diff --git a/Documentation/admin-guide/pm/cpuidle.rst b/Documentation/admin-guide/pm/cpuidle.rst index 106379e2619f..9c58b35a81cb 100644 --- a/Documentation/admin-guide/pm/cpuidle.rst +++ b/Documentation/admin-guide/pm/cpuidle.rst @@ -155,14 +155,14 @@ governor uses that information depends on what algorithm is implemented by it and that is the primary reason for having more than one governor in the ``CPUIdle`` subsystem. -There are two ``CPUIdle`` governors available, ``menu`` and ``ladder``. Which -of them is used depends on the configuration of the kernel and in particular on -whether or not the scheduler tick can be `stopped by the idle -loop `_. It is possible to change the governor at run time -if the ``cpuidle_sysfs_switch`` command line parameter has been passed to the -kernel, but that is not safe in general, so it should not be done on production -systems (that may change in the future, though). The name of the ``CPUIdle`` -governor currently used by the kernel can be read from the +There are three ``CPUIdle`` governors available, ``menu``, `TEO `_ +and ``ladder``. Which of them is used by default depends on the configuration +of the kernel and in particular on whether or not the scheduler tick can be +`stopped by the idle loop `_. It is possible to change the +governor at run time if the ``cpuidle_sysfs_switch`` command line parameter has +been passed to the kernel, but that is not safe in general, so it should not be +done on production systems (that may change in the future, though). The name of +the ``CPUIdle`` governor currently used by the kernel can be read from the :file:`current_governor_ro` (or :file:`current_governor` if ``cpuidle_sysfs_switch`` is present in the kernel command line) file under :file:`/sys/devices/system/cpu/cpuidle/` in ``sysfs``. @@ -256,6 +256,8 @@ the ``menu`` governor by default and if it is not tickless, the default ``CPUIdle`` governor on it will be ``ladder``. +.. _menu-gov: + The ``menu`` Governor ===================== @@ -333,6 +335,92 @@ that time, the governor may need to select a shallower state with a suitable target residency. +.. _teo-gov: + +The Timer Events Oriented (TEO) Governor +======================================== + +The timer events oriented (TEO) governor is an alternative ``CPUIdle`` governor +for tickless systems. It follows the same basic strategy as the ``menu`` `one +`_: it always tries to find the deepest idle state suitable for the +given conditions. However, it applies a different approach to that problem. + +First, it does not use sleep length correction factors, but instead it attempts +to correlate the observed idle duration values with the available idle states +and use that information to pick up the idle state that is most likely to +"match" the upcoming CPU idle interval. Second, it does not take the tasks +that were running on the given CPU in the past and are waiting on some I/O +operations to complete now at all (there is no guarantee that they will run on +the same CPU when they become runnable again) and the pattern detection code in +it avoids taking timer wakeups into account. It also only uses idle duration +values less than the current time till the closest timer (with the scheduler +tick excluded) for that purpose. + +Like in the ``menu`` governor `case `_, the first step is to obtain +the *sleep length*, which is the time until the closest timer event with the +assumption that the scheduler tick will be stopped (that also is the upper bound +on the time until the next CPU wakeup). That value is then used to preselect an +idle state on the basis of three metrics maintained for each idle state provided +by the ``CPUIdle`` driver: ``hits``, ``misses`` and ``early_hits``. + +The ``hits`` and ``misses`` metrics measure the likelihood that a given idle +state will "match" the observed (post-wakeup) idle duration if it "matches" the +sleep length. They both are subject to decay (after a CPU wakeup) every time +the target residency of the idle state corresponding to them is less than or +equal to the sleep length and the target residency of the next idle state is +greater than the sleep length (that is, when the idle state corresponding to +them "matches" the sleep length). The ``hits`` metric is increased if the +former condition is satisfied and the target residency of the given idle state +is less than or equal to the observed idle duration and the target residency of +the next idle state is greater than the observed idle duration at the same time +(that is, it is increased when the given idle state "matches" both the sleep +length and the observed idle duration). In turn, the ``misses`` metric is +increased when the given idle state "matches" the sleep length only and the +observed idle duration is too short for its target residency. + +The ``early_hits`` metric measures the likelihood that a given idle state will +"match" the observed (post-wakeup) idle duration if it does not "match" the +sleep length. It is subject to decay on every CPU wakeup and it is increased +when the idle state corresponding to it "matches" the observed (post-wakeup) +idle duration and the target residency of the next idle state is less than or +equal to the sleep length (i.e. the idle state "matching" the sleep length is +deeper than the given one). + +The governor walks