Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
torvalds
GitHub Repository: torvalds/linux
Path: blob/master/drivers/base/auxiliary.c
26378 views
1
// SPDX-License-Identifier: GPL-2.0-only
2
/*
3
* Copyright (c) 2019-2020 Intel Corporation
4
*
5
* Please see Documentation/driver-api/auxiliary_bus.rst for more information.
6
*/
7
8
#define pr_fmt(fmt) "%s:%s: " fmt, KBUILD_MODNAME, __func__
9
10
#include <linux/device.h>
11
#include <linux/init.h>
12
#include <linux/slab.h>
13
#include <linux/module.h>
14
#include <linux/pm_domain.h>
15
#include <linux/pm_runtime.h>
16
#include <linux/string.h>
17
#include <linux/auxiliary_bus.h>
18
#include "base.h"
19
20
/**
21
* DOC: PURPOSE
22
*
23
* In some subsystems, the functionality of the core device (PCI/ACPI/other) is
24
* too complex for a single device to be managed by a monolithic driver (e.g.
25
* Sound Open Firmware), multiple devices might implement a common intersection
26
* of functionality (e.g. NICs + RDMA), or a driver may want to export an
27
* interface for another subsystem to drive (e.g. SIOV Physical Function export
28
* Virtual Function management). A split of the functionality into child-
29
* devices representing sub-domains of functionality makes it possible to
30
* compartmentalize, layer, and distribute domain-specific concerns via a Linux
31
* device-driver model.
32
*
33
* An example for this kind of requirement is the audio subsystem where a
34
* single IP is handling multiple entities such as HDMI, Soundwire, local
35
* devices such as mics/speakers etc. The split for the core's functionality
36
* can be arbitrary or be defined by the DSP firmware topology and include
37
* hooks for test/debug. This allows for the audio core device to be minimal
38
* and focused on hardware-specific control and communication.
39
*
40
* Each auxiliary_device represents a part of its parent functionality. The
41
* generic behavior can be extended and specialized as needed by encapsulating
42
* an auxiliary_device within other domain-specific structures and the use of
43
* .ops callbacks. Devices on the auxiliary bus do not share any structures and
44
* the use of a communication channel with the parent is domain-specific.
45
*
46
* Note that ops are intended as a way to augment instance behavior within a
47
* class of auxiliary devices, it is not the mechanism for exporting common
48
* infrastructure from the parent. Consider EXPORT_SYMBOL_NS() to convey
49
* infrastructure from the parent module to the auxiliary module(s).
50
*/
51
52
/**
53
* DOC: USAGE
54
*
55
* The auxiliary bus is to be used when a driver and one or more kernel
56
* modules, who share a common header file with the driver, need a mechanism to
57
* connect and provide access to a shared object allocated by the
58
* auxiliary_device's registering driver. The registering driver for the
59
* auxiliary_device(s) and the kernel module(s) registering auxiliary_drivers
60
* can be from the same subsystem, or from multiple subsystems.
61
*
62
* The emphasis here is on a common generic interface that keeps subsystem
63
* customization out of the bus infrastructure.
64
*
65
* One example is a PCI network device that is RDMA-capable and exports a child
66
* device to be driven by an auxiliary_driver in the RDMA subsystem. The PCI
67
* driver allocates and registers an auxiliary_device for each physical
68
* function on the NIC. The RDMA driver registers an auxiliary_driver that
69
* claims each of these auxiliary_devices. This conveys data/ops published by
70
* the parent PCI device/driver to the RDMA auxiliary_driver.
71
*
72
* Another use case is for the PCI device to be split out into multiple sub
73
* functions. For each sub function an auxiliary_device is created. A PCI sub
74
* function driver binds to such devices that creates its own one or more class
75
* devices. A PCI sub function auxiliary device is likely to be contained in a
76
* struct with additional attributes such as user defined sub function number
77
* and optional attributes such as resources and a link to the parent device.
78
* These attributes could be used by systemd/udev; and hence should be
79
* initialized before a driver binds to an auxiliary_device.
