1/* 2 * Linux WiMAX 3 * Kernel space API for accessing WiMAX devices 4 * 5 * 6 * Copyright (C) 2007-2008 Intel Corporation <linux-wimax@intel.com> 7 * Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com> 8 * 9 * This program is free software; you can redistribute it and/or 10 * modify it under the terms of the GNU General Public License version 11 * 2 as published by the Free Software Foundation. 12 * 13 * This program is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 * GNU General Public License for more details. 17 * 18 * You should have received a copy of the GNU General Public License 19 * along with this program; if not, write to the Free Software 20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 21 * 02110-1301, USA. 22 * 23 * 24 * The WiMAX stack provides an API for controlling and managing the 25 * system's WiMAX devices. This API affects the control plane; the 26 * data plane is accessed via the network stack (netdev). 27 * 28 * Parts of the WiMAX stack API and notifications are exported to 29 * user space via Generic Netlink. In user space, libwimax (part of 30 * the wimax-tools package) provides a shim layer for accessing those 31 * calls. 32 * 33 * The API is standarized for all WiMAX devices and different drivers 34 * implement the backend support for it. However, device-specific 35 * messaging pipes are provided that can be used to issue commands and 36 * receive notifications in free form. 37 * 38 * Currently the messaging pipes are the only means of control as it 39 * is not known (due to the lack of more devices in the market) what 40 * will be a good abstraction layer. Expect this to change as more 41 * devices show in the market. This API is designed to be growable in 42 * order to address this problem. 43 * 44 * USAGE 45 * 46 * Embed a `struct wimax_dev` at the beginning of the the device's 47 * private structure, initialize and register it. For details, see 48 * `struct wimax_dev`s documentation. 49 * 50 * Once this is done, wimax-tools's libwimaxll can be used to 51 * communicate with the driver from user space. You user space 52 * application does not have to forcibily use libwimaxll and can talk 53 * the generic netlink protocol directly if desired. 54 * 55 * Remember this is a very low level API that will to provide all of 56 * WiMAX features. Other daemons and services running in user space 57 * are the expected clients of it. They offer a higher level API that 58 * applications should use (an example of this is the Intel's WiMAX 59 * Network Service for the i2400m). 60 * 61 * DESIGN 62 * 63 * Although not set on final stone, this very basic interface is 64 * mostly completed. Remember this is meant to grow as new common 65 * operations are decided upon. New operations will be added to the 66 * interface, intent being on keeping backwards compatibility as much 67 * as possible. 68 * 69 * This layer implements a set of calls to control a WiMAX device, 70 * exposing a frontend to the rest of the kernel and user space (via 71 * generic netlink) and a backend implementation in the driver through 72 * function pointers. 73 * 74 * WiMAX devices have a state, and a kernel-only API allows the 75 * drivers to manipulate that state. State transitions are atomic, and 76 * only some of them are allowed (see `enum wimax_st`). 77 * 78 * Most API calls will set the state automatically; in most cases 79 * drivers have to only report state changes due to external 80 * conditions. 81 * 82 * All API operations are 'atomic', serialized through a mutex in the 83 * `struct wimax_dev`. 84 * 85 * EXPORTING TO USER SPACE THROUGH GENERIC NETLINK 86 * 87 * The API is exported to user space using generic netlink (other 88 * methods can be added as needed). 89 * 90 * There is a Generic Netlink Family named "WiMAX", where interfaces 91 * supporting the WiMAX interface receive commands and broadcast their 92 * signals over a multicast group named "msg". 93 * 94 * Mapping to the source/destination interface is done by an interface 95 * index attribute. 96 * 97 * For user-to-kernel traffic (commands) we use a function call 98 * marshalling mechanism, where a message X with attributes A, B, C 99 * sent from user space to kernel space means executing the WiMAX API 100 * call wimax_X(A, B, C), sending the results back as a message. 