specific driver |
| 106 | structure), they would use driver_register and pass a pointer to their |
| 107 | struct device_driver object. |
| 108 | |
| 109 | Most drivers, however, will have a bus-specific structure and will |
| 110 | need to register with the bus using something like pci_driver_register. |
| 111 | |
| 112 | It is important that drivers register their driver structure as early as |
| 113 | possible. Registration with the core initializes several fields in the |
| 114 | struct device_driver object, including the reference count and the |
| 115 | lock. These fields are assumed to be valid at all times and may be |
| 116 | used by the device model core or the bus driver. |
| 117 | |
| 118 | |
| 119 | Transition Bus Drivers |
| 120 | ~~~~~~~~~~~~~~~~~~~~~~ |
| 121 | |
| 122 | By defining wrapper functions, the transition to the new model can be |
| 123 | made easier. Drivers can ignore the generic structure altogether and |
| 124 | let the bus wrapper fill in the fields. For the callbacks, the bus can |
| 125 | define generic callbacks that forward the call to the bus-specific |
| 126 | callbacks of the drivers. |
| 127 | |
| 128 | This solution is intended to be only temporary. In order to get class |
| 129 | information in the driver, the drivers must be modified anyway. Since |
| 130 | converting drivers to the new model should reduce some infrastructural |
| 131 | complexity and code size, it is recommended that they are converted as |
| 132 | class information is added. |
| 133 | |
| 134 | Access |
| 135 | ~~~~~~ |
| 136 | |
| 137 | Once the object has been registered, it may access the common fields of |
| 138 | the object, like the lock and the list of devices. |
| 139 | |
| 140 | int driver_for_each_dev(struct device_driver * drv, void * data, |
| 141 | int (*callback)(struct device * dev, void * data)); |
| 142 | |
| 143 | The devices field is a list of all the devices that have been bound to |
| 144 | the driver. The LDM core provides a helper function to operate on all |
| 145 | the devices a driver controls. This helper locks the driver on each |
| 146 | node access, and does proper reference counting on each device as it |
| 147 | accesses it. |
| 148 | |
| 149 | |
| 150 | sysfs |
| 151 | ~~~~~ |
| 152 | |
| 153 | When a driver is registered, a sysfs directory is created in its |
| 154 | bus's directory. In this directory, the driver can export an interface |
| 155 | to userspace to control operation of the driver on a global basis; |
| 156 | e.g. toggling debugging output in the driver. |
| 157 | |
| 158 | A future feature of this directory will be a 'devices' directory. This |
| 159 | directory will contain symlinks to the directories of devices it |
| 160 | supports. |
| 161 | |
| 162 | |
| 163 | |
| 164 | Callbacks |
| 165 | ~~~~~~~~~ |
| 166 | |
| 167 | int (*probe) (struct device * dev); |
| 168 | |
| 169 | probe is called to verify the existence of a certain type of |
| 170 | hardware. This is called during the driver binding process, after the |
| 171 | bus has verified that the device ID of a device matches one of the |
| 172 | device IDs supported by the driver. |
| 173 | |
| 174 | This callback only verifies that there actually is supported hardware |
| 175 | present. It may allocate a driver-specific structure, but it should |
| 176 | not do any initialization of the hardware itself. The device-specific |
| 177 | structure may be stored in the device's driver_data field. |
| 178 | |
| 179 | int (*init) (struct device * dev); |
| 180 | |
| 181 | init is called during the binding stage. It is called after probe has |
| 182 | successfully returned and the device has been registered with its |
| 183 | class. It is responsible for initializing the hardware. |
| 184 | |
| 185 | int (*remove) (struct device * dev); |
| 186 | |
| 187 | remove is called to dissociate a driver with a device. This may be |
| 188 | called if a device is physically removed from the system, if the |
| 189 | driver module is being unloaded, or during a reboot sequence. |
| 190 | |
| 191 | It is up to the driver to determine if the device is present or |
| 192 | not. It should free any resources allocated specifically for the |
| 193 | device; i.e. anything in the device's driver_data field. |
| 194 | |
| 195 | If the device is still present, it should quiesce the device and place |
| 196 | it into a supported low-power state. |
| 197 | |
Pavel Machek | 438510f | 2005-04-16 15:25:24 -0700 | [diff] [blame^] | 198 | int (*suspend) (struct device * dev, pm_message_t state, u32 level); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | |
