From dddcab78bdda9d63b5cf6ddcdc30f815d98e8d98 Mon Sep 17 00:00:00 2001 From: Syed Rameez Mustafa Date: Wed, 7 Sep 2016 16:18:27 -0700 Subject: [PATCH] sched: Introduce HMP scheduling Add a snapshot of all changes associated with the HMP scheduler ported over to the latest kernel version. This will allow proper big Little scheduling for targets that will use this kernel version. The msm-4-4 snapshot is taken at commitcc60f0790f27 (sched: constrain HMP scheduler tunable range with in better way). Please note that this is not an exact snapshot; certain HMP related code has been removed. This includes redundant code, useless CONFIG options, rarely taken error code paths, runtime enable/disable support for the HMP scheduler. Change-Id: I923f35904008f1fb74ac9c1a39481810cdea5790 Signed-off-by: Syed Rameez Mustafa --- Documentation/scheduler/sched-hmp.txt | 1432 +++++++++ include/linux/cpu.h | 15 + include/linux/sched.h | 185 +- include/linux/sched/sysctl.h | 38 + include/linux/tick.h | 2 + include/trace/events/sched.h | 664 ++++- init/Kconfig | 17 + kernel/sched/Makefile | 3 +- kernel/sched/core.c | 277 +- kernel/sched/cputime.c | 12 +- kernel/sched/deadline.c | 44 + kernel/sched/debug.c | 43 +- kernel/sched/fair.c | 1515 +++++++++- kernel/sched/features.h | 2 +- kernel/sched/hmp.c | 3975 +++++++++++++++++++++++++ kernel/sched/idle_task.c | 11 + kernel/sched/rt.c | 181 +- kernel/sched/sched.h | 690 +++++ kernel/sched/sched_avg.c | 128 + kernel/sched/stop_task.c | 40 + kernel/sysctl.c | 184 ++ kernel/time/tick-sched.c | 15 + lib/Kconfig.debug | 19 + 23 files changed, 9418 insertions(+), 74 deletions(-) create mode 100644 Documentation/scheduler/sched-hmp.txt create mode 100644 kernel/sched/hmp.c create mode 100644 kernel/sched/sched_avg.c diff --git a/Documentation/scheduler/sched-hmp.txt b/Documentation/scheduler/sched-hmp.txt new file mode 100644 index 000000000000..22449aec5558 --- /dev/null +++ b/Documentation/scheduler/sched-hmp.txt @@ -0,0 +1,1432 @@ +CONTENTS + +1. Introduction + 1.1 Heterogeneous Systems + 1.2 CPU Frequency Guidance +2. Window-Based Load Tracking Scheme + 2.1 Synchronized Windows + 2.2 struct ravg + 2.3 Scaling Load Statistics + 2.4 sched_window_stats_policy + 2.5 Task Events + 2.6 update_task_ravg() + 2.7 update_history() + 2.8 Per-task 'initial task load' +3. CPU Capacity + 3.1 Load scale factor + 3.2 CPU Power +4. CPU Power +5. HMP Scheduler + 5.1 Classification of Tasks and CPUs + 5.2 select_best_cpu() + 5.2.1 sched_boost + 5.2.2 task_will_fit() + 5.2.3 Tunables affecting select_best_cpu() + 5.2.4 Wakeup Logic + 5.3 Scheduler Tick + 5.4 Load Balancer + 5.5 Real Time Tasks + 5.6 Task packing +6. Frequency Guidance + 6.1 Per-CPU Window-Based Stats + 6.2 Per-task Window-Based Stats + 6.3 Effect of various task events +7. Tunables +8. HMP Scheduler Trace Points + 8.1 sched_enq_deq_task + 8.2 sched_task_load + 8.3 sched_cpu_load_* + 8.4 sched_update_task_ravg + 8.5 sched_update_history + 8.6 sched_reset_all_windows_stats + 8.7 sched_migration_update_sum + 8.8 sched_get_busy + 8.9 sched_freq_alert + 8.10 sched_set_boost + +=============== +1. INTRODUCTION +=============== + +Scheduler extensions described in this document serves two goals: + +1) handle heterogeneous multi-processor (HMP) systems +2) guide cpufreq governor on proactive changes to cpu frequency + +*** 1.1 Heterogeneous systems + +Heterogeneous systems have cpus that differ with regard to their performance and +power characteristics. Some cpus could offer peak performance better than +others, although at cost of consuming more power. We shall refer such cpus as +"high performance" or "performance efficient" cpus. Other cpus that offer lesser +peak performance are referred to as "power efficient". + +In this situation the scheduler is tasked with the responsibility of assigning +tasks to run on the right cpus where their performance requirements can be met +at the least expense of power. + +Achieving that goal is made complicated by the fact that the scheduler has +little clue about performance requirements of tasks and how they may change by +running on power or performance efficient cpus! One simplifying assumption here +could be that a task's desire for more performance is expressed by its cpu +utilization. A task demanding high cpu utilization on a power-efficient cpu +would likely improve in its performance by running on a performance-efficient +cpu. This idea forms the basis for HMP-related scheduler extensions. + +Key inputs required by the HMP scheduler for its task placement decisions are: + +a) task load - this reflects cpu utilization or demand of tasks +b) CPU capacity - this reflects peak performance offered by cpus +c) CPU power - this reflects power or energy cost of cpus + +Once all 3 pieces of information are available, the HMP scheduler can place +tasks on the lowest power cpus where their demand can be satisfied. + +*** 1.2 CPU Frequency guidance + +A somewhat separate but related goal of the scheduler extensions described here +is to provide guidance to the cpufreq governor on the need to change cpu +frequency. Most governors that control cpu frequency work on a reactive basis. +CPU utilization is sampled at regular intervals, based on which the need to +change frequency is determined. Higher utilization leads to a frequency increase +and vice-versa. There are several problems with this approach that scheduler +can help resolve. + +a) latency + + Reactive nature introduces latency for cpus to ramp up to desired speed + which can hurt application performance. This is inevitable as cpufreq + governors can only track cpu utilization as a whole and not tasks which + are driving that demand. Scheduler can however keep track of individual + task demand and can alert the governor on changing task activity. For + example, request raise in frequency when tasks activity is increasing on + a cpu because of wakeup or migration or request frequency to be lowered + when task activity is decreasing because of sleep/exit or migration. + +b) part-picture + + Most governors track utilization of each CPU independently. When a task + migrates from one cpu to another the task's execution time is split + across the two cpus. The governor can fail to see the full picture of + task demand in this case and thus the need for increasing frequency, + affecting the task's performance. Scheduler can keep track of task + migrations, fix up busy time upon migration and report per-cpu busy time + to the governor that reflects task demand accurately. + +The rest of this document explains key enhancements made to the scheduler to +accomplish both of the aforementioned goals. + +==================================== +2. WINDOW-BASED LOAD TRACKING SCHEME +==================================== + +As mentioned in the introduction section, knowledge of the CPU demand exerted by +a task is a prerequisite to knowing where to best place the task in an HMP +system. The per-entity load tracking (PELT) scheme, present in Linux kernel +since v3.7, has some perceived shortcomings when used to place tasks on HMP +systems or provide recommendations on CPU frequency. + +Per-entity load tracking does not make a distinction between the ramp up +vs ramp down time of task load. It also decays task load without exception when +a task sleeps. As an example, a cpu bound task at its peak load (LOAD_AVG_MAX or +47742) can see its load decay to 0 after a sleep of just 213ms! A cpu-bound task +running on a performance-efficient cpu could thus get re-classified as not +requiring such a cpu after a short sleep. In the case of mobile workloads, tasks +could go to sleep due to a lack of user input. When they wakeup it is very +likely their cpu utilization pattern repeats. Resetting their load across sleep +and incurring latency to reclassify them as requiring a high performance cpu can +hurt application performance. + +The window-based load tracking scheme described in this document avoids these +drawbacks. It keeps track of N windows of execution for every task. Windows +where a task had no activity are ignored and not recorded. N can be tuned at +compile time (RAVG_HIST_SIZE defined in include/linux/sched.h) or at runtime +(/proc/sys/kernel/sched_ravg_hist_size). The window size, W, is common for all +tasks and currently defaults to 10ms ('sched_ravg_window' defined in +kernel/sched/core.c). The window size can be tuned at boot time via the +sched_ravg_window=W argument to kernel. Alternately it can be tuned after boot +via tunables provided by the interactive governor. More on this later. + +Based on the N samples available per-task, a per-task "demand" attribute is +calculated which represents the cpu demand of that task. The demand attribute is +used to classify tasks as to whether or not they need a performance-efficient +CPU and also serves to provide inputs on frequency to the cpufreq governor. More +on this later. The 'sched_window_stats_policy' tunable (defined in +kernel/sched/core.c) controls how the demand field for a task is derived from +its N past samples. + +*** 2.1 Synchronized windows + +Windows of observation for task activity are synchronized across cpus. This +greatly aids in the scheduler's frequency guidance feature. Scheduler currently +relies on a synchronized clock (sched_clock()) for this feature to work. It may +be possible to extend this feature to work on systems having an unsynchronized +sched_clock(). + +struct rq { + + .. + + u64 window_start; + + .. +}; + +The 'window_start' attribute represents the time when current window began on a +cpu. It is updated when key task events such as wakeup or context-switch call +update_task_ravg() to record task activity. The window_start value is expected +to be the same for all cpus, although it could be behind on some cpus where it +has not yet been updated because update_task_ravg() has not been recently +called. For example, when a cpu is idle for a long time its window_start could +be stale. The window_start value for such cpus is rolled forward upon +occurrence of a task event resulting in a call to update_task_ravg(). + +*** 2.2 struct ravg + +The ravg struct contains information tracked per-task. + +struct ravg { + u64 mark_start; + u32 sum, demand; + u32 sum_history[RAVG_HIST_SIZE]; +}; + +struct task_struct { + + .. + + struct ravg ravg; + + .. +}; + +sum_history[] - stores cpu utilization samples from N previous windows + where task had activity + +sum - stores cpu utilization of the task in its most recently + tracked window. Once the corresponding window terminates, + 'sum' will be pushed into the sum_history[] array and is then + reset to 0. It is possible that the window corresponding to + sum is not the current window being tracked on a cpu. For + example, a task could go to sleep in window X and wakeup in + window Y (Y > X). In this case, sum would correspond to the + task's activity seen in window X. When update_task_ravg() is + called during the task's wakeup event it will be seen that + window X has elapsed. The sum value will be pushed to + 'sum_history[]' array before being reset to 0. + +demand - represents task's cpu demand and is derived from the + elements in sum_history[]. The section on + 'sched_window_stats_policy' provides more details on how + 'demand' is derived from elements in sum_history[] array + +mark_start - records timestamp of the beginning of the most recent task + event. See section on 'Task events' for possible events that + update 'mark_start' + +curr_window - this is described in the section on 'Frequency guidance' + +prev_window - this is described in the section on 'Frequency guidance' + + +*** 2.3 Scaling load statistics + +Time required for a task to complete its work (and hence its load) depends on, +among various other factors, cpu frequency and its efficiency. In a HMP system, +some cpus are more performance efficient than others. Performance efficiency of +a cpu can be described by its "instructions-per-cycle" (IPC) attribute. History +of task execution could involve task having run at different frequencies and on +cpus with different IPC attributes. To avoid ambiguity of how task load relates +to the frequency and IPC of cpus on which a task has run, task load is captured +in a scaled form, with scaling being done in reference to an "ideal" cpu that +has best possible IPC and frequency. Such an "ideal" cpu, having the best +possible frequency and IPC, may or may not exist in system. + +As an example, consider a HMP system, with two types of cpus, A53 and A57. A53 +has IPC count of 1024 and can run at maximum frequency of 1 GHz, while A57 has +IPC count of 2048 and can run at maximum frequency of 2 GHz. Ideal cpu in this +case is A57 running at 2 GHz. + +A unit of work that takes 100ms to finish on A53 running at 100MHz would get +done in 10ms on A53 running at 1GHz, in 5 ms running on A57 at 1 GHz and 2.5ms +on A57 running at 2 GHz. Thus a load of 100ms can be expressed as 2.5ms in +reference to ideal cpu of A57 running at 2 GHz. + +In order to understand how much load a task will consume on a given cpu, its +scaled load needs to be multiplied by a factor (load scale factor). In above +example, scaled load of 2.5ms needs to be multiplied by a factor of 4 in order +to estimate the load of task on A53 running at 1 GHz. + +/proc/sched_debug provides IPC attribute and load scale factor for every cpu. + +In summary, task load information stored in a task's sum_history[] array is +scaled for both frequency and efficiency. If a task runs for X ms, then the +value stored in its 'sum' field is derived as: + + X_s = X * (f_cur / max_possible_freq) * + (efficiency / max_possible_efficiency) + +where: + +X = cpu utilization that needs to be accounted +X_s = Scaled derivative of X +f_cur = current frequency of the cpu where the task was + running +max_possible_freq = maximum possible frequency (across all cpus) +efficiency = instructions per cycle (IPC) of cpu where task was + running +max_possible_efficiency = maximum IPC offered by any cpu in system + + +*** 2.4 sched_window_stats_policy + +sched_window_stats_policy controls how the 'demand' attribute for a task is +derived from elements in its 'sum_history[]' array. + +WINDOW_STATS_RECENT (0) + demand = recent + +WINDOW_STATS_MAX (1) + demand = max + +WINDOW_STATS_MAX_RECENT_AVG (2) + demand = maximum(average, recent) + +WINDOW_STATS_AVG (3) + demand = average + +where: + M = history size specified by + /proc/sys/kernel/sched_ravg_hist_size + average = average of first M samples found in the sum_history[] array + max = maximum value of first M samples found in the sum_history[] + array + recent = most recent sample (sum_history[0]) + demand = demand attribute found in 'struct ravg' + +This policy can be changed at runtime via +/proc/sys/kernel/sched_window_stats_policy. For example, the command +below would select WINDOW_STATS_USE_MAX policy + +echo 1 > /proc/sys/kernel/sched_window_stats_policy + +*** 2.5 Task events + +A number of events results in the window-based stats of a task being +updated. These are: + +PICK_NEXT_TASK - the task is about to start running on a cpu +PUT_PREV_TASK - the task stopped running on a cpu +TASK_WAKE - the task is waking from sleep +TASK_MIGRATE - the task is migrating from one cpu to another +TASK_UPDATE - this event is invoked on a currently running task to + update the task's window-stats and also the cpu's + window-stats such as 'window_start' +IRQ_UPDATE - event to record the busy time spent by an idle cpu + processing interrupts + +*** 2.6 update_task_ravg() + +update_task_ravg() is called to mark the beginning of an event for a task or a +cpu. It serves to accomplish these functions: + +a. Update a cpu's window_start value +b. Update a task's window-stats (sum, sum_history[], demand and mark_start) + +In addition update_task_ravg() updates the busy time information for the given +cpu, which is used for frequency guidance. This is described further in section +6. + +*** 2.7 update_history() + +update_history() is called on a task to record its activity in an elapsed +window. 'sum', which represents task's cpu demand in its elapsed window is +pushed onto sum_history[] array and its 'demand' attribute is updated based on +the sched_window_stats_policy in effect. + +*** 2.8 Initial task load attribute for a task (init_load_pct) + +In some cases, it may be desirable for children of a task to be assigned a +"high" load so that they can start running on best capacity cluster. By default, +newly created tasks are assigned a load defined by tunable sched_init_task_load +(Sec 7.8). Some specialized tasks may need a higher value than the global +default for their child tasks. This will let child tasks run on cpus with best +capacity. This is accomplished by setting the 'initial task load' attribute +(init_load_pct) for a task. Child tasks starting load (ravg.demand and +ravg.sum_history[]) is initialized from their parent's 'initial task load' +attribute. Note that child task's 'initial task load' attribute itself will be 0 +by default (i.e it is not inherited from parent). + +A task's 'initial task load' attribute can be set in two ways: + +**** /proc interface + +/proc/[pid]/sched_init_task_load can be written to for setting a task's 'initial +task load' attribute. A numeric value between 0 - 100 (in percent scale) is +accepted for task's 'initial task load' attribute. + +Reading /proc/[pid]/sched_init_task_load returns the 'initial task load' +attribute for the given task. + +**** kernel API + +Following kernel APIs are provided to set or retrieve a given task's 'initial +task load' attribute: + +int sched_set_init_task_load(struct task_struct *p, int init_load_pct); +int sched_get_init_task_load(struct task_struct *p); + + +=============== +3. CPU CAPACITY +=============== + +CPU capacity reflects peak performance offered by a cpu. It is defined both by +maximum frequency at which cpu can run and its efficiency attribute. Capacity of +a cpu is defined in reference to "least" performing cpu such that "least" +performing cpu has capacity of 1024. + + capacity = 1024 * (fmax_cur * / min_max_freq) * + (efficiency / min_possible_efficiency) + +where: + + fmax_cur = maximum frequency at which cpu is currently + allowed to run at + efficiency = IPC of cpu + min_max_freq = max frequency at which "least" performing cpu + can run + min_possible_efficiency = IPC of "least" performing cpu + +'fmax_cur' reflects the fact that a cpu may be constrained at runtime to run at +a maximum frequency less than what is supported. This may be a constraint placed +by user or drivers such as thermal that intends to reduce temperature of a cpu +by restricting its maximum frequency. + +'max_possible_capacity' reflects the maximum capacity of a cpu based on the +maximum frequency it supports. + +max_possible_capacity = 1024 * (fmax * / min_max_freq) * + (efficiency / min_possible_efficiency) + +where: + fmax = maximum frequency supported by a cpu + +/proc/sched_debug lists capacity and maximum_capacity information for a cpu. + +In the example HMP system quoted in Sec 2.3, "least" performing CPU is A53 and +thus min_max_freq = 1GHz and min_possible_efficiency = 1024. + +Capacity of A57 = 1024 * (2GHz / 1GHz) * (2048 / 1024) = 4096 +Capacity of A53 = 1024 * (1GHz / 1GHz) * (1024 / 1024) = 1024 + +Capacity of A57 when constrained to run at maximum frequency of 500MHz can be +calculated as: + +Capacity of A57 = 1024 * (500MHz / 1GHz) * (2048 / 1024) = 1024 + +*** 3.1 load_scale_factor + +'lsf' or load scale factor attribute of a cpu is used to estimate load of a task +on that cpu when running at its fmax_cur frequency. 'lsf' is defined in +reference to "best" performing cpu such that it's lsf is 1024. 'lsf' for a cpu +is defined as: + + lsf = 1024 * (max_possible_freq / fmax_cur) * + (max_possible_efficiency / ipc) + +where: + fmax_cur = maximum frequency at which cpu is currently + allowed to run at + ipc = IPC of cpu + max_possible_freq = max frequency at which "best" performing cpu + can run + max_possible_efficiency = IPC of "best" performing cpu + +In the example HMP system quoted in Sec 2.3, "best" performing CPU is A57 and +thus max_possible_freq = 2 GHz, max_possible_efficiency = 2048 + +lsf of A57 = 1024 * (2GHz / 2GHz) * (2048 / 2048) = 1024 +lsf of A53 = 1024 * (2GHz / 1 GHz) * (2048 / 1024) = 4096 + +lsf of A57 constrained to run at maximum frequency of 500MHz can be calculated +as: + +lsf of A57 = 1024 * (2GHz / 500Mhz) * (2048 / 2048) = 4096 + +To estimate load of a task on a given cpu running at its fmax_cur: + + load = scaled_load * lsf / 1024 + +A task with scaled load of 20% would thus be estimated to consume 80% bandwidth +of A53 running at 1GHz. The same task with scaled load of 20% would be estimated +to consume 160% bandwidth on A53 constrained to run at maximum frequency of +500MHz. + +load_scale_factor, thus, is very useful to estimate load of a task on a given +cpu and thus to decide whether it can fit in a cpu or not. + +*** 3.2 cpu_power + +A metric 'cpu_power' related to 'capacity' is also listed in /proc/sched_debug. +'cpu_power' is ideally same for all cpus (1024) when they are idle and running +at the same frequency. 'cpu_power' of a cpu can be scaled down from its ideal +value to reflect reduced frequency it is operating at and also to reflect the +amount of cpu bandwidth consumed by real-time tasks executing on it. +'cpu_power' metric is used by scheduler to decide task load distribution among +cpus. CPUs with low 'cpu_power' will be assigned less task load compared to cpus +with higher 'cpu_power' + +============ +4. CPU POWER +============ + +The HMP scheduler extensions currently depend on an architecture-specific driver +to provide runtime information on cpu power. In the absence of an +architecture-specific driver, the scheduler will resort to using the +max_possible_capacity metric of a cpu as a measure of its power. + +================ +5. HMP SCHEDULER +================ + +For normal (SCHED_OTHER/fair class) tasks there are three paths in the +scheduler which these HMP extensions affect. The task wakeup path, the +load balancer, and the scheduler tick are each modified. + +Real-time and stop-class tasks are served by different code +paths. These will be discussed separately. + +Prior to delving further into the algorithm and implementation however +some definitions are required. + +*** 5.1 Classification of Tasks and CPUs + +With the extensions described thus far, the following information is +available to the HMP scheduler: + +- per-task CPU demand information from either Per-Entity Load Tracking + (PELT) or the window-based algorithm described above + +- a power value for each frequency supported by each CPU via the API + described in section 4 + +- current CPU frequency, maximum CPU frequency (may be throttled by at + runtime due to thermal conditions), maximum possible CPU frequency supported + by hardware + +- data previously maintained within the scheduler such as the number + of currently runnable tasks on each CPU + +Combined with tunable parameters, this information can be used to classify +both tasks and CPUs to aid in the placement of tasks. + +- big task + + A big task is one that exerts a CPU demand too high for a particular + CPU to satisfy. The scheduler will attempt to find a CPU with more + capacity for such a task. + + The definition of "big" is specific to a task *and* a CPU. A task + may be considered big on one CPU in the system and not big on + another if the first CPU has less capacity than the second. + + What task demand is "too high" for a particular CPU? One obvious + answer would be a task demand which, as measured by PELT or + window-based load tracking, matches or exceeds the capacity of that + CPU. A task which runs on a CPU for a long time, for example, might + meet this criteria as it would report 100% demand of that CPU. It + may be desirable however to classify tasks which use less than 100% + of a particular CPU as big so that the task has some "headroom" to grow + without its CPU bandwidth getting capped and its performance requirements + not being met. This task demand is therefore a tunable parameter: + + /proc/sys/kernel/sched_upmigrate + + This value is a percentage. If a task consumes more than this much of a + particular CPU, that CPU will be considered too small for the task. The task + will thus be seen as a "big" task on the cpu and will reflect in nr_big_tasks + statistics maintained for that cpu. Note that certain tasks (whose nice + value exceeds SCHED_UPMIGRATE_MIN_NICE value or those that belong to a cgroup + whose upmigrate_discourage flag is set) will never be classified as big tasks + despite their high demand. + + As the load scale factor is calculated against current fmax, it gets boosted + when a lower capacity CPU is restricted to run at lower fmax. The task + demand is inflated in this scenario and the task upmigrates early to the + maximum capacity CPU. Hence this threshold is auto-adjusted by a factor + equal to max_possible_frequency/current_frequency of a lower capacity CPU. + This adjustment happens only when the lower capacity CPU frequency is + restricted. The same adjustment is applied to the downmigrate threshold + as well. + + When the frequency restriction is relaxed, the previous values are restored. + sched_up_down_migrate_auto_update macro defined in kernel/sched/core.c + controls this auto-adjustment behavior and it is enabled by default. + + If the adjusted upmigrate threshold exceeds the window size, it is clipped to + the window size. If the adjusted downmigrate threshold decreases the difference + between the upmigrate and downmigrate, it is clipped to a value such that the + difference between the modified and the original thresholds is same. + +- spill threshold + + Tasks will normally be placed on lowest power-cost cluster where they can fit. + This could result in power-efficient cluster becoming overcrowded when there + are "too" many low-demand tasks. Spill threshold provides a spill over + criteria, wherein low-demand task are allowed to be placed on idle or + busy cpus in high-performance cluster. + + Scheduler will avoid placing a task on a cpu if it can result in cpu exceeding + its spill threshold, which is defined by two tunables: + + /proc/sys/kernel/sched_spill_nr_run (default: 10) + /proc/sys/kernel/sched_spill_load (default : 100%) + + A cpu is considered to be above its spill level if it already has 10 tasks or + if the sum of task load (scaled in reference to given cpu) and + rq->cumulative_runnable_avg exceeds 'sched_spill_load'. + +- power band + + The scheduler may be faced with a tradeoff between power and performance when + placing a task. If the scheduler sees two CPUs which can accommodate a task: + + CPU 1, power cost of 20, load of 10 + CPU 2, power cost of 10, load of 15 + + It is not clear what the right choice of CPU is. The HMP scheduler + offers the sched_powerband_limit tunable to determine how this + situation should be handled. When the power delta between two CPUs + is less than sched_powerband_limit_pct, load will be prioritized as + the deciding factor as to which CPU is selected. If the power delta + between two CPUs exceeds that, the lower power CPU is considered to + be in a different "band" and it is selected, despite perhaps having + a higher current task load. + +*** 5.2 select_best_cpu() + +CPU placement decisions for a task at its wakeup or creation time are the +most important decisions made by the HMP scheduler. This section will describe +the call flow and algorithm used in detail. + +The primary entry point for a task wakeup operation is try_to_wake_up(), +located in kernel/sched/core.c. This function relies on select_task_rq() to +determine the target CPU for the waking task. For fair-class (SCHED_OTHER) +tasks, that request will be routed to select_task_rq_fair() in +kernel/sched/fair.c. As part of these scheduler extensions a hook has been +inserted into the top of that function. If HMP scheduling is enabled the normal +scheduling behavior will be replaced by a call to select_best_cpu(). This +function, select_best_cpu(), represents the heart of the HMP scheduling +algorithm described in this document. Note that select_best_cpu() is also +invoked for a task being created. + +The behavior of select_best_cpu() depends on several factors such as boost +setting, choice of several tunables and on task demand. + +**** 5.2.1 Boost + +The task placement policy changes signifincantly when scheduler boost is in +effect. When boost is in effect the scheduler ignores the power cost of +placing tasks on CPUs. Instead it figures out the load on