the list of idle states provided by the ``CPUIdle`` driver +and finds the last (deepest) one with the target residency less than or equal +to the sleep length. Then, the ``hits`` and ``misses`` metrics of that idle +state are compared with each other and it is preselected if the ``hits`` one is +greater (which means that that idle state is likely to "match" the observed idle +duration after CPU wakeup). If the ``misses`` one is greater, the governor +preselects the shallower idle state with the maximum ``early_hits`` metric +(or if there are multiple shallower idle states with equal ``early_hits`` +metric which also is the maximum, the shallowest of them will be preselected). +[If there is a wakeup latency constraint coming from the `PM QoS framework +`_ which is hit before reaching the deepest idle state with the +target residency within the sleep length, the deepest idle state with the exit +latency within the constraint is preselected without consulting the ``hits``, +``misses`` and ``early_hits`` metrics.] + +Next, the governor takes several idle duration values observed most recently +into consideration and if at least a half of them are greater than or equal to +the target residency of the preselected idle state, that idle state becomes the +final candidate to ask for. Otherwise, the average of the most recent idle +duration values below the target residency of the preselected idle state is +computed and the governor walks the idle states shallower than the preselected +one and finds the deepest of them with the target residency within that average. +That idle state is then taken as the final candidate to ask for. + +Still, at this point the governor may need to refine the idle state selection if +it has not decided to `stop the scheduler tick `_. That +generally happens if the target residency of the idle state selected so far is +less than the tick period and the tick has not been stopped already (in a +previous iteration of the idle loop). Then, like in the ``menu`` governor +`case `_, the sleep length used in the previous computations may not +reflect the real time until the closest timer event and if it really is greater +than that time, a shallower state with a suitable target residency may need to +be selected. + + .. _idle-states-representation: Representation of Idle States diff --git a/Documentation/cpuidle/driver.txt b/Documentation/cpuidle/driver.txt deleted file mode 100644 index 1b0d81d92583..000000000000 --- a/Documentation/cpuidle/driver.txt +++ /dev/null @@ -1,37 +0,0 @@ - - - Supporting multiple CPU idle levels in kernel - - cpuidle drivers - - - - -cpuidle driver hooks into the cpuidle infrastructure and handles the -architecture/platform dependent part of CPU idle states. Driver -provides the platform idle state detection capability and also -has mechanisms in place to support actual entry-exit into CPU idle states. - -cpuidle driver initializes the cpuidle_device structure for each CPU device -and registers with cpuidle using cpuidle_register_device. - -If all the idle states are the same, the wrapper function cpuidle_register -could be used instead. - -It can also support the dynamic changes (like battery <-> AC), by using -cpuidle_pause_and_lock, cpuidle_disable_device and cpuidle_enable_device, -cpuidle_resume_and_unlock. - -Interfaces: -extern int cpuidle_register(struct cpuidle_driver *drv, - const struct cpumask *const coupled_cpus); -extern int cpuidle_unregister(struct cpuidle_driver *drv); -extern int cpuidle_register_driver(struct cpuidle_driver *drv); -extern void cpuidle_unregister_driver(struct cpuidle_driver *drv); -extern int cpuidle_register_device(struct cpuidle_device *dev); -extern void cpuidle_unregister_device(struct cpuidle_device *dev); - -extern void cpuidle_pause_and_lock(void); -extern void cpuidle_resume_and_unlock(void); -extern int cpuidle_enable_device(struct cpuidle_device *dev); -extern void cpuidle_disable_device(struct cpuidle_device *dev); diff --git a/Documentation/cpuidle/governor.txt b/Documentation/cpuidle/governor.txt deleted file mode 100644 index d9020f5e847b..000000000000 --- a/Documentation/cpuidle/governor.txt +++ /dev/null @@ -1,28 +0,0 @@ - - - - Supporting multiple CPU idle levels in kernel - - cpuidle governors - - - - -cpuidle governor is policy routine that decides what idle state to enter at -any given time. cpuidle core uses different callbacks to the governor. - -* enable() to enable governor for a particular device -* disable() to disable governor for a particular device -* select() to select an idle state to enter -* reflect() called after returning from the idle state, which can be used - by the governor for some record keeping. - -More than one governor can be registered at the same time and -users can switch between drivers using /sysfs interface (when enabled). -More than one governor part is supported for developers to easily experiment -with different governors. By default, most optimal governor based on your -kernel configuration and platform will be selected by cpuidle. - -Interfaces: -extern int cpuidle_register_governor(struct cpuidle_governor *gov); -struct cpuidle_governor diff --git a/Documentation/driver-api/pm/cpuidle.rst b/Documentation/driver-api/pm/cpuidle.rst new file mode 100644 index 000000000000..5842ab621a58 --- /dev/null +++ b/Documentation/driver-api/pm/cpuidle.rst @@ -0,0 +1,282 @@ +.. |struct cpuidle_governor| replace:: :c:type:`struct cpuidle_governor ` +.. |struct cpuidle_device| replace:: :c:type:`struct cpuidle_device ` +.. |struct cpuidle_driver| replace:: :c:type:`struct cpuidle_driver ` +.. |struct cpuidle_state| replace:: :c:type:`struct cpuidle_state ` + +======================== +CPU Idle Time Management +======================== + +:: + + Copyright (c) 2019 Intel Corp., Rafael J. Wysocki + + +CPU Idle Time Management Subsystem +================================== + +Every time one of the logical CPUs in the system (the entities that appear to +fetch and execute instructions: hardware threads, if present, or processor +cores) is idle after an interrupt or equivalent wakeup event, which means that +there are no tasks to run on it except for the special "idle" task associated +with it, there is an opportunity to save energy for the processor that it +belongs to. That can be done by making the idle logical CPU stop fetching +instructions from memory and putting some of the processor's functional units +depended on by it into an idle state in which they will draw less power. + +However, there may be multiple different idle states that can be used in such a +situation in principle, so it may be necessary to find the most suitable one +(from the kernel perspective) and ask the processor to use (or "enter") that +particular idle state. That is the role of the CPU idle time management +subsystem in the kernel, called ``CPUIdle``. + +The design of ``CPUIdle`` is modular and based on the code duplication avoidance +principle, so the generic code that in principle need not depend on the hardware +or platform design details in it is separate from the code that interacts with +the hardware. It generally is divided into three categories of functional +units: *governors* responsible for selecting idle states to ask the processor +to enter, *drivers* that pass the governors' decisions on to the hardware and +the *core* providing a common framework for them. + + +CPU Idle Time Governors +======================= + +A CPU idle time (``CPUIdle``) governor is a bundle of policy code invoked when +one of the logical CPUs in the system turns out to be idle. Its role is to +select an idle state to ask the processor to enter in order to save some energy. + +``CPUIdle`` governors are generic and each of them can be used on any hardware +platform that the Linux kernel can run on. For this reason, data structures +operated on by them cannot depend on any hardware architecture or platform +design details as well. + +The governor itself is represented by a |struct cpuidle_governor| object +containing four callback pointers, :c:member:`enable`, :c:member:`disable`, +:c:member:`select`, :c:member:`reflect`, a :c:member:`rating` field described +below, and a name (string) used for identifying it. + +For the governor to be available at all, that object needs to be registered +with the ``CPUIdle`` core by calling :c:func:`cpuidle_register_governor()` with +a pointer to it passed as the argument. If successful, that causes the core to +add the governor to the global list of available governors and, if it is the +only one in the list (that is, the list was empty before) or the value of its +:c:member:`rating` field is greater than the value of that field for the +governor currently in use, or the name of the new governor was passed to the +kernel as the value of the ``cpuidle.governor=`` command line parameter, the new +governor will be used from that point on (there can be only one ``CPUIdle`` +governor in use at a time). Also, if ``cpuidle_sysfs_switch`` is passed to the +kernel in the command line, user space can choose the ``CPUIdle`` governor to +use at run time via ``sysfs``. + +Once registered, ``CPUIdle`` governors cannot be unregistered, so it is not +practical to put them into loadable kernel modules. + +The interface between ``CPUIdle`` governors and the core consists of four +callbacks: + +:c:member:`enable` + :: + + int (*enable) (struct cpuidle_driver *drv, struct cpuidle_device *dev); + + The role of this callback is to prepare the governor for handling the + (logical) CPU represented by the |struct cpuidle_device| object pointed + to by the ``dev`` argument. The |struct cpuidle_driver| object pointed + to by the ``drv`` argument represents the ``CPUIdle`` driver to be used + with that CPU (among other things, it should contain the list of + |struct cpuidle_state| objects representing idle states that the + processor holding the given CPU can be asked to enter). + + It may fail, in which case it is expected to return a negative error + code, and that causes the kernel to run the architecture-specific + default code for idle CPUs on the CPU in question instead of ``CPUIdle`` + until the ``->enable()`` governor callback is invoked for that CPU + again. + +:c:member:`disable` + :: + + void (*disable) (struct cpuidle_driver *drv, struct cpuidle_device *dev); + + Called to make the governor stop handling the (logical) CPU represented + by the |struct cpuidle_device| object pointed to by the ``dev`` + argument. + + It is expected to reverse any changes made by the ``->enable()`` + callback when it was last invoked for the target CPU, free all memory + allocated by that