101 * 102 * Kernel-to-user (notifications or signals) communication is sent 103 * over multicast groups. This allows to have multiple applications 104 * monitoring them. 105 * 106 * Each command/signal gets assigned it's own attribute policy. This 107 * way the validator will verify that all the attributes in there are 108 * only the ones that should be for each command/signal. Thing of an 109 * attribute mapping to a type+argumentname for each command/signal. 110 * 111 * If we had a single policy for *all* commands/signals, after running 112 * the validator we'd have to check "does this attribute belong in 113 * here"? for each one. It can be done manually, but it's just easier 114 * to have the validator do that job with multiple policies. As well, 115 * it makes it easier to later expand each command/signal signature 116 * without affecting others and keeping the namespace more or less 117 * sane. Not that it is too complicated, but it makes it even easier. 118 * 119 * No state information is maintained in the kernel for each user 120 * space connection (the connection is stateless). 121 * 122 * TESTING FOR THE INTERFACE AND VERSIONING 123 * 124 * If network interface X is a WiMAX device, there will be a Generic 125 * Netlink family named "WiMAX X" and the device will present a 126 * "wimax" directory in it's network sysfs directory 127 * (/sys/class/net/DEVICE/wimax) [used by HAL]. 128 * 129 * The inexistence of any of these means the device does not support 130 * this WiMAX API. 131 * 132 * By querying the generic netlink controller, versioning information 133 * and the multicast groups available can be found. Applications using 134 * the interface can either rely on that or use the generic netlink 135 * controller to figure out which generic netlink commands/signals are 136 * supported. 137 * 138 * NOTE: this versioning is a last resort to avoid hard 139 * incompatibilities. It is the intention of the design of this 140 * stack not to introduce backward incompatible changes. 141 * 142 * The version code has to fit in one byte (restrictions imposed by 143 * generic netlink); we use `version / 10` for the major version and 144 * `version % 10` for the minor. This gives 9 minors for each major 145 * and 25 majors. 146 * 147 * The version change protocol is as follow: 148 * 149 * - Major versions: needs to be increased if an existing message/API 150 * call is changed or removed. Doesn't need to be changed if a new 151 * message is added. 152 * 153 * - Minor version: needs to be increased if new messages/API calls are 154 * being added or some other consideration that doesn't impact the 155 * user-kernel interface too much (like some kind of bug fix) and 156 * that is kind of left up in the air to common sense. 157 * 158 * User space code should not try to work if the major version it was 159 * compiled for differs from what the kernel offers. As well, if the 160 * minor version of the kernel interface is lower than the one user 161 * space is expecting (the one it was compiled for), the kernel 162 * might be missing API calls; user space shall be ready to handle 163 * said condition. Use the generic netlink controller operations to 164 * find which ones are supported and which not. 165 * 166 * libwimaxll:wimaxll_open() takes care of checking versions. 167 * 168 * THE OPERATIONS: 169 * 170 * Each operation is defined in its on file (drivers/net/wimax/op-*.c) 171 * for clarity. The parts needed for an operation are: 172 * 173 * - a function pointer in `struct wimax_dev`: optional, as the 174 * operation might be implemented by the stack and not by the 175 * driver. 176 * 177 * All function pointers are named wimax_dev->op_*(), and drivers 178 * must implement them except where noted otherwise. 179 * 180 * - When exported to user space, a `struct nla_policy` to define the 181 * attributes of the generic netlink command and a `struct genl_ops` 182 * to define the operation. 183 * 184 * All the declarations for the operation codes (WIMAX_GNL_OP_<NAME>) 185 * and generic netlink attributes (WIMAX_GNL_<NAME>_*) are declared in 186 * include/linux/wimax.h; this file is intended to be cloned by user 187 * space to gain access to those declarations. 