each CPU and then +places task on the least loaded CPU. If the load of two or more CPUs is the +same (generally when CPUs are idle) the task prefers to go highest capacity +CPU in the system. + +A further enhancement during boost is the scheduler' early detection feature. +While boost is in effect the scheduler checks for the precence of tasks that +have been runnable for over some period of time within the tick. For such +tasks the scheduler informs the governor of imminent need for high frequency. +If there exists a task on the runqueue at the tick that has been runnable +for greater than SCHED_EARLY_DETECTION_DURATION amount of time, it notifies +the governor with a fabricated load of the full window at the highest +frequency. The fabricated load is maintained until the task is no longer +runnable or until the next tick. + +Boost can be set via either /proc/sys/kernel/sched_boost or by invoking +kernel API sched_set_boost(). + + int sched_set_boost(int enable); + +Once turned on, boost will remain in effect until it is explicitly turned off. +To allow for boost to be controlled by multiple external entities (application +or kernel module) at same time, boost setting is reference counted. This means +that two applications can turn on boost and the effect of boost is eliminated +only after both applications have turned off boost. boost_refcount variable +represents this reference count. + +**** 5.2.2 task_will_fit() + +The overall goal of select_best_cpu() is to place a task on the least power +cluster where it can "fit" i.e where its cpu usage shall be below the capacity +offered by cluster. Criteria for a task to be considered as fitting in a cluster +is: + + i) A low-priority task, whose nice value is greater than + SCHED_UPMIGRATE_MIN_NICE or whose cgroup has its + upmigrate_discourage flag set, is considered to be fitting in all clusters, + irrespective of their capacity and task's cpu demand. + + ii) All tasks are considered to fit in highest capacity cluster. + + iii) Task demand scaled in reference to the given cluster should be less than a + threshold. See section on load_scale_factor to know more about how task + demand is scaled in reference to a given cpu (cluster). The threshold used + is normally sched_upmigrate. Its possible for a task's demand to exceed + sched_upmigrate threshold in reference to a cluster when its upmigrated to + higher capacity cluster. To prevent it from coming back immediately to + lower capacity cluster, the task is not considered to "fit" on its earlier + cluster until its demand has dropped below sched_downmigrate in reference + to that earlier cluster. sched_downmigrate thus provides for some + hysteresis control. + + +**** 5.2.3 Factors affecting select_best_cpu() + +Behavior of select_best_cpu() is further controlled by several tunables and +synchronous nature of wakeup. + +a. /proc/sys/kernel/sched_cpu_high_irqload + A cpu whose irq load is greater than this threshold will not be + considered eligible for placement. This threshold value in expressed in + nanoseconds scale, with default threshold being 10000000 (10ms). See + notes on sched_cpu_high_irqload tunable to understand how irq load on a + cpu is measured. + +b. Synchronous nature of wakeup + Synchronous wakeup is a hint to scheduler that the task issuing wakeup + (i.e task currently running on cpu where wakeup is being processed by + scheduler) will "soon" relinquish CPU. A simple example is two tasks + communicating with each other using a pipe structure. When reader task + blocks waiting for data, its woken by writer task after it has written + data to pipe. Writer task usually blocks waiting for reader task to + consume data in pipe (which may not have any more room for writes). + + Synchronous wakeup is accounted for by adjusting load of a cpu to not + include load of currently running task. As a result, a cpu that has only + one runnable task and which is currently processing synchronous wakeup + will be considered idle. + +c. PF_WAKE_UP_IDLE + Any task with this flag set will be woken up to an idle cpu (if one is + available) independent of sched_prefer_idle flag setting, its demand and + synchronous nature of wakeup. Similarly idle cpu is preferred during + wakeup for any task that does not have this flag set but is being woken + by a task with PF_WAKE_UP_IDLE flag set. For simplicity, we will use the + term "PF_WAKE_UP_IDLE wakeup" to signify wakeups involving a task with + PF_WAKE_UP_IDLE flag set. + +d. /proc/sys/kernel/sched_select_prev_cpu_us + This threshold controls whether task placement goes through fast path or + not. If task's wakeup time since last sleep is short there are high + chances that it's better to place the task on its previous CPU. This + reduces task placement latency, cache miss and number of migrations. + Default value of sched_select_prev_cpu_us is 2000 (2ms). This can be + turned off by setting it to 0. + +**** 5.2.4 Wakeup Logic for Task "p" + +Wakeup task placement logic is as follows: + +1) Eliminate CPUs with high irq load based on sched_cpu_high_irqload tunable. + +2) Eliminate CPUs where either the task does not fit or CPUs where placement +will result in exceeding the spill threshold tunables. CPUs elimiated at this +stage will be considered as backup choices incase none of the CPUs get past +this stage. + +3) Find out and return the least power CPU that satisfies all conditions above. + +4) If two or more CPUs are projected to have the same power, break ties in the +following preference order: + a) The CPU is the task's previous CPU. + b) The CPU is in the same cluster as the task's previous CPU. + c) The CPU has the least load + +The placement logic described above does not apply when PF_WAKE_UP_IDLE is set +for either the waker task or the wakee task. Instead the scheduler chooses the +most power efficient idle CPU. + +5) If no CPU is found after step 2, resort to backup CPU selection logic +whereby the CPU with highest amount of spare capacity is selected. + +6) If none of the CPUs have any spare capacity, return the task's previous +CPU. + +*** 5.3 Scheduler Tick + +Every CPU is interrupted periodically to let kernel update various statistics +and possibly preempt the currently running task in favor of a waiting task. This +periodicity, determined by CONFIG_HZ value, is set at 10ms. There are various +optimizations by which a CPU however can skip taking these interrupts (ticks). +A cpu going idle for considerable time in one such case. + +HMP scheduler extensions brings in a change in processing of tick +(scheduler_tick()) that can result in task migration. In case the currently +running task on a cpu belongs to fair_sched class, a check is made if it needs +to be migrated. Possible reasons for migrating task could be: + +a) A big task is running on a power-efficient cpu and a high-performance cpu is +available (idle) to service it + +b) A task is starving on a CPU with high irq load. + +c) A task with upmigration discouraged is running on a performance cluster. +See notes on 'cpu.upmigrate_discourage'. + +In case the test for migration turns out positive (which is expected to be rare +event), a candidate cpu is identified for task migration. To avoid multiple task +migrations to the same candidate cpu(s), identification of candidate cpu is +serialized via global spinlock (migration_lock). + +*** 5.4 Load Balancer + +Load balance is a key functionality of scheduler that strives to distribute task +across available cpus in a "fair" manner. Most of the complexity associated with +this feature involves balancing fair_sched class tasks. Changes made to load +balance code serve these goals: + +1. Restrict flow of tasks from power-efficient cpus to high-performance cpu. + Provide a spill-over threshold, defined in terms of number of tasks + (sched_spill_nr_run) and cpu demand (sched_spill_load), beyond which tasks + can spill over from power-efficient cpu to high-performance cpus. + +2. Allow idle power-efficient cpus to pick up extra load from over-loaded + performance-efficient cpu + +3. Allow idle high-performance cpu to pick up big tasks from power-efficient cpu + +*** 5.5 Real Time Tasks + +Minimal changes introduced in treatment of real-time tasks by HMP scheduler +aims at preferring scheduling of real-time tasks on cpus with low load on +a power efficient cluster. + +Prior to HMP scheduler, the fast-path cpu selection for placing a real-time task +(at wakeup) is its previous cpu, provided the currently running task on its +previous cpu is not a real-time task or a real-time task with lower priority. +Failing this, cpu selection in slow-path involves building a list of candidate +cpus where the waking real-time task will be of highest priority and thus can be +run immediately. The first cpu from this candidate list is chosen for the waking +real-time task. Much of the premise for this simple approach is the assumption +that real-time tasks often execute for very short intervals and thus the focus +is to place them on a cpu where they can be run immediately. + +HMP scheduler brings in a change which avoids fast-path and always resorts to +slow-path. Further cpu with lowest load in a power efficient cluster from +candidate list of cpus is chosen as cpu for placing waking real-time task. + +- PF_WAKE_UP_IDLE + +Idle cpu is preferred for any waking task that has this flag set in its +'task_struct.flags' field. Further idle cpu is preferred for any task woken by +such tasks. PF_WAKE_UP_IDLE flag of a task is inherited by it's children. It can +be modified for a task in two ways: + + > kernel-space interface + set_wake_up_idle() needs to be called in the context of a task + to set or clear its PF_WAKE_UP_IDLE flag. + + > user-space interface + /proc/[pid]/sched_wake_up_idle file needs to be written to for + setting or clearing PF_WAKE_UP_IDLE flag for a given task + +===================== +6. FREQUENCY GUIDANCE +===================== + +As mentioned in the introduction section the scheduler is in a unique +position to assist with the determination of CPU frequency. Because +the scheduler now maintains an estimate of per-task CPU demand, task +activity can be tracked, aggregated and provided to the CPUfreq +governor as a replacement for simple CPU busy time. + +Two of the most popular CPUfreq governors, interactive and ondemand, +utilize a window-based approach for measuring CPU busy time. This +works well with the window-based load tracking scheme previously +described. The following APIs are provided to allow the CPUfreq +governor to query busy time from the scheduler instead of using the +basic CPU busy time value derived via get_cpu_idle_time_us() and +get_cpu_iowait_time_us() APIs. + + int sched_set_window(u64 window_start, unsigned int window_size) + + This API is invoked by governor at initialization time or whenever + window size is changed. 'window_size' argument (in jiffy units) + indicates the size of window to be used. The first window of size + 'window_size' is set to begin at jiffy 'window_start' + + -EINVAL is returned if per-entity load tracking is in use rather + than window-based load tracking, otherwise a success value of 0 + is returned. + + int sched_get_busy(int cpu) + + Returns the busy time for the given CPU in the most recent + complete window. The value returned is microseconds of busy + time at fmax of given CPU. + +The values returned by sched_get_busy() take a bit of explanation, +both in what they mean and also how they are derived. + +*** 6.1 Per-CPU Window-Based Stats + +In addition to the per-task window-based demand, the HMP scheduler +extensions also track the aggregate demand seen on each CPU. This is +done using the same windows that the task demand is tracked with +(which is in turn set by the governor when frequency guidance is in +use). There are four quantities maintained for each CPU by the HMP scheduler: + + curr_runnable_sum: aggregate demand from all tasks which executed during + the current (not yet completed) window + + prev_runnable_sum: aggregate demand from all tasks which executed during + the most recent completed window + + nt_curr_runnable_sum: aggregate demand from all 'new' tasks which executed + during the current (not yet completed) window + + nt_prev_runnable_sum: aggregate demand from all 'new' tasks which executed + during the most recent completed window. + +When the scheduler is updating a task's window-based stats it also +updates these values. Like per-task window-based demand these +quantities are normalized against the max possible frequency and max +efficiency (instructions per cycle) in the system. If an update occurs +and a window rollover is observed, curr_runnable_sum is copied into +prev_runnable_sum before being reset to 0. The sched_get_busy() API +returns prev_runnable_sum, scaled to the efficiency and fmax of given +CPU. The same applies to nt_curr_runnable_sum and nt_prev_runnable_sum. + +A 'new' task is defined as a task whose number of active windows since fork is +less than sysctl_sched_new_task_windows. An active window is defined as a window +where a task was observed to be runnable. + +*** 6.2 Per-task window-based stats + +Corresponding to curr_runnable_sum and prev_runnable_sum, two counters are +maintained per-task + +curr_window - represents cpu demand of task in its most recently tracked + window +prev_window - represents cpu demand of task in the window prior to the one + being tracked by curr_window + +The above counters are resued for nt_curr_runnable_sum and +nt_prev_runnable_sum. + +"cpu demand" of a task includes its execution time and can also include its +wait time. 'SCHED_FREQ_ACCOUNT_WAIT_TIME' controls whether task's wait +time is included in its 'curr_window' and 'prev_window' counters or not. + +Needless to say, curr_runnable_sum counter of a cpu is derived from curr_window +counter of various tasks that ran on it in its most recent window. + +*** 6.3 Effect of various task events + +We now consider various events and how they affect above mentioned counters. + +PICK_NEXT_TASK + This represents beginning of execution for a task. Provided the task + refers to a non-idle task, a portion of task's wait time that + corresponds to the current window being tracked on a cpu is added to + task's curr_window counter, provided SCHED_FREQ_ACCOUNT_WAIT_TIME is + set. The same quantum is also added to cpu's curr_runnable_sum counter. + The remaining portion, which corresponds to task's wait time in previous + window is added to task's prev_window and cpu's prev_runnable_sum + counters. + +PUT_PREV_TASK + This represents end of execution of a time-slice for a task, where the + task could refer to a cpu's idle task also. In case the task is non-idle + or (in case of task being idle with cpu having non-zero rq->nr_iowait + count and sched_io_is_busy =1), a portion of task's execution time, that + corresponds to current window being tracked on a cpu is added to task's + curr_window_counter and also to cpu's curr_runnable_sum counter. Portion + of task's execution that corresponds to the previous window is added to + task's prev_window and cpu's prev_runnable_sum counters. + +TASK_UPDATE + This event is called on a cpu's currently running task and hence + behaves effectively as PUT_PREV_TASK. Task continues executing after + this event, until PUT_PREV_TASK event occurs on the task (during + context switch). + +TASK_WAKE + This event signifies a task waking from sleep. Since many windows + could have elapsed since the task went to sleep, its curr_window + and prev_window are updated to reflect task's demand in the most + recent and its previous window that is being tracked on a cpu. + +TASK_MIGRATE + This event signifies task migration across cpus. It is invoked on the + task prior to being moved. Thus at the time of this event, the task + can be considered to be in "waiting" state on src_cpu. In that way + this event reflects actions taken under PICK_NEXT_TASK (i.e its + wait time is added to task's curr/prev_window counters as well + as src_cpu's curr/prev_runnable_sum counters, provided + SCHED_FREQ_ACCOUNT_WAIT_TIME is non-zero). After that update, + src_cpu's curr_runnable_sum is reduced by task's curr_window value + and dst_cpu's curr_runnable_sum is increased by task's curr_window + value. Similarly, src_cpu's prev_runnable_sum is reduced by task's + prev_window value and dst_cpu's prev_runnable_sum is increased by + task's prev_window value. + +IRQ_UPDATE + This event signifies end of execution of an interrupt handler. This + event results in update of cpu's busy time counters, curr_runnable_sum + and prev_runnable_sum, provided cpu was idle. + When sched_io_is_busy = 0, only the interrupt handling time is added + to cpu's curr_runnable_sum and prev_runnable_sum counters. When + sched_io_is_busy = 1, the event mirrors actions taken under + TASK_UPDATED event i.e time since last accounting of idle task's cpu + usage is added to cpu's curr_runnable_sum and prev_runnable_sum + counters. + +=========== +7. TUNABLES +=========== + +*** 7.1 sched_spill_load + +Appears at: /proc/sys/kernel/sched_spill_load + +Default value: 100 + +CPU selection criteria for fair-sched class tasks is the lowest power cpu where +they can fit. When the most power-efficient cpu where a task can fit is +overloaded (aggregate demand of tasks currently queued on it exceeds +sched_spill_load), a task can be placed on a higher-performance cpu, even though +the task strictly doesn't need one. + +*** 7.2 sched_spill_nr_run + +Appears at: /proc/sys/kernel/sched_spill_nr_run + +Default value: 10 + +The intent of this tunable is similar to sched_spill_load, except it applies to +nr_running count of a cpu. A task can spill over to a higher-performance cpu +when the most power-efficient cpu where it can normally fit has more tasks than +sched_spill_nr_run. + +*** 7.3 sched_upmigrate + +Appears at: /proc/sys/kernel/sched_upmigrate + +Default value: 80 + +This tunable is a percentage. If a task consumes more than this much +of a CPU, the CPU is considered too small for the task and the +scheduler will try to find a bigger CPU to place the task on. + +*** 7.4 sched_init_task_load + +Appears at: /proc/sys/kernel/sched_init_task_load + +Default value: 15 + +This tunable is a percentage. When a task is first created it has no +history, so the task load tracking mechanism cannot determine a +historical load value to assign to it. This tunable specifies the +initial load value for newly created tasks. Also see Sec 2.8 on per-task +'initial task load' attribute. + +*** 7.5 sched_ravg_hist_size + +Appears at: /proc/sys/kernel/sched_ravg_hist_size + +Default value: 5 + +This tunable controls the number of samples used from task's sum_history[] +array for determination of its demand. + +*** 7.6 sched_window_stats_policy + +Appears at: /proc/sys/kernel/sched_window_stats_policy + +Default value: 2 + +This tunable controls the policy in how window-based load tracking +calculates an overall demand value based on the windows of CPU +utilization it has collected for a task. + +Possible values for this tunable are: +0: Just use the most recent window sample of task activity when calculating + task demand. +1: Use the maximum value of first M samples found in task's cpu demand + history (sum_history[] array), where M = sysctl_sched_ravg_hist_size +2: Use the maximum of (the most recent window sample, average of first M + samples), where M = sysctl_sched_ravg_hist_size +3. Use average of first M samples, where M = sysctl_sched_ravg_hist_size + +*** 7.7 sched_ravg_window + +Appears at: kernel command line argument + +Default value: 10000000 (10ms, units of tunable are nanoseconds) + +This specifies the duration of each window in window-based load +tracking. By default each window is 10ms long. This quantity must +currently be set at boot time on the kernel command line (or the +default value of 10ms can be used). + +*** 7.8 RAVG_HIST_SIZE + +Appears at: compile time only (see RAVG_HIST_SIZE in include/linux/sched.h) + +Default value: 5 + +This macro specifies the number of windows the window-based load +tracking mechanism maintains per task. If default values are used for +both this and sched_ravg_window then a total of 50ms of task history +would be maintained in 5 10ms windows. + +*** 7.9 sched_freq_inc_notify + +Appears at: /proc/sys/kernel/sched_freq_inc_notify + +Default value: 10 * 1024 * 1024 (10 Ghz) + +When scheduler detects that cur_freq of a cluster is insufficient to meet +demand, it sends notification to governor, provided (freq_required - cur_freq) +exceeds sched_freq_inc_notify, where freq_required is the frequency calculated +by scheduler to meet current task demand. Note that sched_freq_inc_notify is +specified in kHz units. + +*** 7.10 sched_freq_dec_notify + +Appears at: /proc/sys/kernel/sched_freq_dec_notify + +Default value: 10 * 1024 * 1024 (10 Ghz) + +When scheduler detects that cur_freq of a cluster is far greater than what is +needed to serve current task demand, it will send notification to governor. +More specifically, notification is sent when (cur_freq - freq_required) +exceeds sched_freq_dec_notify, where freq_required is the frequency calculated +by scheduler to meet current task demand. Note that sched_freq_dec_notify is +specified in kHz units. + +*** 7.11 sched_cpu_high_irqload + +Appears at: /proc/sys/kernel/sched_cpu_high_irqload + +Default value: 10000000 (10ms) + +The scheduler keeps a decaying average of the amount of irq and softirq activity +seen on each CPU within a ten millisecond window. Note that this "irqload" +(reported in the sched_cpu_load_* tracepoint) will be higher than the typical load +in a single window since every time the window rolls over, the value is decayed +by some fraction and then added to the irq/softirq time spent in the next +window. + +When the irqload on a CPU exceeds the value of this tunable, the CPU is no +longer eligible for placement. This will affect the task placement logic +described above, causing the scheduler to try and steer tasks away from +the CPU. + +*** 7.12 cpu.upmigrate_discourage + +Default value : 0 + +This is a cgroup attribute supported by the cpu resource controller. It normally +appears at [root_cpu]/[name1]/../[name2]/cpu.upmigrate_discourage. Here +"root_cpu" is the mount point for cgroup (cpu resource control) filesystem +and name1, name2 etc are names of cgroups that form a hierarchy. + +Setting this flag to 1 discourages upmigration for all tasks of a cgroup. High +demand tasks of such a cgroup will never be classified as big tasks and hence +not upmigrated. Any task of the cgroup is allowed to upmigrate only under +overcommitted scenario. See notes on sched_spill_nr_run and sched_spill_load for +how overcommitment threshold is defined. + +*** 7.13 sched_static_cpu_pwr_cost + +Default value: 0 + +Appears at /sys/devices/system/cpu/cpu/sched_static_cpu_pwr_cost + +This is the power cost associated with bringing an idle CPU out of low power +mode. It ignores the actual C-state that a CPU may be in and assumes the +worst case power cost of the highest C-state. It is means of biasing task +placement away from idle CPUs when necessary. It can be defined per CPU, +however, a more appropriate usage to define the same value for every CPU +within a cluster and possibly have differing value between clusters as +needed. + + +*** 7.14 sched_static_cluster_pwr_cost + +Default value: 0 + +Appears at /sys/devices/system/cpu/cpu/sched_static_cluster_pwr_cost + +This is the power cost associated with bringing an idle cluster out of low +power mode. It ignores the actual D-state that a cluster may be in and assumes +the worst case power cost of the highest D-state. It is means of biasing task +placement away from idle clusters when necessary. + +*** 7.15 sched_restrict_cluster_spill + +Default value: 0 + +Appears at /proc/sys/kernel/sched_restrict_cluster_spill + +This tunable can be used to restrict tasks spilling to the higher capacity +(higher power) cluster. When this tunable is enabled, + +- Restrict the higher capacity cluster pulling tasks from the lower capacity +cluster in the load balance path. The restriction is lifted if all of the CPUS +in the lower capacity cluster are above spill. The power cost is used to break +the ties if the capacity of clusters are same for applying this restriction. + +- The current CPU selection algorithm for RT tasks looks for the least loaded +CPU across all clusters. When this tunable is enabled, the RT tasks are +restricted to the lowest possible power cluster. + + +*** 7.16 sched_downmigrate + +Appears at: /proc/sys/kernel/sched_downmigrate + +Default value: 60 + +This tunable is a percentage. It exists to control hysteresis. Lets say a task +migrated to a high-performance cpu when it crossed 80% demand on a +power-efficient cpu. We don't let it come back to a power-efficient cpu until +its demand *in reference to the power-efficient cpu* drops less than 60% +(sched_downmigrate). + + +*** 7.17 sched_small_wakee_task_load + +Appears at: /proc/sys/kernel/sched_small_wakee_task_load + +Default value: 10 + +This tunable is a percentage. Configure the maximum demand of small wakee task. +Sync wakee tasks which have demand less than sched_small_wakee_task_load are +categorized as small wakee tasks. Scheduler places small wakee tasks on the +waker's cluster. + + +*** 7.18 sched_big_waker_task_load + +Appears at: /proc/sys/kernel/sched_big_waker_task_load + +Default value: 25 + +This tunable is a percentage. Configure the minimum demand of big sync waker +task. Scheduler places small wakee tasks woken up by big sync waker on the +waker's cluster. + +========================= +8. HMP SCHEDULER TRACE POINTS +========================= + +*** 8.1 sched_enq_deq_task + +Logged when a task is either enqueued or dequeued on a CPU's run queue. + + -0 [004] d.h4 12700.711665: sched_enq_deq_task: cpu=4 enqueue comm=powertop pid=13227 prio=120 nr_running=1 cpu_load=0 rt_nr_running=0 affine=ff demand=13364423 + +- cpu: the CPU that the task is being enqueued on to or dequeued off of +- enqueue/dequeue: whether this was an enqueue or dequeue event +- comm: name of task +- pid: PID of task +- prio: priority of task +- nr_running: number of runnable tasks on this CPU +- cpu_load: current priority-weighted load on the CPU (note, this is *not* + the same as CPU utilization or a metric tracked by PELT/window-based tracking) +- rt_nr_running: number of real-time processes running on this CPU +- affine: CPU affinity mask in hex for this task (so ff is a task eligible to + run on CPUs 0-7) +- demand: window-based task demand computed based on selected policy (recent, + max, or average) (ns) + +*** 8.2 sched_task_load + +Logged when selecting the best CPU to run the task (select_best_cpu()). + +sched_task_load: 4004 (adbd): demand=698425 boost=0 reason=0 sync=0 need_idle=0 best_cpu=0 latency=103177 + +- demand: window-based task demand computed based on selected policy (recent, + max, or average) (ns) +- boost: whether boost is in effect +- reason: reason we are picking a new CPU: + 0: no migration - selecting a CPU for a wakeup or new task wakeup + 1: move to big CPU (migration) + 2: move to little CPU (migration) + 3: move to low irq load CPU (migration) +- sync: is the nature synchronous in nature +- need_idle: is an idle CPU required for this task based on PF_WAKE_UP_IDLE +- best_cpu: The CPU selected by the select_best_cpu() function for placement +- latency: The execution time of the function select_best_cpu() + +*** 8.3 sched_cpu_load_* + +Logged when selecting the best CPU to run a task (select_best_cpu() for fair +class tasks, find_lowest_rq_hmp() for RT tasks) and load balancing +(update_sg_lb_stats()). + +-0 [004] d.h3 12700.711541: sched_cpu_load_*: cpu 0 idle 1 nr_run 0 nr_big 0 lsf 1119 capacity 1024 cr_avg 0 irqload 3301121 fcur 729600 fmax 1459200 power_cost 5 cstate 2 temp 38 + +- cpu: the CPU being described +- idle: boolean indicating whether the CPU is idle +- nr_run: number of tasks running on CPU +- nr_big: number of BIG tasks running on CPU +- lsf: load scale factor - multiply normalized load by this factor to determine + how much load task will exert on CPU +- capacity: capacity of CPU (based on max possible frequency and efficiency) +- cr_avg: cumulative runnable average, instantaneous sum of the demand (either + PELT or window-based) of all the runnable task on a CPU (ns) +- irqload: decaying average of irq activity on CPU (ns) +- fcur: current CPU frequency (Khz) +- fmax: max CPU frequency (but not maximum _possible_ frequency) (KHz) +- power_cost: cost of running this CPU at the current frequency +- cstate: current cstate of CPU +- temp: current temperature of the CPU + +The power_cost value above differs in how it is calculated depending on the +callsite of this tracepoint. The select_best_cpu() call to this tracepoint +finds the minimum frequency required to satisfy the existing load on the CPU +as well as the task being placed, and returns the power cost of that frequency. +The load balance and real time task placement paths used a fixed frequency +(highest frequency common to all CPUs for load balancing, minimum +frequency of the CPU for real time task placement). + +*** 8.4 sched_update_task_ravg + +Logged when window-based stats are updated for a task. The update may happen +for a variety of reasons, see section 2.5, "Task Events." + +-0 [004] d.h4 12700.711513: sched_update_task_ravg: wc 12700711473496 ws 12700691772135 delta 19701361 event TASK_WAKE cpu 4 cur_freq 199200 cur_pid 0 task 13227 (powertop) ms 12640648272532 delta 60063200964 demand 13364423 sum 0 irqtime 0 cs 0 ps 495018 cur_window 0 prev_window 0 + +- wc: wallclock, output of sched_clock(), monotonically increasing time since + boot (will roll over in 585 years) (ns) +- ws: window start, time when the current window started (ns) +- delta: time since the window started (wc - ws) (ns) +- event: What event caused this trace event to occur (see section 2.5 for more + details) +- cpu: which CPU the task is running on +- cur_freq: CPU's current frequency in KHz +- curr_pid: PID of the current running task (current) +- task: PID and name of task being updated +- ms: mark start - timestamp of the beginning of a segment of task activity, + either sleeping or runnable/running (ns) +- delta: time since last event within the window (wc - ms) (ns) +- demand: task demand computed based on selected policy (recent, max, or + average) (ns) +- sum: the task's run time during current window scaled by frequency and + efficiency (ns) +- irqtime: length of interrupt activity (ns). A non-zero irqtime is seen + when an idle cpu handles interrupts, the time for which needs to be + accounted as cpu busy time +- cs: curr_runnable_sum of cpu (ns). See section 6.1 for more details of this + counter. +- ps: prev_runnable_sum of cpu (ns). See section 6.1 for more details of this + counter. +- cur_window: cpu demand of task in its most recently tracked window (ns) +- prev_window: cpu demand of task in the window prior to the one being tracked + by cur_window + +*** 8.5 sched_update_history + +Logged when update_task_ravg() is accounting task activity into one or +more windows that have completed. This may occur more than once for a +single call into update_task_ravg(). A task that ran for 24ms spanning +four 10ms windows (the last 2ms of window 1, all of windows 2 and 3, +and the first 2ms of window 4) would result in two calls into +update_history() from update_task_ravg(). The first call would record activity +in completed window 1 and second call would record activity for windows 2 and 3 +together (samples will be 2 in second call). + +-0 [004] d.h4 12700.711489: sched_update_history: 13227 (powertop): runtime 13364423 samples 1 event TASK_WAKE demand 13364423 (hist: 13364423 9871252 2236009 6162476 10282078) cpu 4 nr_big 0 + +- runtime: task cpu demand in recently completed window(s). This value is scaled + to max_possible_freq and max_possible_efficiency. This value is pushed into + task's demand history array. The number of windows to which runtime applies is + provided by samples field. +- samples: Number of samples (windows), each having value of runtime, that is + recorded in task's demand history array. +- event: What event caused this trace event to occur (see section 2.5 for more + details) - PUT_PREV_TASK, PICK_NEXT_TASK, TASK_WAKE, TASK_MIGRATE, + TASK_UPDATE +- demand: task demand computed based on selected policy (recent, max, or + average) (ns) +- hist: last 5 windows of history for the task with the most recent window + listed first +- cpu: CPU the task is associated with +- nr_big: number of big tasks on the CPU + +*** 8.6 sched_reset_all_windows_stats + +Logged when key parameters controlling window-based statistics collection are +changed. This event signifies that all window-based statistics for tasks and +cpus are being reset. Changes to below attributes result in such a reset: + +* sched_ravg_window (See Sec 2) +* sched_window_stats_policy (See Sec 2.4) +* sched_ravg_hist_size (See Sec 7.11) + +-0 [004] d.h4 12700.711489: sched_reset_all_windows_stats: time_taken 1123 window_start 0 window_size 0 reason POLICY_CHANGE old_val 0 new_val 1 + +- time_taken: time taken for the reset function to complete (ns) +- window_start: Beginning of first window following change to window size (ns) +- window_size: Size of window. Non-zero if window-size is changing (in ticks) +- reason: Reason for reset of statistics. +- old_val: Old value of variable, change of which is triggering reset +- new_val: New value of variable, change of which is triggering reset + +*** 8.7 sched_migration_update_sum + +Logged when a task is migrating to another cpu. + +-0 [000] d..8 5020.404137: sched_migration_update_sum: cpu 0: cs 471278 ps 902463 nt_cs 0 nt_ps 0 pid 2645 + +- cpu: cpu, away from which or to which, task is migrating +- cs: curr_runnable_sum of cpu (ns). See Sec 6.1 for more details of this + counter. +- ps: prev_runnable_sum of cpu (ns). See Sec 6.1 for more details of this + counter. +- nt_cs: nt_curr_runnable_sum of cpu (ns). See Sec 6.1 for more details of + this counter. +- nt_ps: nt_prev_runnable_sum of cpu (ns). See Sec 6.1 for more details of + this counter +- pid: PID of migrating task + +*** 8.8 sched_get_busy + +Logged when scheduler is returning busy time statistics for a cpu. + +<...>-4331 [003] d.s3 313.700108: sched_get_busy: cpu 3 load 19076 new_task_load 0 early 0 + + +- cpu: cpu, for which busy time statistic (prev_runnable_sum) is being + returned (ns) +- load: corresponds to prev_runnable_sum (ns), scaled to fmax of cpu +- new_task_load: corresponds to nt_prev_runnable_sum to fmax of cpu +- early: A flag indicating whether the scheduler is passing regular load or early detection load + 0 - regular load + 1 - early detection load + +*** 8.9 sched_freq_alert + +Logged when scheduler is alerting cpufreq governor about need to change +frequency + +-0 [004] d.h4 12700.711489: sched_freq_alert: cpu 0 old_load=XXX new_load=YYY + +- cpu: cpu in cluster that has highest load (prev_runnable_sum) +- old_load: cpu busy time last reported to governor. This is load scaled in + reference to max_possible_freq and max_possible_efficiency. +- new_load: recent cpu busy time. This is load scaled in + reference to max_possible_freq and max_possible_efficiency. + +*** 8.10 sched_set_boost + +Logged when boost settings are being changed + +-0 [004] d.h4 12700.711489: sched_set_boost: ref_count=1 + +- ref_count: A non-zero value indicates boost is in effect diff --git a/include/linux/cpu.h b/include/linux/cpu.h index 9f237b657018..aae03c4077b4 100644 --- a/include/linux/cpu.h +++ b/include/linux/cpu.h @@ -28,6 +28,21 @@ struct cpu { struct device dev; }; +struct cpu_pstate_pwr { + unsigned int freq; + uint32_t power; +}; + +struct cpu_pwr_stats { + int cpu; + long temp; + struct cpu_pstate_pwr *ptable; + bool throttling; + int len; +}; + +extern struct cpu_pwr_stats *get_cpu_pwr_stats(void); + extern void boot_cpu_init(void); extern void boot_cpu_state_init(void); diff --git a/include/linux/sched.h b/include/linux/sched.h index 348f51b0ec92..cc6f436e8a48 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -174,6 +174,9 @@ extern unsigned long nr_iowait(void); extern unsigned long nr_iowait_cpu(int cpu); extern void get_iowait_load(unsigned long *nr_waiters, unsigned long *load); +extern void sched_update_nr_prod(int cpu, long delta, bool inc); +extern void sched_get_nr_running_avg(int *avg, int *iowait_avg, int *big_avg); + extern void calc_global_load(unsigned long ticks); #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) @@ -315,6 +318,25 @@ extern char ___assert_task_state[1 - 2*!!( /* Task command name length */ #define TASK_COMM_LEN 16 +extern const char *sched_window_reset_reasons[]; + +enum task_event { + PUT_PREV_TASK = 0, + PICK_NEXT_TASK = 1, + TASK_WAKE = 2, + TASK_MIGRATE = 3, + TASK_UPDATE = 4, + IRQ_UPDATE = 5, +}; + +/* Note: this need to be in sync with migrate_type_names array */ +enum migrate_types { + GROUP_TO_RQ, + RQ_TO_GROUP, + RQ_TO_RQ, + GROUP_TO_GROUP, +}; + #include /* @@ -1335,6 +1357,47 @@ struct sched_statistics { }; #endif +#define RAVG_HIST_SIZE_MAX 5 +#define NUM_BUSY_BUCKETS 10 + +/* ravg represents frequency scaled cpu-demand of tasks */ +struct ravg { + /* + * 'mark_start' marks the beginning of an event (task waking up, task + * starting to execute, task being preempted) within a window + * + * 'sum' represents how runnable a task has been within current + * window. It incorporates both running time and wait time and is + * frequency scaled. + * + * 'sum_history' keeps track of history of 'sum' seen over previous + * RAVG_HIST_SIZE windows. Windows where task was entirely sleeping are + * ignored. + * + * 'demand' represents maximum sum seen over previous + * sysctl_sched_ravg_hist_size windows. 'demand' could drive frequency + * demand for tasks. + * + * 'curr_window' represents task's contribution to cpu busy time + * statistics (rq->curr_runnable_sum) in current window + * + * 'prev_window' represents task's contribution to cpu busy time + * statistics (rq->prev_runnable_sum) in previous window + * + * 'pred_demand' represents task's current predicted cpu busy time + * + * 'busy_buckets' groups historical busy time into different buckets + * used for prediction + */ + u64 mark_start; + u32 sum, demand; + u32 sum_history[RAVG_HIST_SIZE_MAX]; + u32 curr_window, prev_window; + u16 active_windows; + u32 pred_demand; + u8 busy_buckets[NUM_BUSY_BUCKETS]; +}; + struct sched_entity { struct load_weight load; /* for load-balancing */ struct rb_node run_node; @@ -1505,6 +1568,20 @@ struct task_struct { const struct sched_class *sched_class; struct sched_entity se; struct sched_rt_entity rt; +#ifdef CONFIG_SCHED_HMP + struct ravg ravg; + /* + * 'init_load_pct' represents the initial task load assigned to children + * of this task + */ + u32 init_load_pct; + u64 last_wake_ts; + u64 last_switch_out_ts; + u64 last_cpu_selected_ts; + struct related_thread_group *grp; + struct list_head grp_list; + u64 cpu_cycles; +#endif #ifdef CONFIG_CGROUP_SCHED struct task_group *sched_task_group; #endif @@ -2254,6 +2331,7 @@ extern void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, /* * Per process flags */ +#define PF_WAKE_UP_IDLE 0x00000002 /* try to wake up on an idle CPU */ #define PF_EXITING 0x00000004 /* getting shut down */ #define PF_EXITPIDONE 0x00000008 /* pi exit done on shut down */ #define PF_VCPU 0x00000010 /* I'm a virtual CPU */ @@ -2436,6 +2514,93 @@ static inline int set_cpus_allowed_ptr(struct task_struct *p, } #endif +struct sched_load { + unsigned long prev_load; + unsigned long new_task_load; + unsigned long predicted_load; +}; + +extern int sched_set_wake_up_idle(struct task_struct *p, int wake_up_idle); +extern u32 sched_get_wake_up_idle(struct task_struct *p); + +struct cpu_cycle_counter_cb { + u64 (*get_cpu_cycle_counter)(int cpu); +}; + +#ifdef CONFIG_SCHED_HMP +extern int sched_set_window(u64 window_start, unsigned int window_size); +extern unsigned long sched_get_busy(int cpu); +extern void sched_get_cpus_busy(struct sched_load *busy, + const struct cpumask *query_cpus); +extern void sched_set_io_is_busy(int val); +extern int sched_set_boost(int enable); +extern int sched_set_init_task_load(struct task_struct *p, int init_load_pct); +extern u32 sched_get_init_task_load(struct task_struct *p); +extern int sched_set_static_cpu_pwr_cost(int cpu, unsigned int cost); +extern unsigned int sched_get_static_cpu_pwr_cost(int cpu); +extern int sched_set_static_cluster_pwr_cost(int cpu, unsigned int cost); +extern unsigned int sched_get_static_cluster_pwr_cost(int cpu); +extern int sched_update_freq_max_load(const cpumask_t *cpumask); +extern void sched_update_cpu_freq_min_max(const cpumask_t *cpus, + u32 fmin, u32 fmax); +extern void sched_set_cpu_cstate(int cpu, int cstate, + int wakeup_energy, int wakeup_latency); +extern void sched_set_cluster_dstate(const cpumask_t *cluster_cpus, int dstate, + int wakeup_energy, int wakeup_latency); +extern int register_cpu_cycle_counter_cb(struct cpu_cycle_counter_cb *cb); +extern u64 sched_ktime_clock(void); +extern int sched_set_group_id(struct task_struct *p, unsigned int group_id); +extern unsigned int sched_get_group_id(struct task_struct *p); + +#else /* CONFIG_SCHED_HMP */ +static inline u64 sched_ktime_clock(void) +{ + return 0; +} + +static inline int +register_cpu_cycle_counter_cb(struct cpu_cycle_counter_cb *cb) +{ + return 0; +} + +static inline int sched_set_window(u64 window_start, unsigned int window_size) +{ + return -EINVAL; +} +static inline unsigned long sched_get_busy(int cpu) +{ + return 0; +} +static inline void sched_get_cpus_busy(struct sched_load *busy, + const struct cpumask *query_cpus) {}; + +static inline void sched_set_io_is_busy(int val) {}; + +static inline int sched_set_boost(int enable) +{ + return -EINVAL; +} + +static inline int sched_update_freq_max_load(const cpumask_t *cpumask) +{ + return 0; +} + +static inline void sched_update_cpu_freq_min_max(const cpumask_t *cpus, + u32 fmin, u32 fmax) { } + +static inline void +sched_set_cpu_cstate(int cpu, int cstate, int wakeup_energy, int wakeup_latency) +{ +} + +static inline void sched_set_cluster_dstate(const cpumask_t *cluster_cpus, + int dstate, int wakeup_energy, int wakeup_latency) +{ +} +#endif /* CONFIG_SCHED_HMP */ + #ifdef CONFIG_NO_HZ_COMMON void calc_load_enter_idle(void); void calc_load_exit_idle(void); @@ -2444,6 +2609,14 @@ static inline void calc_load_enter_idle(void) { } static inline void calc_load_exit_idle(void) { } #endif /* CONFIG_NO_HZ_COMMON */ +static inline void set_wake_up_idle(bool enabled) +{ + if (enabled) + current->flags |= PF_WAKE_UP_IDLE; + else + current->flags &= ~PF_WAKE_UP_IDLE; +} + /* * Do not use outside of architecture code which knows its limitations. * @@ -2461,6 +2634,7 @@ extern u64 sched_clock_cpu(int cpu); extern void sched_clock_init(void); +extern int sched_clock_initialized(void); #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK static inline void sched_clock_tick(void) @@ -2537,7 +2711,7 @@ extern unsigned long long task_sched_runtime(struct task_struct *task); /* sched_exec is called by processes performing an exec */ -#ifdef CONFIG_SMP +#if defined(CONFIG_SMP) && !defined(CONFIG_SCHED_HMP) extern void sched_exec(void); #else #define sched_exec() {} @@ -2670,6 +2844,7 @@ extern void xtime_update(unsigned long ticks); extern int wake_up_state(struct task_struct *tsk, unsigned int state); extern int wake_up_process(struct task_struct *tsk); +extern int wake_up_process_no_notif(struct task_struct *tsk); extern void wake_up_new_task(struct task_struct *tsk); #ifdef CONFIG_SMP extern void kick_process(struct task_struct *tsk); @@ -2678,6 +2853,12 @@ extern void wake_up_new_task(struct task_struct *tsk); #endif extern int sched_fork(unsigned long clone_flags, struct task_struct *p); extern void sched_dead(struct task_struct *p); +#ifdef CONFIG_SCHED_HMP +extern void sched_exit(struct task_struct *p); +#else +static inline void sched_exit(struct task_struct *p) { } +#endif + extern void proc_caches_init(void); extern void flush_signals(struct task_struct *); @@ -3506,6 +3687,8 @@ static inline void set_task_cpu(struct task_struct *p, unsigned int cpu) #endif /* CONFIG_SMP */ +extern struct atomic_notifier_head load_alert_notifier_head; + extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask); extern long sched_getaffinity(pid_t pid, struct cpumask *mask); diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h index 22db1e63707e..6726f05ee4b7 100644 --- a/include/linux/sched/sysctl.h +++ b/include/linux/sched/sysctl.h @@ -18,6 +18,32 @@ extern unsigned int sysctl_sched_latency; extern unsigned int sysctl_sched_min_granularity; extern unsigned int sysctl_sched_wakeup_granularity; extern unsigned int sysctl_sched_child_runs_first; +extern unsigned int sysctl_sched_wake_to_idle; + +#ifdef CONFIG_SCHED_HMP +extern int sysctl_sched_freq_inc_notify; +extern int sysctl_sched_freq_dec_notify; +extern unsigned int sysctl_sched_window_stats_policy; +extern unsigned int sysctl_sched_ravg_hist_size; +extern unsigned int sysctl_sched_cpu_high_irqload; +extern unsigned int sysctl_sched_init_task_load_pct; +extern unsigned int sysctl_sched_spill_nr_run; +extern unsigned int sysctl_sched_spill_load_pct; +extern unsigned int sysctl_sched_upmigrate_pct; +extern unsigned int sysctl_sched_downmigrate_pct; +extern unsigned int sysctl_early_detection_duration; +extern unsigned int sysctl_sched_boost; +extern unsigned int sysctl_sched_small_wakee_task_load_pct; +extern unsigned int sysctl_sched_big_waker_task_load_pct; +extern unsigned int sysctl_sched_select_prev_cpu_us; +extern unsigned int sysctl_sched_enable_colocation; +extern unsigned int sysctl_sched_restrict_cluster_spill; +extern unsigned int sysctl_sched_new_task_windows; +extern unsigned int sysctl_sched_pred_alert_freq; +extern unsigned int sysctl_sched_freq_aggregate; +extern unsigned int sysctl_sched_enable_thread_grouping; +extern unsigned int sysctl_sched_freq_aggregate_threshold_pct; +#endif /* CONFIG_SCHED_HMP */ enum sched_tunable_scaling { SCHED_TUNABLESCALING_NONE, @@ -43,6 +69,18 @@ int sched_proc_update_handler(struct ctl_table *table, int write, loff_t *ppos); #endif +extern int sched_migrate_notify_proc_handler(struct ctl_table *table, + int write, void __user *buffer, size_t *lenp, loff_t *ppos); + +extern int sched_hmp_proc_update_handler(struct ctl_table *table, + int write, void __user *buffer, size_t *lenp, loff_t *ppos); + +extern int sched_boost_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, loff_t *ppos); + +extern int sched_window_update_handler(struct ctl_table *table, + int write, void __user *buffer, size_t *lenp, loff_t *ppos); + /* * control realtime throttling: * diff --git a/include/linux/tick.h b/include/linux/tick.h index 62be0786d6d0..9e207e30a699 100644 --- a/include/linux/tick.h +++ b/include/linux/tick.h @@ -27,6 +27,8 @@ static inline void tick_handover_do_timer(void) { } static inline void tick_cleanup_dead_cpu(int cpu) { } #endif /* !CONFIG_GENERIC_CLOCKEVENTS */ +extern u64 jiffy_to_ktime_ns(u64 *now, u64 *jiffy_ktime_ns); + #if defined(CONFIG_GENERIC_CLOCKEVENTS) && defined(CONFIG_SUSPEND) extern void tick_freeze(void); extern void tick_unfreeze(void); diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h index 3211890ee7d5..a52c343ecf42 100644 --- a/include/trace/events/sched.h +++ b/include/trace/events/sched.h @@ -8,6 +8,8 @@ #include #include +struct rq; + /* * Tracepoint for calling kthread_stop, performed to end a kthread: */ @@ -50,6 +52,561 @@ TRACE_EVENT(sched_kthread_stop_ret, TP_printk("ret=%d", __entry->ret) ); +/* + * Tracepoint for task enqueue/dequeue: + */ +TRACE_EVENT(sched_enq_deq_task, + + TP_PROTO(struct task_struct *p, bool enqueue, unsigned int cpus_allowed), + + TP_ARGS(p, enqueue, cpus_allowed), + + TP_STRUCT__entry( + __array( char, comm, TASK_COMM_LEN ) + __field( pid_t, pid ) + __field( int, prio ) + __field( int, cpu ) + __field( bool, enqueue ) + __field(unsigned int, nr_running ) + __field(unsigned long, cpu_load ) + __field(unsigned int, rt_nr_running ) + __field(unsigned int, cpus_allowed ) +#ifdef CONFIG_SCHED_HMP + __field(unsigned int, demand ) + __field(unsigned int, pred_demand ) +#endif + ), + + TP_fast_assign( + memcpy(__entry->comm, p->comm, TASK_COMM_LEN); + __entry->pid = p->pid; + __entry->prio = p->prio; + __entry->cpu = task_cpu(p); + __entry->enqueue = enqueue; + __entry->nr_running = task_rq(p)->nr_running; + __entry->cpu_load = task_rq(p)->cpu_load[0]; + __entry->rt_nr_running = task_rq(p)->rt.rt_nr_running; + __entry->cpus_allowed = cpus_allowed; +#ifdef CONFIG_SCHED_HMP + __entry->demand = p->ravg.demand; + __entry->pred_demand = p->ravg.pred_demand; +#endif + ), + + TP_printk("cpu=%d %s comm=%s pid=%d prio=%d nr_running=%u cpu_load=%lu rt_nr_running=%u affine=%x" +#ifdef CONFIG_SCHED_HMP + " demand=%u pred_demand=%u" +#endif + , __entry->cpu, + __entry->enqueue ? "enqueue" : "dequeue", + __entry->comm, __entry->pid, + __entry->prio, __entry->nr_running, + __entry->cpu_load, __entry->rt_nr_running, __entry->cpus_allowed +#ifdef CONFIG_SCHED_HMP + , __entry->demand, __entry->pred_demand +#endif + ) +); + +#ifdef CONFIG_SCHED_HMP + +struct group_cpu_time; +struct migration_sum_data; +extern const char *task_event_names[]; +extern const char *migrate_type_names[]; + +TRACE_EVENT(sched_task_load, + + TP_PROTO(struct task_struct *p, bool boost, int reason, + bool sync, bool need_idle, u32 flags, int best_cpu), + + TP_ARGS(p, boost, reason, sync, need_idle, flags, best_cpu), + + TP_STRUCT__entry( + __array( char, comm, TASK_COMM_LEN ) + __field( pid_t, pid ) + __field(unsigned int, demand ) + __field( bool, boost ) + __field( int, reason ) + __field( bool, sync ) + __field( bool, need_idle ) + __field( u32, flags ) + __field( int, best_cpu ) + __field( u64, latency ) + ), + + TP_fast_assign( + memcpy(__entry->comm, p->comm, TASK_COMM_LEN); + __entry->pid = p->pid; + __entry->demand = p->ravg.demand; + __entry->boost = boost; + __entry->reason = reason; + __entry->sync = sync; + __entry->need_idle = need_idle; + __entry->flags = flags; + __entry->best_cpu = best_cpu; + __entry->latency = p->state == TASK_WAKING ? + sched_ktime_clock() - + p->ravg.mark_start : 0; + ), + + TP_printk("%d (%s): demand=%u boost=%d reason=%d sync=%d need_idle=%d flags=%x best_cpu=%d latency=%llu", + __entry->pid, __entry->comm, __entry->demand, + __entry->boost, __entry->reason, __entry->sync, + __entry->need_idle, __entry->flags, + __entry->best_cpu, __entry->latency) +); + +TRACE_EVENT(sched_set_preferred_cluster, + + TP_PROTO(struct related_thread_group *grp, u64 total_demand), + + TP_ARGS(grp, total_demand), + + TP_STRUCT__entry( + __field( int, id ) + __field( u64, demand ) + __field( int, cluster_first_cpu ) + ), + + TP_fast_assign( + __entry->id = grp->id; + __entry->demand = total_demand; + __entry->cluster_first_cpu = grp->preferred_cluster ? + cluster_first_cpu(grp->preferred_cluster) + : -1; + ), + + TP_printk("group_id %d total_demand %llu preferred_cluster_first_cpu %d", + __entry->id, __entry->demand, + __entry->cluster_first_cpu) +); + +DECLARE_EVENT_CLASS(sched_cpu_load, + + TP_PROTO(struct rq *rq, int idle, u64 irqload, unsigned int power_cost, int temp), + + TP_ARGS(rq, idle, irqload, power_cost, temp), + + TP_STRUCT__entry( + __field(unsigned int, cpu ) + __field(unsigned int, idle ) + __field(unsigned int, nr_running ) + __field(unsigned int, nr_big_tasks ) + __field(unsigned int, load_scale_factor ) + __field(unsigned int, capacity ) + __field( u64, cumulative_runnable_avg ) + __field( u64, irqload ) + __field(unsigned int, max_freq ) + __field(unsigned int, power_cost ) + __field( int, cstate ) + __field( int, dstate ) + __field( int, temp ) + ), + + TP_fast_assign( + __entry->cpu = rq->cpu; + __entry->idle = idle; + __entry->nr_running = rq->nr_running; + __entry->nr_big_tasks = rq->hmp_stats.nr_big_tasks; + __entry->load_scale_factor = cpu_load_scale_factor(rq->cpu); + __entry->capacity = cpu_capacity(rq->cpu); + __entry->cumulative_runnable_avg = rq->hmp_stats.cumulative_runnable_avg; + __entry->irqload = irqload; + __entry->max_freq = cpu_max_freq(rq->cpu); + __entry->power_cost = power_cost; + __entry->cstate = rq->cstate; + __entry->dstate = rq->cluster->dstate; + __entry->temp = temp; + ), + + TP_printk("cpu %u idle %d nr_run %u nr_big %u lsf %u capacity %u cr_avg %llu irqload %llu fmax %u power_cost %u cstate %d dstate %d temp %d", + __entry->cpu, __entry->idle, __entry->nr_running, __entry->nr_big_tasks, + __entry->load_scale_factor, __entry->capacity, + __entry->cumulative_runnable_avg, __entry->irqload, + __entry->max_freq, __entry->power_cost, __entry->cstate, + __entry->dstate, __entry->temp) +); + +DEFINE_EVENT(sched_cpu_load, sched_cpu_load_wakeup, + TP_PROTO(struct rq *rq, int idle, u64 irqload, unsigned int power_cost, int temp), + TP_ARGS(rq, idle, irqload, power_cost, temp) +); + +DEFINE_EVENT(sched_cpu_load, sched_cpu_load_lb, + TP_PROTO(struct rq *rq, int idle, u64 irqload, unsigned int power_cost, int temp), + TP_ARGS(rq, idle, irqload, power_cost, temp) +); + +DEFINE_EVENT(sched_cpu_load, sched_cpu_load_cgroup, + TP_PROTO(struct rq *rq, int idle, u64 irqload, unsigned int power_cost, int temp), + TP_ARGS(rq, idle, irqload, power_cost, temp) +); + +TRACE_EVENT(sched_set_boost, + + TP_PROTO(int ref_count), + + TP_ARGS(ref_count), + + TP_STRUCT__entry( + __field(unsigned int, ref_count ) + ), + + TP_fast_assign( + __entry->ref_count = ref_count; + ), + + TP_printk("ref_count=%d", __entry->ref_count) +); + +TRACE_EVENT(sched_update_task_ravg, + + TP_PROTO(struct task_struct *p, struct rq *rq, enum task_event evt, + u64 wallclock, u64 irqtime, u64 cycles, u64 exec_time, + struct group_cpu_time *cpu_time), + + TP_ARGS(p, rq, evt, wallclock, irqtime, cycles, exec_time, cpu_time), + + TP_STRUCT__entry( + __array( char, comm, TASK_COMM_LEN ) + __field( pid_t, pid ) + __field( pid_t, cur_pid ) + __field(unsigned int, cur_freq ) + __field( u64, wallclock ) + __field( u64, mark_start ) + __field( u64, delta_m ) + __field( u64, win_start ) + __field( u64, delta ) + __field( u64, irqtime ) + __field(enum task_event, evt ) + __field(unsigned int, demand ) + __field(unsigned int, sum ) + __field( int, cpu ) + __field(unsigned int, pred_demand ) + __field( u64, rq_cs ) + __field( u64, rq_ps ) + __field( u64, grp_cs ) + __field( u64, grp_ps ) + __field( u64, grp_nt_cs ) + __field( u64, grp_nt_ps ) + __field( u32, curr_window ) + __field( u32, prev_window ) + __field( u64, nt_cs ) + __field( u64, nt_ps ) + __field( u32, active_windows ) + ), + + TP_fast_assign( + __entry->wallclock = wallclock; + __entry->win_start = rq->window_start; + __entry->delta = (wallclock - rq->window_start); + __entry->evt = evt; + __entry->cpu = rq->cpu; + __entry->cur_pid = rq->curr->pid; + __entry->cur_freq = cpu_cycles_to_freq(cycles, exec_time); + memcpy(__entry->comm, p->comm, TASK_COMM_LEN); + __entry->pid = p->pid; + __entry->mark_start = p->ravg.mark_start; + __entry->delta_m = (wallclock - p->ravg.mark_start); + __entry->demand = p->ravg.demand; + __entry->sum = p->ravg.sum; + __entry->irqtime = irqtime; + __entry->pred_demand = p->ravg.pred_demand; + __entry->rq_cs = rq->curr_runnable_sum; + __entry->rq_ps = rq->prev_runnable_sum; + __entry->grp_cs = cpu_time ? cpu_time->curr_runnable_sum : 0; + __entry->grp_ps = cpu_time ? cpu_time->prev_runnable_sum : 0; + __entry->grp_nt_cs = cpu_time ? cpu_time->nt_curr_runnable_sum : 0; + __entry->grp_nt_ps = cpu_time ? cpu_time->nt_prev_runnable_sum : 0; + __entry->curr_window = p->ravg.curr_window; + __entry->prev_window = p->ravg.prev_window; + __entry->nt_cs = rq->nt_curr_runnable_sum; + __entry->nt_ps = rq->nt_prev_runnable_sum; + __entry->active_windows = p->ravg.active_windows; + ), + + TP_printk("wc %llu ws %llu delta %llu event %s cpu %d cur_freq %u cur_pid %d task %d (%s) ms %llu delta %llu demand %u sum %u irqtime %llu pred_demand %u rq_cs %llu rq_ps %llu cur_window %u prev_window %u nt_cs %llu nt_ps %llu active_wins %u grp_cs %lld grp_ps %lld, grp_nt_cs %llu, grp_nt_ps: %llu" + , __entry->wallclock, __entry->win_start, __entry->delta, + task_event_names[__entry->evt], __entry->cpu, + __entry->cur_freq, __entry->cur_pid, + __entry->pid, __entry->comm, __entry->mark_start, + __entry->delta_m, __entry->demand, + __entry->sum, __entry->irqtime, __entry->pred_demand, + __entry->rq_cs, __entry->rq_ps, __entry->curr_window, + __entry->prev_window, __entry->nt_cs, __entry->nt_ps, + __entry->active_windows, __entry->grp_cs, + __entry->grp_ps, __entry->grp_nt_cs, __entry->grp_nt_ps) +); + +TRACE_EVENT(sched_get_task_cpu_cycles, + + TP_PROTO(int cpu, int event, u64 cycles, u64 exec_time), + + TP_ARGS(cpu, event, cycles, exec_time), + + TP_STRUCT__entry( + __field(int, cpu ) + __field(int, event ) + __field(u64, cycles ) + __field(u64, exec_time ) + __field(u32, freq ) + __field(u32, legacy_freq ) + ), + + TP_fast_assign( + __entry->cpu = cpu; + __entry->event = event; + __entry->cycles = cycles; + __entry->exec_time = exec_time; + __entry->freq = cpu_cycles_to_freq(cycles, exec_time); + __entry->legacy_freq = cpu_cur_freq(cpu); + ), + + TP_printk("cpu=%d event=%d cycles=%llu