callback and so on. + +:c:member:`select` + :: + + int (*select) (struct cpuidle_driver *drv, struct cpuidle_device *dev, + bool *stop_tick); + + Called to select an idle state for the processor holding the (logical) + CPU represented by the |struct cpuidle_device| object pointed to by the + ``dev`` argument. + + The list of idle states to take into consideration is represented by the + :c:member:`states` array of |struct cpuidle_state| objects held by the + |struct cpuidle_driver| object pointed to by the ``drv`` argument (which + represents the ``CPUIdle`` driver to be used with the CPU at hand). The + value returned by this callback is interpreted as an index into that + array (unless it is a negative error code). + + The ``stop_tick`` argument is used to indicate whether or not to stop + the scheduler tick before asking the processor to enter the selected + idle state. When the ``bool`` variable pointed to by it (which is set + to ``true`` before invoking this callback) is cleared to ``false``, the + processor will be asked to enter the selected idle state without + stopping the scheduler tick on the given CPU (if the tick has been + stopped on that CPU already, however, it will not be restarted before + asking the processor to enter the idle state). + + This callback is mandatory (i.e. the :c:member:`select` callback pointer + in |struct cpuidle_governor| must not be ``NULL`` for the registration + of the governor to succeed). + +:c:member:`reflect` + :: + + void (*reflect) (struct cpuidle_device *dev, int index); + + Called to allow the governor to evaluate the accuracy of the idle state + selection made by the ``->select()`` callback (when it was invoked last + time) and possibly use the result of that to improve the accuracy of + idle state selections in the future. + +In addition, ``CPUIdle`` governors are required to take power management +quality of service (PM QoS) constraints on the processor wakeup latency into +account when selecting idle states. In order to obtain the current effective +PM QoS wakeup latency constraint for a given CPU, a ``CPUIdle`` governor is +expected to pass the number of the CPU to +:c:func:`cpuidle_governor_latency_req()`. Then, the governor's ``->select()`` +callback must not return the index of an indle state whose +:c:member:`exit_latency` value is greater than the number returned by that +function. + + +CPU Idle Time Management Drivers +================================ + +CPU idle time management (``CPUIdle``) drivers provide an interface between the +other parts of ``CPUIdle`` and the hardware. + +First of all, a ``CPUIdle`` driver has to populate the :c:member:`states` array +of |struct cpuidle_state| objects included in the |struct cpuidle_driver| object +representing it. Going forward this array will represent the list of available +idle states that the processor hardware can be asked to enter shared by all of +the logical CPUs handled by the given driver. + +The entries in the :c:member:`states` array are expected to be sorted by the +value of the :c:member:`target_residency` field in |struct cpuidle_state| in +the ascending order (that is, index 0 should correspond to the idle state with +the minimum value of :c:member:`target_residency`). [Since the +:c:member:`target_residency` value is expected to reflect the "depth" of the +idle state represented by the |struct cpuidle_state| object holding it, this +sorting order should be the same as the ascending sorting order by the idle +state "depth".] + +Three fields in |struct cpuidle_state| are used by the existing ``CPUIdle`` +governors for computations related to idle state selection: + +:c:member:`target_residency` + Minimum time to spend in this idle state including the time needed to + enter it (which may be substantial) to save more energy than could + be saved by staying in a shallower idle state for the same amount of + time, in microseconds. + +:c:member:`exit_latency` + Maximum time it will take a CPU asking the processor to enter this idle + state to start executing the first instruction after a wakeup from it, + in microseconds. + +:c:member:`flags` + Flags representing idle state properties. Currently, governors only use + the ``CPUIDLE_FLAG_POLLING`` flag which is set if the given object + does not represent a real idle state, but an interface to a software + "loop" that can be used in order to avoid asking the processor to enter + any idle state at all. [There are other flags used by the ``CPUIdle`` + core in special situations.] + +The :c:member:`enter` callback pointer in |struct cpuidle_state|, which must not +be ``NULL``, points to the routine to execute in order to ask the processor to +enter this particular idle state: + +:: + + void (*enter) (struct cpuidle_device *dev, struct cpuidle_driver *drv, + int index); + +The first two arguments of it point to the |struct cpuidle_device| object +representing the logical CPU running this callback and the +|struct cpuidle_driver| object representing the driver itself, respectively, +and the last one is an index of the |struct cpuidle_state| entry in the driver's +:c:member:`states` array representing the idle state to ask the processor to +enter. + +The analogous ``->enter_s2idle()`` callback in |struct cpuidle_state| is used +only for implementing the suspend-to-idle system-wide