xref: /linux/Documentation/power/pm_qos_interface.rst (revision b4ada0618eed0fbd1b1630f73deb048c592b06a1)
1===============================
2PM Quality Of Service Interface
3===============================
4
5This interface provides a kernel and user mode interface for registering
6performance expectations by drivers, subsystems and user space applications on
7one of the parameters.
8
9Two different PM QoS frameworks are available:
10 * CPU latency QoS.
11 * The per-device PM QoS framework provides the API to manage the
12   per-device latency constraints and PM QoS flags.
13
14The latency unit used in the PM QoS framework is the microsecond (usec).
15
16
171. PM QoS framework
18===================
19
20A global list of CPU latency QoS requests is maintained along with an aggregated
21(effective) target value.  The aggregated target value is updated with changes
22to the request list or elements of the list.  For CPU latency QoS, the
23aggregated target value is simply the min of the request values held in the list
24elements.
25
26Note: the aggregated target value is implemented as an atomic variable so that
27reading the aggregated value does not require any locking mechanism.
28
29From kernel space the use of this interface is simple:
30
31void cpu_latency_qos_add_request(handle, target_value):
32  Will insert an element into the CPU latency QoS list with the target value.
33  Upon change to this list the new target is recomputed and any registered
34  notifiers are called only if the target value is now different.
35  Clients of PM QoS need to save the returned handle for future use in other
36  PM QoS API functions.
37
38void cpu_latency_qos_update_request(handle, new_target_value):
39  Will update the list element pointed to by the handle with the new target
40  value and recompute the new aggregated target, calling the notification tree
41  if the target is changed.
42
43void cpu_latency_qos_remove_request(handle):
44  Will remove the element.  After removal it will update the aggregate target
45  and call the notification tree if the target was changed as a result of
46  removing the request.
47
48int cpu_latency_qos_limit():
49  Returns the aggregated value for the CPU latency QoS.
50
51int cpu_latency_qos_request_active(handle):
52  Returns if the request is still active, i.e. it has not been removed from the
53  CPU latency QoS list.
54
55
56From user space:
57
58The infrastructure exposes one device node, /dev/cpu_dma_latency, for the CPU
59latency QoS.
60
61Only processes can register a PM QoS request.  To provide for automatic
62cleanup of a process, the interface requires the process to register its
63parameter requests as follows.
64
65To register the default PM QoS target for the CPU latency QoS, the process must
66open /dev/cpu_dma_latency.
67
68As long as the device node is held open that process has a registered
69request on the parameter.
70
71To change the requested target value, the process needs to write an s32 value to
72the open device node.  Alternatively, it can write a hex string for the value
73using the 10 char long format e.g. "0x12345678".  This translates to a
74cpu_latency_qos_update_request() call.
75
76To remove the user mode request for a target value simply close the device
77node.
78
79
802. PM QoS per-device latency and flags framework
81================================================
82
83For each device, there are three lists of PM QoS requests. Two of them are
84maintained along with the aggregated targets of resume latency and active
85state latency tolerance (in microseconds) and the third one is for PM QoS flags.
86Values are updated in response to changes of the request list.
87
88The target values of resume latency and active state latency tolerance are
89simply the minimum of the request values held in the parameter list elements.
90The PM QoS flags aggregate value is a gather (bitwise OR) of all list elements'
91values.  One device PM QoS flag is defined currently: PM_QOS_FLAG_NO_POWER_OFF.
92
93Note: The aggregated target values are implemented in such a way that reading
94the aggregated value does not require any locking mechanism.
95
96
97From kernel mode the use of this interface is the following:
98
99int dev_pm_qos_add_request(device, handle, type, value):
100  Will insert an element into the list for that identified device with the
101  target value.  Upon change to this list the new target is recomputed and any
102  registered notifiers are called only if the target value is now different.
103  Clients of dev_pm_qos need to save the handle for future use in other
104  dev_pm_qos API functions.
105
106int dev_pm_qos_update_request(handle, new_value):
107  Will update the list element pointed to by the handle with the new target
108  value and recompute the new aggregated target, calling the notification
109  trees if the target is changed.
110
111int dev_pm_qos_remove_request(handle):
112  Will remove the element.  After removal it will update the aggregate target
113  and call the notification trees if the target was changed as a result of
114  removing the request.
