1.. |struct dev_pm_domain| replace:: :c:type:`struct dev_pm_domain <dev_pm_domain>`
2.. |struct generic_pm_domain| replace:: :c:type:`struct generic_pm_domain <generic_pm_domain>`
3
4
5.. _device_link:
6
7============
8Device links
9============
10
11By default, the driver core only enforces dependencies between devices
12that are borne out of a parent/child relationship within the device
13hierarchy: When suspending, resuming or shutting down the system, devices
14are ordered based on this relationship, i.e. children are always suspended
15before their parent, and the parent is always resumed before its children.
16
17Sometimes there is a need to represent device dependencies beyond the
18mere parent/child relationship, e.g. between siblings, and have the
19driver core automatically take care of them.
20
21Secondly, the driver core by default does not enforce any driver presence
22dependencies, i.e. that one device must be bound to a driver before
23another one can probe or function correctly.
24
25Often these two dependency types come together, so a device depends on
26another one both with regards to driver presence *and* with regards to
27suspend/resume and shutdown ordering.
28
29Device links allow representation of such dependencies in the driver core.
30
31In its standard or *managed* form, a device link combines *both* dependency
32types:  It guarantees correct suspend/resume and shutdown ordering between a
33"supplier" device and its "consumer" devices, and it guarantees driver
34presence on the supplier.  The consumer devices are not probed before the
35supplier is bound to a driver, and they're unbound before the supplier
36is unbound.
37
38When driver presence on the supplier is irrelevant and only correct
39suspend/resume and shutdown ordering is needed, the device link may
40simply be set up with the ``DL_FLAG_STATELESS`` flag.  In other words,
41enforcing driver presence on the supplier is optional.
42
43Another optional feature is runtime PM integration:  By setting the
44``DL_FLAG_PM_RUNTIME`` flag on addition of the device link, the PM core
45is instructed to runtime resume the supplier and keep it active
46whenever and for as long as the consumer is runtime resumed.
47
48Usage
49=====
50
51The earliest point in time when device links can be added is after
52:c:func:`device_add()` has been called for the supplier and
53:c:func:`device_initialize()` has been called for the consumer.
54
55It is legal to add them later, but care must be taken that the system
56remains in a consistent state:  E.g. a device link cannot be added in
57the midst of a suspend/resume transition, so either commencement of
58such a transition needs to be prevented with :c:func:`lock_system_sleep()`,
59or the device link needs to be added from a function which is guaranteed
60not to run in parallel to a suspend/resume transition, such as from a
61device ``->probe`` callback or a boot-time PCI quirk.
62
63Another example for an inconsistent state would be a device link that
64represents a driver presence dependency, yet is added from the consumer's
65``->probe`` callback while the supplier hasn't started to probe yet:  Had the
66driver core known about the device link earlier, it wouldn't have probed the
67consumer in the first place.  The onus is thus on the consumer to check
68presence of the supplier after adding the link, and defer probing on
69non-presence.  [Note that it is valid to create a link from the consumer's
70``->probe`` callback while the supplier is still probing, but the consumer must
71know that the supplier is functional already at the link creation time (that is
72the case, for instance, if the consumer has just acquired some resources that
73would not have been available had the supplier not been functional then).]
74
75If a device link with ``DL_FLAG_STATELESS`` set (i.e. a stateless device link)
76is added in the ``->probe`` callback of the supplier or consumer driver, it is
77typically deleted in its ``->remove`` callback for symmetry.  That way, if the
78driver is compiled as a module, the device link is added on module load and
79orderly deleted on unload.  The same restrictions that apply to device link
80addition (e.g. exclusion of a parallel suspend/resume transition) apply equally
81to deletion.  Device links managed by the driver core are deleted automatically
82by it.
83
84Several flags may be specified on device link addition, two of which
85have already been mentioned above:  ``DL_FLAG_STATELESS`` to express that no
86driver presence dependency is needed (but only correct suspend/resume and
87shutdown ordering) and ``DL_FLAG_PM_RUNTIME`` to express that runtime PM
88integration is desired.
