Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame] | 1 | PM Quality Of Service Interface. |
| 2 | |
| 3 | This interface provides a kernel and user mode interface for registering |
| 4 | performance expectations by drivers, subsystems and user space applications on |
| 5 | one of the parameters. |
| 6 | |
| 7 | Two different PM QoS frameworks are available: |
| 8 | 1. PM QoS classes for cpu_dma_latency, network_latency, network_throughput, |
| 9 | memory_bandwidth. |
| 10 | 2. the per-device PM QoS framework provides the API to manage the per-device latency |
| 11 | constraints and PM QoS flags. |
| 12 | |
| 13 | Each parameters have defined units: |
| 14 | * latency: usec |
| 15 | * timeout: usec |
| 16 | * throughput: kbs (kilo bit / sec) |
| 17 | * memory bandwidth: mbs (mega bit / sec) |
| 18 | |
| 19 | |
| 20 | 1. PM QoS framework |
| 21 | |
| 22 | The infrastructure exposes multiple misc device nodes one per implemented |
| 23 | parameter. The set of parameters implement is defined by pm_qos_power_init() |
| 24 | and pm_qos_params.h. This is done because having the available parameters |
| 25 | being runtime configurable or changeable from a driver was seen as too easy to |
| 26 | abuse. |
| 27 | |
| 28 | For each parameter a list of performance requests is maintained along with |
| 29 | an aggregated target value. The aggregated target value is updated with |
| 30 | changes to the request list or elements of the list. Typically the |
| 31 | aggregated target value is simply the max or min of the request values held |
| 32 | in the parameter list elements. |
| 33 | Note: the aggregated target value is implemented as an atomic variable so that |
| 34 | reading the aggregated value does not require any locking mechanism. |
| 35 | |
| 36 | |
| 37 | From kernel mode the use of this interface is simple: |
| 38 | |
| 39 | void pm_qos_add_request(handle, param_class, target_value): |
| 40 | Will insert an element into the list for that identified PM QoS class with the |
| 41 | target value. Upon change to this list the new target is recomputed and any |
| 42 | registered notifiers are called only if the target value is now different. |
| 43 | Clients of pm_qos need to save the returned handle for future use in other |
| 44 | pm_qos API functions. |
| 45 | |
| 46 | void pm_qos_update_request(handle, new_target_value): |
| 47 | Will update the list element pointed to by the handle with the new target value |
| 48 | and recompute the new aggregated target, calling the notification tree if the |
| 49 | target is changed. |
| 50 | |
| 51 | void pm_qos_remove_request(handle): |
| 52 | Will remove the element. After removal it will update the aggregate target and |
| 53 | call the notification tree if the target was changed as a result of removing |
| 54 | the request. |
| 55 | |
| 56 | int pm_qos_request(param_class): |
| 57 | Returns the aggregated value for a given PM QoS class. |
| 58 | |
| 59 | int pm_qos_request_active(handle): |
| 60 | Returns if the request is still active, i.e. it has not been removed from a |
| 61 | PM QoS class constraints list. |
| 62 | |
| 63 | int pm_qos_add_notifier(param_class, notifier): |
| 64 | Adds a notification callback function to the PM QoS class. The callback is |
| 65 | called when the aggregated value for the PM QoS class is changed. |
| 66 | |
| 67 | int pm_qos_remove_notifier(int param_class, notifier): |
| 68 | Removes the notification callback function for the PM QoS class. |
| 69 | |
| 70 | |
| 71 | From user mode: |
| 72 | Only processes can register a pm_qos request. To provide for automatic |
| 73 | cleanup of a process, the interface requires the process to register its |
| 74 | parameter requests in the following way: |
| 75 | |
| 76 | To register the default pm_qos target for the specific parameter, the process |
| 77 | must open one of /dev/[cpu_dma_latency, network_latency, network_throughput] |
| 78 | |
| 79 | As long as the device node is held open that process has a registered |
| 80 | request on the parameter. |
| 81 | |
| 82 | To change the requested target value the process needs to write an s32 value to |