80
*
81
* A key requirement for utilizing the auxiliary bus is that there is no
82
* dependency on a physical bus, device, register accesses or regmap support.
83
* These individual devices split from the core cannot live on the platform bus
84
* as they are not physical devices that are controlled by DT/ACPI. The same
85
* argument applies for not using MFD in this scenario as MFD relies on
86
* individual function devices being physical devices.
87
*/
88
89
/**
90
* DOC: EXAMPLE
91
*
92
* Auxiliary devices are created and registered by a subsystem-level core
93
* device that needs to break up its functionality into smaller fragments. One
94
* way to extend the scope of an auxiliary_device is to encapsulate it within a
95
* domain-specific structure defined by the parent device. This structure
96
* contains the auxiliary_device and any associated shared data/callbacks
97
* needed to establish the connection with the parent.
98
*
99
* An example is:
100
*
101
* .. code-block:: c
102
*
103
* struct foo {
104
* struct auxiliary_device auxdev;
105
* void (*connect)(struct auxiliary_device *auxdev);
106
* void (*disconnect)(struct auxiliary_device *auxdev);
107
* void *data;
108
* };
109
*
110
* The parent device then registers the auxiliary_device by calling
111
* auxiliary_device_init(), and then auxiliary_device_add(), with the pointer
112
* to the auxdev member of the above structure. The parent provides a name for
113
* the auxiliary_device that, combined with the parent's KBUILD_MODNAME,
114
* creates a match_name that is be used for matching and binding with a driver.
115
*
116
* Whenever an auxiliary_driver is registered, based on the match_name, the
117
* auxiliary_driver's probe() is invoked for the matching devices. The
118
* auxiliary_driver can also be encapsulated inside custom drivers that make
119
* the core device's functionality extensible by adding additional
120
* domain-specific ops as follows:
121
*
122
* .. code-block:: c
123
*
124
* struct my_ops {
125
* void (*send)(struct auxiliary_device *auxdev);
126
* void (*receive)(struct auxiliary_device *auxdev);
127
* };
128
*
129
*
130
* struct my_driver {
131
* struct auxiliary_driver auxiliary_drv;
132
* const struct my_ops ops;
133
* };
134
*
135
* An example of this type of usage is:
136
*
137
* .. code-block:: c
138
*
139
* const struct auxiliary_device_id my_auxiliary_id_table[] = {
140
* { .name = "foo_mod.foo_dev" },
141
* { },
142
* };
143
*
144
* const struct my_ops my_custom_ops = {
145
* .send = my_tx,
146
* .receive = my_rx,
147
* };
148
*
149
* const struct my_driver my_drv = {
150
* .auxiliary_drv = {
151
* .name = "myauxiliarydrv",
152
* .id_table = my_auxiliary_id_table,
153
* .probe = my_probe,
154
* .remove = my_remove,
155
* .shutdown = my_shutdown,
156
* },
157
* .ops = my_custom_ops,
158
* };
159
*
160
* Please note that such custom ops approach is valid, but it is hard to implement
161
* it right without global locks per-device to protect from auxiliary_drv removal
162
* during call to that ops. In addition, this implementation lacks proper module
163
* dependency, which causes to load/unload races between auxiliary parent and devices
164
* modules.
165
*
166
* The most easiest way to provide these ops reliably without needing to
167
* have a lock is to EXPORT_SYMBOL*() them and rely on already existing
168
* modules infrastructure for validity and correct dependencies chains.