| 200 | suspend is called to put the device in a low power state. There are |
| 201 | several stages to successfully suspending a device, which is denoted in |
| 202 | the @level parameter. Breaking the suspend transition into several |
| 203 | stages affords the platform flexibility in performing device power |
| 204 | management based on the requirements of the system and the |
| 205 | user-defined policy. |
| 206 | |
| 207 | SUSPEND_NOTIFY notifies the device that a suspend transition is about |
| 208 | to happen. This happens on system power state transitions to verify |
| 209 | that all devices can successfully suspend. |
| 210 | |
| 211 | A driver may choose to fail on this call, which should cause the |
| 212 | entire suspend transition to fail. A driver should fail only if it |
| 213 | knows that the device will not be able to be resumed properly when the |
| 214 | system wakes up again. It could also fail if it somehow determines it |
| 215 | is in the middle of an operation too important to stop. |
| 216 | |
| 217 | SUSPEND_DISABLE tells the device to stop I/O transactions. When it |
| 218 | stops transactions, or what it should do with unfinished transactions |
| 219 | is a policy of the driver. After this call, the driver should not |
| 220 | accept any other I/O requests. |
| 221 | |
| 222 | SUSPEND_SAVE_STATE tells the device to save the context of the |
| 223 | hardware. This includes any bus-specific hardware state and |
| 224 | device-specific hardware state. A pointer to this saved state can be |
| 225 | stored in the device's saved_state field. |
| 226 | |
| 227 | SUSPEND_POWER_DOWN tells the driver to place the device in the low |
| 228 | power state requested. |
| 229 | |
| 230 | Whether suspend is called with a given level is a policy of the |
| 231 | platform. Some levels may be omitted; drivers must not assume the |
| 232 | reception of any level. However, all levels must be called in the |
| 233 | order above; i.e. notification will always come before disabling; |
| 234 | disabling the device will come before suspending the device. |
| 235 | |
| 236 | All calls are made with interrupts enabled, except for the |
| 237 | SUSPEND_POWER_DOWN level. |
| 238 | |
| 239 | int (*resume) (struct device * dev, u32 level); |
| 240 | |
| 241 | Resume is used to bring a device back from a low power state. Like the |
| 242 | suspend transition, it happens in several stages. |
| 243 | |
| 244 | RESUME_POWER_ON tells the driver to set the power state to the state |
| 245 | before the suspend call (The device could have already been in a low |
| 246 | power state before the suspend call to put in a lower power state). |
| 247 | |
| 248 | RESUME_RESTORE_STATE tells the driver to restore the state saved by |
| 249 | the SUSPEND_SAVE_STATE suspend call. |
| 250 | |
| 251 | RESUME_ENABLE tells the driver to start accepting I/O transactions |
| 252 | again. Depending on driver policy, the device may already have pending |
| 253 | I/O requests. |
| 254 | |
| 255 | RESUME_POWER_ON is called with interrupts disabled. The other resume |
| 256 | levels are called with interrupts enabled. |
| 257 | |
| 258 | As with the various suspend stages, the driver must not assume that |
| 259 | any other resume calls have been or will be made. Each call should be |
| 260 | self-contained and not dependent on any external state. |
| 261 | |
| 262 | |
| 263 | Attributes |
| 264 | ~~~~~~~~~~ |
| 265 | struct driver_attribute { |
| 266 | struct attribute attr; |
| 267 | ssize_t (*show)(struct device_driver *, char * buf, size_t count, loff_t off); |
| 268 | ssize_t (*store)(struct device_driver *, const char * buf, size_t count, loff_t off); |
| 269 | }; |
| 270 | |
| 271 | Device drivers can export attributes via their sysfs directories. |
| 272 | Drivers can declare attributes using a DRIVER_ATTR macro that works |
| 273 | identically to the DEVICE_ATTR macro. |
| 274 | |
| 275 | Example: |
| 276 | |
| 277 | DRIVER_ATTR(debug,0644,show_debug,store_debug); |
| 278 | |
| 279 | This is equivalent to declaring: |
| 280 | |
| 281 | struct driver_attribute driver_attr_debug; |
| 282 | |
| 283 | This can then be used to add and remove the attribute from the |
| 284 | driver's directory using: |
| 285 | |
| 286 | int driver_create_file(struct device_driver *, struct driver_attribute *); |
| 287 | void driver_remove_file(struct device_driver *, struct driver_attribute *); |