| 83 | the open device node. Alternatively the user mode program could write a hex |
| 84 | string for the value using 10 char long format e.g. "0x12345678". This |
| 85 | translates to a pm_qos_update_request call. |
| 86 | |
| 87 | To remove the user mode request for a target value simply close the device |
| 88 | node. |
| 89 | |
| 90 | |
| 91 | 2. PM QoS per-device latency and flags framework |
| 92 | |
| 93 | For each device, there are three lists of PM QoS requests. Two of them are |
| 94 | maintained along with the aggregated targets of resume latency and active |
| 95 | state latency tolerance (in microseconds) and the third one is for PM QoS flags. |
| 96 | Values are updated in response to changes of the request list. |
| 97 | |
| 98 | The target values of resume latency and active state latency tolerance are |
| 99 | simply the minimum of the request values held in the parameter list elements. |
| 100 | The PM QoS flags aggregate value is a gather (bitwise OR) of all list elements' |
| 101 | values. Two device PM QoS flags are defined currently: PM_QOS_FLAG_NO_POWER_OFF |
| 102 | and PM_QOS_FLAG_REMOTE_WAKEUP. |
| 103 | |
| 104 | Note: The aggregated target values are implemented in such a way that reading |
| 105 | the aggregated value does not require any locking mechanism. |
| 106 | |
| 107 | |
| 108 | From kernel mode the use of this interface is the following: |
| 109 | |
| 110 | int dev_pm_qos_add_request(device, handle, type, value): |
| 111 | Will insert an element into the list for that identified device with the |
| 112 | target value. Upon change to this list the new target is recomputed and any |
| 113 | registered notifiers are called only if the target value is now different. |
| 114 | Clients of dev_pm_qos need to save the handle for future use in other |
| 115 | dev_pm_qos API functions. |
| 116 | |
| 117 | int dev_pm_qos_update_request(handle, new_value): |
| 118 | Will update the list element pointed to by the handle with the new target value |
| 119 | and recompute the new aggregated target, calling the notification trees if the |
| 120 | target is changed. |
| 121 | |
| 122 | int dev_pm_qos_remove_request(handle): |
| 123 | Will remove the element. After removal it will update the aggregate target and |
| 124 | call the notification trees if the target was changed as a result of removing |
| 125 | the request. |
| 126 | |
| 127 | s32 dev_pm_qos_read_value(device): |
| 128 | Returns the aggregated value for a given device's constraints list. |
| 129 | |
| 130 | enum pm_qos_flags_status dev_pm_qos_flags(device, mask) |
| 131 | Check PM QoS flags of the given device against the given mask of flags. |
| 132 | The meaning of the return values is as follows: |
| 133 | PM_QOS_FLAGS_ALL: All flags from the mask are set |
| 134 | PM_QOS_FLAGS_SOME: Some flags from the mask are set |
| 135 | PM_QOS_FLAGS_NONE: No flags from the mask are set |
| 136 | PM_QOS_FLAGS_UNDEFINED: The device's PM QoS structure has not been |
| 137 | initialized or the list of requests is empty. |
| 138 | |
| 139 | int dev_pm_qos_add_ancestor_request(dev, handle, type, value) |
| 140 | Add a PM QoS request for the first direct ancestor of the given device whose |
| 141 | power.ignore_children flag is unset (for DEV_PM_QOS_RESUME_LATENCY requests) |
| 142 | or whose power.set_latency_tolerance callback pointer is not NULL (for |
| 143 | DEV_PM_QOS_LATENCY_TOLERANCE requests). |
| 144 | |
| 145 | int dev_pm_qos_expose_latency_limit(device, value) |
| 146 | Add a request to the device's PM QoS list of resume latency constraints and |
| 147 | create a sysfs attribute pm_qos_resume_latency_us under the device's power |
| 148 | directory allowing user space to manipulate that request. |
| 149 | |
| 150 | void dev_pm_qos_hide_latency_limit(device) |
| 151 | Drop the request added by dev_pm_qos_expose_latency_limit() from the device's |
| 152 | PM QoS list of resume latency constraints and remove sysfs attribute |
| 153 | pm_qos_resume_latency_us from the device's power directory. |
| 154 | |
| 155 | int dev_pm_qos_expose_flags(device, value) |
| 156 | Add a request to the device's PM QoS list of flags and create sysfs attributes |