169
*/
170
171
static const struct auxiliary_device_id *auxiliary_match_id(const struct auxiliary_device_id *id,
172
const struct auxiliary_device *auxdev)
173
{
174
for (; id->name[0]; id++) {
175
const char *p = strrchr(dev_name(&auxdev->dev), '.');
176
int match_size;
177
178
if (!p)
179
continue;
180
match_size = p - dev_name(&auxdev->dev);
181
182
/* use dev_name(&auxdev->dev) prefix before last '.' char to match to */
183
if (strlen(id->name) == match_size &&
184
!strncmp(dev_name(&auxdev->dev), id->name, match_size))
185
return id;
186
}
187
return NULL;
188
}
189
190
static int auxiliary_match(struct device *dev, const struct device_driver *drv)
191
{
192
struct auxiliary_device *auxdev = to_auxiliary_dev(dev);
193
const struct auxiliary_driver *auxdrv = to_auxiliary_drv(drv);
194
195
return !!auxiliary_match_id(auxdrv->id_table, auxdev);
196
}
197
198
static int auxiliary_uevent(const struct device *dev, struct kobj_uevent_env *env)
199
{
200
const char *name, *p;
201
202
name = dev_name(dev);
203
p = strrchr(name, '.');
204
205
return add_uevent_var(env, "MODALIAS=%s%.*s", AUXILIARY_MODULE_PREFIX,
206
(int)(p - name), name);
207
}
208
209
static const struct dev_pm_ops auxiliary_dev_pm_ops = {
210
SET_RUNTIME_PM_OPS(pm_generic_runtime_suspend, pm_generic_runtime_resume, NULL)
211
SET_SYSTEM_SLEEP_PM_OPS(pm_generic_suspend, pm_generic_resume)
212
};
213
214
static int auxiliary_bus_probe(struct device *dev)
215
{
216
const struct auxiliary_driver *auxdrv = to_auxiliary_drv(dev->driver);
217
struct auxiliary_device *auxdev = to_auxiliary_dev(dev);
218
int ret;
219
220
ret = dev_pm_domain_attach(dev, PD_FLAG_ATTACH_POWER_ON);
221
if (ret) {
222
dev_warn(dev, "Failed to attach to PM Domain : %d\n", ret);
223
return ret;
224
}
225
226
ret = auxdrv->probe(auxdev, auxiliary_match_id(auxdrv->id_table, auxdev));
227
if (ret)
228
dev_pm_domain_detach(dev, true);
229
230
return ret;
231
}
232
233
static void auxiliary_bus_remove(struct device *dev)
234
{
235
const struct auxiliary_driver *auxdrv = to_auxiliary_drv(dev->driver);
236
struct auxiliary_device *auxdev = to_auxiliary_dev(dev);
237
238
if (auxdrv->remove)
239
auxdrv->remove(auxdev);
240
dev_pm_domain_detach(dev, true);
241
}
242
243
static void auxiliary_bus_shutdown(struct device *dev)
244
{
245
const struct auxiliary_driver *auxdrv = NULL;
246
struct auxiliary_device *auxdev;
247
248
if (dev->driver) {
249
auxdrv = to_auxiliary_drv(dev->driver);
250
auxdev = to_auxiliary_dev(dev);
251
}
252
253
if (auxdrv && auxdrv->shutdown)
254
auxdrv->shutdown(auxdev);
255
}
256
257
static const struct bus_type auxiliary_bus_type = {
258
.name = "auxiliary",
259
.probe = auxiliary_bus_probe,
260
.remove = auxiliary_bus_remove,
261
.shutdown = auxiliary_bus_shutdown,
262
.match = auxiliary_match,
263
.uevent = auxiliary_uevent,
264
.pm = &auxiliary_dev_pm_ops,
265
};
266
267
/**
268
* auxiliary_device_init - check auxiliary_device and initialize
269
* @auxdev: auxiliary device struct
270
*
271
* This is the second step in the three-step process to register an
272
* auxiliary_device.
273
*
274
* When this function returns an error code, then the device_initialize will
275
* *not* have been performed, and the caller will be responsible to free any
276
* memory allocated for the auxiliary_device in the error path directly.
277
*
278
* It returns 0 on success. On success, the device_initialize has been
279
* performed. After this point any error unwinding will need to include a call
280
* to auxiliary_device_uninit(). In this post-initialize error scenario, a call
281
* to the device's .release callback will be triggered, and all memory clean-up
282
* is expected to be handled there.