89
90Two other flags are specifically targeted at use cases where the device
91link is added from the consumer's ``->probe`` callback:  ``DL_FLAG_RPM_ACTIVE``
92can be specified to runtime resume the supplier and prevent it from suspending
93before the consumer is runtime suspended.  ``DL_FLAG_AUTOREMOVE_CONSUMER``
94causes the device link to be automatically purged when the consumer fails to
95probe or later unbinds.
96
97Similarly, when the device link is added from supplier's ``->probe`` callback,
98``DL_FLAG_AUTOREMOVE_SUPPLIER`` causes the device link to be automatically
99purged when the supplier fails to probe or later unbinds.
100
101If neither ``DL_FLAG_AUTOREMOVE_CONSUMER`` nor ``DL_FLAG_AUTOREMOVE_SUPPLIER``
102is set, ``DL_FLAG_AUTOPROBE_CONSUMER`` can be used to request the driver core
103to probe for a driver for the consumer driver on the link automatically after
104a driver has been bound to the supplier device.
105
106Note, however, that any combinations of ``DL_FLAG_AUTOREMOVE_CONSUMER``,
107``DL_FLAG_AUTOREMOVE_SUPPLIER`` or ``DL_FLAG_AUTOPROBE_CONSUMER`` with
108``DL_FLAG_STATELESS`` are invalid and cannot be used.
109
110Limitations
111===========
112
113Driver authors should be aware that a driver presence dependency for managed
114device links (i.e. when ``DL_FLAG_STATELESS`` is not specified on link addition)
115may cause probing of the consumer to be deferred indefinitely.  This can become
116a problem if the consumer is required to probe before a certain initcall level
117is reached.  Worse, if the supplier driver is blacklisted or missing, the
118consumer will never be probed.
119
120Moreover, managed device links cannot be deleted directly.  They are deleted
121by the driver core when they are not necessary any more in accordance with the
122``DL_FLAG_AUTOREMOVE_CONSUMER`` and ``DL_FLAG_AUTOREMOVE_SUPPLIER`` flags.
123However, stateless device links (i.e. device links with ``DL_FLAG_STATELESS``
124set) are expected to be removed by whoever called :c:func:`device_link_add()`
125to add them with the help of either :c:func:`device_link_del()` or
126:c:func:`device_link_remove()`.
127
128Passing ``DL_FLAG_RPM_ACTIVE`` along with ``DL_FLAG_STATELESS`` to
129:c:func:`device_link_add()` may cause the PM-runtime usage counter of the
130supplier device to remain nonzero after a subsequent invocation of either
131:c:func:`device_link_del()` or :c:func:`device_link_remove()` to remove the
132device link returned by it.  This happens if :c:func:`device_link_add()` is
133called twice in a row for the same consumer-supplier pair without removing the
134link between these calls, in which case allowing the PM-runtime usage counter
135of the supplier to drop on an attempt to remove the link may cause it to be
136suspended while the consumer is still PM-runtime-active and that has to be
137avoided.  [To work around this limitation it is sufficient to let the consumer
138runtime suspend at least once, or call :c:func:`pm_runtime_set_suspended()` for
139it with PM-runtime disabled, between the :c:func:`device_link_add()` and
140:c:func:`device_link_del()` or :c:func:`device_link_remove()` calls.]
141
142Sometimes drivers depend on optional resources.  They are able to operate
143in a degraded mode (reduced feature set or performance) when those resources
144are not present.  An example is an SPI controller that can use a DMA engine
145or work in PIO mode.  The controller can determine presence of the optional
146resources at probe time but on non-presence there is no way to know whether
147they will become available in the near future (due to a supplier driver
148probing) or never.  Consequently it cannot be determined whether to defer
149probing or not.  It would be possible to notify drivers when optional
150resources become available after probing, but it would come at a high cost
151for drivers as switching between modes of operation at runtime based on the
152availability of such resources would be much more complex than a mechanism
153based on probe deferral.  In any case optional resources are beyond the
154scope of device links.