115
116s32 dev_pm_qos_read_value(device, type):
117  Returns the aggregated value for a given device's constraints list.
118
119enum pm_qos_flags_status dev_pm_qos_flags(device, mask)
120  Check PM QoS flags of the given device against the given mask of flags.
121  The meaning of the return values is as follows:
122
123	PM_QOS_FLAGS_ALL:
124		All flags from the mask are set
125	PM_QOS_FLAGS_SOME:
126		Some flags from the mask are set
127	PM_QOS_FLAGS_NONE:
128		No flags from the mask are set
129	PM_QOS_FLAGS_UNDEFINED:
130		The device's PM QoS structure has not been initialized
131		or the list of requests is empty.
132
133int dev_pm_qos_add_ancestor_request(dev, handle, type, value)
134  Add a PM QoS request for the first direct ancestor of the given device whose
135  power.ignore_children flag is unset (for DEV_PM_QOS_RESUME_LATENCY requests)
136  or whose power.set_latency_tolerance callback pointer is not NULL (for
137  DEV_PM_QOS_LATENCY_TOLERANCE requests).
138
139int dev_pm_qos_expose_latency_limit(device, value)
140  Add a request to the device's PM QoS list of resume latency constraints and
141  create a sysfs attribute pm_qos_resume_latency_us under the device's power
142  directory allowing user space to manipulate that request.
143
144void dev_pm_qos_hide_latency_limit(device)
145  Drop the request added by dev_pm_qos_expose_latency_limit() from the device's
146  PM QoS list of resume latency constraints and remove sysfs attribute
147  pm_qos_resume_latency_us from the device's power directory.
148
149int dev_pm_qos_expose_flags(device, value)
150  Add a request to the device's PM QoS list of flags and create sysfs attribute
151  pm_qos_no_power_off under the device's power directory allowing user space to
152  change the value of the PM_QOS_FLAG_NO_POWER_OFF flag.
153
154void dev_pm_qos_hide_flags(device)
155  Drop the request added by dev_pm_qos_expose_flags() from the device's PM QoS
156  list of flags and remove sysfs attribute pm_qos_no_power_off from the device's
157  power directory.
158
159Notification mechanisms:
160
161The per-device PM QoS framework has a per-device notification tree.
162
163int dev_pm_qos_add_notifier(device, notifier, type):
164  Adds a notification callback function for the device for a particular request
165  type.
166
167  The callback is called when the aggregated value of the device constraints
168  list is changed.
169
170int dev_pm_qos_remove_notifier(device, notifier, type):
171  Removes the notification callback function for the device.
172
173
174Active state latency tolerance
175^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
176
177This device PM QoS type is used to support systems in which hardware may switch
178to energy-saving operation modes on the fly.  In those systems, if the operation
179mode chosen by the hardware attempts to save energy in an overly aggressive way,
180it may cause excess latencies to be visible to software, causing it to miss
181certain protocol requirements or target frame or sample rates etc.
182
183If there is a latency tolerance control mechanism for a given device available
184to software, the .set_latency_tolerance callback in that device's dev_pm_info
185structure should be populated.  The routine pointed to by it is should implement
186whatever is necessary to transfer the effective requirement value to the
187hardware.
188
189Whenever the effective latency tolerance changes for the device, its
190.set_latency_tolerance() callback will be executed and the effective value will
191be passed to it.  If that value is negative, which means that the list of
192latency tolerance requirements for the device is empty, the callback is expected
193to switch the underlying hardware latency tolerance control mechanism to an
194autonomous mode if available.  If that value is PM_QOS_LATENCY_ANY, in turn, and
195the hardware supports a special "no requirement" setting, the callback is
196expected to use it.  That allows software to prevent the hardware from
197automatically updating the device's latency tolerance in response to its power
198state changes (e.g. during transitions from D3cold to D0), which generally may
199be done in the autonomous latency tolerance control mode.
200
201If .set_latency_tolerance() is present for the device, sysfs attribute
202pm_qos_latency_tolerance_us will be present in the devivce's power directory.
203Then, user space can use that attribute to specify its latency tolerance
204requirement for the device, if any.  Writing "any" to it means "no requirement,
205but do not let the hardware control latency tolerance" and writing "auto" to it
206allows the hardware to be switched to the autonomous mode if there are no other
207requirements from the kernel side in the device's list.
208
209Kernel code can use the functions described above along with the
210DEV_PM_QOS_LATENCY_TOLERANCE device PM QoS type to add, remove and update
211latency tolerance requirements for devices.
212