exec_time=%llu freq=%u legacy_freq=%u", + __entry->cpu, __entry->event, __entry->cycles, + __entry->exec_time, __entry->freq, __entry->legacy_freq) +); + +TRACE_EVENT(sched_update_history, + + TP_PROTO(struct rq *rq, struct task_struct *p, u32 runtime, int samples, + enum task_event evt), + + TP_ARGS(rq, p, runtime, samples, evt), + + TP_STRUCT__entry( + __array( char, comm, TASK_COMM_LEN ) + __field( pid_t, pid ) + __field(unsigned int, runtime ) + __field( int, samples ) + __field(enum task_event, evt ) + __field(unsigned int, demand ) + __field(unsigned int, pred_demand ) + __array( u32, hist, RAVG_HIST_SIZE_MAX) + __field(unsigned int, nr_big_tasks ) + __field( int, cpu ) + ), + + TP_fast_assign( + memcpy(__entry->comm, p->comm, TASK_COMM_LEN); + __entry->pid = p->pid; + __entry->runtime = runtime; + __entry->samples = samples; + __entry->evt = evt; + __entry->demand = p->ravg.demand; + __entry->pred_demand = p->ravg.pred_demand; + memcpy(__entry->hist, p->ravg.sum_history, + RAVG_HIST_SIZE_MAX * sizeof(u32)); + __entry->nr_big_tasks = rq->hmp_stats.nr_big_tasks; + __entry->cpu = rq->cpu; + ), + + TP_printk("%d (%s): runtime %u samples %d event %s demand %u pred_demand %u" + " (hist: %u %u %u %u %u) cpu %d nr_big %u", + __entry->pid, __entry->comm, + __entry->runtime, __entry->samples, + task_event_names[__entry->evt], + __entry->demand, __entry->pred_demand, + __entry->hist[0], __entry->hist[1], + __entry->hist[2], __entry->hist[3], + __entry->hist[4], __entry->cpu, __entry->nr_big_tasks) +); + +TRACE_EVENT(sched_reset_all_window_stats, + + TP_PROTO(u64 window_start, u64 window_size, u64 time_taken, + int reason, unsigned int old_val, unsigned int new_val), + + TP_ARGS(window_start, window_size, time_taken, + reason, old_val, new_val), + + TP_STRUCT__entry( + __field( u64, window_start ) + __field( u64, window_size ) + __field( u64, time_taken ) + __field( int, reason ) + __field(unsigned int, old_val ) + __field(unsigned int, new_val ) + ), + + TP_fast_assign( + __entry->window_start = window_start; + __entry->window_size = window_size; + __entry->time_taken = time_taken; + __entry->reason = reason; + __entry->old_val = old_val; + __entry->new_val = new_val; + ), + + TP_printk("time_taken %llu window_start %llu window_size %llu reason %s old_val %u new_val %u", + __entry->time_taken, __entry->window_start, + __entry->window_size, + sched_window_reset_reasons[__entry->reason], + __entry->old_val, __entry->new_val) +); + +TRACE_EVENT(sched_update_pred_demand, + + TP_PROTO(struct rq *rq, struct task_struct *p, u32 runtime, int pct, + unsigned int pred_demand), + + TP_ARGS(rq, p, runtime, pct, pred_demand), + + TP_STRUCT__entry( + __array( char, comm, TASK_COMM_LEN ) + __field( pid_t, pid ) + __field(unsigned int, runtime ) + __field( int, pct ) + __field(unsigned int, pred_demand ) + __array( u8, bucket, NUM_BUSY_BUCKETS) + __field( int, cpu ) + ), + + TP_fast_assign( + memcpy(__entry->comm, p->comm, TASK_COMM_LEN); + __entry->pid = p->pid; + __entry->runtime = runtime; + __entry->pct = pct; + __entry->pred_demand = pred_demand; + memcpy(__entry->bucket, p->ravg.busy_buckets, + NUM_BUSY_BUCKETS * sizeof(u8)); + __entry->cpu = rq->cpu; + ), + + TP_printk("%d (%s): runtime %u pct %d cpu %d pred_demand %u (buckets: %u %u %u %u %u %u %u %u %u %u)", + __entry->pid, __entry->comm, + __entry->runtime, __entry->pct, __entry->cpu, + __entry->pred_demand, __entry->bucket[0], __entry->bucket[1], + __entry->bucket[2], __entry->bucket[3] ,__entry->bucket[4], + __entry->bucket[5], __entry->bucket[6], __entry->bucket[7], + __entry->bucket[8], __entry->bucket[9]) +); + +TRACE_EVENT(sched_migration_update_sum, + + TP_PROTO(struct task_struct *p, enum migrate_types migrate_type, struct migration_sum_data *d), + + TP_ARGS(p, migrate_type, d), + + TP_STRUCT__entry( + __field(int, tcpu ) + __field(int, pid ) + __field( u64, cs ) + __field( u64, ps ) + __field( s64, nt_cs ) + __field( s64, nt_ps ) + __field(enum migrate_types, migrate_type ) + __field( s64, src_cs ) + __field( s64, src_ps ) + __field( s64, dst_cs ) + __field( s64, dst_ps ) + __field( s64, src_nt_cs ) + __field( s64, src_nt_ps ) + __field( s64, dst_nt_cs ) + __field( s64, dst_nt_ps ) + ), + + TP_fast_assign( + __entry->tcpu = task_cpu(p); + __entry->pid = p->pid; + __entry->migrate_type = migrate_type; + __entry->src_cs = d->src_rq ? + d->src_rq->curr_runnable_sum : + d->src_cpu_time->curr_runnable_sum; + __entry->src_ps = d->src_rq ? + d->src_rq->prev_runnable_sum : + d->src_cpu_time->prev_runnable_sum; + __entry->dst_cs = d->dst_rq ? + d->dst_rq->curr_runnable_sum : + d->dst_cpu_time->curr_runnable_sum; + __entry->dst_ps = d->dst_rq ? + d->dst_rq->prev_runnable_sum : + d->dst_cpu_time->prev_runnable_sum; + __entry->src_nt_cs = d->src_rq ? + d->src_rq->nt_curr_runnable_sum : + d->src_cpu_time->nt_curr_runnable_sum; + __entry->src_nt_ps = d->src_rq ? + d->src_rq->nt_prev_runnable_sum : + d->src_cpu_time->nt_prev_runnable_sum; + __entry->dst_nt_cs = d->dst_rq ? + d->dst_rq->nt_curr_runnable_sum : + d->dst_cpu_time->nt_curr_runnable_sum; + __entry->dst_nt_ps = d->dst_rq ? + d->dst_rq->nt_prev_runnable_sum : + d->dst_cpu_time->nt_prev_runnable_sum; + ), + + TP_printk("pid %d task_cpu %d migrate_type %s src_cs %llu src_ps %llu dst_cs %lld dst_ps %lld src_nt_cs %llu src_nt_ps %llu dst_nt_cs %lld dst_nt_ps %lld", + __entry->pid, __entry->tcpu, migrate_type_names[__entry->migrate_type], + __entry->src_cs, __entry->src_ps, __entry->dst_cs, __entry->dst_ps, + __entry->src_nt_cs, __entry->src_nt_ps, __entry->dst_nt_cs, __entry->dst_nt_ps) +); + +TRACE_EVENT(sched_get_busy, + + TP_PROTO(int cpu, u64 load, u64 nload, u64 pload, int early), + + TP_ARGS(cpu, load, nload, pload, early), + + TP_STRUCT__entry( + __field( int, cpu ) + __field( u64, load ) + __field( u64, nload ) + __field( u64, pload ) + __field( int, early ) + ), + + TP_fast_assign( + __entry->cpu = cpu; + __entry->load = load; + __entry->nload = nload; + __entry->pload = pload; + __entry->early = early; + ), + + TP_printk("cpu %d load %lld new_task_load %lld predicted_load %lld early %d", + __entry->cpu, __entry->load, __entry->nload, + __entry->pload, __entry->early) +); + +TRACE_EVENT(sched_freq_alert, + + TP_PROTO(int cpu, int pd_notif, int check_groups, struct rq *rq, + u64 new_load), + + TP_ARGS(cpu, pd_notif, check_groups, rq, new_load), + + TP_STRUCT__entry( + __field( int, cpu ) + __field( int, pd_notif ) + __field( int, check_groups ) + __field( u64, old_busy_time ) + __field( u64, ps ) + __field( u64, new_load ) + __field( u64, old_pred ) + __field( u64, new_pred ) + ), + + TP_fast_assign( + __entry->cpu = cpu; + __entry->pd_notif = pd_notif; + __entry->check_groups = check_groups; + __entry->old_busy_time = rq->old_busy_time; + __entry->ps = rq->prev_runnable_sum; + __entry->new_load = new_load; + __entry->old_pred = rq->old_estimated_time; + __entry->new_pred = rq->hmp_stats.pred_demands_sum; + ), + + TP_printk("cpu %d pd_notif=%d check_groups %d old_busy_time=%llu prev_sum=%lld new_load=%llu old_pred=%llu new_pred=%llu", + __entry->cpu, __entry->pd_notif, __entry->check_groups, + __entry->old_busy_time, __entry->ps, __entry->new_load, + __entry->old_pred, __entry->new_pred) +); + +#endif /* CONFIG_SCHED_HMP */ + /* * Tracepoint for waking up a task: */ @@ -166,14 +723,15 @@ TRACE_EVENT(sched_switch, */ TRACE_EVENT(sched_migrate_task, - TP_PROTO(struct task_struct *p, int dest_cpu), + TP_PROTO(struct task_struct *p, int dest_cpu, unsigned int load), - TP_ARGS(p, dest_cpu), + TP_ARGS(p, dest_cpu, load), TP_STRUCT__entry( __array( char, comm, TASK_COMM_LEN ) __field( pid_t, pid ) __field( int, prio ) + __field(unsigned int, load ) __field( int, orig_cpu ) __field( int, dest_cpu ) ), @@ -182,15 +740,91 @@ TRACE_EVENT(sched_migrate_task, memcpy(__entry->comm, p->comm, TASK_COMM_LEN); __entry->pid = p->pid; __entry->prio = p->prio; + __entry->load = load; __entry->orig_cpu = task_cpu(p); __entry->dest_cpu = dest_cpu; ), - TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d", - __entry->comm, __entry->pid, __entry->prio, + TP_printk("comm=%s pid=%d prio=%d load=%d orig_cpu=%d dest_cpu=%d", + __entry->comm, __entry->pid, __entry->prio, __entry->load, __entry->orig_cpu, __entry->dest_cpu) ); +/* + * Tracepoint for a CPU going offline/online: + */ +TRACE_EVENT(sched_cpu_hotplug, + + TP_PROTO(int affected_cpu, int error, int status), + + TP_ARGS(affected_cpu, error, status), + + TP_STRUCT__entry( + __field( int, affected_cpu ) + __field( int, error ) + __field( int, status ) + ), + + TP_fast_assign( + __entry->affected_cpu = affected_cpu; + __entry->error = error; + __entry->status = status; + ), + + TP_printk("cpu %d %s error=%d", __entry->affected_cpu, + __entry->status ? "online" : "offline", __entry->error) +); + +/* + * Tracepoint for load balancing: + */ +#if NR_CPUS > 32 +#error "Unsupported NR_CPUS for lb tracepoint." +#endif +TRACE_EVENT(sched_load_balance, + + TP_PROTO(int cpu, enum cpu_idle_type idle, int balance, + unsigned long group_mask, int busiest_nr_running, + unsigned long imbalance, unsigned int env_flags, int ld_moved, + unsigned int balance_interval), + + TP_ARGS(cpu, idle, balance, group_mask, busiest_nr_running, + imbalance, env_flags, ld_moved, balance_interval), + + TP_STRUCT__entry( + __field( int, cpu) + __field( enum cpu_idle_type, idle) + __field( int, balance) + __field( unsigned long, group_mask) + __field( int, busiest_nr_running) + __field( unsigned long, imbalance) + __field( unsigned int, env_flags) + __field( int, ld_moved) + __field( unsigned int, balance_interval) + ), + + TP_fast_assign( + __entry->cpu = cpu; + __entry->idle = idle; + __entry->balance = balance; + __entry->group_mask = group_mask; + __entry->busiest_nr_running = busiest_nr_running; + __entry->imbalance = imbalance; + __entry->env_flags = env_flags; + __entry->ld_moved = ld_moved; + __entry->balance_interval = balance_interval; + ), + + TP_printk("cpu=%d state=%s balance=%d group=%#lx busy_nr=%d imbalance=%ld flags=%#x ld_moved=%d bal_int=%d", + __entry->cpu, + __entry->idle == CPU_IDLE ? "idle" : + (__entry->idle == CPU_NEWLY_IDLE ? "newly_idle" : "busy"), + __entry->balance, + __entry->group_mask, __entry->busiest_nr_running, + __entry->imbalance, __entry->env_flags, __entry->ld_moved, + __entry->balance_interval) +); + DECLARE_EVENT_CLASS(sched_process_template, TP_PROTO(struct task_struct *p), @@ -586,6 +1220,28 @@ TRACE_EVENT(sched_wake_idle_without_ipi, TP_printk("cpu=%d", __entry->cpu) ); + +TRACE_EVENT(sched_get_nr_running_avg, + + TP_PROTO(int avg, int big_avg, int iowait_avg), + + TP_ARGS(avg, big_avg, iowait_avg), + + TP_STRUCT__entry( + __field( int, avg ) + __field( int, big_avg ) + __field( int, iowait_avg ) + ), + + TP_fast_assign( + __entry->avg = avg; + __entry->big_avg = big_avg; + __entry->iowait_avg = iowait_avg; + ), + + TP_printk("avg=%d big_avg=%d iowait_avg=%d", + __entry->avg, __entry->big_avg, __entry->iowait_avg) +); #endif /* _TRACE_SCHED_H */ /* This part must be outside protection */ diff --git a/init/Kconfig b/init/Kconfig index 34407f15e6d3..f595c26258fd 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1165,6 +1165,23 @@ config CGROUP_DEBUG endif # CGROUPS +config SCHED_HMP + bool "Scheduler support for heterogenous multi-processor systems" + depends on SMP && FAIR_GROUP_SCHED + help + This feature will let the scheduler optimize task placement on + systems made of heterogeneous cpus i.e cpus that differ either + in their instructions per-cycle capability or the maximum + frequency they can attain. + +config SCHED_HMP_CSTATE_AWARE + bool "CPU C-state aware scheduler" + depends on SCHED_HMP + help + This feature will let the HMP scheduler optimize task placement + with CPUs C-state. If this is enabled, scheduler places tasks + onto the shallowest C-state CPU among the most power efficient CPUs. + config CHECKPOINT_RESTORE bool "Checkpoint/restore support" if EXPERT select PROC_CHILDREN diff --git a/kernel/sched/Makefile b/kernel/sched/Makefile index 5e59b832ae2b..1302fffe0d50 100644 --- a/kernel/sched/Makefile +++ b/kernel/sched/Makefile @@ -17,8 +17,9 @@ endif obj-y += core.o loadavg.o clock.o cputime.o obj-y += idle_task.o fair.o rt.o deadline.o stop_task.o -obj-y += wait.o swait.o completion.o idle.o +obj-y += wait.o swait.o completion.o idle.o sched_avg.o obj-$(CONFIG_SMP) += cpupri.o cpudeadline.o +obj-$(CONFIG_SCHED_HMP) += hmp.o obj-$(CONFIG_SCHED_AUTOGROUP) += auto_group.o obj-$(CONFIG_SCHEDSTATS) += stats.o obj-$(CONFIG_SCHED_DEBUG) += debug.o diff --git a/kernel/sched/core.c b/kernel/sched/core.c index ae4c93ffe84e..a662a7f753b2 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -91,6 +91,8 @@ #define CREATE_TRACE_POINTS #include +ATOMIC_NOTIFIER_HEAD(load_alert_notifier_head); + DEFINE_MUTEX(sched_domains_mutex); DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); @@ -758,6 +760,7 @@ static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) if (!(flags & ENQUEUE_RESTORE)) sched_info_queued(rq, p); p->sched_class->enqueue_task(rq, p, flags); + trace_sched_enq_deq_task(p, 1, cpumask_bits(&p->cpus_allowed)[0]); } static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) @@ -766,6 +769,7 @@ static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) if (!(flags & DEQUEUE_SAVE)) sched_info_dequeued(rq, p); p->sched_class->dequeue_task(rq, p, flags); + trace_sched_enq_deq_task(p, 0, cpumask_bits(&p->cpus_allowed)[0]); } void activate_task(struct rq *rq, struct task_struct *p, int flags) @@ -998,8 +1002,9 @@ static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new p->on_rq = TASK_ON_RQ_MIGRATING; dequeue_task(rq, p, 0); + double_lock_balance(rq, cpu_rq(new_cpu)); set_task_cpu(p, new_cpu); - raw_spin_unlock(&rq->lock); + double_rq_unlock(cpu_rq(new_cpu), rq); rq = cpu_rq(new_cpu); @@ -1050,6 +1055,8 @@ static int migration_cpu_stop(void *data) struct migration_arg *arg = data; struct task_struct *p = arg->task; struct rq *rq = this_rq(); + int src_cpu = cpu_of(rq); + bool moved = false; /* * The original target cpu might have gone down and we might @@ -1071,15 +1078,21 @@ static int migration_cpu_stop(void *data) * we're holding p->pi_lock. */ if (task_rq(p) == rq) { - if (task_on_rq_queued(p)) + if (task_on_rq_queued(p)) { rq = __migrate_task(rq, p, arg->dest_cpu); - else + moved = true; + } else { p->wake_cpu = arg->dest_cpu; + } } raw_spin_unlock(&rq->lock); raw_spin_unlock(&p->pi_lock); local_irq_enable(); + + if (moved) + notify_migration(src_cpu, arg->dest_cpu, false, p); + return 0; } @@ -1161,7 +1174,8 @@ static int __set_cpus_allowed_ptr(struct task_struct *p, if (cpumask_equal(&p->cpus_allowed, new_mask)) goto out; - if (!cpumask_intersects(new_mask, cpu_valid_mask)) { + dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); + if (dest_cpu >= nr_cpu_ids) { ret = -EINVAL; goto out; } @@ -1246,13 +1260,15 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) #endif #endif - trace_sched_migrate_task(p, new_cpu); + trace_sched_migrate_task(p, new_cpu, pct_task_load(p)); if (task_cpu(p) != new_cpu) { if (p->sched_class->migrate_task_rq) p->sched_class->migrate_task_rq(p); p->se.nr_migrations++; perf_event_task_migrate(p); + + fixup_busy_time(p, new_cpu); } __set_task_cpu(p, new_cpu); @@ -1456,7 +1472,7 @@ unsigned long wait_task_inactive(struct task_struct *p, long match_state) * yield - it could be a while. */ if (unlikely(queued)) { - ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); + ktime_t to = ktime_set(0, NSEC_PER_SEC); set_current_state(TASK_UNINTERRUPTIBLE); schedule_hrtimeout(&to, HRTIMER_MODE_REL); @@ -1798,6 +1814,8 @@ void sched_ttwu_pending(void) void scheduler_ipi(void) { + int cpu = smp_processor_id(); + /* * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting * TIF_NEED_RESCHED remotely (for the first time) will also send @@ -1805,9 +1823,18 @@ void scheduler_ipi(void) */ preempt_fold_need_resched(); - if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) + if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick() && + !got_boost_kick()) return; + if (got_boost_kick()) { + struct rq *rq = cpu_rq(cpu); + + if (rq->curr->sched_class == &fair_sched_class) + check_for_migration(rq, rq->curr); + clear_boost_kick(cpu); + } + /* * Not all reschedule IPI handlers call irq_enter/irq_exit, since * traditionally all their work was done from the interrupt return @@ -2009,6 +2036,18 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) { unsigned long flags; int cpu, success = 0; +#ifdef CONFIG_SMP + unsigned int old_load; + struct rq *rq; + u64 wallclock; + struct related_thread_group *grp = NULL; + int src_cpu; + bool notif_required = false; +#endif + bool freq_notif_allowed = !(wake_flags & WF_NO_NOTIFIER); + bool check_group = false; + + wake_flags &= ~WF_NO_NOTIFIER; /* * If we are going to wake up a thread waiting for CONDITION we @@ -2082,14 +2121,33 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) */ smp_cond_load_acquire(&p->on_cpu, !VAL); + rq = cpu_rq(task_cpu(p)); + raw_spin_lock(&rq->lock); + old_load = task_load(p); + wallclock = sched_ktime_clock(); + update_task_ravg(rq->curr, rq, TASK_UPDATE, wallclock, 0); + update_task_ravg(p, rq, TASK_WAKE, wallclock, 0); + raw_spin_unlock(&rq->lock); + + rcu_read_lock(); + grp = task_related_thread_group(p); + if (update_preferred_cluster(grp, p, old_load)) + set_preferred_cluster(grp); + rcu_read_unlock(); + check_group = grp != NULL; + p->sched_contributes_to_load = !!task_contributes_to_load(p); p->state = TASK_WAKING; cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); - if (task_cpu(p) != cpu) { + src_cpu = task_cpu(p); + if (src_cpu != cpu) { wake_flags |= WF_MIGRATED; set_task_cpu(p, cpu); + notif_required = true; } + + set_task_last_wake(p, wallclock); #endif /* CONFIG_SMP */ ttwu_queue(p, cpu, wake_flags); @@ -2098,6 +2156,17 @@ stat: out: raw_spin_unlock_irqrestore(&p->pi_lock, flags); + if (freq_notif_allowed) { + if (notif_required && !same_freq_domain(src_cpu, cpu)) { + check_for_freq_change(cpu_rq(cpu), + false, check_group); + check_for_freq_change(cpu_rq(src_cpu), + false, check_group); + } else if (success) { + check_for_freq_change(cpu_rq(cpu), true, false); + } + } + return success; } @@ -2114,9 +2183,13 @@ static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie { struct rq *rq = task_rq(p); - if (WARN_ON_ONCE(rq != this_rq()) || - WARN_ON_ONCE(p == current)) + if (rq != this_rq() || p == current) { + printk_deferred("%s: Failed to wakeup task %d (%s), rq = %p," + " this_rq = %p, p = %p, current = %p\n", __func__, + task_pid_nr(p), p->comm, rq, this_rq(), p, current); + return; + } lockdep_assert_held(&rq->lock); @@ -2139,8 +2212,14 @@ static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie trace_sched_waking(p); - if (!task_on_rq_queued(p)) + if (!task_on_rq_queued(p)) { + u64 wallclock = sched_ktime_clock(); + + update_task_ravg(rq->curr, rq, TASK_UPDATE, wallclock, 0); + update_task_ravg(p, rq, TASK_WAKE, wallclock, 0); ttwu_activate(rq, p, ENQUEUE_WAKEUP); + set_task_last_wake(p, wallclock); + } ttwu_do_wakeup(rq, p, 0, cookie); ttwu_stat(p, smp_processor_id(), 0); @@ -2166,6 +2245,26 @@ int wake_up_process(struct task_struct *p) } EXPORT_SYMBOL(wake_up_process); +/** + * wake_up_process_no_notif - Wake up a specific process without notifying + * governor + * @p: The process to be woken up. + * + * Attempt to wake up the nominated process and move it to the set of runnable + * processes. + * + * Return: 1 if the process was woken up, 0 if it was already running. + * + * It may be assumed that this function implies a write memory barrier before + * changing the task state if and only if any tasks are woken up. + */ +int wake_up_process_no_notif(struct task_struct *p) +{ + WARN_ON(task_is_stopped_or_traced(p)); + return try_to_wake_up(p, TASK_NORMAL, WF_NO_NOTIFIER); +} +EXPORT_SYMBOL(wake_up_process_no_notif); + int wake_up_state(struct task_struct *p, unsigned int state) { return try_to_wake_up(p, state, 0); @@ -2188,6 +2287,44 @@ void __dl_clear_params(struct task_struct *p) dl_se->dl_yielded = 0; } +#ifdef CONFIG_SCHED_HMP +/* + * sched_exit() - Set EXITING_TASK_MARKER in task's ravg.demand field + * + * Stop accounting (exiting) task's future cpu usage + * + * We need this so that reset_all_windows_stats() can function correctly. + * reset_all_window_stats() depends on do_each_thread/for_each_thread task + * iterators to reset *all* task's statistics. Exiting tasks however become + * invisible to those iterators. sched_exit() is called on a exiting task prior + * to being removed from task_list, which will let reset_all_window_stats() + * function correctly. + */ +void sched_exit(struct task_struct *p) +{ + unsigned long flags; + int cpu = get_cpu(); + struct rq *rq = cpu_rq(cpu); + u64 wallclock; + + sched_set_group_id(p, 0); + + raw_spin_lock_irqsave(&rq->lock, flags); + /* rq->curr == p */ + wallclock = sched_ktime_clock(); + update_task_ravg(rq->curr, rq, TASK_UPDATE, wallclock, 0); + dequeue_task(rq, p, 0); + reset_task_stats(p); + p->ravg.mark_start = wallclock; + p->ravg.sum_history[0] = EXITING_TASK_MARKER; + enqueue_task(rq, p, 0); + clear_ed_task(p, rq); + raw_spin_unlock_irqrestore(&rq->lock, flags); + + put_cpu(); +} +#endif /* CONFIG_SCHED_HMP */ + /* * Perform scheduler related setup for a newly forked process p. * p is forked by current. @@ -2564,6 +2701,8 @@ void wake_up_new_task(struct task_struct *p) struct rq *rq; raw_spin_lock_irqsave(&p->pi_lock, rf.flags); + init_new_task_load(p); + add_new_task_to_grp(p); p->state = TASK_RUNNING; #ifdef CONFIG_SMP /* @@ -2579,6 +2718,7 @@ void wake_up_new_task(struct task_struct *p) rq = __task_rq_lock(p, &rf); post_init_entity_util_avg(&p->se); + mark_task_starting(p); activate_task(rq, p, 0); p->on_rq = TASK_ON_RQ_QUEUED; trace_sched_wakeup_new(p); @@ -2971,7 +3111,7 @@ void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) *load = rq->load.weight; } -#ifdef CONFIG_SMP +#if defined(CONFIG_SMP) && !defined(CONFIG_SCHED_HMP) /* * sched_exec - execve() is a valuable balancing opportunity, because at @@ -3077,16 +3217,32 @@ void scheduler_tick(void) int cpu = smp_processor_id(); struct rq *rq = cpu_rq(cpu); struct task_struct *curr = rq->curr; + u64 wallclock; + bool early_notif; + u32 old_load; + struct related_thread_group *grp; sched_clock_tick(); raw_spin_lock(&rq->lock); + + old_load = task_load(curr); + set_window_start(rq); + update_rq_clock(rq); curr->sched_class->task_tick(rq, curr, 0); cpu_load_update_active(rq); calc_global_load_tick(rq); + + wallclock = sched_ktime_clock(); + update_task_ravg(rq->curr, rq, TASK_UPDATE, wallclock, 0); + early_notif = early_detection_notify(rq, wallclock); + raw_spin_unlock(&rq->lock); + if (early_notif) + atomic_notifier_call_chain(&load_alert_notifier_head, + 0, (void *)(long)cpu); perf_event_task_tick(); #ifdef CONFIG_SMP @@ -3094,6 +3250,15 @@ void scheduler_tick(void) trigger_load_balance(rq); #endif rq_last_tick_reset(rq); + + rcu_read_lock(); + grp = task_related_thread_group(curr); + if (update_preferred_cluster(grp, curr, old_load)) + set_preferred_cluster(grp); + rcu_read_unlock(); + + if (curr->sched_class == &fair_sched_class) + check_for_migration(rq, curr); } #ifdef CONFIG_NO_HZ_FULL @@ -3227,6 +3392,9 @@ static noinline void __schedule_bug(struct task_struct *prev) if (panic_on_warn) panic("scheduling while atomic\n"); +#ifdef CONFIG_PANIC_ON_SCHED_BUG + BUG(); +#endif dump_stack(); add_taint(TAINT_WARN, LOCKDEP_STILL_OK); } @@ -3337,6 +3505,7 @@ static void __sched notrace __schedule(bool preempt) struct pin_cookie cookie; struct rq *rq; int cpu; + u64 wallclock; cpu = smp_processor_id(); rq = cpu_rq(cpu); @@ -3389,6 +3558,11 @@ static void __sched notrace __schedule(bool preempt) update_rq_clock(rq); next = pick_next_task(rq, prev, cookie); + + wallclock = sched_ktime_clock(); + update_task_ravg(prev, rq, PUT_PREV_TASK, wallclock, 0); + update_task_ravg(next, rq, PICK_NEXT_TASK, wallclock, 0); + clear_tsk_need_resched(prev); clear_preempt_need_resched(); rq->clock_skip_update = 0; @@ -3398,6 +3572,8 @@ static void __sched notrace __schedule(bool preempt) rq->curr = next; ++*switch_count; + set_task_last_switch_out(prev, wallclock); + trace_sched_switch(preempt, prev, next); rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */ } else { @@ -5273,10 +5449,11 @@ void init_idle(struct task_struct *idle, int cpu) struct rq *rq = cpu_rq(cpu); unsigned long flags; + __sched_fork(0, idle); + raw_spin_lock_irqsave(&idle->pi_lock, flags); raw_spin_lock(&rq->lock); - __sched_fork(0, idle); idle->state = TASK_RUNNING; idle->se.exec_start = sched_clock(); @@ -5586,7 +5763,10 @@ static void migrate_tasks(struct rq *dead_rq) rq = __migrate_task(rq, next, dest_cpu); if (rq != dead_rq) { raw_spin_unlock(&rq->lock); + raw_spin_unlock(&next->pi_lock); + notify_migration(dead_rq->cpu, dest_cpu, true, next); rq = dead_rq; + raw_spin_lock(&next->pi_lock); raw_spin_lock(&rq->lock); } raw_spin_unlock(&next->pi_lock); @@ -6030,6 +6210,7 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) { struct rq *rq = cpu_rq(cpu); struct sched_domain *tmp; + unsigned long next_balance = rq->next_balance; /* Remove the sched domains which do not contribute to scheduling. */ for (tmp = sd; tmp; ) { @@ -6061,6 +6242,17 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) sd->child = NULL; } + for (tmp = sd; tmp; ) { + unsigned long interval; + + interval = msecs_to_jiffies(tmp->balance_interval); + if (time_after(next_balance, tmp->last_balance + interval)) + next_balance = tmp->last_balance + interval; + + tmp = tmp->parent; + } + rq->next_balance = next_balance; + sched_domain_debug(sd, cpu); rq_attach_root(rq, rd); @@ -7386,7 +7578,11 @@ int sched_cpu_deactivate(unsigned int cpu) static void sched_rq_cpu_starting(unsigned int cpu) { struct rq *rq = cpu_rq(cpu); + unsigned long flags; + raw_spin_lock_irqsave(&rq->lock, flags); + set_window_start(rq); + raw_spin_unlock_irqrestore(&rq->lock, flags); rq->calc_load_update = calc_load_update; update_max_interval(); } @@ -7407,6 +7603,8 @@ int sched_cpu_dying(unsigned int cpu) /* Handle pending wakeups and then migrate everything off */ sched_ttwu_pending(); raw_spin_lock_irqsave(&rq->lock, flags); + migrate_sync_cpu(cpu); + if (rq->rd) { BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); set_rq_offline(rq); @@ -7414,6 +7612,9 @@ int sched_cpu_dying(unsigned int cpu) migrate_tasks(rq); BUG_ON(rq->nr_running != 1); raw_spin_unlock_irqrestore(&rq->lock, flags); + + clear_hmp_request(cpu); + calc_load_migrate(rq); update_max_interval(); nohz_balance_exit_idle(cpu); @@ -7459,6 +7660,8 @@ void __init sched_init_smp(void) cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); mutex_unlock(&sched_domains_mutex); + update_cluster_topology(); + /* Move init over to a non-isolated CPU */ if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) BUG(); @@ -7530,6 +7733,11 @@ void __init sched_init(void) for (i = 0; i < WAIT_TABLE_SIZE; i++) init_waitqueue_head(bit_wait_table + i); +#ifdef CONFIG_SCHED_HMP + pr_info("HMP scheduling enabled.