155
156Examples
157========
158
159* An MMU device exists alongside a busmaster device, both are in the same
160  power domain.  The MMU implements DMA address translation for the busmaster
161  device and shall be runtime resumed and kept active whenever and as long
162  as the busmaster device is active.  The busmaster device's driver shall
163  not bind before the MMU is bound.  To achieve this, a device link with
164  runtime PM integration is added from the busmaster device (consumer)
165  to the MMU device (supplier).  The effect with regards to runtime PM
166  is the same as if the MMU was the parent of the master device.
167
168  The fact that both devices share the same power domain would normally
169  suggest usage of a |struct dev_pm_domain| or |struct generic_pm_domain|,
170  however these are not independent devices that happen to share a power
171  switch, but rather the MMU device serves the busmaster device and is
172  useless without it.  A device link creates a synthetic hierarchical
173  relationship between the devices and is thus more apt.
174
175* A Thunderbolt host controller comprises a number of PCIe hotplug ports
176  and an NHI device to manage the PCIe switch.  On resume from system sleep,
177  the NHI device needs to re-establish PCI tunnels to attached devices
178  before the hotplug ports can resume.  If the hotplug ports were children
179  of the NHI, this resume order would automatically be enforced by the
180  PM core, but unfortunately they're aunts.  The solution is to add
181  device links from the hotplug ports (consumers) to the NHI device
182  (supplier).  A driver presence dependency is not necessary for this
183  use case.
184
185* Discrete GPUs in hybrid graphics laptops often feature an HDA controller
186  for HDMI/DP audio.  In the device hierarchy the HDA controller is a sibling
187  of the VGA device, yet both share the same power domain and the HDA
188  controller is only ever needed when an HDMI/DP display is attached to the
189  VGA device.  A device link from the HDA controller (consumer) to the
190  VGA device (supplier) aptly represents this relationship.
191
192* ACPI allows definition of a device start order by way of _DEP objects.
193  A classical example is when ACPI power management methods on one device
194  are implemented in terms of I\ :sup:`2`\ C accesses and require a specific
195  I\ :sup:`2`\ C controller to be present and functional for the power
196  management of the device in question to work.
197
198* In some SoCs a functional dependency exists from display, video codec and
199  video processing IP cores on transparent memory access IP cores that handle
200  burst access and compression/decompression.
201
202Alternatives
203============
204
205* A |struct dev_pm_domain| can be used to override the bus,
206  class or device type callbacks.  It is intended for devices sharing
207  a single on/off switch, however it does not guarantee a specific
208  suspend/resume ordering, this needs to be implemented separately.
209  It also does not by itself track the runtime PM status of the involved
210  devices and turn off the power switch only when all of them are runtime
211  suspended.  Furthermore it cannot be used to enforce a specific shutdown
212  ordering or a driver presence dependency.
213
214* A |struct generic_pm_domain| is a lot more heavyweight than a
215  device link and does not allow for shutdown ordering or driver presence
216  dependencies.  It also cannot be used on ACPI systems.
217
218Implementation
219==============
220
221The device hierarchy, which -- as the name implies -- is a tree,
222becomes a directed acyclic graph once device links are added.
223
224Ordering of these devices during suspend/resume is determined by the
225dpm_list.  During shutdown it is determined by the devices_kset.  With
226no device links present, the two lists are a flattened, one-dimensional
227representations of the device tree such that a device is placed behind
228all its ancestors.  That is achieved by traversing the ACPI namespace
229or OpenFirmware device tree top-down and appending devices to the lists
230as they are discovered.
231
232Once device links are added, the lists need to satisfy the additional
233constraint that a device is placed behind all its suppliers, recursively.
234To ensure this, upon addition of the device link the consumer and the
235entire sub-graph below it (all children and consumers of the consumer)
236are moved to the end of the list.  (Call to :c:func:`device_reorder_to_tail()`
237from :c:func:`device_link_add()`.)
238
239To prevent introduction of dependency loops into the graph, it is
240verified upon device link addition that the supplier is not dependent
241on the consumer or any children or consumers of the consumer.