power management feature. +The difference between in and ``->enter()`` is that it must not re-enable +interrupts at any point (even temporarily) or attempt to change the states of +clock event devices, which the ``->enter()`` callback may do sometimes. + +Once the :c:member:`states` array has been populated, the number of valid +entries in it has to be stored in the :c:member:`state_count` field of the +|struct cpuidle_driver| object representing the driver. Moreover, if any +entries in the :c:member:`states` array represent "coupled" idle states (that +is, idle states that can only be asked for if multiple related logical CPUs are +idle), the :c:member:`safe_state_index` field in |struct cpuidle_driver| needs +to be the index of an idle state that is not "coupled" (that is, one that can be +asked for if only one logical CPU is idle). + +In addition to that, if the given ``CPUIdle`` driver is only going to handle a +subset of logical CPUs in the system, the :c:member:`cpumask` field in its +|struct cpuidle_driver| object must point to the set (mask) of CPUs that will be +handled by it. + +A ``CPUIdle`` driver can only be used after it has been registered. If there +are no "coupled" idle state entries in the driver's :c:member:`states` array, +that can be accomplished by passing the driver's |struct cpuidle_driver| object +to :c:func:`cpuidle_register_driver()`. Otherwise, :c:func:`cpuidle_register()` +should be used for this purpose. + +However, it also is necessary to register |struct cpuidle_device| objects for +all of the logical CPUs to be handled by the given ``CPUIdle`` driver with the +help of :c:func:`cpuidle_register_device()` after the driver has been registered +and :c:func:`cpuidle_register_driver()`, unlike :c:func:`cpuidle_register()`, +does not do that automatically. For this reason, the drivers that use +:c:func:`cpuidle_register_driver()` to register themselves must also take care +of registering the |struct cpuidle_device| objects as needed, so it is generally +recommended to use :c:func:`cpuidle_register()` for ``CPUIdle`` driver +registration in all cases. + +The registration of a |struct cpuidle_device| object causes the ``CPUIdle`` +``sysfs`` interface to be created and the governor's ``->enable()`` callback to +be invoked for the logical CPU represented by it, so it must take place after +registering the driver that will handle the CPU in question. + +``CPUIdle`` drivers and |struct cpuidle_device| objects can be unregistered +when they are not necessary any more which allows some resources associated with +them to be released. Due to dependencies between them, all of the +|struct cpuidle_device| objects representing CPUs handled by the given +``CPUIdle`` driver must be unregistered, with the help of +:c:func:`cpuidle_unregister_device()`, before calling +:c:func:`cpuidle_unregister_driver()` to unregister the driver. Alternatively, +:c:func:`cpuidle_unregister()` can be called to unregister a ``CPUIdle`` driver +along with all of the |struct cpuidle_device| objects representing CPUs handled +by it. + +``CPUIdle`` drivers can respond to runtime system configuration changes that +lead to modifications of the list of available processor idle states (which can +happen, for example, when the system's power source is switched from AC to +battery or the other way around). Upon a notification of such a change, +a ``CPUIdle`` driver is expected to call :c:func:`cpuidle_pause_and_lock()` to +turn ``CPUIdle`` off temporarily and then :c:func:`cpuidle_disable_device()` for +all of the |struct cpuidle_device| objects representing CPUs affected by that +change. Next, it can update its :c:member:`states` array in accordance with +the new configuration of the system, call :c:func:`cpuidle_enable_device()` for +all of the relevant |struct cpuidle_device| objects and invoke +:c:func:`cpuidle_resume_and_unlock()` to allow ``CPUIdle`` to be used again. diff --git a/Documentation/driver-api/pm/index.rst b/Documentation/driver-api/pm/index.rst index 2f6d0e9cf6b7..56975c6bc789 100644 --- a/Documentation/driver-api/pm/index.rst +++ b/Documentation/driver-api/pm/index.rst @@ -1,9 +1,10 @@ -======================= -Device Power Management -======================= +=============================== +CPU and Device Power Management +=============================== .. toctree:: + cpuidle devices notifiers types diff --git a/MAINTAINERS b/MAINTAINERS index 9f64f8d3740e..b053a355894e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -4016,6 +4016,7 @@ S: Maintained T: git git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm.git B: https://bugzilla.kernel.org F: Documentation/admin-guide/pm/cpuidle.rst +F: Documentation/driver-api/pm/cpuidle.rst F: drivers/cpuidle/* F: include/linux/cpuidle.h diff --git a/drivers/cpuidle/Kconfig b/drivers/cpuidle/Kconfig index 7e48eb5bf0a7..8caccbbd7353 100644 --- a/drivers/cpuidle/Kconfig +++ b/drivers/cpuidle/Kconfig @@ -4,7 +4,7 @@ config CPU_IDLE bool "CPU idle PM support" default y if ACPI || PPC_PSERIES select CPU_IDLE_GOV_LADDER if (!NO_HZ && !NO_HZ_IDLE) - select CPU_IDLE_GOV_MENU if (NO_HZ || NO_HZ_IDLE) + select CPU_IDLE_GOV_MENU if (NO_HZ || NO_HZ_IDLE) && !CPU_IDLE_GOV_TEO help CPU idle is a generic framework for supporting software-controlled idle processor power management. It includes