\n"); + init_clusters(); +#endif + #ifdef CONFIG_FAIR_GROUP_SCHED alloc_size += 2 * nr_cpu_ids * sizeof(void **); #endif @@ -7646,7 +7854,36 @@ void __init sched_init(void) rq->idle_stamp = 0; rq->avg_idle = 2*sysctl_sched_migration_cost; rq->max_idle_balance_cost = sysctl_sched_migration_cost; + rq->push_task = NULL; +#ifdef CONFIG_SCHED_HMP + cpumask_set_cpu(i, &rq->freq_domain_cpumask); + rq->hmp_stats.cumulative_runnable_avg = 0; + rq->window_start = 0; + rq->hmp_stats.nr_big_tasks = 0; + rq->hmp_flags = 0; + rq->cur_irqload = 0; + rq->avg_irqload = 0; + rq->irqload_ts = 0; + rq->static_cpu_pwr_cost = 0; + rq->cc.cycles = 1; + rq->cc.time = 1; + rq->cstate = 0; + rq->wakeup_latency = 0; + rq->wakeup_energy = 0; + /* + * All cpus part of same cluster by default. This avoids the + * need to check for rq->cluster being non-NULL in hot-paths + * like select_best_cpu() + */ + rq->cluster = &init_cluster; + rq->curr_runnable_sum = rq->prev_runnable_sum = 0; + rq->nt_curr_runnable_sum = rq->nt_prev_runnable_sum = 0; + rq->old_busy_time = 0; + rq->old_estimated_time = 0; + rq->old_busy_time_group = 0; + rq->hmp_stats.pred_demands_sum = 0; +#endif INIT_LIST_HEAD(&rq->cfs_tasks); rq_attach_root(rq, &def_root_domain); @@ -7662,6 +7899,8 @@ void __init sched_init(void) atomic_set(&rq->nr_iowait, 0); } + set_hmp_defaults(); + set_load_weight(&init_task); /* @@ -7768,6 +8007,9 @@ void ___might_sleep(const char *file, int line, int preempt_offset) print_ip_sym(preempt_disable_ip); pr_cont("\n"); } +#ifdef CONFIG_PANIC_ON_SCHED_BUG + BUG(); +#endif dump_stack(); add_taint(TAINT_WARN, LOCKDEP_STILL_OK); } @@ -8369,7 +8611,7 @@ int sched_rr_handler(struct ctl_table *table, int write, #ifdef CONFIG_CGROUP_SCHED -static inline struct task_group *css_tg(struct cgroup_subsys_state *css) +inline struct task_group *css_tg(struct cgroup_subsys_state *css) { return css ? container_of(css, struct task_group, css) : NULL; } @@ -8756,6 +8998,13 @@ static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, #endif /* CONFIG_RT_GROUP_SCHED */ static struct cftype cpu_files[] = { +#ifdef CONFIG_SCHED_HMP + { + .name = "upmigrate_discourage", + .read_u64 = cpu_upmigrate_discourage_read_u64, + .write_u64 = cpu_upmigrate_discourage_write_u64, + }, +#endif #ifdef CONFIG_FAIR_GROUP_SCHED { .name = "shares", diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c index 5ebee3164e64..6ce492fbedcf 100644 --- a/kernel/sched/cputime.c +++ b/kernel/sched/cputime.c @@ -46,12 +46,15 @@ void irqtime_account_irq(struct task_struct *curr) struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime); s64 delta; int cpu; + u64 wallclock; + bool account = true; if (!sched_clock_irqtime) return; cpu = smp_processor_id(); - delta = sched_clock_cpu(cpu) - irqtime->irq_start_time; + wallclock = sched_clock_cpu(cpu); + delta = wallclock - irqtime->irq_start_time; irqtime->irq_start_time += delta; u64_stats_update_begin(&irqtime->sync); @@ -65,6 +68,13 @@ void irqtime_account_irq(struct task_struct *curr) irqtime->hardirq_time += delta; else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) irqtime->softirq_time += delta; + else + account = false; + + if (account) + sched_account_irqtime(cpu, curr, delta, wallclock); + else if (curr != this_cpu_ksoftirqd()) + sched_account_irqstart(cpu, curr, wallclock); u64_stats_update_end(&irqtime->sync); } diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index 37e2449186c4..ffca4787dac2 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -820,6 +820,41 @@ static inline void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {} #endif /* CONFIG_SMP */ +#ifdef CONFIG_SCHED_HMP + +static void +inc_hmp_sched_stats_dl(struct rq *rq, struct task_struct *p) +{ + inc_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +static void +dec_hmp_sched_stats_dl(struct rq *rq, struct task_struct *p) +{ + dec_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +static void +fixup_hmp_sched_stats_dl(struct rq *rq, struct task_struct *p, + u32 new_task_load, u32 new_pred_demand) +{ + s64 task_load_delta = (s64)new_task_load - task_load(p); + s64 pred_demand_delta = PRED_DEMAND_DELTA; + + fixup_cumulative_runnable_avg(&rq->hmp_stats, p, task_load_delta, + pred_demand_delta); +} + +#else /* CONFIG_SCHED_HMP */ + +static inline void +inc_hmp_sched_stats_dl(struct rq *rq, struct task_struct *p) { } + +static inline void +dec_hmp_sched_stats_dl(struct rq *rq, struct task_struct *p) { } + +#endif /* CONFIG_SCHED_HMP */ + static inline void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq) { @@ -829,6 +864,7 @@ void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq) WARN_ON(!dl_prio(prio)); dl_rq->dl_nr_running++; add_nr_running(rq_of_dl_rq(dl_rq), 1); + inc_hmp_sched_stats_dl(rq_of_dl_rq(dl_rq), dl_task_of(dl_se)); inc_dl_deadline(dl_rq, deadline); inc_dl_migration(dl_se, dl_rq); @@ -843,6 +879,7 @@ void dec_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq) WARN_ON(!dl_rq->dl_nr_running); dl_rq->dl_nr_running--; sub_nr_running(rq_of_dl_rq(dl_rq), 1); + dec_hmp_sched_stats_dl(rq_of_dl_rq(dl_rq), dl_task_of(dl_se)); dec_dl_deadline(dl_rq, dl_se->deadline); dec_dl_migration(dl_se, dl_rq); @@ -1500,9 +1537,11 @@ retry: goto retry; } + next_task->on_rq = TASK_ON_RQ_MIGRATING; deactivate_task(rq, next_task, 0); set_task_cpu(next_task, later_rq->cpu); activate_task(later_rq, next_task, 0); + next_task->on_rq = TASK_ON_RQ_QUEUED; ret = 1; resched_curr(later_rq); @@ -1588,9 +1627,11 @@ static void pull_dl_task(struct rq *this_rq) resched = true; + p->on_rq = TASK_ON_RQ_MIGRATING; deactivate_task(src_rq, p, 0); set_task_cpu(p, this_cpu); activate_task(this_rq, p, 0); + p->on_rq = TASK_ON_RQ_QUEUED; dmin = p->dl.deadline; /* Is there any other task even earlier? */ @@ -1803,6 +1844,9 @@ const struct sched_class dl_sched_class = { .switched_to = switched_to_dl, .update_curr = update_curr_dl, +#ifdef CONFIG_SCHED_HMP + .fixup_hmp_sched_stats = fixup_hmp_sched_stats_dl, +#endif }; #ifdef CONFIG_SCHED_DEBUG diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c index fa178b62ea79..5671f2606702 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -534,6 +534,14 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) cfs_rq->throttled); SEQ_printf(m, " .%-30s: %d\n", "throttle_count", cfs_rq->throttle_count); + SEQ_printf(m, " .%-30s: %d\n", "runtime_enabled", + cfs_rq->runtime_enabled); +#ifdef CONFIG_SCHED_HMP + SEQ_printf(m, " .%-30s: %d\n", "nr_big_tasks", + cfs_rq->hmp_stats.nr_big_tasks); + SEQ_printf(m, " .%-30s: %llu\n", "cumulative_runnable_avg", + cfs_rq->hmp_stats.cumulative_runnable_avg); +#endif #endif #ifdef CONFIG_FAIR_GROUP_SCHED @@ -622,6 +630,23 @@ do { \ P(cpu_load[2]); P(cpu_load[3]); P(cpu_load[4]); +#ifdef CONFIG_SMP + P(cpu_capacity); +#endif +#ifdef CONFIG_SCHED_HMP + P(static_cpu_pwr_cost); + P(cluster->static_cluster_pwr_cost); + P(cluster->load_scale_factor); + P(cluster->capacity); + P(cluster->max_possible_capacity); + P(cluster->efficiency); + P(cluster->cur_freq); + P(cluster->max_freq); + P(cluster->exec_scale_factor); + P(hmp_stats.nr_big_tasks); + SEQ_printf(m, " .%-30s: %llu\n", "hmp_stats.cumulative_runnable_avg", + rq->hmp_stats.cumulative_runnable_avg); +#endif #undef P #undef PN @@ -700,6 +725,14 @@ static void sched_debug_header(struct seq_file *m) PN(sysctl_sched_wakeup_granularity); P(sysctl_sched_child_runs_first); P(sysctl_sched_features); +#ifdef CONFIG_SCHED_HMP + P(sched_upmigrate); + P(sched_downmigrate); + P(sched_init_task_load_windows); + P(min_capacity); + P(max_capacity); + P(sched_ravg_window); +#endif #undef PN #undef P @@ -861,6 +894,9 @@ static void sched_show_numa(struct task_struct *p, struct seq_file *m) void proc_sched_show_task(struct task_struct *p, struct seq_file *m) { unsigned long nr_switches; + unsigned int load_avg; + + load_avg = pct_task_load(p); SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr(p), get_nr_threads(p)); @@ -918,7 +954,12 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m) P_SCHEDSTAT(se.statistics.nr_wakeups_affine_attempts); P_SCHEDSTAT(se.statistics.nr_wakeups_passive); P_SCHEDSTAT(se.statistics.nr_wakeups_idle); - +#if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED) + __P(load_avg); +#ifdef CONFIG_SCHED_HMP + P(ravg.demand); +#endif +#endif avg_atom = p->se.sum_exec_runtime; if (nr_switches) avg_atom = div64_ul(avg_atom, nr_switches); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index bdb03b0104b7..e84ec63a3974 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -31,9 +31,8 @@ #include #include -#include - #include "sched.h" +#include /* * Targeted preemption latency for CPU-bound tasks: @@ -80,6 +79,14 @@ static unsigned int sched_nr_latency = 8; */ unsigned int sysctl_sched_child_runs_first __read_mostly; +/* + * Controls whether, when SD_SHARE_PKG_RESOURCES is on, if all + * tasks go to idle CPUs when woken. If this is off, note that the + * per-task flag PF_WAKE_UP_IDLE can still cause a task to go to an + * idle CPU upon being woken. + */ +unsigned int __read_mostly sysctl_sched_wake_to_idle; + /* * SCHED_OTHER wake-up granularity. * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds) @@ -242,6 +249,9 @@ static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight return mul_u64_u32_shr(delta_exec, fact, shift); } +#ifdef CONFIG_SMP +static int active_load_balance_cpu_stop(void *data); +#endif const struct sched_class fair_sched_class; @@ -2671,6 +2681,25 @@ static inline void update_cfs_shares(struct cfs_rq *cfs_rq) #endif /* CONFIG_FAIR_GROUP_SCHED */ #ifdef CONFIG_SMP +u32 sched_get_wake_up_idle(struct task_struct *p) +{ + u32 enabled = p->flags & PF_WAKE_UP_IDLE; + + return !!enabled; +} + +int sched_set_wake_up_idle(struct task_struct *p, int wake_up_idle) +{ + int enable = !!wake_up_idle; + + if (enable) + p->flags |= PF_WAKE_UP_IDLE; + else + p->flags &= ~PF_WAKE_UP_IDLE; + + return 0; +} + /* Precomputed fixed inverse multiplies for multiplication by y^n */ static const u32 runnable_avg_yN_inv[] = { 0xffffffff, 0xfa83b2da, 0xf5257d14, 0xefe4b99a, 0xeac0c6e6, 0xe5b906e6, @@ -2756,6 +2785,920 @@ static u32 __compute_runnable_contrib(u64 n) return contrib + runnable_avg_yN_sum[n]; } +#ifdef CONFIG_SCHED_HMP + +/* CPU selection flag */ +#define SBC_FLAG_PREV_CPU 0x1 +#define SBC_FLAG_BEST_CAP_CPU 0x2 +#define SBC_FLAG_CPU_COST 0x4 +#define SBC_FLAG_MIN_COST 0x8 +#define SBC_FLAG_IDLE_LEAST_LOADED 0x10 +#define SBC_FLAG_IDLE_CSTATE 0x20 +#define SBC_FLAG_COST_CSTATE_TIE_BREAKER 0x40 +#define SBC_FLAG_COST_CSTATE_PREV_CPU_TIE_BREAKER 0x80 +#define SBC_FLAG_CSTATE_LOAD 0x100 +#define SBC_FLAG_BEST_SIBLING 0x200 + +/* Cluster selection flag */ +#define SBC_FLAG_COLOC_CLUSTER 0x10000 +#define SBC_FLAG_WAKER_CLUSTER 0x20000 +#define SBC_FLAG_BACKUP_CLUSTER 0x40000 + +struct cpu_select_env { + struct task_struct *p; + struct related_thread_group *rtg; + u8 reason; + u8 need_idle:1; + u8 need_waker_cluster:1; + u8 sync:1; + u8 ignore_prev_cpu:1; + enum sched_boost_type boost_type; + int prev_cpu; + DECLARE_BITMAP(candidate_list, NR_CPUS); + DECLARE_BITMAP(backup_list, NR_CPUS); + u64 task_load; + u64 cpu_load; + u32 sbc_best_flag; + u32 sbc_best_cluster_flag; +}; + +struct cluster_cpu_stats { + int best_idle_cpu, least_loaded_cpu; + int best_capacity_cpu, best_cpu, best_sibling_cpu; + int min_cost, best_sibling_cpu_cost; + int best_cpu_cstate; + u64 min_load, best_load, best_sibling_cpu_load; + s64 highest_spare_capacity; +}; + +static int spill_threshold_crossed(struct cpu_select_env *env, struct rq *rq) +{ + u64 total_load; + + total_load = env->task_load + env->cpu_load; + + if (total_load > sched_spill_load || + (rq->nr_running + 1) > sysctl_sched_spill_nr_run) + return 1; + + return 0; +} + +static int skip_cpu(int cpu, struct cpu_select_env *env) +{ + int tcpu = task_cpu(env->p); + int skip = 0; + + if (!env->reason) + return 0; + + if (is_reserved(cpu)) + return 1; + + switch (env->reason) { + case UP_MIGRATION: + skip = !idle_cpu(cpu); + break; + case IRQLOAD_MIGRATION: + /* Purposely fall through */ + default: + skip = (cpu == tcpu); + break; + } + + return skip; +} + +static inline int +acceptable_capacity(struct sched_cluster *cluster, struct cpu_select_env *env) +{ + int tcpu; + + if (!env->reason) + return 1; + + tcpu = task_cpu(env->p); + switch (env->reason) { + case UP_MIGRATION: + return cluster->capacity > cpu_capacity(tcpu); + + case DOWN_MIGRATION: + return cluster->capacity < cpu_capacity(tcpu); + + default: + break; + } + + return 1; +} + +static int +skip_cluster(struct sched_cluster *cluster, struct cpu_select_env *env) +{ + if (!test_bit(cluster->id, env->candidate_list)) + return 1; + + if (!acceptable_capacity(cluster, env)) { + __clear_bit(cluster->id, env->candidate_list); + return 1; + } + + return 0; +} + +static struct sched_cluster * +select_least_power_cluster(struct cpu_select_env *env) +{ + struct sched_cluster *cluster; + + if (env->rtg) { + env->task_load = scale_load_to_cpu(task_load(env->p), + cluster_first_cpu(env->rtg->preferred_cluster)); + env->sbc_best_cluster_flag |= SBC_FLAG_COLOC_CLUSTER; + return env->rtg->preferred_cluster; + } + + for_each_sched_cluster(cluster) { + if (!skip_cluster(cluster, env)) { + int cpu = cluster_first_cpu(cluster); + + env->task_load = scale_load_to_cpu(task_load(env->p), + cpu); + if (task_load_will_fit(env->p, env->task_load, cpu, + env->boost_type)) + return cluster; + + __set_bit(cluster->id, env->backup_list); + __clear_bit(cluster->id, env->candidate_list); + } + } + + return NULL; +} + +static struct sched_cluster * +next_candidate(const unsigned long *list, int start, int end) +{ + int cluster_id; + + cluster_id = find_next_bit(list, end, start - 1 + 1); + if (cluster_id >= end) + return NULL; + + return sched_cluster[cluster_id]; +} + +static void +update_spare_capacity(struct cluster_cpu_stats *stats, + struct cpu_select_env *env, int cpu, int capacity, + u64 cpu_load) +{ + s64 spare_capacity = sched_ravg_window - cpu_load; + + if (spare_capacity > 0 && + (spare_capacity > stats->highest_spare_capacity || + (spare_capacity == stats->highest_spare_capacity && + ((!env->need_waker_cluster && + capacity > cpu_capacity(stats->best_capacity_cpu)) || + (env->need_waker_cluster && + cpu_rq(cpu)->nr_running < + cpu_rq(stats->best_capacity_cpu)->nr_running))))) { + /* + * If sync waker is the only runnable of CPU, cr_avg of the + * CPU is 0 so we have high chance to place the wakee on the + * waker's CPU which likely causes preemtion of the waker. + * This can lead migration of preempted waker. Place the + * wakee on the real idle CPU when it's possible by checking + * nr_running to avoid such preemption. + */ + stats->highest_spare_capacity = spare_capacity; + stats->best_capacity_cpu = cpu; + } +} + +static inline void find_backup_cluster( +struct cpu_select_env *env, struct cluster_cpu_stats *stats) +{ + struct sched_cluster *next = NULL; + int i; + + while (!bitmap_empty(env->backup_list, num_clusters)) { + next = next_candidate(env->backup_list, 0, num_clusters); + __clear_bit(next->id, env->backup_list); + for_each_cpu_and(i, &env->p->cpus_allowed, &next->cpus) { + trace_sched_cpu_load_wakeup(cpu_rq(i), idle_cpu(i), + sched_irqload(i), power_cost(i, task_load(env->p) + + cpu_cravg_sync(i, env->sync)), 0); + + update_spare_capacity(stats, env, i, next->capacity, + cpu_load_sync(i, env->sync)); + } + env->sbc_best_cluster_flag = SBC_FLAG_BACKUP_CLUSTER; + } +} + +struct sched_cluster * +next_best_cluster(struct sched_cluster *cluster, struct cpu_select_env *env, + struct cluster_cpu_stats *stats) +{ + struct sched_cluster *next = NULL; + + __clear_bit(cluster->id, env->candidate_list); + + if (env->rtg && preferred_cluster(cluster, env->p)) + return NULL; + + do { + if (bitmap_empty(env->candidate_list, num_clusters)) + return NULL; + + next = next_candidate(env->candidate_list, 0, num_clusters); + if (next) { + if (next->min_power_cost > stats->min_cost) { + clear_bit(next->id, env->candidate_list); + next = NULL; + continue; + } + + if (skip_cluster(next, env)) + next = NULL; + } + } while (!next); + + env->task_load = scale_load_to_cpu(task_load(env->p), + cluster_first_cpu(next)); + return next; +} + +#ifdef CONFIG_SCHED_HMP_CSTATE_AWARE +static void __update_cluster_stats(int cpu, struct cluster_cpu_stats *stats, + struct cpu_select_env *env, int cpu_cost) +{ + int cpu_cstate; + int prev_cpu = env->prev_cpu; + + cpu_cstate = cpu_rq(cpu)->cstate; + + if (env->need_idle) { + stats->min_cost = cpu_cost; + if (idle_cpu(cpu)) { + if (cpu_cstate < stats->best_cpu_cstate || + (cpu_cstate == stats->best_cpu_cstate && + cpu == prev_cpu)) { + stats->best_idle_cpu = cpu; + stats->best_cpu_cstate = cpu_cstate; + } + } else { + if (env->cpu_load < stats->min_load || + (env->cpu_load == stats->min_load && + cpu == prev_cpu)) { + stats->least_loaded_cpu = cpu; + stats->min_load = env->cpu_load; + } + } + + return; + } + + if (cpu_cost < stats->min_cost) { + stats->min_cost = cpu_cost; + stats->best_cpu_cstate = cpu_cstate; + stats->best_load = env->cpu_load; + stats->best_cpu = cpu; + env->sbc_best_flag = SBC_FLAG_CPU_COST; + return; + } + + /* CPU cost is the same. Start breaking the tie by C-state */ + + if (cpu_cstate > stats->best_cpu_cstate) + return; + + if (cpu_cstate < stats->best_cpu_cstate) { + stats->best_cpu_cstate = cpu_cstate; + stats->best_load = env->cpu_load; + stats->best_cpu = cpu; + env->sbc_best_flag = SBC_FLAG_COST_CSTATE_TIE_BREAKER; + return; + } + + /* C-state is the same. Use prev CPU to break the tie */ + if (cpu == prev_cpu) { + stats->best_cpu = cpu; + env->sbc_best_flag = SBC_FLAG_COST_CSTATE_PREV_CPU_TIE_BREAKER; + return; + } + + if (stats->best_cpu != prev_cpu && + ((cpu_cstate == 0 && env->cpu_load < stats->best_load) || + (cpu_cstate > 0 && env->cpu_load > stats->best_load))) { + stats->best_load = env->cpu_load; + stats->best_cpu = cpu; + env->sbc_best_flag = SBC_FLAG_CSTATE_LOAD; + } +} +#else /* CONFIG_SCHED_HMP_CSTATE_AWARE */ +static void __update_cluster_stats(int cpu, struct cluster_cpu_stats *stats, + struct cpu_select_env *env, int cpu_cost) +{ + int prev_cpu = env->prev_cpu; + + if (cpu != prev_cpu && cpus_share_cache(prev_cpu, cpu)) { + if (stats->best_sibling_cpu_cost > cpu_cost || + (stats->best_sibling_cpu_cost == cpu_cost && + stats->best_sibling_cpu_load > env->cpu_load)) { + stats->best_sibling_cpu_cost = cpu_cost; + stats->best_sibling_cpu_load = env->cpu_load; + stats->best_sibling_cpu = cpu; + } + } + + if ((cpu_cost < stats->min_cost) || + ((stats->best_cpu != prev_cpu && + stats->min_load > env->cpu_load) || cpu == prev_cpu)) { + if (env->need_idle) { + if (idle_cpu(cpu)) { + stats->min_cost = cpu_cost; + stats->best_idle_cpu = cpu; + } + } else { + stats->min_cost = cpu_cost; + stats->min_load = env->cpu_load; + stats->best_cpu = cpu; + env->sbc_best_flag = SBC_FLAG_MIN_COST; + } + } +} +#endif /* CONFIG_SCHED_HMP_CSTATE_AWARE */ + +static void update_cluster_stats(int cpu, struct cluster_cpu_stats *stats, + struct cpu_select_env *env) +{ + int cpu_cost; + + cpu_cost = power_cost(cpu, task_load(env->p) + + cpu_cravg_sync(cpu, env->sync)); + if (cpu_cost <= stats->min_cost) + __update_cluster_stats(cpu, stats, env, cpu_cost); +} + +static void find_best_cpu_in_cluster(struct sched_cluster *c, + struct cpu_select_env *env, struct cluster_cpu_stats *stats) +{ + int i; + struct cpumask search_cpus; + + cpumask_and(&search_cpus, tsk_cpus_allowed(env->p), &c->cpus); + if (env->ignore_prev_cpu) + cpumask_clear_cpu(env->prev_cpu, &search_cpus); + + for_each_cpu(i, &search_cpus) { + env->cpu_load = cpu_load_sync(i, env->sync); + + trace_sched_cpu_load_wakeup(cpu_rq(i), idle_cpu(i), + sched_irqload(i), + power_cost(i, task_load(env->p) + + cpu_cravg_sync(i, env->sync)), 0); + + if (unlikely(!cpu_active(i)) || skip_cpu(i, env)) + continue; + + update_spare_capacity(stats, env, i, c->capacity, + env->cpu_load); + + if (env->boost_type == SCHED_BOOST_ON_ALL || + env->need_waker_cluster || + sched_cpu_high_irqload(i) || + spill_threshold_crossed(env, cpu_rq(i))) + continue; + + update_cluster_stats(i, stats, env); + } +} + +static inline void init_cluster_cpu_stats(struct cluster_cpu_stats *stats) +{ + stats->best_cpu = stats->best_idle_cpu = -1; + stats->best_capacity_cpu = stats->best_sibling_cpu = -1; + stats->min_cost = stats->best_sibling_cpu_cost = INT_MAX; + stats->min_load = stats->best_sibling_cpu_load = ULLONG_MAX; + stats->highest_spare_capacity = 0; + stats->least_loaded_cpu = -1; + stats->best_cpu_cstate = INT_MAX; + /* No need to initialize stats->best_load */ +} + +/* + * Should task be woken to any available idle cpu? + * + * Waking tasks to idle cpu has mixed implications on both performance and + * power. In many cases, scheduler can't estimate correctly impact of using idle + * cpus on either performance or power. PF_WAKE_UP_IDLE allows external kernel + * module to pass a strong hint to scheduler that the task in question should be + * woken to idle cpu, generally to improve performance. + */ +static inline int wake_to_idle(struct task_struct *p) +{ + return (current->flags & PF_WAKE_UP_IDLE) || + (p->flags & PF_WAKE_UP_IDLE) || sysctl_sched_wake_to_idle; +} + +static inline bool +bias_to_prev_cpu(struct cpu_select_env *env, struct cluster_cpu_stats *stats) +{ + int prev_cpu; + struct task_struct *task = env->p; + struct sched_cluster *cluster; + + if (env->boost_type != SCHED_BOOST_NONE || env->reason || + !task->ravg.mark_start || + env->need_idle || !sched_short_sleep_task_threshold) + return false; + + prev_cpu = env->prev_cpu; + if (!cpumask_test_cpu(prev_cpu, tsk_cpus_allowed(task)) || + unlikely(!cpu_active(prev_cpu))) + return false; + + if (task->ravg.mark_start - task->last_cpu_selected_ts >= + sched_long_cpu_selection_threshold) + return false; + + /* + * This function should be used by task wake up path only as it's + * assuming p->last_switch_out_ts as last sleep time. + * p->last_switch_out_ts can denote last preemption time as well as + * last sleep time. + */ + if (task->ravg.mark_start - task->last_switch_out_ts >= + sched_short_sleep_task_threshold) + return false; + + env->task_load = scale_load_to_cpu(task_load(task), prev_cpu); + cluster = cpu_rq(prev_cpu)->cluster; + + if (!task_load_will_fit(task, env->task_load, prev_cpu, + sched_boost_type())) { + + __set_bit(cluster->id, env->backup_list); + __clear_bit(cluster->id, env->candidate_list); + return false; + } + + env->cpu_load = cpu_load_sync(prev_cpu, env->sync); + if (sched_cpu_high_irqload(prev_cpu) || + spill_threshold_crossed(env, cpu_rq(prev_cpu))) { + update_spare_capacity(stats, env, prev_cpu, + cluster->capacity, env->cpu_load); + env->ignore_prev_cpu = 1; + return false; + } + + return true; +} + +static inline bool +wake_to_waker_cluster(struct cpu_select_env *env) +{ + return !env->need_idle && !env->reason && env->sync && + task_load(current) > sched_big_waker_task_load && + task_load(env->p) < sched_small_wakee_task_load; +} + +static inline int +cluster_allowed(struct task_struct *p, struct sched_cluster *cluster) +{ + cpumask_t tmp_mask; + + cpumask_and(&tmp_mask, &cluster->cpus, cpu_active_mask); + cpumask_and(&tmp_mask, &tmp_mask, &p->cpus_allowed); + + return !cpumask_empty(&tmp_mask); +} + + +/* return cheapest cpu that can fit this task */ +static int select_best_cpu(struct task_struct *p, int target, int reason, + int sync) +{ + struct sched_cluster *cluster, *pref_cluster = NULL; + struct cluster_cpu_stats stats; + struct related_thread_group *grp; + unsigned int sbc_flag = 0; + + struct cpu_select_env env = { + .p = p, + .reason = reason, + .need_idle = wake_to_idle(p), + .need_waker_cluster = 0, + .boost_type = sched_boost_type(), + .sync = sync, + .prev_cpu = target, + .ignore_prev_cpu = 0, + .rtg = NULL, + .sbc_best_flag = 0, + .sbc_best_cluster_flag = 0, + }; + + bitmap_copy(env.candidate_list, all_cluster_ids, NR_CPUS); + bitmap_zero(env.backup_list, NR_CPUS); + + init_cluster_cpu_stats(&stats); + + rcu_read_lock(); + + grp = task_related_thread_group(p); + + if (grp && grp->preferred_cluster) { + pref_cluster = grp->preferred_cluster; + if (!cluster_allowed(p, pref_cluster)) + clear_bit(pref_cluster->id, env.candidate_list); + else + env.rtg = grp; + } else { + cluster = cpu_rq(smp_processor_id())->cluster; + if (wake_to_waker_cluster(&env) && + cluster_allowed(p, cluster)) { + env.need_waker_cluster = 1; + bitmap_zero(env.candidate_list, NR_CPUS); + __set_bit(cluster->id, env.candidate_list); + env.sbc_best_cluster_flag = SBC_FLAG_WAKER_CLUSTER; + + } else if (bias_to_prev_cpu(&env, &stats)) { + sbc_flag = SBC_FLAG_PREV_CPU; + goto out; + } + } + +retry: + cluster = select_least_power_cluster(&env); + + if (!cluster) + goto out; + + /* + * 'cluster' now points to the minimum power cluster which can satisfy + * task's perf goals. Walk down the cluster list starting with that + * cluster. For non-small tasks, skip clusters that don't have + * mostly_idle/idle cpus + */ + + do { + find_best_cpu_in_cluster(cluster, &env, &stats); + + } while ((cluster = next_best_cluster(cluster, &env, &stats))); + + if (env.need_idle) { + if (stats.best_idle_cpu >= 0) { + target = stats.best_idle_cpu; + sbc_flag |= SBC_FLAG_IDLE_CSTATE; + } else if (stats.least_loaded_cpu >= 0) { + target = stats.least_loaded_cpu; + sbc_flag |= SBC_FLAG_IDLE_LEAST_LOADED; + } + } else if (stats.best_cpu >= 0) { + if (stats.best_cpu != task_cpu(p) && + stats.min_cost == stats.best_sibling_cpu_cost) { + stats.best_cpu = stats.best_sibling_cpu; + sbc_flag |= SBC_FLAG_BEST_SIBLING; + } + sbc_flag |= env.sbc_best_flag; + target = stats.best_cpu; + } else { + if (env.rtg) { + env.rtg = NULL; + goto retry; + } + + find_backup_cluster(&env, &stats); + if (stats.best_capacity_cpu >= 0) { + target = stats.best_capacity_cpu; + sbc_flag |= SBC_FLAG_BEST_CAP_CPU; + } + } + p->last_cpu_selected_ts = sched_ktime_clock(); + sbc_flag |= env.sbc_best_cluster_flag; +out: + rcu_read_unlock(); + trace_sched_task_load(p, sched_boost(), env.reason, env.sync, + env.need_idle, sbc_flag, target); + return target; +} + +#ifdef CONFIG_CFS_BANDWIDTH + +static inline struct task_group *next_task_group(struct task_group *tg) +{ + tg = list_entry_rcu(tg->list.next, typeof(struct task_group), list); + + return (&tg->list == &task_groups) ? NULL : tg; +} + +/* Iterate over all cfs_rq in a cpu */ +#define for_each_cfs_rq(cfs_rq, tg, cpu) \ + for (tg = container_of(&task_groups, struct task_group, list); \ + ((tg = next_task_group(tg)) && (cfs_rq = tg->cfs_rq[cpu]));) + +void reset_cfs_rq_hmp_stats(int cpu, int reset_cra) +{ + struct task_group *tg; + struct cfs_rq *cfs_rq; + + rcu_read_lock(); + + for_each_cfs_rq(cfs_rq, tg, cpu) + reset_hmp_stats(&cfs_rq->hmp_stats, reset_cra); + + rcu_read_unlock(); +} + +static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq); + +static void inc_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra); +static void dec_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra); + +/* Add task's contribution to a cpu' HMP statistics */ +void inc_hmp_sched_stats_fair(struct rq *rq, + struct task_struct *p, int change_cra) +{ + struct cfs_rq *cfs_rq; + struct sched_entity *se = &p->se; + + /* + * Although below check is not strictly required (as + * inc/dec_nr_big_task and inc/dec_cumulative_runnable_avg called + * from inc_cfs_rq_hmp_stats() have similar checks), we gain a bit on + * efficiency by short-circuiting for_each_sched_entity() loop when + * sched_disable_window_stats + */ + if (sched_disable_window_stats) + return; + + for_each_sched_entity(se) { + cfs_rq = cfs_rq_of(se); + inc_cfs_rq_hmp_stats(cfs_rq, p, change_cra); + if (cfs_rq_throttled(cfs_rq)) + break; + } + + /* Update rq->hmp_stats only if we didn't find any throttled cfs_rq */ + if (!se) + inc_rq_hmp_stats(rq, p, change_cra); +} + +static void fixup_hmp_sched_stats_fair(struct rq *rq, struct task_struct *p, + u32 new_task_load, u32 new_pred_demand) +{ + struct cfs_rq *cfs_rq; + struct sched_entity *se = &p->se; + s64 task_load_delta = (s64)new_task_load - task_load(p); + s64 pred_demand_delta = PRED_DEMAND_DELTA; + + for_each_sched_entity(se) { + cfs_rq = cfs_rq_of(se); + + fixup_cumulative_runnable_avg(&cfs_rq->hmp_stats, p, + task_load_delta, + pred_demand_delta); + fixup_nr_big_tasks(&cfs_rq->hmp_stats, p, task_load_delta); + if (cfs_rq_throttled(cfs_rq)) + break; + } + + /* Fix up rq->hmp_stats only if we didn't find any throttled cfs_rq */ + if (!se) { + fixup_cumulative_runnable_avg(&rq->hmp_stats, p, + task_load_delta, + pred_demand_delta); + fixup_nr_big_tasks(&rq->hmp_stats, p, task_load_delta); + } +} + +static int task_will_be_throttled(struct task_struct *p); + +#else /* CONFIG_CFS_BANDWIDTH */ + +inline void reset_cfs_rq_hmp_stats(int cpu, int reset_cra) { } + +static void +fixup_hmp_sched_stats_fair(struct rq *rq, struct task_struct *p, + u32 new_task_load, u32 new_pred_demand) +{ + s64 task_load_delta = (s64)new_task_load - task_load(p); + s64 pred_demand_delta = PRED_DEMAND_DELTA; + + fixup_cumulative_runnable_avg(&rq->hmp_stats, p, task_load_delta, + pred_demand_delta); + fixup_nr_big_tasks(&rq->hmp_stats, p, task_load_delta); +} + +static inline int task_will_be_throttled(struct task_struct *p) +{ + return 0; +} + +void inc_hmp_sched_stats_fair(struct rq *rq, + struct task_struct *p, int change_cra) +{ + inc_nr_big_task(&rq->hmp_stats, p); +} + +#endif /* CONFIG_CFS_BANDWIDTH */ + +/* + * Reset balance_interval at all sched_domain levels of given cpu, so that it + * honors kick. + */ +static inline void reset_balance_interval(int cpu) +{ + struct sched_domain *sd; + + if (cpu >= nr_cpu_ids) + return; + + rcu_read_lock(); + for_each_domain(cpu, sd) + sd->balance_interval = 0; + rcu_read_unlock(); +} + +/* + * Check if a task is on the "wrong" cpu (i.e its current cpu is not the ideal + * cpu as per its demand or priority) + * + * Returns reason why task needs to be migrated + */ +static inline int migration_needed(struct task_struct *p, int cpu) +{ + int nice; + struct related_thread_group *grp; + + if (p->state != TASK_RUNNING || p->nr_cpus_allowed == 1) + return 0; + + /* No need to migrate task that is about to be throttled */ + if (task_will_be_throttled(p)) + return 0; + + if (sched_boost_type() == SCHED_BOOST_ON_BIG) { + if (cpu_capacity(cpu) != max_capacity) + return UP_MIGRATION; + return 0; + } + + if (sched_cpu_high_irqload(cpu)) + return