242(Call to :c:func:`device_is_dependent()` from :c:func:`device_link_add()`.)
243If that constraint is violated, :c:func:`device_link_add()` will return
244``NULL`` and a ``WARNING`` will be logged.
245
246Notably this also prevents the addition of a device link from a parent
247device to a child.  However the converse is allowed, i.e. a device link
248from a child to a parent.  Since the driver core already guarantees
249correct suspend/resume and shutdown ordering between parent and child,
250such a device link only makes sense if a driver presence dependency is
251needed on top of that.  In this case driver authors should weigh
252carefully if a device link is at all the right tool for the purpose.
253A more suitable approach might be to simply use deferred probing or
254add a device flag causing the parent driver to be probed before the
255child one.
256
257State machine
258=============
259
260.. kernel-doc:: include/linux/device.h
261   :functions: device_link_state
262
263::
264
265                 .=============================.
266                 |                             |
267                 v                             |
268 DORMANT <=> AVAILABLE <=> CONSUMER_PROBE => ACTIVE
269    ^                                          |
270    |                                          |
271    '============ SUPPLIER_UNBIND <============'
272
273* The initial state of a device link is automatically determined by
274  :c:func:`device_link_add()` based on the driver presence on the supplier
275  and consumer.  If the link is created before any devices are probed, it
276  is set to ``DL_STATE_DORMANT``.
277
278* When a supplier device is bound to a driver, links to its consumers
279  progress to ``DL_STATE_AVAILABLE``.
280  (Call to :c:func:`device_links_driver_bound()` from
281  :c:func:`driver_bound()`.)
282
283* Before a consumer device is probed, presence of supplier drivers is
284  verified by checking the consumer device is not in the wait_for_suppliers
285  list and by checking that links to suppliers are in ``DL_STATE_AVAILABLE``
286  state.  The state of the links is updated to ``DL_STATE_CONSUMER_PROBE``.
287  (Call to :c:func:`device_links_check_suppliers()` from
288  :c:func:`really_probe()`.)
289  This prevents the supplier from unbinding.
290  (Call to :c:func:`wait_for_device_probe()` from
291  :c:func:`device_links_unbind_consumers()`.)
292
293* If the probe fails, links to suppliers revert back to ``DL_STATE_AVAILABLE``.
294  (Call to :c:func:`device_links_no_driver()` from :c:func:`really_probe()`.)
295
296* If the probe succeeds, links to suppliers progress to ``DL_STATE_ACTIVE``.
297  (Call to :c:func:`device_links_driver_bound()` from :c:func:`driver_bound()`.)
298
299* When the consumer's driver is later on removed, links to suppliers revert
300  back to ``DL_STATE_AVAILABLE``.
301  (Call to :c:func:`__device_links_no_driver()` from
302  :c:func:`device_links_driver_cleanup()`, which in turn is called from
303  :c:func:`__device_release_driver()`.)
304
305* Before a supplier's driver is removed, links to consumers that are not
306  bound to a driver are updated to ``DL_STATE_SUPPLIER_UNBIND``.
307  (Call to :c:func:`device_links_busy()` from
308  :c:func:`__device_release_driver()`.)
309  This prevents the consumers from binding.
310  (Call to :c:func:`device_links_check_suppliers()` from
311  :c:func:`really_probe()`.)
312  Consumers that are bound are freed from their driver; consumers that are
313  probing are waited for until they are done.
314  (Call to :c:func:`device_links_unbind_consumers()` from
315  :c:func:`__device_release_driver()`.)
316  Once all links to consumers are in ``DL_STATE_SUPPLIER_UNBIND`` state,
317  the supplier driver is released and the links revert to ``DL_STATE_DORMANT``.
318  (Call to :c:func:`device_links_driver_cleanup()` from
319  :c:func:`__device_release_driver()`.)
320
321API
322===
323
324.. kernel-doc:: drivers/base/core.c
325   :functions: device_link_add device_link_del device_link_remove
326