283
*/
284
int auxiliary_device_init(struct auxiliary_device *auxdev)
285
{
286
struct device *dev = &auxdev->dev;
287
288
if (!dev->parent) {
289
pr_err("auxiliary_device has a NULL dev->parent\n");
290
return -EINVAL;
291
}
292
293
if (!auxdev->name) {
294
pr_err("auxiliary_device has a NULL name\n");
295
return -EINVAL;
296
}
297
298
dev->bus = &auxiliary_bus_type;
299
device_initialize(&auxdev->dev);
300
mutex_init(&auxdev->sysfs.lock);
301
return 0;
302
}
303
EXPORT_SYMBOL_GPL(auxiliary_device_init);
304
305
/**
306
* __auxiliary_device_add - add an auxiliary bus device
307
* @auxdev: auxiliary bus device to add to the bus
308
* @modname: name of the parent device's driver module
309
*
310
* This is the third step in the three-step process to register an
311
* auxiliary_device.
312
*
313
* This function must be called after a successful call to
314
* auxiliary_device_init(), which will perform the device_initialize. This
315
* means that if this returns an error code, then a call to
316
* auxiliary_device_uninit() must be performed so that the .release callback
317
* will be triggered to free the memory associated with the auxiliary_device.
318
*
319
* The expectation is that users will call the "auxiliary_device_add" macro so
320
* that the caller's KBUILD_MODNAME is automatically inserted for the modname
321
* parameter. Only if a user requires a custom name would this version be
322
* called directly.
323
*/
324
int __auxiliary_device_add(struct auxiliary_device *auxdev, const char *modname)
325
{
326
struct device *dev = &auxdev->dev;
327
int ret;
328
329
if (!modname) {
330
dev_err(dev, "auxiliary device modname is NULL\n");
331
return -EINVAL;
332
}
333
334
ret = dev_set_name(dev, "%s.%s.%d", modname, auxdev->name, auxdev->id);
335
if (ret) {
336
dev_err(dev, "auxiliary device dev_set_name failed: %d\n", ret);
337
return ret;
338
}
339
340
ret = device_add(dev);
341
if (ret)
342
dev_err(dev, "adding auxiliary device failed!: %d\n", ret);
343
344
return ret;
345
}
346
EXPORT_SYMBOL_GPL(__auxiliary_device_add);
347
348
/**
349
* __auxiliary_driver_register - register a driver for auxiliary bus devices
350
* @auxdrv: auxiliary_driver structure
351
* @owner: owning module/driver
352
* @modname: KBUILD_MODNAME for parent driver
353
*
354
* The expectation is that users will call the "auxiliary_driver_register"
355
* macro so that the caller's KBUILD_MODNAME is automatically inserted for the
356
* modname parameter. Only if a user requires a custom name would this version
357
* be called directly.
358
*/
359
int __auxiliary_driver_register(struct auxiliary_driver *auxdrv,
360
struct module *owner, const char *modname)
361
{
362
int ret;
363
364
if (WARN_ON(!auxdrv->probe) || WARN_ON(!auxdrv->id_table))
365
return -EINVAL;
366
367
if (auxdrv->name)
368
auxdrv->driver.name = kasprintf(GFP_KERNEL, "%s.%s", modname,
369
auxdrv->name);
370
else
371
auxdrv->driver.name = kasprintf(GFP_KERNEL, "%s", modname);
372
if (!auxdrv->driver.name)
373
return -ENOMEM;
374
375
auxdrv->driver.owner = owner;
376
auxdrv->driver.bus = &auxiliary_bus_type;
377
auxdrv->driver.mod_name = modname;
378
379
ret = driver_register(&auxdrv->driver);
380
if (ret)
381
kfree(auxdrv->driver.name);
382
383
return ret;
384
}
385
EXPORT_SYMBOL_GPL(__auxiliary_driver_register);
386
387
/**
388
* auxiliary_driver_unregister - unregister a driver
389
* @auxdrv: auxiliary_driver structure
390
*/
391
void auxiliary_driver_unregister(struct auxiliary_driver *auxdrv)
392
{
393
driver_unregister(&auxdrv->driver);
394
kfree(auxdrv->driver.name);
395
}
396
EXPORT_SYMBOL_GPL(auxiliary_driver_unregister);
397
398
static void auxiliary_device_release(struct device *dev)
399
{
400
struct auxiliary_device *auxdev = to_auxiliary_dev(dev);
401
402
of_node_put(dev->of_node);
403
kfree(auxdev);
404
}
405
406
/**
407
* auxiliary_device_create - create a device on the auxiliary bus
408
* @dev: parent device
409
* @modname: module name used to create the auxiliary driver name.