IRQLOAD_MIGRATION; + + nice = task_nice(p); + rcu_read_lock(); + grp = task_related_thread_group(p); + if (!grp && (nice > SCHED_UPMIGRATE_MIN_NICE || + upmigrate_discouraged(p)) && cpu_capacity(cpu) > min_capacity) { + rcu_read_unlock(); + return DOWN_MIGRATION; + } + + if (!grp && !task_will_fit(p, cpu)) { + rcu_read_unlock(); + return UP_MIGRATION; + } + rcu_read_unlock(); + + return 0; +} + +static inline int +kick_active_balance(struct rq *rq, struct task_struct *p, int new_cpu) +{ + unsigned long flags; + int rc = 0; + + /* Invoke active balance to force migrate currently running task */ + raw_spin_lock_irqsave(&rq->lock, flags); + if (!rq->active_balance) { + rq->active_balance = 1; + rq->push_cpu = new_cpu; + get_task_struct(p); + rq->push_task = p; + rc = 1; + } + raw_spin_unlock_irqrestore(&rq->lock, flags); + + return rc; +} + +static DEFINE_RAW_SPINLOCK(migration_lock); + +/* + * Check if currently running task should be migrated to a better cpu. + * + * Todo: Effect this via changes to nohz_balancer_kick() and load balance? + */ +void check_for_migration(struct rq *rq, struct task_struct *p) +{ + int cpu = cpu_of(rq), new_cpu; + int active_balance = 0, reason; + + reason = migration_needed(p, cpu); + if (!reason) + return; + + raw_spin_lock(&migration_lock); + new_cpu = select_best_cpu(p, cpu, reason, 0); + + if (new_cpu != cpu) { + active_balance = kick_active_balance(rq, p, new_cpu); + if (active_balance) + mark_reserved(new_cpu); + } + + raw_spin_unlock(&migration_lock); + + if (active_balance) + stop_one_cpu_nowait(cpu, active_load_balance_cpu_stop, rq, + &rq->active_balance_work); +} + +#ifdef CONFIG_CFS_BANDWIDTH + +static void init_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq) +{ + cfs_rq->hmp_stats.nr_big_tasks = 0; + cfs_rq->hmp_stats.cumulative_runnable_avg = 0; + cfs_rq->hmp_stats.pred_demands_sum = 0; +} + +static void inc_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) +{ + inc_nr_big_task(&cfs_rq->hmp_stats, p); + if (change_cra) + inc_cumulative_runnable_avg(&cfs_rq->hmp_stats, p); +} + +static void dec_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) +{ + dec_nr_big_task(&cfs_rq->hmp_stats, p); + if (change_cra) + dec_cumulative_runnable_avg(&cfs_rq->hmp_stats, p); +} + +static void inc_throttled_cfs_rq_hmp_stats(struct hmp_sched_stats *stats, + struct cfs_rq *cfs_rq) +{ + stats->nr_big_tasks += cfs_rq->hmp_stats.nr_big_tasks; + stats->cumulative_runnable_avg += + cfs_rq->hmp_stats.cumulative_runnable_avg; + stats->pred_demands_sum += cfs_rq->hmp_stats.pred_demands_sum; +} + +static void dec_throttled_cfs_rq_hmp_stats(struct hmp_sched_stats *stats, + struct cfs_rq *cfs_rq) +{ + stats->nr_big_tasks -= cfs_rq->hmp_stats.nr_big_tasks; + stats->cumulative_runnable_avg -= + cfs_rq->hmp_stats.cumulative_runnable_avg; + stats->pred_demands_sum -= cfs_rq->hmp_stats.pred_demands_sum; + + BUG_ON(stats->nr_big_tasks < 0 || + (s64)stats->cumulative_runnable_avg < 0); + verify_pred_demands_sum(stats); +} + +#else /* CONFIG_CFS_BANDWIDTH */ + +static inline void inc_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) { } + +static inline void dec_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) { } + +#endif /* CONFIG_CFS_BANDWIDTH */ + +#else /* CONFIG_SCHED_HMP */ + +static inline void init_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq) { } + +static inline void inc_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) { } + +static inline void dec_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) { } + +static inline void inc_throttled_cfs_rq_hmp_stats(struct hmp_sched_stats *stats, + struct cfs_rq *cfs_rq) +{ +} + +static inline void dec_throttled_cfs_rq_hmp_stats(struct hmp_sched_stats *stats, + struct cfs_rq *cfs_rq) +{ +} +#endif /* CONFIG_SCHED_HMP */ + #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) /* @@ -3273,6 +4216,12 @@ static inline int idle_balance(struct rq *rq) return 0; } +static inline void inc_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) { } + +static inline void dec_cfs_rq_hmp_stats(struct cfs_rq *cfs_rq, + struct task_struct *p, int change_cra) { } + #endif /* CONFIG_SMP */ static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) @@ -3884,6 +4833,33 @@ static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq) return cfs_bandwidth_used() && cfs_rq->throttled; } +/* + * Check if task is part of a hierarchy where some cfs_rq does not have any + * runtime left. + * + * We can't rely on throttled_hierarchy() to do this test, as + * cfs_rq->throttle_count will not be updated yet when this function is called + * from scheduler_tick() + */ +static int task_will_be_throttled(struct task_struct *p) +{ + struct sched_entity *se = &p->se; + struct cfs_rq *cfs_rq; + + if (!cfs_bandwidth_used()) + return 0; + + for_each_sched_entity(se) { + cfs_rq = cfs_rq_of(se); + if (!cfs_rq->runtime_enabled) + continue; + if (cfs_rq->runtime_remaining <= 0) + return 1; + } + + return 0; +} + /* check whether cfs_rq, or any parent, is throttled */ static inline int throttled_hierarchy(struct cfs_rq *cfs_rq) { @@ -3961,13 +4937,20 @@ static void throttle_cfs_rq(struct cfs_rq *cfs_rq) if (dequeue) dequeue_entity(qcfs_rq, se, DEQUEUE_SLEEP); qcfs_rq->h_nr_running -= task_delta; +#ifdef CONFIG_SCHED_HMP + dec_throttled_cfs_rq_hmp_stats(&qcfs_rq->hmp_stats, cfs_rq); +#endif if (qcfs_rq->load.weight) dequeue = 0; } - if (!se) + if (!se) { sub_nr_running(rq, task_delta); +#ifdef CONFIG_SCHED_HMP + dec_throttled_cfs_rq_hmp_stats(&rq->hmp_stats, cfs_rq); +#endif + } cfs_rq->throttled = 1; cfs_rq->throttled_clock = rq_clock(rq); @@ -3988,6 +4971,12 @@ static void throttle_cfs_rq(struct cfs_rq *cfs_rq) start_cfs_bandwidth(cfs_b); raw_spin_unlock(&cfs_b->lock); + + /* Log effect on hmp stats after throttling */ + trace_sched_cpu_load_cgroup(rq, idle_cpu(cpu_of(rq)), + sched_irqload(cpu_of(rq)), + power_cost(cpu_of(rq), 0), + cpu_temp(cpu_of(rq))); } void unthrottle_cfs_rq(struct cfs_rq *cfs_rq) @@ -3997,6 +4986,7 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq) struct sched_entity *se; int enqueue = 1; long task_delta; + struct cfs_rq *tcfs_rq = cfs_rq; se = cfs_rq->tg->se[cpu_of(rq)]; @@ -4024,17 +5014,30 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq) if (enqueue) enqueue_entity(cfs_rq, se, ENQUEUE_WAKEUP); cfs_rq->h_nr_running += task_delta; +#ifdef CONFIG_SCHED_HMP + inc_throttled_cfs_rq_hmp_stats(&cfs_rq->hmp_stats, tcfs_rq); +#endif if (cfs_rq_throttled(cfs_rq)) break; } - if (!se) + if (!se) { add_nr_running(rq, task_delta); +#ifdef CONFIG_SCHED_HMP + inc_throttled_cfs_rq_hmp_stats(&rq->hmp_stats, tcfs_rq); +#endif + } /* determine whether we need to wake up potentially idle cpu */ if (rq->curr == rq->idle && rq->cfs.nr_running) resched_curr(rq); + + /* Log effect on hmp stats after un-throttling */ + trace_sched_cpu_load_cgroup(rq, idle_cpu(cpu_of(rq)), + sched_irqload(cpu_of(rq)), + power_cost(cpu_of(rq), 0), + cpu_temp(cpu_of(rq))); } static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b, @@ -4372,6 +5375,7 @@ static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) { cfs_rq->runtime_enabled = 0; INIT_LIST_HEAD(&cfs_rq->throttled_list); + init_cfs_rq_hmp_stats(cfs_rq); } void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b) @@ -4562,6 +5566,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags) if (cfs_rq_throttled(cfs_rq)) break; cfs_rq->h_nr_running++; + inc_cfs_rq_hmp_stats(cfs_rq, p, 1); flags = ENQUEUE_WAKEUP; } @@ -4569,6 +5574,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags) for_each_sched_entity(se) { cfs_rq = cfs_rq_of(se); cfs_rq->h_nr_running++; + inc_cfs_rq_hmp_stats(cfs_rq, p, 1); if (cfs_rq_throttled(cfs_rq)) break; @@ -4577,8 +5583,10 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags) update_cfs_shares(cfs_rq); } - if (!se) + if (!se) { add_nr_running(rq, 1); + inc_rq_hmp_stats(rq, p, 1); + } hrtick_update(rq); } @@ -4609,6 +5617,7 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) if (cfs_rq_throttled(cfs_rq)) break; cfs_rq->h_nr_running--; + dec_cfs_rq_hmp_stats(cfs_rq, p, 1); /* Don't dequeue parent if it has other entities besides us */ if (cfs_rq->load.weight) { @@ -4628,6 +5637,7 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) for_each_sched_entity(se) { cfs_rq = cfs_rq_of(se); cfs_rq->h_nr_running--; + dec_cfs_rq_hmp_stats(cfs_rq, p, 1); if (cfs_rq_throttled(cfs_rq)) break; @@ -4636,8 +5646,10 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) update_cfs_shares(cfs_rq); } - if (!se) + if (!se) { sub_nr_running(rq, 1); + dec_rq_hmp_stats(rq, p, 1); + } hrtick_update(rq); } @@ -5632,6 +6644,10 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_f int want_affine = 0; int sync = wake_flags & WF_SYNC; +#ifdef CONFIG_SCHED_HMP + return select_best_cpu(p, prev_cpu, 0, sync); +#endif + if (sd_flag & SD_BALANCE_WAKE) { record_wakee(p); want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu) @@ -6247,6 +7263,13 @@ enum fbq_type { regular, remote, all }; #define LBF_NEED_BREAK 0x02 #define LBF_DST_PINNED 0x04 #define LBF_SOME_PINNED 0x08 +#define LBF_SCHED_BOOST_ACTIVE_BALANCE 0x40 +#define LBF_BIG_TASK_ACTIVE_BALANCE 0x80 +#define LBF_HMP_ACTIVE_BALANCE (LBF_SCHED_BOOST_ACTIVE_BALANCE | \ + LBF_BIG_TASK_ACTIVE_BALANCE) +#define LBF_IGNORE_BIG_TASKS 0x100 +#define LBF_IGNORE_PREFERRED_CLUSTER_TASKS 0x200 +#define LBF_MOVED_RELATED_THREAD_GROUP_TASK 0x400 struct lb_env { struct sched_domain *sd; @@ -6263,6 +7286,8 @@ struct lb_env { long imbalance; /* The set of CPUs under consideration for load-balancing */ struct cpumask *cpus; + unsigned int busiest_grp_capacity; + unsigned int busiest_nr_running; unsigned int flags; @@ -6369,6 +7394,7 @@ static int can_migrate_task(struct task_struct *p, struct lb_env *env) { int tsk_cache_hot; + int twf, group_cpus; lockdep_assert_held(&env->src_rq->lock); @@ -6415,6 +7441,34 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env) /* Record that we found atleast one task that could run on dst_cpu */ env->flags &= ~LBF_ALL_PINNED; + if (cpu_capacity(env->dst_cpu) > cpu_capacity(env->src_cpu) && + nr_big_tasks(env->src_rq) && !is_big_task(p)) + return 0; + + twf = task_will_fit(p, env->dst_cpu); + + /* + * Attempt to not pull tasks that don't fit. We may get lucky and find + * one that actually fits. + */ + if (env->flags & LBF_IGNORE_BIG_TASKS && !twf) + return 0; + + if (env->flags & LBF_IGNORE_PREFERRED_CLUSTER_TASKS && + !preferred_cluster(rq_cluster(cpu_rq(env->dst_cpu)), p)) + return 0; + + /* + * Group imbalance can sometimes cause work to be pulled across groups + * even though the group could have managed the imbalance on its own. + * Prevent inter-cluster migrations for big tasks when the number of + * tasks is lower than the capacity of the group. + */ + group_cpus = DIV_ROUND_UP(env->busiest_grp_capacity, + SCHED_CAPACITY_SCALE); + if (!twf && env->busiest_nr_running <= group_cpus) + return 0; + if (task_running(env->src_rq, p)) { schedstat_inc(p->se.statistics.nr_failed_migrations_running); return 0; @@ -6422,15 +7476,16 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env) /* * Aggressive migration if: - * 1) destination numa is preferred - * 2) task is cache cold, or - * 3) too many balance attempts have failed. + * 1) IDLE or NEWLY_IDLE balance. + * 2) destination numa is preferred + * 3) task is cache cold, or + * 4) too many balance attempts have failed. */ tsk_cache_hot = migrate_degrades_locality(p, env); if (tsk_cache_hot == -1) tsk_cache_hot = task_hot(p, env); - if (tsk_cache_hot <= 0 || + if (env->idle != CPU_NOT_IDLE || tsk_cache_hot <= 0 || env->sd->nr_balance_failed > env->sd->cache_nice_tries) { if (tsk_cache_hot == 1) { schedstat_inc(env->sd->lb_hot_gained[env->idle]); @@ -6452,7 +7507,11 @@ static void detach_task(struct task_struct *p, struct lb_env *env) p->on_rq = TASK_ON_RQ_MIGRATING; deactivate_task(env->src_rq, p, 0); + double_lock_balance(env->src_rq, env->dst_rq); set_task_cpu(p, env->dst_cpu); + if (task_in_related_thread_group(p)) + env->flags |= LBF_MOVED_RELATED_THREAD_GROUP_TASK; + double_unlock_balance(env->src_rq, env->dst_rq); } /* @@ -6499,12 +7558,20 @@ static int detach_tasks(struct lb_env *env) struct task_struct *p; unsigned long load; int detached = 0; + int orig_loop = env->loop; lockdep_assert_held(&env->src_rq->lock); if (env->imbalance <= 0) return 0; + if (cpu_capacity(env->dst_cpu) < cpu_capacity(env->src_cpu) && + !sched_boost()) + env->flags |= LBF_IGNORE_BIG_TASKS; + else if (!same_cluster(env->dst_cpu, env->src_cpu)) + env->flags |= LBF_IGNORE_PREFERRED_CLUSTER_TASKS; + +redo: while (!list_empty(tasks)) { /* * We don't want to steal all, otherwise we may be treated likewise, @@ -6566,6 +7633,15 @@ next: list_move_tail(&p->se.group_node, tasks); } + if (env->flags & (LBF_IGNORE_BIG_TASKS | + LBF_IGNORE_PREFERRED_CLUSTER_TASKS) && !detached) { + tasks = &env->src_rq->cfs_tasks; + env->flags &= ~(LBF_IGNORE_BIG_TASKS | + LBF_IGNORE_PREFERRED_CLUSTER_TASKS); + env->loop = orig_loop; + goto redo; + } + /* * Right now, this is one of only two places we collect this stat * so we can safely collect detach_one_task() stats here rather @@ -6730,6 +7806,10 @@ struct sg_lb_stats { unsigned long group_capacity; unsigned long group_util; /* Total utilization of the group */ unsigned int sum_nr_running; /* Nr tasks running in the group */ +#ifdef CONFIG_SCHED_HMP + unsigned long sum_nr_big_tasks; + u64 group_cpu_load; /* Scaled load of all CPUs of the group */ +#endif unsigned int idle_cpus; unsigned int group_weight; enum group_type group_type; @@ -6772,10 +7852,64 @@ static inline void init_sd_lb_stats(struct sd_lb_stats *sds) .avg_load = 0UL, .sum_nr_running = 0, .group_type = group_other, +#ifdef CONFIG_SCHED_HMP + .sum_nr_big_tasks = 0UL, + .group_cpu_load = 0ULL, +#endif }, }; } +#ifdef CONFIG_SCHED_HMP + +static int +bail_inter_cluster_balance(struct lb_env *env, struct sd_lb_stats *sds) +{ + int local_cpu, busiest_cpu; + int local_capacity, busiest_capacity; + int local_pwr_cost, busiest_pwr_cost; + int nr_cpus; + + if (!sysctl_sched_restrict_cluster_spill || sched_boost()) + return 0; + + local_cpu = group_first_cpu(sds->local); + busiest_cpu = group_first_cpu(sds->busiest); + + local_capacity = cpu_max_possible_capacity(local_cpu); + busiest_capacity = cpu_max_possible_capacity(busiest_cpu); + + local_pwr_cost = cpu_max_power_cost(local_cpu); + busiest_pwr_cost = cpu_max_power_cost(busiest_cpu); + + if (local_capacity < busiest_capacity || + (local_capacity == busiest_capacity && + local_pwr_cost <= busiest_pwr_cost)) + return 0; + + if (local_capacity > busiest_capacity && + sds->busiest_stat.sum_nr_big_tasks) + return 0; + + nr_cpus = cpumask_weight(sched_group_cpus(sds->busiest)); + if ((sds->busiest_stat.group_cpu_load < nr_cpus * sched_spill_load) && + (sds->busiest_stat.sum_nr_running < + nr_cpus * sysctl_sched_spill_nr_run)) + return 1; + + return 0; +} + +#else /* CONFIG_SCHED_HMP */ + +static inline int +bail_inter_cluster_balance(struct lb_env *env, struct sd_lb_stats *sds) +{ + return 0; +} + +#endif /* CONFIG_SCHED_HMP */ + /** * get_sd_load_idx - Obtain the load index for a given sched domain. * @sd: The sched_domain whose load_idx is to be obtained. @@ -7038,6 +8172,11 @@ static inline void update_sg_lb_stats(struct lb_env *env, for_each_cpu_and(i, sched_group_cpus(group), env->cpus) { struct rq *rq = cpu_rq(i); + trace_sched_cpu_load_lb(cpu_rq(i), idle_cpu(i), + sched_irqload(i), + power_cost(i, 0), + cpu_temp(i)); + /* Bias balancing toward cpus of our domain */ if (local_group) load = target_load(i, load_idx); @@ -7052,6 +8191,11 @@ static inline void update_sg_lb_stats(struct lb_env *env, if (nr_running > 1) *overload = true; +#ifdef CONFIG_SCHED_HMP + sgs->sum_nr_big_tasks += rq->hmp_stats.nr_big_tasks; + sgs->group_cpu_load += cpu_load(i); +#endif + #ifdef CONFIG_NUMA_BALANCING sgs->nr_numa_running += rq->nr_numa_running; sgs->nr_preferred_running += rq->nr_preferred_running; @@ -7077,6 +8221,38 @@ static inline void update_sg_lb_stats(struct lb_env *env, sgs->group_type = group_classify(group, sgs); } +#ifdef CONFIG_SCHED_HMP +static bool update_sd_pick_busiest_active_balance(struct lb_env *env, + struct sd_lb_stats *sds, + struct sched_group *sg, + struct sg_lb_stats *sgs) +{ + if (env->idle != CPU_NOT_IDLE && + cpu_capacity(env->dst_cpu) > group_rq_capacity(sg)) { + if (sched_boost() && !sds->busiest && sgs->sum_nr_running) { + env->flags |= LBF_SCHED_BOOST_ACTIVE_BALANCE; + return true; + } + + if (sgs->sum_nr_big_tasks > + sds->busiest_stat.sum_nr_big_tasks) { + env->flags |= LBF_BIG_TASK_ACTIVE_BALANCE; + return true; + } + } + + return false; +} +#else +static bool update_sd_pick_busiest_active_balance(struct lb_env *env, + struct sd_lb_stats *sds, + struct sched_group *sg, + struct sg_lb_stats *sgs) +{ + return false; +} +#endif + /** * update_sd_pick_busiest - return 1 on busiest group * @env: The load balancing environment. @@ -7097,6 +8273,9 @@ static bool update_sd_pick_busiest(struct lb_env *env, { struct sg_lb_stats *busiest = &sds->busiest_stat; + if (update_sd_pick_busiest_active_balance(env, sds, sg, sgs)) + return true; + if (sgs->group_type > busiest->group_type) return true; @@ -7218,6 +8397,8 @@ static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sd if (update_sd_pick_busiest(env, sds, sg, sgs)) { sds->busiest = sg; sds->busiest_stat = *sgs; + env->busiest_nr_running = sgs->sum_nr_running; + env->busiest_grp_capacity = sgs->group_capacity; } next_group: @@ -7466,6 +8647,12 @@ static struct sched_group *find_busiest_group(struct lb_env *env) if (!sds.busiest || busiest->sum_nr_running == 0) goto out_balanced; + if (env->flags & LBF_HMP_ACTIVE_BALANCE) + goto force_balance; + + if (bail_inter_cluster_balance(env, &sds)) + goto out_balanced; + sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load) / sds.total_capacity; @@ -7527,6 +8714,57 @@ out_balanced: return NULL; } +#ifdef CONFIG_SCHED_HMP +static struct rq *find_busiest_queue_hmp(struct lb_env *env, + struct sched_group *group) +{ + struct rq *busiest = NULL, *busiest_big = NULL; + u64 max_runnable_avg = 0, max_runnable_avg_big = 0; + int max_nr_big = 0, nr_big; + bool find_big = !!(env->flags & LBF_BIG_TASK_ACTIVE_BALANCE); + int i; + + for_each_cpu(i, sched_group_cpus(group)) { + struct rq *rq = cpu_rq(i); + u64 cumulative_runnable_avg = + rq->hmp_stats.cumulative_runnable_avg; + + if (!cpumask_test_cpu(i, env->cpus)) + continue; + + + if (find_big) { + nr_big = nr_big_tasks(rq); + if (nr_big > max_nr_big || + (nr_big > 0 && nr_big == max_nr_big && + cumulative_runnable_avg > max_runnable_avg_big)) { + max_runnable_avg_big = cumulative_runnable_avg; + busiest_big = rq; + max_nr_big = nr_big; + continue; + } + } + + if (cumulative_runnable_avg > max_runnable_avg) { + max_runnable_avg = cumulative_runnable_avg; + busiest = rq; + } + } + + if (busiest_big) + return busiest_big; + + env->flags &= ~LBF_BIG_TASK_ACTIVE_BALANCE; + return busiest; +} +#else +static inline struct rq *find_busiest_queue_hmp(struct lb_env *env, + struct sched_group *group) +{ + return NULL; +} +#endif + /* * find_busiest_queue - find the busiest runqueue among the cpus in group. */ @@ -7537,6 +8775,10 @@ static struct rq *find_busiest_queue(struct lb_env *env, unsigned long busiest_load = 0, busiest_capacity = 1; int i; +#ifdef CONFIG_SCHED_HMP + return find_busiest_queue_hmp(env, group); +#endif + for_each_cpu_and(i, sched_group_cpus(group), env->cpus) { unsigned long capacity, wl; enum fbq_type rt; @@ -7604,12 +8846,16 @@ static struct rq *find_busiest_queue(struct lb_env *env, * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but * so long as it is large enough. */ -#define MAX_PINNED_INTERVAL 512 +#define MAX_PINNED_INTERVAL 16 +#define NEED_ACTIVE_BALANCE_THRESHOLD 10 static int need_active_balance(struct lb_env *env) { struct sched_domain *sd = env->sd; + if (env->flags & LBF_HMP_ACTIVE_BALANCE) + return 1; + if (env->idle == CPU_NEWLY_IDLE) { /* @@ -7634,7 +8880,8 @@ static int need_active_balance(struct lb_env *env) return 1; } - return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2); + return unlikely(sd->nr_balance_failed > + sd->cache_nice_tries + NEED_ACTIVE_BALANCE_THRESHOLD); } static int active_load_balance_cpu_stop(void *data); @@ -7681,10 +8928,10 @@ static int load_balance(int this_cpu, struct rq *this_rq, struct sched_domain *sd, enum cpu_idle_type idle, int *continue_balancing) { - int ld_moved, cur_ld_moved, active_balance = 0; + int ld_moved = 0, cur_ld_moved, active_balance = 0; struct sched_domain *sd_parent = sd->parent; - struct sched_group *group; - struct rq *busiest; + struct sched_group *group = NULL; + struct rq *busiest = NULL; unsigned long flags; struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask); @@ -7698,6 +8945,11 @@ static int load_balance(int this_cpu, struct rq *this_rq, .cpus = cpus, .fbq_type = all, .tasks = LIST_HEAD_INIT(env.tasks), + .imbalance = 0, + .flags = 0, + .loop = 0, + .busiest_nr_running = 0, + .busiest_grp_capacity = 0, }; /* @@ -7750,6 +9002,13 @@ redo: more_balance: raw_spin_lock_irqsave(&busiest->lock, flags); + /* The world might have changed. Validate assumptions */ + if (busiest->nr_running <= 1) { + raw_spin_unlock_irqrestore(&busiest->lock, flags); + env.flags &= ~LBF_ALL_PINNED; + goto no_move; + } + /* * cur_ld_moved - load moved in current iteration * ld_moved - cumulative load moved across iterations @@ -7837,15 +9096,18 @@ more_balance: } } +no_move: if (!ld_moved) { - schedstat_inc(sd->lb_failed[idle]); + if (!(env.flags & LBF_HMP_ACTIVE_BALANCE)) + schedstat_inc(sd->lb_failed[idle]); /* * Increment the failure counter only on periodic balance. * We do not want newidle balance, which can be very * frequent, pollute the failure counter causing * excessive cache_hot migrations and active balances. */ - if (idle != CPU_NEWLY_IDLE) + if (idle != CPU_NEWLY_IDLE && + !(env.flags & LBF_HMP_ACTIVE_BALANCE)) sd->nr_balance_failed++; if (need_active_balance(&env)) { @@ -7879,14 +9141,28 @@ more_balance: stop_one_cpu_nowait(cpu_of(busiest), active_load_balance_cpu_stop, busiest, &busiest->active_balance_work); + *continue_balancing = 0; } /* We've kicked active balancing, force task migration. */ - sd->nr_balance_failed = sd->cache_nice_tries+1; + sd->nr_balance_failed = sd->cache_nice_tries + + NEED_ACTIVE_BALANCE_THRESHOLD - 1; } - } else + } else { sd->nr_balance_failed = 0; + /* Assumes one 'busiest' cpu that we pulled tasks from */ + if (!same_freq_domain(this_cpu, cpu_of(busiest))) { + int check_groups = !!(env.flags & + LBF_MOVED_RELATED_THREAD_GROUP_TASK); + + check_for_freq_change(this_rq, false, check_groups); + check_for_freq_change(busiest, false, check_groups); + } else { + check_for_freq_change(this_rq, true, false); + } + } + if (likely(!active_balance)) { /* We were unbalanced, so reset the balancing interval */ sd->balance_interval = sd->min_interval; @@ -7934,6 +9210,11 @@ out_one_pinned: ld_moved = 0; out: + trace_sched_load_balance(this_cpu, idle, *continue_balancing, + group ? group->cpumask[0] : 0, + busiest ? busiest->nr_running : 0, + env.imbalance, env.flags, ld_moved, + sd->balance_interval); return ld_moved; } @@ -8028,9 +9309,12 @@ static int idle_balance(struct rq *this_rq) /* * Stop searching for tasks to pull if there are - * now runnable tasks on this rq. + * now runnable tasks on the balance rq or if + * continue_balancing has been unset (only possible + * due to active migration). */ - if (pulled_task || this_rq->nr_running > 0) + if (pulled_task || this_rq->nr_running > 0 || + !continue_balancing) break; } rcu_read_unlock(); @@ -8075,8 +9359,23 @@ static int active_load_balance_cpu_stop(void *data) int busiest_cpu = cpu_of(busiest_rq); int target_cpu = busiest_rq->push_cpu; struct rq *target_rq = cpu_rq(target_cpu); - struct sched_domain *sd; + struct sched_domain *sd = NULL; struct task_struct *p = NULL; + struct task_struct *push_task; + int push_task_detached = 0; + struct lb_env env = { + .sd = sd, + .dst_cpu = target_cpu, + .dst_rq = target_rq, + .src_cpu = busiest_rq->cpu, + .src_rq = busiest_rq, + .idle = CPU_IDLE, + .busiest_nr_running = 0, + .busiest_grp_capacity = 0, + .flags = 0, + .loop = 0, + }; + bool moved = false; raw_spin_lock_irq(&busiest_rq->lock); @@ -8096,6 +9395,20 @@ static int active_load_balance_cpu_stop(void *data) */ BUG_ON(busiest_rq == target_rq); + push_task = busiest_rq->push_task; + target_cpu = busiest_rq->push_cpu; + if (push_task) { + if (task_on_rq_queued(push_task) && + push_task->state == TASK_RUNNING && + task_cpu(push_task) == busiest_cpu && + cpu_online(target_cpu)) { + detach_task(push_task, &env); + push_task_detached = 1; + moved = true; + } + goto out_unlock; + } + /* Search for an sd spanning us and the target CPU. */ rcu_read_lock(); for_each_domain(target_cpu, sd) { @@ -8105,15 +9418,7 @@ static int active_load_balance_cpu_stop(void *data) } if (likely(sd)) { - struct lb_env env = { - .sd = sd, - .dst_cpu = target_cpu, - .dst_rq = target_rq, - .src_cpu = busiest_rq->cpu, - .src_rq = busiest_rq, - .idle = CPU_IDLE, - }; - + env.sd = sd; schedstat_inc(sd->alb_count); p = detach_one_task(&env); @@ -8121,6 +9426,7 @@ static int active_load_balance_cpu_stop(void *data) schedstat_inc(sd->alb_pushed); /* Active balancing done, reset the failure counter. */ sd->nr_balance_failed = 0; + moved = true; } else { schedstat_inc(sd->alb_failed); } @@ -8128,6 +9434,12 @@ static int active_load_balance_cpu_stop(void *data) rcu_read_unlock(); out_unlock: busiest_rq->active_balance = 0; + push_task = busiest_rq->push_task; + target_cpu = busiest_rq->push_cpu; + + if (push_task) + busiest_rq->push_task = NULL; + raw_spin_unlock(&busiest_rq->lock); if (p) @@ -8135,6 +9447,15 @@ out_unlock: local_irq_enable(); + if (moved && !same_freq_domain(busiest_cpu, target_cpu)) { + int check_groups = !!(env.flags & + LBF_MOVED_RELATED_THREAD_GROUP_TASK); + check_for_freq_change(busiest_rq, false, check_groups); + check_for_freq_change(target_rq, false, check_groups); + } else if (moved) { + check_for_freq_change(target_rq, true, false); + } + return 0; } @@ -8156,10 +9477,45 @@ static struct { unsigned long next_balance; /* in jiffy units */ } nohz ____cacheline_aligned; -static inline int find_new_ilb(void) +#ifdef CONFIG_SCHED_HMP +static inline int find_new_hmp_ilb(int type) +{ + int call_cpu = raw_smp_processor_id(); + struct sched_domain *sd; + int ilb; + + rcu_read_lock(); + + /* Pick an idle cpu "closest" to call_cpu */ + for_each_domain(call_cpu, sd) { + for_each_cpu_and(ilb, nohz.idle_cpus_mask, + sched_domain_span(sd)) { + if (idle_cpu(ilb) && (type != NOHZ_KICK_RESTRICT || + (hmp_capable() && + cpu_max_possible_capacity(ilb) <= + cpu_max_possible_capacity(call_cpu)) || + cpu_max_power_cost(ilb) <= + cpu_max_power_cost(call_cpu))) { + rcu_read_unlock(); + reset_balance_interval(ilb); + return ilb; + } + } + } + + rcu_read_unlock(); + return nr_cpu_ids; +} +#endif /* CONFIG_SCHED_HMP */ + +static inline int find_new_ilb(int type) { int ilb = cpumask_first(nohz.idle_cpus_mask); +#ifdef CONFIG_SCHED_HMP + return find_new_hmp_ilb(type); +#endif + if (ilb < nr_cpu_ids && idle_cpu(ilb)) return ilb; @@ -8171,13 +9527,13 @@ static inline int find_new_ilb(void) * nohz_load_balancer CPU (if there is one) otherwise fallback to any idle * CPU (if there is one). */ -static void nohz_balancer_kick(void) +static void nohz_balancer_kick(int type) { int ilb_cpu; nohz.next_balance++; - ilb_cpu = find_new_ilb(); + ilb_cpu = find_new_ilb(type); if (ilb_cpu >= nr_cpu_ids) return; @@ -8450,6 +9806,66 @@ end: clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)); } +#ifdef CONFIG_SCHED_HMP +static inline int _nohz_kick_needed_hmp(struct rq *rq, int cpu, int *type) +{ + struct sched_domain *sd; + int i; + + if (rq->nr_running < 2) + return 0; + + if (!sysctl_sched_restrict_cluster_spill || sched_boost()) + return 1; + + if (hmp_capable() && cpu_max_possible_capacity(cpu) == + max_possible_capacity) + return 1; + + rcu_read_lock(); + sd = rcu_dereference_check_sched_domain(rq->sd); + if (!sd) { + rcu_read_unlock(); + return 0; + } + + for_each_cpu(i, sched_domain_span(sd)) { + if (cpu_load(i) < sched_spill_load && + cpu_rq(i)->nr_running < + sysctl_sched_spill_nr_run) { + /* Change the kick type to limit to CPUs that + * are of equal or lower capacity. + */ + *type = NOHZ_KICK_RESTRICT; + break; + } + } + rcu_read_unlock(); + return 1; +} +#endif /* CONFIG_SCHED_HMP */ + +static inline int _nohz_kick_needed(struct rq *rq, int cpu, int *type) +{ + unsigned long now = jiffies; + + /* + * None are in tickless mode and hence no need for NOHZ idle load + * balancing. + */ + if (likely(!atomic_read(&nohz.nr_cpus))) + return 0; + +#ifdef CONFIG_SCHED_HMP + return _nohz_kick_needed_hmp(rq, cpu, type); +#endif + + if (time_before(now, nohz.next_balance)) + return 0; + + return (rq->nr_running >= 2); +} + /* * Current heuristic for kicking the idle load balancer in the presence * of an idle cpu in the system. @@ -8461,12 +9877,14 @@ end: * - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler * domain span are idle. */ -static inline bool nohz_kick_needed(struct rq *rq) +static inline bool nohz_kick_needed(struct rq *rq, int *type) { - unsigned long now = jiffies; +#ifndef CONFIG_SCHED_HMP struct sched_domain_shared *sds; struct sched_domain *sd; - int nr_busy, cpu = rq->cpu; + int nr_busy; +#endif + int cpu = rq->cpu; bool kick = false; if (unlikely(rq->idle_balance)) @@ -8479,19 +9897,10 @@ static inline bool nohz_kick_needed(struct rq *rq) set_cpu_sd_state_busy(); nohz_balance_exit_idle(cpu); - /* - * None are in tickless mode and hence no need for NOHZ idle load - * balancing. - */ - if (likely(!atomic_read(&nohz.nr_cpus))) - return false; - - if (time_before(now, nohz.next_balance)) - return false; - - if (rq->nr_running >= 2) + if (_nohz_kick_needed(rq, cpu, type)) return true; +#ifndef CONFIG_SCHED_HMP rcu_read_lock(); sds = rcu_dereference(per_cpu(sd_llc_shared, cpu)); if (sds) { @@ -8525,6 +9934,7 @@ static inline bool nohz_kick_needed(struct rq *rq) unlock: rcu_read_unlock(); +#endif return kick; } #else @@ -8558,6 +9968,8 @@ static __latent_entropy void run_rebalance_domains(struct softirq_action *h) */ void trigger_load_balance(struct rq *rq) { + int type = NOHZ_KICK_ANY; + /* Don't need to rebalance while attached to NULL domain */ if (unlikely(on_null_domain(rq))) return; @@ -8565,8 +9977,8 @@ void trigger_load_balance(struct rq *rq) if (time_after_eq(jiffies, rq->next_balance)) raise_softirq(SCHED_SOFTIRQ); #ifdef CONFIG_NO_HZ_COMMON - if (nohz_kick_needed(rq)) - nohz_balancer_kick(); + if (nohz_kick_needed(rq, &type)) + nohz_balancer_kick(type); #endif } @@ -9061,6 +10473,9 @@ const struct sched_class fair_sched_class = { #ifdef CONFIG_FAIR_GROUP_SCHED .task_change_group = task_change_group_fair, #endif +#ifdef CONFIG_SCHED_HMP + .fixup_hmp_sched_stats = fixup_hmp_sched_stats_fair, +#endif }; #ifdef CONFIG_SCHED_DEBUG diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 69631fa46c2f..6651a37ba057 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -49,7 +49,7 @@ SCHED_FEAT(NONTASK_CAPACITY, true) * Queue remote wakeups on the target CPU and process them * using the scheduler IPI. Reduces rq->lock contention/bounces. */ -SCHED_FEAT(TTWU_QUEUE, true) +SCHED_FEAT(TTWU_QUEUE, false) #ifdef HAVE_RT_PUSH_IPI /* diff --git a/kernel/sched/hmp.c b/kernel/sched/hmp.c new file mode 100644 index 000000000000..9bc1bacd52e5 --- /dev/null +++ b/kernel/sched/hmp.c @@ -0,0 +1,3975 @@ +/* Copyright (c) 2012-2016, The Linux Foundation. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 and + * only version 2 as published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * Implementation credits: Srivatsa Vaddagiri, Steve Muckle + * Syed Rameez Mustafa, Olav haugan, Joonwoo Park, Pavan Kumar Kondeti + * and Vikram Mulukutla + */ + +#include +#include +#include + +#include "sched.h" + +#include + +const char *task_event_names[] = {"PUT_PREV_TASK", "PICK_NEXT_TASK", + "TASK_WAKE", "TASK_MIGRATE", "TASK_UPDATE", "IRQ_UPDATE"}; + +const char *migrate_type_names[] = {"GROUP_TO_RQ", "RQ_TO_GROUP", + "RQ_TO_RQ", "GROUP_TO_GROUP"}; + + +static ktime_t ktime_last; +static bool sched_ktime_suspended; + +static bool use_cycle_counter; +static struct cpu_cycle_counter_cb cpu_cycle_counter_cb; + +u64 sched_ktime_clock(void) +{ + if (unlikely(sched_ktime_suspended)) + return ktime_to_ns(ktime_last); + return ktime_get_ns(); +} + +static void sched_resume(void) +{ + sched_ktime_suspended = false; +} + +static int sched_suspend(void) +{ + ktime_last = ktime_get(); + sched_ktime_suspended = true; + return 0; +} + +static struct syscore_ops sched_syscore_ops = { + .resume = sched_resume, + .suspend = sched_suspend +}; + +static int __init sched_init_ops(void) +{ + register_syscore_ops(&sched_syscore_ops); + return 0; +} +late_initcall(sched_init_ops); + +inline void clear_ed_task(struct task_struct *p, struct rq *rq) +{ + if (p == rq->ed_task) + rq->ed_task = NULL; +} + +inline void set_task_last_wake(struct task_struct *p, u64 wallclock) +{ + p->last_wake_ts = wallclock; +} + +inline void set_task_last_switch_out(struct task_struct *p, u64 wallclock) +{ + p->last_switch_out_ts = wallclock; +} + +/* + * Note C-state for (idle) cpus. + * + * @cstate = cstate index, 0 -> active state + * @wakeup_energy = energy spent in waking up cpu + * @wakeup_latency = latency to wakeup from cstate + * + */ +void +sched_set_cpu_cstate(int cpu, int cstate, int wakeup_energy, int wakeup_latency) +{ + struct rq *rq = cpu_rq(cpu); + + rq->cstate = cstate; /* C1, C2 etc */ + rq->wakeup_energy = wakeup_energy; + rq->wakeup_latency = wakeup_latency; +} + +/* + * Note D-state for (idle) cluster. + * + * @dstate = dstate index, 0 -> active state + * @wakeup_energy = energy spent in waking up cluster + * @wakeup_latency = latency to wakeup from cluster + * + */ +void sched_set_cluster_dstate(const cpumask_t *cluster_cpus, int dstate, + int wakeup_energy, int wakeup_latency) +{ + struct sched_cluster *cluster = + cpu_rq(cpumask_first(cluster_cpus))->cluster; + cluster->dstate = dstate; + cluster->dstate_wakeup_energy = wakeup_energy; + cluster->dstate_wakeup_latency = wakeup_latency; +} + +u32 __weak get_freq_max_load(int cpu, u32 freq) +{ + /* 100% by default */ + return 100; +} + +struct freq_max_load_entry { + /* The maximum load which has accounted governor's headroom. */ + u64 hdemand; +}; + +struct freq_max_load { + struct rcu_head rcu; + int length; + struct freq_max_load_entry freqs[0]; +}; + +static DEFINE_PER_CPU(struct freq_max_load *, freq_max_load); +static DEFINE_SPINLOCK(freq_max_load_lock); + +struct cpu_pwr_stats __weak *get_cpu_pwr_stats(void) +{ + return NULL; +} + +int sched_update_freq_max_load(const cpumask_t *cpumask) +{ + int i, cpu, ret; + unsigned int freq; + struct cpu_pstate_pwr *costs; + struct cpu_pwr_stats *per_cpu_info = get_cpu_pwr_stats(); + struct freq_max_load *max_load, *old_max_load; + struct freq_max_load_entry *entry; + u64 max_demand_capacity, max_demand; + unsigned long flags; + u32 hfreq; + int hpct; + + if (!per_cpu_info) + return 0; + + spin_lock_irqsave(&freq_max_load_lock, flags); + max_demand_capacity = div64_u64(max_task_load(), max_possible_capacity); + for_each_cpu(cpu, cpumask) { + if (!per_cpu_info[cpu].ptable) { + ret = -EINVAL; + goto fail; + } + + old_max_load = rcu_dereference(per_cpu(freq_max_load, cpu)); + + /* + * allocate len + 1 and leave the last power cost as 0 for + * power_cost() can stop iterating index when + * per_cpu_info[cpu].len > len of max_load due to race between + * cpu power stats update and get_cpu_pwr_stats(). + */ + max_load = kzalloc(sizeof(struct freq_max_load) + + sizeof(struct freq_max_load_entry) * + (per_cpu_info[cpu].len + 1), GFP_ATOMIC); + if (unlikely(!max_load)) { + ret = -ENOMEM; + goto fail; + } + + max_load->length = per_cpu_info[cpu].len; + + max_demand = max_demand_capacity * + cpu_max_possible_capacity(cpu); + + i = 0; + costs = per_cpu_info[cpu].ptable; + while (costs[i].freq) { + entry = &max_load->freqs[i]; + freq = costs[i].freq; + hpct = get_freq_max_load(cpu, freq); + if (hpct <= 0 && hpct > 100) + hpct = 100; + hfreq = div64_u64((u64)freq * hpct, 100); + entry->hdemand = + div64_u64(max_demand * hfreq, + cpu_max_possible_freq(cpu)); + i++; + } + + rcu_assign_pointer(per_cpu(freq_max_load, cpu), max_load); + if (old_max_load) + kfree_rcu(old_max_load, rcu); + } + + spin_unlock_irqrestore(&freq_max_load_lock, flags); + return 0; + +fail: + for_each_cpu(cpu, cpumask) { + max_load = rcu_dereference(per_cpu(freq_max_load, cpu)); + if (max_load) { + rcu_assign_pointer(per_cpu(freq_max_load, cpu), NULL); + kfree_rcu(max_load, rcu); + } + } + + spin_unlock_irqrestore(&freq_max_load_lock, flags); + return ret; +} + +unsigned int max_possible_efficiency = 1; +unsigned int min_possible_efficiency = UINT_MAX; + +unsigned long __weak arch_get_cpu_efficiency(int cpu) +{ + return SCHED_CAPACITY_SCALE; +} + +/* Keep track of max/min capacity possible across CPUs "currently" */ +static void __update_min_max_capacity(void) +{ + int i; + int max_cap = 0, min_cap = INT_MAX; + + for_each_online_cpu(i) { + max_cap = max(max_cap, cpu_capacity(i)); + min_cap = min(min_cap, cpu_capacity(i)); + } + + max_capacity = max_cap; + min_capacity = min_cap; +} + +static void update_min_max_capacity(void) +{ + unsigned long flags; + int i; + + local_irq_save(flags); + for_each_possible_cpu(i) + raw_spin_lock(&cpu_rq(i)->lock); + + __update_min_max_capacity(); + + for_each_possible_cpu(i) + raw_spin_unlock(&cpu_rq(i)->lock); + local_irq_restore(flags); +} + +/* + * Return 'capacity' of a cpu in reference to "least" efficient cpu, such that + * least efficient cpu gets capacity of 1024 + */ +static unsigned long +capacity_scale_cpu_efficiency(struct sched_cluster *cluster) +{ + return (1024 * cluster->efficiency) / min_possible_efficiency; +} + +/* + * Return 'capacity' of a cpu in reference to cpu with lowest max_freq + * (min_max_freq), such that one with lowest max_freq gets capacity of 1024. + */ +static unsigned long capacity_scale_cpu_freq(struct sched_cluster *cluster) +{ + return (1024 * cluster_max_freq(cluster)) / min_max_freq; +} + +/* + * Return load_scale_factor of a cpu in reference to "most" efficient cpu, so + * that "most" efficient cpu gets a load_scale_factor of 1 + */ +static inline unsigned long +load_scale_cpu_efficiency(struct sched_cluster *cluster) +{ + return DIV_ROUND_UP(1024 * max_possible_efficiency, + cluster->efficiency); +} + +/* + * Return load_scale_factor of a cpu in reference to cpu with best max_freq + * (max_possible_freq), so that one with best max_freq gets a load_scale_factor + * of 1. + */ +static inline unsigned long load_scale_cpu_freq(struct sched_cluster *cluster) +{ + return DIV_ROUND_UP(1024 * max_possible_freq, + cluster_max_freq(cluster)); +} + +static int compute_capacity(struct sched_cluster *cluster) +{ + int capacity = 1024; + + capacity *= capacity_scale_cpu_efficiency(cluster); + capacity >>= 10; + + capacity *= capacity_scale_cpu_freq(cluster); + capacity >>= 10; + + return capacity; +} + +static int compute_max_possible_capacity(struct sched_cluster *cluster) +{ + int capacity = 1024; + + capacity *= capacity_scale_cpu_efficiency(cluster); + capacity >>= 10; + + capacity *= (1024 * cluster->max_possible_freq) / min_max_freq; + capacity >>= 10; + + return capacity; +} + +static int compute_load_scale_factor(struct sched_cluster *cluster) +{ + int load_scale = 1024; + + /* + * load_scale_factor accounts for the fact that task load + * is in reference to "best" performing cpu. Task's load will need to be + * scaled (up) by a factor to determine suitability to be placed on a + * (little) cpu. + */ + load_scale *= load_scale_cpu_efficiency(cluster); + load_scale >>= 10; + + load_scale *= load_scale_cpu_freq(cluster); + load_scale >>= 10; + + return load_scale; +} + +struct list_head cluster_head; +static DEFINE_MUTEX(cluster_lock); +static cpumask_t all_cluster_cpus = CPU_MASK_NONE; +DECLARE_BITMAP(all_cluster_ids, NR_CPUS); +struct sched_cluster *sched_cluster[NR_CPUS]; +int num_clusters; + +struct sched_cluster init_cluster = { + .list = LIST_HEAD_INIT(init_cluster.list), + .id = 0, + .max_power_cost = 1, + .min_power_cost = 1, + .capacity = 1024, + .max_possible_capacity = 1024, + .efficiency = 1, + .load_scale_factor = 1024, + .cur_freq = 1, + .max_freq = 1, + .max_mitigated_freq = UINT_MAX, + .min_freq = 1, + .max_possible_freq = 1, + .dstate = 0, + .dstate_wakeup_energy = 0, + .dstate_wakeup_latency = 0, + .exec_scale_factor = 1024, + .notifier_sent = 0, +}; + +static void update_all_clusters_stats(void) +{ + struct sched_cluster *cluster; + u64 highest_mpc = 0, lowest_mpc = U64_MAX; + + pre_big_task_count_change(cpu_possible_mask); + + for_each_sched_cluster(cluster) { + u64 mpc; + + cluster->capacity = compute_capacity(cluster); + mpc = cluster->max_possible_capacity = + compute_max_possible_capacity(cluster); + cluster->load_scale_factor = compute_load_scale_factor(cluster); + + cluster->exec_scale_factor = + DIV_ROUND_UP(cluster->efficiency * 1024, + max_possible_efficiency); + + if (mpc > highest_mpc) + highest_mpc = mpc; + + if (mpc < lowest_mpc) + lowest_mpc = mpc; + } + + max_possible_capacity = highest_mpc; + min_max_possible_capacity = lowest_mpc; + + __update_min_max_capacity(); + sched_update_freq_max_load(cpu_possible_mask); + post_big_task_count_change(cpu_possible_mask); +} + +static void assign_cluster_ids(struct list_head *head) +{ + struct sched_cluster *cluster; + int pos = 0; + + list_for_each_entry(cluster, head, list) { + cluster->id = pos; + sched_cluster[pos++] = cluster; + } +} + +static void +move_list(struct list_head *dst, struct list_head *src, bool sync_rcu) +{ + struct list_head *first, *last; + + first = src->next; + last = src->prev; + + if (sync_rcu) { + INIT_LIST_HEAD_RCU(src); + synchronize_rcu(); + } + + first->prev = dst; + dst->prev = last; + last->next = dst; + + /* Ensure list sanity before making the head visible to all CPUs. */ + smp_mb(); + dst->next = first; +} + +static int +compare_clusters(void *priv, struct list_head *a, struct list_head *b) +{ + struct sched_cluster *cluster1, *cluster2; + int ret; + + cluster1 = container_of(a, struct sched_cluster, list); + cluster2 = container_of(b, struct sched_cluster, list); + + ret = cluster1->max_power_cost > cluster2->max_power_cost || + (cluster1->max_power_cost == cluster2->max_power_cost && + cluster1->max_possible_capacity < + cluster2->max_possible_capacity); + + return ret; +} + +static void sort_clusters(void) +{ + struct sched_cluster *cluster; + struct list_head new_head; + + INIT_LIST_HEAD(&new_head); + + for_each_sched_cluster(cluster) { + cluster->max_power_cost = power_cost(cluster_first_cpu(cluster), + max_task_load()); + cluster->min_power_cost = power_cost(cluster_first_cpu(cluster), + 0); + } + + move_list(&new_head, &cluster_head, true); + + list_sort(NULL, &new_head, compare_clusters); + assign_cluster_ids(&new_head); + + /* + * Ensure cluster ids are visible to all CPUs before making + * cluster_head visible. + */ + move_list(&cluster_head, &new_head, false); +} + +static void +insert_cluster(struct sched_cluster *cluster, struct list_head *head) +{ + struct sched_cluster *tmp; + struct list_head *iter = head; + + list_for_each_entry(tmp, head, list) { + if (cluster->max_power_cost < tmp->max_power_cost) + break; + iter = &tmp->list; + } + + list_add(&cluster->list, iter); +} + +static struct sched_cluster *alloc_new_cluster(const struct cpumask *cpus) +{ + struct sched_cluster *cluster = NULL; + + cluster = kzalloc(sizeof(struct sched_cluster), GFP_ATOMIC); + if (!cluster) { + __WARN_printf("Cluster allocation failed. \ + Possible bad scheduling\n"); + return NULL; + } + + INIT_LIST_HEAD(&cluster->list); + cluster->max_power_cost = 1; + cluster->min_power_cost = 1; + cluster->capacity = 1024; + cluster->max_possible_capacity = 1024; + cluster->efficiency = 1; + cluster->load_scale_factor = 1024; + cluster->cur_freq = 1; + cluster->max_freq = 1; + cluster->max_mitigated_freq = UINT_MAX; + cluster->min_freq = 1; + cluster->max_possible_freq = 1; + cluster->dstate = 0; + cluster->dstate_wakeup_energy = 0; + cluster->dstate_wakeup_latency = 0; + cluster->freq_init_done = false; + + cluster->cpus = *cpus; + cluster->efficiency = arch_get_cpu_efficiency(cpumask_first(cpus)); + + if (cluster->efficiency > max_possible_efficiency) + max_possible_efficiency = cluster->efficiency; + if (cluster->efficiency < min_possible_efficiency) + min_possible_efficiency = cluster->efficiency; + + cluster->notifier_sent = 0; + return cluster; +} + +static void add_cluster(const struct cpumask *cpus, struct list_head *head) +{ + struct sched_cluster *cluster = alloc_new_cluster(cpus); + int i; + + if (!cluster) + return; + + for_each_cpu(i, cpus) + cpu_rq(i)->cluster = cluster; + + insert_cluster(cluster, head); + set_bit(num_clusters, all_cluster_ids); + num_clusters++; +} + +void update_cluster_topology(void) +{ + struct cpumask cpus = *cpu_possible_mask; + const struct cpumask *cluster_cpus; + struct list_head new_head; + int i; + + INIT_LIST_HEAD(&new_head); + + for_each_cpu(i, &cpus) { + cluster_cpus = cpu_coregroup_mask(i); + cpumask_or(&all_cluster_cpus, &all_cluster_cpus, cluster_cpus); + cpumask_andnot(&cpus, &cpus, cluster_cpus); + add_cluster(cluster_cpus, &new_head); + } + + assign_cluster_ids(&new_head); + + /* + * Ensure cluster ids are visible to all CPUs before making + * cluster_head visible. + */ + move_list(&cluster_head, &new_head, false); +} + +void init_clusters(void) +{ + bitmap_clear(all_cluster_ids, 0, NR_CPUS); + init_cluster.cpus = *cpu_possible_mask; + INIT_LIST_HEAD(&cluster_head); +} + +int register_cpu_cycle_counter_cb(struct cpu_cycle_counter_cb *cb) +{ + mutex_lock(&cluster_lock); + if (!cb->get_cpu_cycle_counter) { + mutex_unlock(&cluster_lock); + return -EINVAL; + } + + cpu_cycle_counter_cb = *cb; + use_cycle_counter = true; + mutex_unlock(&cluster_lock); + + return 0; +} + +int got_boost_kick(void) +{ + int cpu = smp_processor_id(); + struct rq *rq = cpu_rq(cpu); + + return test_bit(BOOST_KICK, &rq->hmp_flags); +} + +inline void clear_boost_kick(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + + clear_bit(BOOST_KICK, &rq->hmp_flags); +} + +inline void boost_kick(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + + if (!test_and_set_bit(BOOST_KICK, &rq->hmp_flags)) + smp_send_reschedule(cpu); +} + +/* Clear any HMP scheduler related requests pending from or on cpu */ +void clear_hmp_request(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + unsigned long flags; + + clear_boost_kick(cpu); + clear_reserved(cpu); + if (rq->push_task) { + raw_spin_lock_irqsave(&rq->lock, flags); + if (rq->push_task) { + clear_reserved(rq->push_cpu); + put_task_struct(rq->push_task); + rq->push_task = NULL; + } + rq->active_balance = 0; + raw_spin_unlock_irqrestore(&rq->lock, flags); + } +} + +int sched_set_static_cpu_pwr_cost(int cpu, unsigned int cost) +{ + struct rq *rq = cpu_rq(cpu); + + rq->static_cpu_pwr_cost = cost; + return 0; +} + +unsigned int sched_get_static_cpu_pwr_cost(int cpu) +{ + return cpu_rq(cpu)->static_cpu_pwr_cost; +} + +int sched_set_static_cluster_pwr_cost(int cpu, unsigned int cost) +{ + struct sched_cluster *cluster = cpu_rq(cpu)->cluster; + + cluster->static_cluster_pwr_cost = cost; + return 0; +} + +unsigned int sched_get_static_cluster_pwr_cost(int cpu) +{ + return cpu_rq(cpu)->cluster->static_cluster_pwr_cost; +} + +/* + * sched_window_stats_policy and sched_ravg_hist_size have a 'sysctl' copy + * associated with them. This is required for atomic update of those variables + * when being modifed via sysctl interface. + * + * IMPORTANT: Initialize both copies to same value!! + */ + +/* + * Tasks that are runnable continuously for a period greather than + * EARLY_DETECTION_DURATION can be flagged early as potential + * high load tasks. + */ +#define EARLY_DETECTION_DURATION 9500000 + +static __read_mostly unsigned int sched_ravg_hist_size = 5; +__read_mostly unsigned int sysctl_sched_ravg_hist_size = 5; + +static __read_mostly unsigned int sched_window_stats_policy = + WINDOW_STATS_MAX_RECENT_AVG; +__read_mostly unsigned int sysctl_sched_window_stats_policy = + WINDOW_STATS_MAX_RECENT_AVG; + +#define SCHED_ACCOUNT_WAIT_TIME 1 + +__read_mostly unsigned int sysctl_sched_cpu_high_irqload = (10 * NSEC_PER_MSEC); + +unsigned int __read_mostly sysctl_sched_enable_colocation = 1; + +/* + * Enable colocation and frequency aggregation for all threads in a process. + * The children inherits the group id from the parent. + */ +unsigned int __read_mostly sysctl_sched_enable_thread_grouping; + + +__read_mostly unsigned int sysctl_sched_new_task_windows = 5; + +#define SCHED_FREQ_ACCOUNT_WAIT_TIME 0 + +/* + * For increase, send notification if + * freq_required - cur_freq > sysctl_sched_freq_inc_notify + */ +__read_mostly int sysctl_sched_freq_inc_notify = 10 * 1024 * 1024; /* + 10GHz */ + +/* + * For decrease, send notification if + * cur_freq - freq_required > sysctl_sched_freq_dec_notify + */ +__read_mostly int sysctl_sched_freq_dec_notify = 10 * 1024 * 1024; /* - 10GHz */ + +static __read_mostly unsigned int sched_io_is_busy; + +__read_mostly unsigned int sysctl_sched_pred_alert_freq = 10 * 1024 * 1024; + +/* + * Maximum possible frequency across all cpus. Task demand and cpu + * capacity (cpu_power) metrics are scaled in reference to it. + */ +unsigned int max_possible_freq = 1; + +/* + * Minimum possible max_freq across all cpus. This will be same as + * max_possible_freq on homogeneous systems and could be different from + * max_possible_freq on heterogenous systems. min_max_freq is used to derive + * capacity (cpu_power) of cpus. + */ +unsigned int min_max_freq = 1; + +unsigned int max_capacity = 1024; /* max(rq->capacity) */ +unsigned int min_capacity = 1024; /* min(rq->capacity) */ +unsigned int max_possible_capacity = 1024; /* max(rq->max_possible_capacity) */ +unsigned int +min_max_possible_capacity = 1024; /* min(rq->max_possible_capacity) */ + +/* Window size (in ns) */ +__read_mostly unsigned int sched_ravg_window = 10000000; + +/* Min window size (in ns) = 10ms */ +#define MIN_SCHED_RAVG_WINDOW 10000000 + +/* Max window size (in ns) = 1s */ +#define MAX_SCHED_RAVG_WINDOW 1000000000 + +/* Temporarily disable window-stats activity on all cpus */ +unsigned int __read_mostly sched_disable_window_stats; + +/* + * Major task runtime. If a task runs for more than sched_major_task_runtime + * in a window, it's considered to be generating majority of workload + * for this window. Prediction could be adjusted for such tasks. + */ +__read_mostly unsigned int sched_major_task_runtime = 10000000; + +static unsigned int sync_cpu; + +static LIST_HEAD(related_thread_groups); +static DEFINE_RWLOCK(related_thread_group_lock); + +#define for_each_related_thread_group(grp) \ + list_for_each_entry(grp, &related_thread_groups, list) + +/* + * Demand aggregation for frequency purpose: + * + * 'sched_freq_aggregate' controls aggregation of cpu demand of related threads + * for frequency determination purpose. This aggregation is done per-cluster. + * + * CPU demand of tasks from various related groups is aggregated per-cluster and + * added to the "max_busy_cpu" in that cluster, where max_busy_cpu is determined + * by just rq->prev_runnable_sum. + * + * Some examples follow, which assume: + * Cluster0 = CPU0-3, Cluster1 = CPU4-7 + * One related thread group A that has tasks A0, A1, A2 + * + * A->cpu_time[X].curr/prev_sum = counters in which cpu execution stats of + * tasks belonging to group A are accumulated when they run on cpu X. + * + * CX->curr/prev_sum = counters in which cpu execution stats of all tasks + * not belonging to group A are accumulated when they run on cpu X + * + * Lets say the stats for window M was as below: + * + * C0->prev_sum = 1ms, A->cpu_time[0].prev_sum = 5ms + * Task A0 ran 5ms on CPU0 + * Task B0 ran 1ms on CPU0 + * + * C1->prev_sum = 5ms, A->cpu_time[1].prev_sum = 6ms + * Task A1 ran 4ms on CPU1 + * Task A2 ran 2ms on CPU1 + * Task B1 ran 5ms on CPU1 + * + * C2->prev_sum = 0ms, A->cpu_time[2].prev_sum = 0 + * CPU2 idle + * + * C3->prev_sum = 0ms, A->cpu_time[3].prev_sum = 0 + * CPU3 idle + * + * In this case, CPU1 was most busy going by just its prev_sum counter. Demand + * from all group A tasks are added to CPU1. IOW, at end of window M, cpu busy + * time reported to governor will be: + * + * + * C0 busy time = 1ms + * C1 busy time = 5 + 5 + 6 = 16ms + * + */ +static __read_mostly unsigned int sched_freq_aggregate; +__read_mostly unsigned int sysctl_sched_freq_aggregate; + +unsigned int __read_mostly sysctl_sched_freq_aggregate_threshold_pct; +static unsigned int __read_mostly sched_freq_aggregate_threshold; + +/* Initial task load. Newly created tasks are assigned this load. */ +unsigned int __read_mostly sched_init_task_load_windows; +unsigned int __read_mostly sysctl_sched_init_task_load_pct = 15; + +unsigned int max_task_load(void) +{ + return sched_ravg_window; +} + +/* + * Scheduler boost is a mechanism to temporarily place tasks on CPUs + * with higher capacity than those where a task would have normally + * ended up with their load characteristics. Any entity enabling + * boost is responsible for disabling it as well. + */ +unsigned int sysctl_sched_boost; + +/* A cpu can no longer accommodate more tasks if: + * + * rq->nr_running > sysctl_sched_spill_nr_run || + * rq->hmp_stats.cumulative_runnable_avg > sched_spill_load + */ +unsigned int __read_mostly sysctl_sched_spill_nr_run = 10; + +/* + * Place sync wakee tasks those have less than configured demand to the waker's + * cluster. + */ +unsigned int __read_mostly sched_small_wakee_task_load; +unsigned int __read_mostly sysctl_sched_small_wakee_task_load_pct = 10; + +unsigned int __read_mostly sched_big_waker_task_load; +unsigned int __read_mostly sysctl_sched_big_waker_task_load_pct = 25; + +/* + * CPUs with load greater than the sched_spill_load_threshold are not + * eligible for task placement. When all CPUs in a cluster achieve a + * load higher than this level, tasks becomes eligible for inter + * cluster migration. + */ +unsigned int __read_mostly sched_spill_load; +unsigned int __read_mostly sysctl_sched_spill_load_pct = 100; + +/* + * Tasks whose bandwidth consumption on a cpu is more than + * sched_upmigrate are considered "big" tasks. Big tasks will be + * considered for "up" migration, i.e migrating to a cpu with better + * capacity. + */ +unsigned int __read_mostly sched_upmigrate; +unsigned int __read_mostly sysctl_sched_upmigrate_pct = 80; + +/* + * Big tasks, once migrated, will need to drop their bandwidth + * consumption to less than sched_downmigrate before they are "down" + * migrated. + */ +unsigned int __read_mostly sched_downmigrate; +unsigned int __read_mostly sysctl_sched_downmigrate_pct = 60; + +/* + * The load scale factor of a CPU gets boosted when its max frequency + * is restricted due to which the tasks are migrating to higher capacity + * CPUs early. The sched_upmigrate threshold is auto-upgraded by + * rq->max_possible_freq/rq->max_freq of a lower capacity CPU. + */ +unsigned int up_down_migrate_scale_factor = 1024; + +/* + * Scheduler selects and places task to its previous CPU if sleep time is + * less than sysctl_sched_select_prev_cpu_us. + */ +unsigned int __read_mostly +sched_short_sleep_task_threshold = 2000 * NSEC_PER_USEC; + +unsigned int __read_mostly sysctl_sched_select_prev_cpu_us = 2000; + +unsigned int __read_mostly +sched_long_cpu_selection_threshold = 100 * NSEC_PER_MSEC; + +unsigned int __read_mostly sysctl_sched_restrict_cluster_spill; + +void update_up_down_migrate(void) +{ + unsigned int up_migrate = pct_to_real(sysctl_sched_upmigrate_pct); + unsigned int down_migrate = pct_to_real(sysctl_sched_downmigrate_pct); + unsigned int delta; + + if (up_down_migrate_scale_factor == 1024) + goto done; + + delta = up_migrate - down_migrate; + + up_migrate /= NSEC_PER_USEC; + up_migrate *= up_down_migrate_scale_factor; + up_migrate >>= 10; + up_migrate *= NSEC_PER_USEC; + + up_migrate = min(up_migrate, sched_ravg_window); + + down_migrate /= NSEC_PER_USEC; + down_migrate *= up_down_migrate_scale_factor; + down_migrate >>= 10; + down_migrate *= NSEC_PER_USEC; + + down_migrate = min(down_migrate, up_migrate - delta); +done: + sched_upmigrate = up_migrate; + sched_downmigrate = down_migrate; +} + +void set_hmp_defaults(void) +{ + sched_spill_load = + pct_to_real(sysctl_sched_spill_load_pct); + + update_up_down_migrate(); + + sched_major_task_runtime = + mult_frac(sched_ravg_window, MAJOR_TASK_PCT, 100); + + sched_init_task_load_windows = + div64_u64((u64)sysctl_sched_init_task_load_pct * + (u64)sched_ravg_window, 100); + + sched_short_sleep_task_threshold = sysctl_sched_select_prev_cpu_us * + NSEC_PER_USEC; + + sched_small_wakee_task_load = + div64_u64((u64)sysctl_sched_small_wakee_task_load_pct * + (u64)sched_ravg_window, 100); + + sched_big_waker_task_load = + div64_u64((u64)sysctl_sched_big_waker_task_load_pct * + (u64)sched_ravg_window, 100); + + sched_freq_aggregate_threshold = + pct_to_real(sysctl_sched_freq_aggregate_threshold_pct); +} + +u32 sched_get_init_task_load(struct task_struct *p) +{ + return p->init_load_pct; +} + +int sched_set_init_task_load(struct task_struct *p, int