188 * 189 * A few caveats to remember: 190 * 191 * - Need to define attribute numbers starting in 1; otherwise it 192 * fails. 193 * 194 * - the `struct genl_family` requires a maximum attribute id; when 195 * defining the `struct nla_policy` for each message, it has to have 196 * an array size of WIMAX_GNL_ATTR_MAX+1. 197 * 198 * The op_*() function pointers will not be called if the wimax_dev is 199 * in a state <= %WIMAX_ST_UNINITIALIZED. The exception is: 200 * 201 * - op_reset: can be called at any time after wimax_dev_add() has 202 * been called. 203 * 204 * THE PIPE INTERFACE: 205 * 206 * This interface is kept intentionally simple. The driver can send 207 * and receive free-form messages to/from user space through a 208 * pipe. See drivers/net/wimax/op-msg.c for details. 209 * 210 * The kernel-to-user messages are sent with 211 * wimax_msg(). user-to-kernel messages are delivered via 212 * wimax_dev->op_msg_from_user(). 213 * 214 * RFKILL: 215 * 216 * RFKILL support is built into the wimax_dev layer; the driver just 217 * needs to call wimax_report_rfkill_{hw,sw}() to inform of changes in 218 * the hardware or software RF kill switches. When the stack wants to 219 * turn the radio off, it will call wimax_dev->op_rfkill_sw_toggle(), 220 * which the driver implements. 221 * 222 * User space can set the software RF Kill switch by calling 223 * wimax_rfkill(). 224 * 225 * The code for now only supports devices that don't require polling; 226 * If the device needs to be polled, create a self-rearming delayed 227 * work struct for polling or look into adding polled support to the 228 * WiMAX stack. 229 * 230 * When initializing the hardware (_probe), after calling 231 * wimax_dev_add(), query the device for it's RF Kill switches status 232 * and feed it back to the WiMAX stack using 233 * wimax_report_rfkill_{hw,sw}(). If any switch is missing, always 234 * report it as ON. 235 * 236 * NOTE: the wimax stack uses an inverted terminology to that of the 237 * RFKILL subsystem: 238 * 239 * - ON: radio is ON, RFKILL is DISABLED or OFF. 240 * - OFF: radio is OFF, RFKILL is ENABLED or ON. 241 * 242 * MISCELLANEOUS OPS: 243 * 244 * wimax_reset() can be used to reset the device to power on state; by 245 * default it issues a warm reset that maintains the same device 246 * node. If that is not possible, it falls back to a cold reset 247 * (device reconnect). The driver implements the backend to this 248 * through wimax_dev->op_reset(). 249 */ 250 251#ifndef __NET__WIMAX_H__ 252#define __NET__WIMAX_H__ 253 254#include <linux/wimax.h> 255#include <net/genetlink.h> 256#include <linux/netdevice.h> 257 258struct net_device; 259struct genl_info; 260struct wimax_dev; 261 262/** 263 * struct wimax_dev - Generic WiMAX device 264 * 265 * @net_dev: [fill] Pointer to the &struct net_device this WiMAX 266 * device implements. 267 * 268 * @op_msg_from_user: [fill] Driver-specific operation to 269 * handle a raw message from user space to the driver. The 270 * driver can send messages to user space using with 271 * wimax_msg_to_user(). 272 * 273 * @op_rfkill_sw_toggle: [fill] Driver-specific operation to act on 274 * userspace (or any other agent) requesting the WiMAX device to 275 * change the RF Kill software switch (WIMAX_RF_ON or 276 * WIMAX_RF_OFF). 277 * If such hardware support is not present, it is assumed the 278 * radio cannot be switched off and it is always on (and the stack 279 * will error out when trying to switch it off). In such case, 280 * this function pointer can be left as NULL. 281 * 282 * @op_reset: [fill] Driver specific operation to reset the 283 * device. 284 * This operation should always attempt first a warm reset that 285 * does not disconnect the device from the bus and return 0. 286 * If that fails, it should resort to some sort of cold or bus 287 * reset (even if it implies a bus disconnection and device 288 * disappearance). In that case, -ENODEV should be returned to 289 * indicate the device is gone. 290 * This operation has to be synchronous, and return only when the 291 * reset is complete. In case of having had to resort to bus/cold 292 * reset implying a device disconnection, the call is allowed to 293 * return immediately. 294 * NOTE: wimax_dev->mutex is NOT locked when this op is being 295 * called; however, wimax_dev->mutex_reset IS locked to ensure 296 * serialization of calls to wimax_reset(). 