| 157 | pm_qos_no_power_off and pm_qos_remote_wakeup under the device's power directory |
| 158 | allowing user space to change these flags' value. |
| 159 | |
| 160 | void dev_pm_qos_hide_flags(device) |
| 161 | Drop the request added by dev_pm_qos_expose_flags() from the device's PM QoS list |
| 162 | of flags and remove sysfs attributes pm_qos_no_power_off and pm_qos_remote_wakeup |
| 163 | under the device's power directory. |
| 164 | |
| 165 | Notification mechanisms: |
| 166 | The per-device PM QoS framework has 2 different and distinct notification trees: |
| 167 | a per-device notification tree and a global notification tree. |
| 168 | |
| 169 | int dev_pm_qos_add_notifier(device, notifier): |
| 170 | Adds a notification callback function for the device. |
| 171 | The callback is called when the aggregated value of the device constraints list |
| 172 | is changed (for resume latency device PM QoS only). |
| 173 | |
| 174 | int dev_pm_qos_remove_notifier(device, notifier): |
| 175 | Removes the notification callback function for the device. |
| 176 | |
| 177 | int dev_pm_qos_add_global_notifier(notifier): |
| 178 | Adds a notification callback function in the global notification tree of the |
| 179 | framework. |
| 180 | The callback is called when the aggregated value for any device is changed |
| 181 | (for resume latency device PM QoS only). |
| 182 | |
| 183 | int dev_pm_qos_remove_global_notifier(notifier): |
| 184 | Removes the notification callback function from the global notification tree |
| 185 | of the framework. |
| 186 | |
| 187 | |
| 188 | Active state latency tolerance |
| 189 | |
| 190 | This device PM QoS type is used to support systems in which hardware may switch |
| 191 | to energy-saving operation modes on the fly. In those systems, if the operation |
| 192 | mode chosen by the hardware attempts to save energy in an overly aggressive way, |
| 193 | it may cause excess latencies to be visible to software, causing it to miss |
| 194 | certain protocol requirements or target frame or sample rates etc. |
| 195 | |
| 196 | If there is a latency tolerance control mechanism for a given device available |
| 197 | to software, the .set_latency_tolerance callback in that device's dev_pm_info |
| 198 | structure should be populated. The routine pointed to by it is should implement |
| 199 | whatever is necessary to transfer the effective requirement value to the |
| 200 | hardware. |
| 201 | |
| 202 | Whenever the effective latency tolerance changes for the device, its |
| 203 | .set_latency_tolerance() callback will be executed and the effective value will |
| 204 | be passed to it. If that value is negative, which means that the list of |
| 205 | latency tolerance requirements for the device is empty, the callback is expected |
| 206 | to switch the underlying hardware latency tolerance control mechanism to an |
| 207 | autonomous mode if available. If that value is PM_QOS_LATENCY_ANY, in turn, and |
| 208 | the hardware supports a special "no requirement" setting, the callback is |
| 209 | expected to use it. That allows software to prevent the hardware from |
| 210 | automatically updating the device's latency tolerance in response to its power |
| 211 | state changes (e.g. during transitions from D3cold to D0), which generally may |
| 212 | be done in the autonomous latency tolerance control mode. |
| 213 | |
| 214 | If .set_latency_tolerance() is present for the device, sysfs attribute |
| 215 | pm_qos_latency_tolerance_us will be present in the devivce's power directory. |
| 216 | Then, user space can use that attribute to specify its latency tolerance |
| 217 | requirement for the device, if any. Writing "any" to it means "no requirement, |
| 218 | but do not let the hardware control latency tolerance" and writing "auto" to it |
| 219 | allows the hardware to be switched to the autonomous mode if there are no other |
| 220 | requirements from the kernel side in the device's list. |
| 221 | |
| 222 | Kernel code can use the functions described above along with the |
| 223 | DEV_PM_QOS_LATENCY_TOLERANCE device PM QoS type to add, remove and update |
| 224 | latency tolerance requirements for devices. |