410
* @devname: auxiliary bus device name
411
* @platform_data: auxiliary bus device platform data
412
* @id: auxiliary bus device id
413
*
414
* Helper to create an auxiliary bus device.
415
* The device created matches driver 'modname.devname' on the auxiliary bus.
416
*/
417
struct auxiliary_device *auxiliary_device_create(struct device *dev,
418
const char *modname,
419
const char *devname,
420
void *platform_data,
421
int id)
422
{
423
struct auxiliary_device *auxdev;
424
int ret;
425
426
auxdev = kzalloc(sizeof(*auxdev), GFP_KERNEL);
427
if (!auxdev)
428
return NULL;
429
430
auxdev->id = id;
431
auxdev->name = devname;
432
auxdev->dev.parent = dev;
433
auxdev->dev.platform_data = platform_data;
434
auxdev->dev.release = auxiliary_device_release;
435
device_set_of_node_from_dev(&auxdev->dev, dev);
436
437
ret = auxiliary_device_init(auxdev);
438
if (ret) {
439
of_node_put(auxdev->dev.of_node);
440
kfree(auxdev);
441
return NULL;
442
}
443
444
ret = __auxiliary_device_add(auxdev, modname);
445
if (ret) {
446
/*
447
* It may look odd but auxdev should not be freed here.
448
* auxiliary_device_uninit() calls device_put() which call
449
* the device release function, freeing auxdev.
450
*/
451
auxiliary_device_uninit(auxdev);
452
return NULL;
453
}
454
455
return auxdev;
456
}
457
EXPORT_SYMBOL_GPL(auxiliary_device_create);
458
459
/**
460
* auxiliary_device_destroy - remove an auxiliary device
461
* @auxdev: pointer to the auxdev to be removed
462
*
463
* Helper to remove an auxiliary device created with
464
* auxiliary_device_create()
465
*/
466
void auxiliary_device_destroy(void *auxdev)
467
{
468
struct auxiliary_device *_auxdev = auxdev;
469
470
auxiliary_device_delete(_auxdev);
471
auxiliary_device_uninit(_auxdev);
472
}
473
EXPORT_SYMBOL_GPL(auxiliary_device_destroy);
474
475
/**
476
* __devm_auxiliary_device_create - create a managed device on the auxiliary bus
477
* @dev: parent device
478
* @modname: module name used to create the auxiliary driver name.
479
* @devname: auxiliary bus device name
480
* @platform_data: auxiliary bus device platform data
481
* @id: auxiliary bus device id
482
*
483
* Device managed helper to create an auxiliary bus device.
484
* The device created matches driver 'modname.devname' on the auxiliary bus.
485
*/
486
struct auxiliary_device *__devm_auxiliary_device_create(struct device *dev,
487
const char *modname,
488
const char *devname,
489
void *platform_data,
490
int id)
491
{
492
struct auxiliary_device *auxdev;
493
int ret;
494
495
auxdev = auxiliary_device_create(dev, modname, devname, platform_data, id);
496
if (!auxdev)
497
return NULL;
498
499
ret = devm_add_action_or_reset(dev, auxiliary_device_destroy,
500
auxdev);
501
if (ret)
502
return NULL;
503
504
return auxdev;
505
}
506
EXPORT_SYMBOL_GPL(__devm_auxiliary_device_create);
507
508
void __init auxiliary_bus_init(void)
509
{
510
WARN_ON(bus_register(&auxiliary_bus_type));
511
}
512
513