modular cross-platform @@ -23,6 +23,15 @@ config CPU_IDLE_GOV_LADDER config CPU_IDLE_GOV_MENU bool "Menu governor (for tickless system)" +config CPU_IDLE_GOV_TEO + bool "Timer events oriented (TEO) governor (for tickless systems)" + help + This governor implements a simplified idle state selection method + focused on timer events and does not do any interactivity boosting. + + Some workloads benefit from using it and it generally should be safe + to use. Say Y here if you are not happy with the alternatives. + config DT_IDLE_STATES bool diff --git a/drivers/cpuidle/governors/Makefile b/drivers/cpuidle/governors/Makefile index 1b512722689f..4d8aff5248a8 100644 --- a/drivers/cpuidle/governors/Makefile +++ b/drivers/cpuidle/governors/Makefile @@ -4,3 +4,4 @@ obj-$(CONFIG_CPU_IDLE_GOV_LADDER) += ladder.o obj-$(CONFIG_CPU_IDLE_GOV_MENU) += menu.o +obj-$(CONFIG_CPU_IDLE_GOV_TEO) += teo.o diff --git a/drivers/cpuidle/governors/teo.c b/drivers/cpuidle/governors/teo.c new file mode 100644 index 000000000000..7d05efdbd3c6 --- /dev/null +++ b/drivers/cpuidle/governors/teo.c @@ -0,0 +1,444 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Timer events oriented CPU idle governor + * + * Copyright (C) 2018 Intel Corporation + * Author: Rafael J. Wysocki + * + * The idea of this governor is based on the observation that on many systems + * timer events are two or more orders of magnitude more frequent than any + * other interrupts, so they are likely to be the most significant source of CPU + * wakeups from idle states. Moreover, information about what happened in the + * (relatively recent) past can be used to estimate whether or not the deepest + * idle state with target residency within the time to the closest timer is + * likely to be suitable for the upcoming idle time of the CPU and, if not, then + * which of the shallower idle states to choose. + * + * Of course, non-timer wakeup sources are more important in some use cases and + * they can be covered by taking a few most recent idle time intervals of the + * CPU into account. However, even in that case it is not necessary to consider + * idle duration values greater than the time till the closest timer, as the + * patterns that they may belong to produce average values close enough to + * the time till the closest timer (sleep length) anyway. + * + * Thus this governor estimates whether or not the upcoming idle time of the CPU + * is likely to be significantly shorter than the sleep length and selects an + * idle state for it in accordance with that, as follows: + * + * - Find an idle state on the basis of the sleep length and state statistics + * collected over time: + * + * o Find the deepest idle state whose target residency is less than or equal + * to the sleep length. + * + * o Select it if it matched both the sleep length and the observed idle + * duration in the past more often than it matched the sleep length alone + * (i.e. the observed idle duration was significantly shorter than the sleep + * length matched by it). + * + * o Otherwise, select the shallower state with the greatest matched "early" + * wakeups metric. + * + * - If the majority of the most recent idle duration values are below the + * target residency of the idle state selected so far, use those values to + * compute the new expected idle duration and find an idle state matching it + * (which has to be shallower than the one selected so far). + */ + +#include +#include +#include +#include +#include + +/* + * The PULSE value is added to metrics when they grow and the DECAY_SHIFT value + * is used for decreasing metrics on a regular basis. + */ +#define PULSE 1024 +#define DECAY_SHIFT 3 + +/* + * Number of the most recent idle duration values to take into consideration for + * the detection of wakeup patterns. + */ +#define INTERVALS 8 + +/** + * struct teo_idle_state - Idle state data used by the TEO cpuidle governor. + * @early_hits: "Early" CPU wakeups "matching" this state. + * @hits: "On time" CPU wakeups "matching" this state. + * @misses: CPU wakeups "missing" this state. + * + * A CPU wakeup is "matched" by a given idle state if the idle duration measured + * after the wakeup is between the target residency of that state and the target + * residency of the next one (or if this is the deepest available idle state, it + * "matches" a CPU wakeup when the measured idle duration is at least equal to + * its target residency). + * + * Also, from the TEO governor perspective, a CPU wakeup from idle is "early" if + * it occurs significantly earlier than the closest expected timer event (that + * is, early enough to match an idle state shallower than the one matching the + * time till the closest timer event). Otherwise, the wakeup is "on time", or + * it is a "hit". + * + * A "miss" occurs when the given state doesn't match the wakeup, but it matches + * the time till the closest timer event used for idle state selection. + */ +struct teo_idle_state { + unsigned int early_hits; + unsigned int hits; + unsigned int misses; +}; + +/** + * struct teo_cpu - CPU data used by the TEO cpuidle governor. + * @time_span_ns: Time between idle state selection and post-wakeup update. + * @sleep_length_ns: Time till the closest timer event (at the selection time). + * @states: Idle states data corresponding to this