297 * See wimax_reset()'s documentation. 298 * 299 * @name: [fill] A way to identify this device. We need to register a 300 * name with many subsystems (rfkill, workqueue creation, etc). 301 * We can't use the network device name as that 302 * might change and in some instances we don't know it yet (until 303 * we don't call register_netdev()). So we generate an unique one 304 * using the driver name and device bus id, place it here and use 305 * it across the board. Recommended naming: 306 * DRIVERNAME-BUSNAME:BUSID (dev->bus->name, dev->bus_id). 307 * 308 * @id_table_node: [private] link to the list of wimax devices kept by 309 * id-table.c. Protected by it's own spinlock. 310 * 311 * @mutex: [private] Serializes all concurrent access and execution of 312 * operations. 313 * 314 * @mutex_reset: [private] Serializes reset operations. Needs to be a 315 * different mutex because as part of the reset operation, the 316 * driver has to call back into the stack to do things such as 317 * state change, that require wimax_dev->mutex. 318 * 319 * @state: [private] Current state of the WiMAX device. 320 * 321 * @rfkill: [private] integration into the RF-Kill infrastructure. 322 * 323 * @rf_sw: [private] State of the software radio switch (OFF/ON) 324 * 325 * @rf_hw: [private] State of the hardware radio switch (OFF/ON) 326 * 327 * @debugfs_dentry: [private] Used to hook up a debugfs entry. This 328 * shows up in the debugfs root as wimax\:DEVICENAME. 329 * 330 * Description: 331 * This structure defines a common interface to access all WiMAX 332 * devices from different vendors and provides a common API as well as 333 * a free-form device-specific messaging channel. 334 * 335 * Usage: 336 * 1. Embed a &struct wimax_dev at *the beginning* the network 337 * device structure so that netdev_priv() points to it. 338 * 339 * 2. memset() it to zero 340 * 341 * 3. Initialize with wimax_dev_init(). This will leave the WiMAX 342 * device in the %__WIMAX_ST_NULL state. 343 * 344 * 4. Fill all the fields marked with [fill]; once called 345 * wimax_dev_add(), those fields CANNOT be modified. 346 * 347 * 5. Call wimax_dev_add() *after* registering the network 348 * device. This will leave the WiMAX device in the %WIMAX_ST_DOWN 349 * state. 350 * Protect the driver's net_device->open() against succeeding if 351 * the wimax device state is lower than %WIMAX_ST_DOWN. 352 * 353 * 6. Select when the device is going to be turned on/initialized; 354 * for example, it could be initialized on 'ifconfig up' (when the 355 * netdev op 'open()' is called on the driver). 356 * 357 * When the device is initialized (at `ifconfig up` time, or right 358 * after calling wimax_dev_add() from _probe(), make sure the 359 * following steps are taken 360 * 361 * a. Move the device to %WIMAX_ST_UNINITIALIZED. This is needed so 362 * some API calls that shouldn't work until the device is ready 363 * can be blocked. 364 * 365 * b. Initialize the device. Make sure to turn the SW radio switch 366 * off and move the device to state %WIMAX_ST_RADIO_OFF when 367 * done. When just initialized, a device should be left in RADIO 368 * OFF state until user space devices to turn it on. 369 * 370 * c. Query the device for the state of the hardware rfkill switch 371 * and call wimax_rfkill_report_hw() and wimax_rfkill_report_sw() 372 * as needed. See below. 373 * 374 * wimax_dev_rm() undoes before unregistering the network device. Once 375 * wimax_dev_add() is called, the driver can get called on the 376 * wimax_dev->op_* function pointers 377 * 378 * CONCURRENCY: 379 * 380 * The stack provides a mutex for each device that will disallow API 381 * calls happening concurrently; thus, op calls into the driver 382 * through the wimax_dev->op*() function pointers will always be 383 * serialized and *never* concurrent. 384 * 385 * For locking, take wimax_dev->mutex is taken; (most) operations in 386 * the API have to check for wimax_dev_is_ready() to return 0 before 387 * continuing (this is done internally). 388 * 389 * REFERENCE COUNTING: 390 * 391 * The WiMAX device is reference counted by the associated network 392 * device. The only operation that can be used to reference the device 393 * is wimax_dev_get_by_genl_info(), and the reference it acquires has 394 * to be released with dev_put(wimax_dev->net_dev). 