CPU. + * @last_state: Idle state entered by the CPU last time. + * @interval_idx: Index of the most recent saved idle interval. + * @intervals: Saved idle duration values. + */ +struct teo_cpu { + u64 time_span_ns; + u64 sleep_length_ns; + struct teo_idle_state states[CPUIDLE_STATE_MAX]; + int last_state; + int interval_idx; + unsigned int intervals[INTERVALS]; +}; + +static DEFINE_PER_CPU(struct teo_cpu, teo_cpus); + +/** + * teo_update - Update CPU data after wakeup. + * @drv: cpuidle driver containing state data. + * @dev: Target CPU. + */ +static void teo_update(struct cpuidle_driver *drv, struct cpuidle_device *dev) +{ + struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu); + unsigned int sleep_length_us = ktime_to_us(cpu_data->sleep_length_ns); + int i, idx_hit = -1, idx_timer = -1; + unsigned int measured_us; + + if (cpu_data->time_span_ns >= cpu_data->sleep_length_ns) { + /* + * One of the safety nets has triggered or this was a timer + * wakeup (or equivalent). + */ + measured_us = sleep_length_us; + } else { + unsigned int lat = drv->states[cpu_data->last_state].exit_latency; + + measured_us = ktime_to_us(cpu_data->time_span_ns); + /* + * The delay between the wakeup and the first instruction + * executed by the CPU is not likely to be worst-case every + * time, so take 1/2 of the exit latency as a very rough + * approximation of the average of it. + */ + if (measured_us >= lat) + measured_us -= lat / 2; + else + measured_us /= 2; + } + + /* + * Decay the "early hits" metric for all of the states and find the + * states matching the sleep length and the measured idle duration. + */ + for (i = 0; i < drv->state_count; i++) { + unsigned int early_hits = cpu_data->states[i].early_hits; + + cpu_data->states[i].early_hits -= early_hits >> DECAY_SHIFT; + + if (drv->states[i].target_residency <= sleep_length_us) { + idx_timer = i; + if (drv->states[i].target_residency <= measured_us) + idx_hit = i; + } + } + + /* + * Update the "hits" and "misses" data for the state matching the sleep + * length. If it matches the measured idle duration too, this is a hit, + * so increase the "hits" metric for it then. Otherwise, this is a + * miss, so increase the "misses" metric for it. In the latter case + * also increase the "early hits" metric for the state that actually + * matches the measured idle duration. + */ + if (idx_timer >= 0) { + unsigned int hits = cpu_data->states[idx_timer].hits; + unsigned int misses = cpu_data->states[idx_timer].misses; + + hits -= hits >> DECAY_SHIFT; + misses -= misses >> DECAY_SHIFT; + + if (idx_timer > idx_hit) { + misses += PULSE; + if (idx_hit >= 0) + cpu_data->states[idx_hit].early_hits += PULSE; + } else { + hits += PULSE; + } + + cpu_data->states[idx_timer].misses = misses; + cpu_data->states[idx_timer].hits = hits; + } + + /* + * If the total time span between idle state selection and the "reflect" + * callback is greater than or equal to the sleep length determined at + * the idle state selection time, the wakeup is likely to be due to a + * timer event. + */ + if (cpu_data->time_span_ns >= cpu_data->sleep_length_ns) + measured_us = UINT_MAX; + + /* + * Save idle duration values corresponding to non-timer wakeups for + * pattern detection. + */ + cpu_data->intervals[cpu_data->interval_idx++] = measured_us; + if (cpu_data->interval_idx > INTERVALS) + cpu_data->interval_idx = 0; +} + +/** + * teo_find_shallower_state - Find shallower idle state matching given duration. + * @drv: cpuidle driver containing state data. + * @dev: Target CPU. + * @state_idx: Index of the capping idle state. + * @duration_us: Idle duration value to match. + */ +static int teo_find_shallower_state(struct cpuidle_driver *drv, + struct cpuidle_device *dev, int state_idx, + unsigned int duration_us) +{ + int i; + + for (i = state_idx - 1; i >= 0; i--) { + if (drv->states[i].disabled || dev->states_usage[i].disable) + continue; + + state_idx = i; + if (drv->states[i].target_residency <= duration_us) + break; + } + return state_idx; +} + +/** + * teo_select - Selects the next idle state to enter. + * @drv: cpuidle driver containing state data. + * @dev: Target CPU. + * @stop_tick: Indication on whether or not to stop the scheduler tick. + */ +static int teo_select(struct cpuidle_driver *drv, struct cpuidle_device *dev, + bool *stop_tick) +{ + struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu); + int latency_req = cpuidle_governor_latency_req(dev->cpu); + unsigned int duration_us, count; + int max_early_idx, idx, i; + ktime_t delta_tick; + + if (cpu_data->last_state >= 0) { + teo_update(drv, dev); + cpu_data->last_state = -1; + } + + cpu_data->time_span_ns = local_clock(); + + cpu_data->sleep_length_ns = tick_nohz_get_sleep_length(&delta_tick); + duration_us = ktime_to_us(cpu_data->sleep_length_ns); + + count = 0; + max_early_idx = -1; + idx = -1; + + for (i = 0; i < drv->state_count; i++) { + struct cpuidle_state *s = &drv->states[i]; + struct cpuidle_state_usage *su = &dev->states_usage[i]; + + if (s->disabled || su->disable) { + /* + * If the "early hits" metric of a disabled state is + * greater than the current maximum, it should be taken + * into account, because it would be a mistake to select + * a deeper state with lower "early hits" metric. The + * index cannot be changed to point to it, however, so + * just increase the max count alone and let the index + * still point to a shallower idle state. + */ + if (max_early_idx >= 0 && + count < cpu_data->states[i].early_hits) + count = cpu_data->states[i].early_hits; + + continue; + } + + if (idx < 0) + idx = i; /* first enabled state */ + + if (s->target_residency > duration_us) + break; + + if (s->exit_latency > latency_req) { + /* + * If we break out of the loop for latency reasons, use + * the target residency of the selected state as the + * expected idle duration to avoid stopping the tick + * as long as that target residency is low enough. + */ + duration_us = drv->states[idx].target_residency; + goto refine; + } + + idx = i; + + if (count < cpu_data->states[i].early_hits && + !(tick_nohz_tick_stopped() && + drv->states[i].target_residency < TICK_USEC)) { + count = cpu_data->states[i].early_hits; + max_early_idx = i; + } + } + + /* + * If the "hits" metric of the idle state matching the sleep length is + * greater than its "misses" metric, that is the one to use. Otherwise, + * it is more likely that one of the shallower states will match the + * idle duration observed after wakeup, so take the one with the maximum + * "early hits" metric, but if that cannot be determined, just use the + * state selected so far. + */ + if (cpu_data->states[idx].hits <= cpu_data->states[idx].misses && + max_early_idx >= 0) { + idx = max_early_idx; + duration_us = drv->states[idx].target_residency; + } + +refine: + if (idx < 0) { + idx = 0; /* No states enabled. Must use 0. */ + } else if (idx > 0) { + u64 sum = 0; + + count = 0; + + /* + * Count and sum the most recent idle duration values less than + * the target residency of the state selected so far, find the + * max. + */ + for (i = 0; i < INTERVALS; i++) { + unsigned int val = cpu_data->intervals[i]; + + if (val >= drv->states[idx].target_residency) + continue; + + count++; + sum += val; + } + + /* + * Give up unless the majority of the most recent idle duration + * values are in the interesting range. + */ + if (count > INTERVALS / 2) { + unsigned int avg_us = div64_u64(sum, count); + + /* + * Avoid spending too much time in an idle state that + * would be too shallow. + */ + if (!(tick_nohz_tick_stopped() && avg_us < TICK_USEC)) { + idx = teo_find_shallower_state(drv, dev, idx, avg_us); + duration_us = avg_us; + } + } + } + + /* + * Don't stop the tick if the selected state is a polling one or if the + * expected idle duration is shorter than the tick period length. + */ + if (((drv->states[idx].flags & CPUIDLE_FLAG_POLLING) || + duration_us < TICK_USEC) && !tick_nohz_tick_stopped()) { + unsigned int delta_tick_us = ktime_to_us(delta_tick); + + *stop_tick = false; + + /* + * The tick is not going to be stopped, so if the target + * residency of the state to be returned is not within the time + * till the closest timer including the tick, try to correct + * that. + */ + if (idx > 0 && drv->states[idx].target_residency > delta_tick_us) + idx = teo_find_shallower_state(drv, dev, idx, delta_tick_us); + } + + return idx; +} + +/** + * teo_reflect - Note that governor data for the CPU need to be updated. + * @dev: Target CPU. + * @state: Entered state. + */ +static void teo_reflect(struct cpuidle_device *dev, int state) +{ + struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu); + + cpu_data->last_state = state; + /* + * If the wakeup was not "natural", but triggered by one of the safety + * nets, assume that the CPU might have been idle for the entire sleep + * length time. + */ + if (dev->poll_time_limit || + (tick_nohz_idle_got_tick() && cpu_data->sleep_length_ns > TICK_NSEC)) { + dev->poll_time_limit = false; + cpu_data->time_span_ns = cpu_data->sleep_length_ns; + } else { + cpu_data->time_span_ns = local_clock() - cpu_data->time_span_ns; + } +} + +/** + * teo_enable_device - Initialize the governor's data for the target CPU. + * @drv: cpuidle driver (not used). + * @dev: Target CPU. + */ +static int teo_enable_device(struct cpuidle_driver *drv, + struct cpuidle_device *dev) +{ + struct teo_cpu *cpu_data = per_cpu_ptr(&teo_cpus, dev->cpu); + int i; + + memset(cpu_data, 0, sizeof(*cpu_data)); + + for (i = 0; i < INTERVALS; i++) + cpu_data->intervals[i] = UINT_MAX; + + return 0; +} + +static struct cpuidle_governor teo_governor = { + .name = "teo", + .rating = 19, + .enable = teo_enable_device, + .select = teo_select, + .reflect = teo_reflect, +}; + +static int __init teo_governor_init(void) +{ + return cpuidle_register_governor(&teo_governor); +} + +postcore_initcall(teo_governor_init); diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h index 4dff74f48d4b..3b39472324a3 100644 --- a/include/linux/cpuidle.h +++ b/include/linux/cpuidle.h @@ -69,11 +69,9 @@ struct cpuidle_state { /* Idle State Flags */ #define CPUIDLE_FLAG_NONE (0x00) -#define CPUIDLE_FLAG_POLLING (0x01) /* polling state */ -#define CPUIDLE_FLAG_COUPLED (0x02) /* state applies to multiple cpus */ -#define CPUIDLE_FLAG_TIMER_STOP (0x04) /* timer is stopped on this state */ - -#define CPUIDLE_DRIVER_FLAGS_MASK (0xFFFF0000) +#define CPUIDLE_FLAG_POLLING BIT(0) /* polling state */ +#define CPUIDLE_FLAG_COUPLED BIT(1) /* state applies to multiple cpus */ +#define CPUIDLE_FLAG_TIMER_STOP BIT(2) /* timer is stopped on this state */ struct cpuidle_device_kobj; struct cpuidle_state_kobj;