395 * 396 * RFKILL: 397 * 398 * At startup, both HW and SW radio switchess are assumed to be off. 399 * 400 * At initialization time [after calling wimax_dev_add()], have the 401 * driver query the device for the status of the software and hardware 402 * RF kill switches and call wimax_report_rfkill_hw() and 403 * wimax_rfkill_report_sw() to indicate their state. If any is 404 * missing, just call it to indicate it is ON (radio always on). 405 * 406 * Whenever the driver detects a change in the state of the RF kill 407 * switches, it should call wimax_report_rfkill_hw() or 408 * wimax_report_rfkill_sw() to report it to the stack. 409 */ 410struct wimax_dev { 411 struct net_device *net_dev; 412 struct list_head id_table_node; 413 struct mutex mutex; /* Protects all members and API calls */ 414 struct mutex mutex_reset; 415 enum wimax_st state; 416 417 int (*op_msg_from_user)(struct wimax_dev *wimax_dev, 418 const char *, 419 const void *, size_t, 420 const struct genl_info *info); 421 int (*op_rfkill_sw_toggle)(struct wimax_dev *wimax_dev, 422 enum wimax_rf_state); 423 int (*op_reset)(struct wimax_dev *wimax_dev); 424 425 struct rfkill *rfkill; 426 unsigned int rf_hw; 427 unsigned int rf_sw; 428 char name[32]; 429 430 struct dentry *debugfs_dentry; 431}; 432 433 434 435/* 436 * WiMAX stack public API for device drivers 437 * ----------------------------------------- 438 * 439 * These functions are not exported to user space. 440 */ 441void wimax_dev_init(struct wimax_dev *); 442int wimax_dev_add(struct wimax_dev *, struct net_device *); 443void wimax_dev_rm(struct wimax_dev *); 444 445static inline 446struct wimax_dev *net_dev_to_wimax(struct net_device *net_dev) 447{ 448 return netdev_priv(net_dev); 449} 450 451static inline 452struct device *wimax_dev_to_dev(struct wimax_dev *wimax_dev) 453{ 454 return wimax_dev->net_dev->dev.parent; 455} 456 457void wimax_state_change(struct wimax_dev *, enum wimax_st); 458enum wimax_st wimax_state_get(struct wimax_dev *); 459 460/* 461 * Radio Switch state reporting. 462 * 463 * enum wimax_rf_state is declared in linux/wimax.h so the exports 464 * to user space can use it. 465 */ 466void wimax_report_rfkill_hw(struct wimax_dev *, enum wimax_rf_state); 467void wimax_report_rfkill_sw(struct wimax_dev *, enum wimax_rf_state); 468 469 470/* 471 * Free-form messaging to/from user space 472 * 473 * Sending a message: 474 * 475 * wimax_msg(wimax_dev, pipe_name, buf, buf_size, GFP_KERNEL); 476 * 477 * Broken up: 478 * 479 * skb = wimax_msg_alloc(wimax_dev, pipe_name, buf_size, GFP_KERNEL); 480 * ...fill up skb... 481 * wimax_msg_send(wimax_dev, pipe_name, skb); 482 * 483 * Be sure not to modify skb->data in the middle (ie: don't use 484 * skb_push()/skb_pull()/skb_reserve() on the skb). 485 * 486 * "pipe_name" is any string, that can be interpreted as the name of 487 * the pipe or recipient; the interpretation of it is driver 488 * specific, so the recipient can multiplex it as wished. It can be 489 * NULL, it won't be used - an example is using a "diagnostics" tag to 490 * send diagnostics information that a device-specific diagnostics 491 * tool would be interested in. 492 */ 493struct sk_buff *wimax_msg_alloc(struct wimax_dev *, const char *, const void *, 494 size_t, gfp_t); 495int wimax_msg_send(struct wimax_dev *, struct sk_buff *); 496int wimax_msg(struct wimax_dev *, const char *, const void *, size_t, gfp_t); 497 498const void *wimax_msg_data_len(struct sk_buff *, size_t *); 499const void *wimax_msg_data(struct sk_buff *); 500ssize_t wimax_msg_len(struct sk_buff *); 501 502 503/* 504 * WiMAX stack user space API 505 * -------------------------- 506 * 507 * This API is what gets exported to user space for general 508 * operations. As well, they can be called from within the kernel, 509 * (with a properly referenced `struct wimax_dev`). 510 * 511 * Properly referenced means: the 'struct net_device' that embeds the 512 * device's control structure and (as such) the 'struct wimax_dev' is 513 * referenced by the caller. 514 */ 515int wimax_rfkill(struct wimax_dev *, enum wimax_rf_state); 516int wimax_reset(struct wimax_dev *); 517 518#endif /* #ifndef __NET__WIMAX_H__ */ 519