init_load_pct) +{ + if (init_load_pct < 0 || init_load_pct > 100) + return -EINVAL; + + p->init_load_pct = init_load_pct; + + return 0; +} + +#ifdef CONFIG_CGROUP_SCHED + +int upmigrate_discouraged(struct task_struct *p) +{ + return task_group(p)->upmigrate_discouraged; +} + +#else + +static inline int upmigrate_discouraged(struct task_struct *p) +{ + return 0; +} + +#endif + +/* Is a task "big" on its current cpu */ +static inline int __is_big_task(struct task_struct *p, u64 scaled_load) +{ + int nice = task_nice(p); + + if (nice > SCHED_UPMIGRATE_MIN_NICE || upmigrate_discouraged(p)) + return 0; + + return scaled_load > sched_upmigrate; +} + +int is_big_task(struct task_struct *p) +{ + return __is_big_task(p, scale_load_to_cpu(task_load(p), task_cpu(p))); +} + +u64 cpu_load(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + + return scale_load_to_cpu(rq->hmp_stats.cumulative_runnable_avg, cpu); +} + +u64 cpu_load_sync(int cpu, int sync) +{ + return scale_load_to_cpu(cpu_cravg_sync(cpu, sync), cpu); +} + +static int boost_refcount; +static DEFINE_SPINLOCK(boost_lock); +static DEFINE_MUTEX(boost_mutex); + +static void boost_kick_cpus(void) +{ + int i; + + for_each_online_cpu(i) { + if (cpu_capacity(i) != max_capacity) + boost_kick(i); + } +} + +int sched_boost(void) +{ + return boost_refcount > 0; +} + +int sched_set_boost(int enable) +{ + unsigned long flags; + int ret = 0; + int old_refcount; + + spin_lock_irqsave(&boost_lock, flags); + + old_refcount = boost_refcount; + + if (enable == 1) { + boost_refcount++; + } else if (!enable) { + if (boost_refcount >= 1) + boost_refcount--; + else + ret = -EINVAL; + } else { + ret = -EINVAL; + } + + if (!old_refcount && boost_refcount) + boost_kick_cpus(); + + trace_sched_set_boost(boost_refcount); + spin_unlock_irqrestore(&boost_lock, flags); + + return ret; +} + +int sched_boost_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, + loff_t *ppos) +{ + int ret; + + mutex_lock(&boost_mutex); + if (!write) + sysctl_sched_boost = sched_boost(); + + ret = proc_dointvec(table, write, buffer, lenp, ppos); + if (ret || !write) + goto done; + + ret = (sysctl_sched_boost <= 1) ? + sched_set_boost(sysctl_sched_boost) : -EINVAL; + +done: + mutex_unlock(&boost_mutex); + return ret; +} + +/* + * Task will fit on a cpu if it's bandwidth consumption on that cpu + * will be less than sched_upmigrate. A big task that was previously + * "up" migrated will be considered fitting on "little" cpu if its + * bandwidth consumption on "little" cpu will be less than + * sched_downmigrate. This will help avoid frequenty migrations for + * tasks with load close to the upmigrate threshold + */ +int task_load_will_fit(struct task_struct *p, u64 task_load, int cpu, + enum sched_boost_type boost_type) +{ + int upmigrate; + + if (cpu_capacity(cpu) == max_capacity) + return 1; + + if (boost_type != SCHED_BOOST_ON_BIG) { + if (task_nice(p) > SCHED_UPMIGRATE_MIN_NICE || + upmigrate_discouraged(p)) + return 1; + + upmigrate = sched_upmigrate; + if (cpu_capacity(task_cpu(p)) > cpu_capacity(cpu)) + upmigrate = sched_downmigrate; + + if (task_load < upmigrate) + return 1; + } + + return 0; +} + +enum sched_boost_type sched_boost_type(void) +{ + if (sched_boost()) { + if (min_possible_efficiency != max_possible_efficiency) + return SCHED_BOOST_ON_BIG; + else + return SCHED_BOOST_ON_ALL; + } + return SCHED_BOOST_NONE; +} + +int task_will_fit(struct task_struct *p, int cpu) +{ + u64 tload = scale_load_to_cpu(task_load(p), cpu); + + return task_load_will_fit(p, tload, cpu, sched_boost_type()); +} + +int group_will_fit(struct sched_cluster *cluster, + struct related_thread_group *grp, u64 demand) +{ + int cpu = cluster_first_cpu(cluster); + int prev_capacity = 0; + unsigned int threshold = sched_upmigrate; + u64 load; + + if (cluster->capacity == max_capacity) + return 1; + + if (grp->preferred_cluster) + prev_capacity = grp->preferred_cluster->capacity; + + if (cluster->capacity < prev_capacity) + threshold = sched_downmigrate; + + load = scale_load_to_cpu(demand, cpu); + if (load < threshold) + return 1; + + return 0; +} + +/* + * Return the cost of running task p on CPU cpu. This function + * currently assumes that task p is the only task which will run on + * the CPU. + */ +unsigned int power_cost(int cpu, u64 demand) +{ + int first, mid, last; + struct cpu_pwr_stats *per_cpu_info = get_cpu_pwr_stats(); + struct cpu_pstate_pwr *costs; + struct freq_max_load *max_load; + int total_static_pwr_cost = 0; + struct rq *rq = cpu_rq(cpu); + unsigned int pc; + + if (!per_cpu_info || !per_cpu_info[cpu].ptable) + /* + * When power aware scheduling is not in use, or CPU + * power data is not available, just use the CPU + * capacity as a rough stand-in for real CPU power + * numbers, assuming bigger CPUs are more power + * hungry. + */ + return cpu_max_possible_capacity(cpu); + + rcu_read_lock(); + max_load = rcu_dereference(per_cpu(freq_max_load, cpu)); + if (!max_load) { + pc = cpu_max_possible_capacity(cpu); + goto unlock; + } + + costs = per_cpu_info[cpu].ptable; + + if (demand <= max_load->freqs[0].hdemand) { + pc = costs[0].power; + goto unlock; + } else if (demand > max_load->freqs[max_load->length - 1].hdemand) { + pc = costs[max_load->length - 1].power; + goto unlock; + } + + first = 0; + last = max_load->length - 1; + mid = (last - first) >> 1; + while (1) { + if (demand <= max_load->freqs[mid].hdemand) + last = mid; + else + first = mid; + + if (last - first == 1) + break; + mid = first + ((last - first) >> 1); + } + + pc = costs[last].power; + +unlock: + rcu_read_unlock(); + + if (idle_cpu(cpu) && rq->cstate) { + total_static_pwr_cost += rq->static_cpu_pwr_cost; + if (rq->cluster->dstate) + total_static_pwr_cost += + rq->cluster->static_cluster_pwr_cost; + } + + return pc + total_static_pwr_cost; + +} + +void inc_nr_big_task(struct hmp_sched_stats *stats, struct task_struct *p) +{ + if (sched_disable_window_stats) + return; + + if (is_big_task(p)) + stats->nr_big_tasks++; +} + +void dec_nr_big_task(struct hmp_sched_stats *stats, struct task_struct *p) +{ + if (sched_disable_window_stats) + return; + + if (is_big_task(p)) + stats->nr_big_tasks--; + + BUG_ON(stats->nr_big_tasks < 0); +} + +void inc_rq_hmp_stats(struct rq *rq, struct task_struct *p, int change_cra) +{ + inc_nr_big_task(&rq->hmp_stats, p); + if (change_cra) + inc_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +void dec_rq_hmp_stats(struct rq *rq, struct task_struct *p, int change_cra) +{ + dec_nr_big_task(&rq->hmp_stats, p); + if (change_cra) + dec_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +static void reset_hmp_stats(struct hmp_sched_stats *stats, int reset_cra) +{ + stats->nr_big_tasks = 0; + if (reset_cra) { + stats->cumulative_runnable_avg = 0; + stats->pred_demands_sum = 0; + } +} + +/* + * Invoked from three places: + * 1) try_to_wake_up() -> ... -> select_best_cpu() + * 2) scheduler_tick() -> ... -> migration_needed() -> select_best_cpu() + * 3) can_migrate_task() + * + * Its safe to de-reference p->grp in first case (since p->pi_lock is held) + * but not in other cases. p->grp is hence freed after a RCU grace period and + * accessed under rcu_read_lock() + */ +int preferred_cluster(struct sched_cluster *cluster, struct task_struct *p) +{ + struct related_thread_group *grp; + int rc = 0; + + rcu_read_lock(); + + grp = task_related_thread_group(p); + if (!grp || !sysctl_sched_enable_colocation) + rc = 1; + else + rc = (grp->preferred_cluster == cluster); + + rcu_read_unlock(); + return rc; +} + +struct sched_cluster *rq_cluster(struct rq *rq) +{ + return rq->cluster; +} + +/* + * reset_cpu_hmp_stats - reset HMP stats for a cpu + * nr_big_tasks + * cumulative_runnable_avg (iff reset_cra is true) + */ +void reset_cpu_hmp_stats(int cpu, int reset_cra) +{ + reset_cfs_rq_hmp_stats(cpu, reset_cra); + reset_hmp_stats(&cpu_rq(cpu)->hmp_stats, reset_cra); +} + +void fixup_nr_big_tasks(struct hmp_sched_stats *stats, + struct task_struct *p, s64 delta) +{ + u64 new_task_load; + u64 old_task_load; + + if (sched_disable_window_stats) + return; + + old_task_load = scale_load_to_cpu(task_load(p), task_cpu(p)); + new_task_load = scale_load_to_cpu(delta + task_load(p), task_cpu(p)); + + if (__is_big_task(p, old_task_load) && !__is_big_task(p, new_task_load)) + stats->nr_big_tasks--; + else if (!__is_big_task(p, old_task_load) && + __is_big_task(p, new_task_load)) + stats->nr_big_tasks++; + + BUG_ON(stats->nr_big_tasks < 0); +} + +/* + * Walk runqueue of cpu and re-initialize 'nr_big_tasks' counters. + */ +static void update_nr_big_tasks(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + struct task_struct *p; + + /* Do not reset cumulative_runnable_avg */ + reset_cpu_hmp_stats(cpu, 0); + + list_for_each_entry(p, &rq->cfs_tasks, se.group_node) + inc_hmp_sched_stats_fair(rq, p, 0); +} + +/* Disable interrupts and grab runqueue lock of all cpus listed in @cpus */ +void pre_big_task_count_change(const struct cpumask *cpus) +{ + int i; + + local_irq_disable(); + + for_each_cpu(i, cpus) + raw_spin_lock(&cpu_rq(i)->lock); +} + +/* + * Reinitialize 'nr_big_tasks' counters on all affected cpus + */ +void post_big_task_count_change(const struct cpumask *cpus) +{ + int i; + + /* Assumes local_irq_disable() keeps online cpumap stable */ + for_each_cpu(i, cpus) + update_nr_big_tasks(i); + + for_each_cpu(i, cpus) + raw_spin_unlock(&cpu_rq(i)->lock); + + local_irq_enable(); +} + +DEFINE_MUTEX(policy_mutex); + +static inline int invalid_value_freq_input(unsigned int *data) +{ + if (data == &sysctl_sched_freq_aggregate) + return !(*data == 0 || *data == 1); + + return 0; +} + +static inline int invalid_value(unsigned int *data) +{ + unsigned int val = *data; + + if (data == &sysctl_sched_ravg_hist_size) + return (val < 2 || val > RAVG_HIST_SIZE_MAX); + + if (data == &sysctl_sched_window_stats_policy) + return val >= WINDOW_STATS_INVALID_POLICY; + + return invalid_value_freq_input(data); +} + +/* + * Handle "atomic" update of sysctl_sched_window_stats_policy, + * sysctl_sched_ravg_hist_size and sched_freq_legacy_mode variables. + */ +int sched_window_update_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, + loff_t *ppos) +{ + int ret; + unsigned int *data = (unsigned int *)table->data; + unsigned int old_val; + + mutex_lock(&policy_mutex); + + old_val = *data; + + ret = proc_dointvec(table, write, buffer, lenp, ppos); + if (ret || !write || (write && (old_val == *data))) + goto done; + + if (invalid_value(data)) { + *data = old_val; + ret = -EINVAL; + goto done; + } + + reset_all_window_stats(0, 0); + +done: + mutex_unlock(&policy_mutex); + + return ret; +} + +/* + * Convert percentage value into absolute form. This will avoid div() operation + * in fast path, to convert task load in percentage scale. + */ +int sched_hmp_proc_update_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, + loff_t *ppos) +{ + int ret; + unsigned int old_val; + unsigned int *data = (unsigned int *)table->data; + int update_min_nice = 0; + + mutex_lock(&policy_mutex); + + old_val = *data; + + ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); + + if (ret || !write) + goto done; + + if (write && (old_val == *data)) + goto done; + + if (sysctl_sched_downmigrate_pct > sysctl_sched_upmigrate_pct) { + *data = old_val; + ret = -EINVAL; + goto done; + } + + /* + * Big task tunable change will need to re-classify tasks on + * runqueue as big and set their counters appropriately. + * sysctl interface affects secondary variables (*_pct), which is then + * "atomically" carried over to the primary variables. Atomic change + * includes taking runqueue lock of all online cpus and re-initiatizing + * their big counter values based on changed criteria. + */ + if ((data == &sysctl_sched_upmigrate_pct || update_min_nice)) { + get_online_cpus(); + pre_big_task_count_change(cpu_online_mask); + } + + set_hmp_defaults(); + + if ((data == &sysctl_sched_upmigrate_pct || update_min_nice)) { + post_big_task_count_change(cpu_online_mask); + put_online_cpus(); + } + +done: + mutex_unlock(&policy_mutex); + return ret; +} + +inline int nr_big_tasks(struct rq *rq) +{ + return rq->hmp_stats.nr_big_tasks; +} + +unsigned int cpu_temp(int cpu) +{ + struct cpu_pwr_stats *per_cpu_info = get_cpu_pwr_stats(); + + if (per_cpu_info) + return per_cpu_info[cpu].temp; + else + return 0; +} + +void init_new_task_load(struct task_struct *p) +{ + int i; + u32 init_load_windows = sched_init_task_load_windows; + u32 init_load_pct = current->init_load_pct; + + p->init_load_pct = 0; + rcu_assign_pointer(p->grp, NULL); + INIT_LIST_HEAD(&p->grp_list); + memset(&p->ravg, 0, sizeof(struct ravg)); + p->cpu_cycles = 0; + + if (init_load_pct) + init_load_windows = div64_u64((u64)init_load_pct * + (u64)sched_ravg_window, 100); + + p->ravg.demand = init_load_windows; + p->ravg.pred_demand = 0; + for (i = 0; i < RAVG_HIST_SIZE_MAX; ++i) + p->ravg.sum_history[i] = init_load_windows; +} + +/* Return task demand in percentage scale */ +unsigned int pct_task_load(struct task_struct *p) +{ + unsigned int load; + + load = div64_u64((u64)task_load(p) * 100, (u64)max_task_load()); + + return load; +} + +/* + * Return total number of tasks "eligible" to run on highest capacity cpu + * + * This is simply nr_big_tasks for cpus which are not of max_capacity and + * nr_running for cpus of max_capacity + */ +unsigned int nr_eligible_big_tasks(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + int nr_big = rq->hmp_stats.nr_big_tasks; + int nr = rq->nr_running; + + if (cpu_max_possible_capacity(cpu) != max_possible_capacity) + return nr_big; + + return nr; +} + +static inline int exiting_task(struct task_struct *p) +{ + return (p->ravg.sum_history[0] == EXITING_TASK_MARKER); +} + +static int __init set_sched_ravg_window(char *str) +{ + unsigned int window_size; + + get_option(&str, &window_size); + + if (window_size < MIN_SCHED_RAVG_WINDOW || + window_size > MAX_SCHED_RAVG_WINDOW) { + WARN_ON(1); + return -EINVAL; + } + + sched_ravg_window = window_size; + return 0; +} + +early_param("sched_ravg_window", set_sched_ravg_window); + +static inline void +update_window_start(struct rq *rq, u64 wallclock) +{ + s64 delta; + int nr_windows; + + delta = wallclock - rq->window_start; + BUG_ON(delta < 0); + if (delta < sched_ravg_window) + return; + + nr_windows = div64_u64(delta, sched_ravg_window); + rq->window_start += (u64)nr_windows * (u64)sched_ravg_window; +} + +#define DIV64_U64_ROUNDUP(X, Y) div64_u64((X) + (Y - 1), Y) + +static inline u64 scale_exec_time(u64 delta, struct rq *rq) +{ + u32 freq; + + freq = cpu_cycles_to_freq(rq->cc.cycles, rq->cc.time); + delta = DIV64_U64_ROUNDUP(delta * freq, max_possible_freq); + delta *= rq->cluster->exec_scale_factor; + delta >>= 10; + + return delta; +} + +static inline int cpu_is_waiting_on_io(struct rq *rq) +{ + if (!sched_io_is_busy) + return 0; + + return atomic_read(&rq->nr_iowait); +} + +/* Does freq_required sufficiently exceed or fall behind cur_freq? */ +static inline int +nearly_same_freq(unsigned int cur_freq, unsigned int freq_required) +{ + int delta = freq_required - cur_freq; + + if (freq_required > cur_freq) + return delta < sysctl_sched_freq_inc_notify; + + delta = -delta; + + return delta < sysctl_sched_freq_dec_notify; +} + +/* Convert busy time to frequency equivalent */ +static inline unsigned int load_to_freq(struct rq *rq, u64 load) +{ + unsigned int freq; + + load = scale_load_to_cpu(load, cpu_of(rq)); + load *= 128; + load = div64_u64(load, max_task_load()); + + freq = load * cpu_max_possible_freq(cpu_of(rq)); + freq /= 128; + + return freq; +} + +static inline struct group_cpu_time * +_group_cpu_time(struct related_thread_group *grp, int cpu); + +/* + * Return load from all related group in given cpu. + * Caller must ensure that related_thread_group_lock is held. + */ +static void _group_load_in_cpu(int cpu, u64 *grp_load, u64 *new_grp_load) +{ + struct related_thread_group *grp; + + for_each_related_thread_group(grp) { + struct group_cpu_time *cpu_time; + + cpu_time = _group_cpu_time(grp, cpu); + *grp_load += cpu_time->prev_runnable_sum; + if (new_grp_load) + *new_grp_load += cpu_time->nt_prev_runnable_sum; + } +} + +/* + * Return load from all related groups in given frequency domain. + * Caller must ensure that related_thread_group_lock is held. + */ +static void group_load_in_freq_domain(struct cpumask *cpus, + u64 *grp_load, u64 *new_grp_load) +{ + struct related_thread_group *grp; + int j; + + for_each_related_thread_group(grp) { + for_each_cpu(j, cpus) { + struct group_cpu_time *cpu_time; + + cpu_time = _group_cpu_time(grp, j); + *grp_load += cpu_time->prev_runnable_sum; + *new_grp_load += cpu_time->nt_prev_runnable_sum; + } + } +} + +/* + * Should scheduler alert governor for changing frequency? + * + * @check_pred - evaluate frequency based on the predictive demand + * @check_groups - add load from all related groups on given cpu + * + * check_groups is set to 1 if a "related" task movement/wakeup is triggering + * the notification check. To avoid "re-aggregation" of demand in such cases, + * we check whether the migrated/woken tasks demand (along with demand from + * existing tasks on the cpu) can be met on target cpu + * + */ + +static int send_notification(struct rq *rq, int check_pred, int check_groups) +{ + unsigned int cur_freq, freq_required; + unsigned long flags; + int rc = 0; + u64 group_load = 0, new_load = 0; + + if (check_pred) { + u64 prev = rq->old_busy_time; + u64 predicted = rq->hmp_stats.pred_demands_sum; + + if (rq->cluster->cur_freq == cpu_max_freq(cpu_of(rq))) + return 0; + + prev = max(prev, rq->old_estimated_time); + if (prev > predicted) + return 0; + + cur_freq = load_to_freq(rq, prev); + freq_required = load_to_freq(rq, predicted); + + if (freq_required < cur_freq + sysctl_sched_pred_alert_freq) + return 0; + } else { + read_lock(&related_thread_group_lock); + /* + * Protect from concurrent update of rq->prev_runnable_sum and + * group cpu load + */ + raw_spin_lock_irqsave(&rq->lock, flags); + if (check_groups) + _group_load_in_cpu(cpu_of(rq), &group_load, NULL); + + new_load = rq->prev_runnable_sum + group_load; + + raw_spin_unlock_irqrestore(&rq->lock, flags); + read_unlock(&related_thread_group_lock); + + cur_freq = load_to_freq(rq, rq->old_busy_time); + freq_required = load_to_freq(rq, new_load); + + if (nearly_same_freq(cur_freq, freq_required)) + return 0; + } + + raw_spin_lock_irqsave(&rq->lock, flags); + if (!rq->cluster->notifier_sent) { + rq->cluster->notifier_sent = 1; + rc = 1; + trace_sched_freq_alert(cpu_of(rq), check_pred, check_groups, rq, + new_load); + } + raw_spin_unlock_irqrestore(&rq->lock, flags); + + return rc; +} + +/* Alert governor if there is a need to change frequency */ +void check_for_freq_change(struct rq *rq, bool check_pred, bool check_groups) +{ + int cpu = cpu_of(rq); + + if (!send_notification(rq, check_pred, check_groups)) + return; + + atomic_notifier_call_chain( + &load_alert_notifier_head, 0, + (void *)(long)cpu); +} + +void notify_migration(int src_cpu, int dest_cpu, bool src_cpu_dead, + struct task_struct *p) +{ + bool check_groups; + + rcu_read_lock(); + check_groups = task_in_related_thread_group(p); + rcu_read_unlock(); + + if (!same_freq_domain(src_cpu, dest_cpu)) { + if (!src_cpu_dead) + check_for_freq_change(cpu_rq(src_cpu), false, + check_groups); + check_for_freq_change(cpu_rq(dest_cpu), false, check_groups); + } else { + check_for_freq_change(cpu_rq(dest_cpu), true, check_groups); + } +} + +static int account_busy_for_cpu_time(struct rq *rq, struct task_struct *p, + u64 irqtime, int event) +{ + if (is_idle_task(p)) { + /* TASK_WAKE && TASK_MIGRATE is not possible on idle task! */ + if (event == PICK_NEXT_TASK) + return 0; + + /* PUT_PREV_TASK, TASK_UPDATE && IRQ_UPDATE are left */ + return irqtime || cpu_is_waiting_on_io(rq); + } + + if (event == TASK_WAKE) + return 0; + + if (event == PUT_PREV_TASK || event == IRQ_UPDATE) + return 1; + + /* + * TASK_UPDATE can be called on sleeping task, when its moved between + * related groups + */ + if (event == TASK_UPDATE) { + if (rq->curr == p) + return 1; + + return p->on_rq ? SCHED_FREQ_ACCOUNT_WAIT_TIME : 0; + } + + /* TASK_MIGRATE, PICK_NEXT_TASK left */ + return SCHED_FREQ_ACCOUNT_WAIT_TIME; +} + +static inline bool is_new_task(struct task_struct *p) +{ + return p->ravg.active_windows < sysctl_sched_new_task_windows; +} + +#define INC_STEP 8 +#define DEC_STEP 2 +#define CONSISTENT_THRES 16 +#define INC_STEP_BIG 16 +/* + * bucket_increase - update the count of all buckets + * + * @buckets: array of buckets tracking busy time of a task + * @idx: the index of bucket to be incremented + * + * Each time a complete window finishes, count of bucket that runtime + * falls in (@idx) is incremented. Counts of all other buckets are + * decayed. The rate of increase and decay could be different based + * on current count in the bucket. + */ +static inline void bucket_increase(u8 *buckets, int idx) +{ + int i, step; + + for (i = 0; i < NUM_BUSY_BUCKETS; i++) { + if (idx != i) { + if (buckets[i] > DEC_STEP) + buckets[i] -= DEC_STEP; + else + buckets[i] = 0; + } else { + step = buckets[i] >= CONSISTENT_THRES ? + INC_STEP_BIG : INC_STEP; + if (buckets[i] > U8_MAX - step) + buckets[i] = U8_MAX; + else + buckets[i] += step; + } + } +} + +static inline int busy_to_bucket(u32 normalized_rt) +{ + int bidx; + + bidx = mult_frac(normalized_rt, NUM_BUSY_BUCKETS, max_task_load()); + bidx = min(bidx, NUM_BUSY_BUCKETS - 1); + + /* + * Combine lowest two buckets. The lowest frequency falls into + * 2nd bucket and thus keep predicting lowest bucket is not + * useful. + */ + if (!bidx) + bidx++; + + return bidx; +} + +static inline u64 +scale_load_to_freq(u64 load, unsigned int src_freq, unsigned int dst_freq) +{ + return div64_u64(load * (u64)src_freq, (u64)dst_freq); +} + +#define HEAVY_TASK_SKIP 2 +#define HEAVY_TASK_SKIP_LIMIT 4 +/* + * get_pred_busy - calculate predicted demand for a task on runqueue + * + * @rq: runqueue of task p + * @p: task whose prediction is being updated + * @start: starting bucket. returned prediction should not be lower than + * this bucket. + * @runtime: runtime of the task. returned prediction should not be lower + * than this runtime. + * Note: @start can be derived from @runtime. It's passed in only to + * avoid duplicated calculation in some cases. + * + * A new predicted busy time is returned for task @p based on @runtime + * passed in. The function searches through buckets that represent busy + * time equal to or bigger than @runtime and attempts to find the bucket to + * to use for prediction. Once found, it searches through historical busy + * time and returns the latest that falls into the bucket. If no such busy + * time exists, it returns the medium of that bucket. + */ +static u32 get_pred_busy(struct rq *rq, struct task_struct *p, + int start, u32 runtime) +{ + int i; + u8 *buckets = p->ravg.busy_buckets; + u32 *hist = p->ravg.sum_history; + u32 dmin, dmax; + u64 cur_freq_runtime = 0; + int first = NUM_BUSY_BUCKETS, final, skip_to; + u32 ret = runtime; + + /* skip prediction for new tasks due to lack of history */ + if (unlikely(is_new_task(p))) + goto out; + + /* find minimal bucket index to pick */ + for (i = start; i < NUM_BUSY_BUCKETS; i++) { + if (buckets[i]) { + first = i; + break; + } + } + /* if no higher buckets are filled, predict runtime */ + if (first >= NUM_BUSY_BUCKETS) + goto out; + + /* compute the bucket for prediction */ + final = first; + if (first < HEAVY_TASK_SKIP_LIMIT) { + /* compute runtime at current CPU frequency */ + cur_freq_runtime = mult_frac(runtime, max_possible_efficiency, + rq->cluster->efficiency); + cur_freq_runtime = scale_load_to_freq(cur_freq_runtime, + max_possible_freq, rq->cluster->cur_freq); + /* + * if the task runs for majority of the window, try to + * pick higher buckets. + */ + if (cur_freq_runtime >= sched_major_task_runtime) { + int next = NUM_BUSY_BUCKETS; + /* + * if there is a higher bucket that's consistently + * hit, don't jump beyond that. + */ + for (i = start + 1; i <= HEAVY_TASK_SKIP_LIMIT && + i < NUM_BUSY_BUCKETS; i++) { + if (buckets[i] > CONSISTENT_THRES) { + next = i; + break; + } + } + skip_to = min(next, start + HEAVY_TASK_SKIP); + /* don't jump beyond HEAVY_TASK_SKIP_LIMIT */ + skip_to = min(HEAVY_TASK_SKIP_LIMIT, skip_to); + /* don't go below first non-empty bucket, if any */ + final = max(first, skip_to); + } + } + + /* determine demand range for the predicted bucket */ + if (final < 2) { + /* lowest two buckets are combined */ + dmin = 0; + final = 1; + } else { + dmin = mult_frac(final, max_task_load(), NUM_BUSY_BUCKETS); + } + dmax = mult_frac(final + 1, max_task_load(), NUM_BUSY_BUCKETS); + + /* + * search through runtime history and return first runtime that falls + * into the range of predicted bucket. + */ + for (i = 0; i < sched_ravg_hist_size; i++) { + if (hist[i] >= dmin && hist[i] < dmax) { + ret = hist[i]; + break; + } + } + /* no historical runtime within bucket found, use average of the bin */ + if (ret < dmin) + ret = (dmin + dmax) / 2; + /* + * when updating in middle of a window, runtime could be higher + * than all recorded history. Always predict at least runtime. + */ + ret = max(runtime, ret); +out: + trace_sched_update_pred_demand(rq, p, runtime, + mult_frac((unsigned int)cur_freq_runtime, 100, + sched_ravg_window), ret); + return ret; +} + +static inline u32 calc_pred_demand(struct rq *rq, struct task_struct *p) +{ + if (p->ravg.pred_demand >= p->ravg.curr_window) + return p->ravg.pred_demand; + + return get_pred_busy(rq, p, busy_to_bucket(p->ravg.curr_window), + p->ravg.curr_window); +} + +/* + * predictive demand of a task is calculated at the window roll-over. + * if the task current window busy time exceeds the predicted + * demand, update it here to reflect the task needs. + */ +void update_task_pred_demand(struct rq *rq, struct task_struct *p, int event) +{ + u32 new, old; + + if (is_idle_task(p) || exiting_task(p)) + return; + + if (event != PUT_PREV_TASK && event != TASK_UPDATE && + (!SCHED_FREQ_ACCOUNT_WAIT_TIME || + (event != TASK_MIGRATE && + event != PICK_NEXT_TASK))) + return; + + /* + * TASK_UPDATE can be called on sleeping task, when its moved between + * related groups + */ + if (event == TASK_UPDATE) { + if (!p->on_rq && !SCHED_FREQ_ACCOUNT_WAIT_TIME) + return; + } + + new = calc_pred_demand(rq, p); + old = p->ravg.pred_demand; + + if (old >= new) + return; + + if (task_on_rq_queued(p) && (!task_has_dl_policy(p) || + !p->dl.dl_throttled)) + p->sched_class->fixup_hmp_sched_stats(rq, p, + p->ravg.demand, + new); + + p->ravg.pred_demand = new; +} + +/* + * Account cpu activity in its busy time counters (rq->curr/prev_runnable_sum) + */ +static void update_cpu_busy_time(struct task_struct *p, struct rq *rq, + int event, u64 wallclock, u64 irqtime) +{ + int new_window, full_window = 0; + int p_is_curr_task = (p == rq->curr); + u64 mark_start = p->ravg.mark_start; + u64 window_start = rq->window_start; + u32 window_size = sched_ravg_window; + u64 delta; + u64 *curr_runnable_sum = &rq->curr_runnable_sum; + u64 *prev_runnable_sum = &rq->prev_runnable_sum; + u64 *nt_curr_runnable_sum = &rq->nt_curr_runnable_sum; + u64 *nt_prev_runnable_sum = &rq->nt_prev_runnable_sum; + int flip_counters = 0; + int prev_sum_reset = 0; + bool new_task; + struct related_thread_group *grp; + + new_window = mark_start < window_start; + if (new_window) { + full_window = (window_start - mark_start) >= window_size; + if (p->ravg.active_windows < USHRT_MAX) + p->ravg.active_windows++; + } + + new_task = is_new_task(p); + + grp = p->grp; + if (grp && sched_freq_aggregate) { + /* cpu_time protected by rq_lock */ + struct group_cpu_time *cpu_time = + _group_cpu_time(grp, cpu_of(rq)); + + curr_runnable_sum = &cpu_time->curr_runnable_sum; + prev_runnable_sum = &cpu_time->prev_runnable_sum; + + nt_curr_runnable_sum = &cpu_time->nt_curr_runnable_sum; + nt_prev_runnable_sum = &cpu_time->nt_prev_runnable_sum; + + if (cpu_time->window_start != rq->window_start) { + int nr_windows; + + delta = rq->window_start - cpu_time->window_start; + nr_windows = div64_u64(delta, window_size); + if (nr_windows > 1) + prev_sum_reset = 1; + + cpu_time->window_start = rq->window_start; + flip_counters = 1; + } + + if (p_is_curr_task && new_window) { + u64 curr_sum = rq->curr_runnable_sum; + u64 nt_curr_sum = rq->nt_curr_runnable_sum; + + if (full_window) + curr_sum = nt_curr_sum = 0; + + rq->prev_runnable_sum = curr_sum; + rq->nt_prev_runnable_sum = nt_curr_sum; + + rq->curr_runnable_sum = 0; + rq->nt_curr_runnable_sum = 0; + } + } else { + if (p_is_curr_task && new_window) { + flip_counters = 1; + if (full_window) + prev_sum_reset = 1; + } + } + + /* + * Handle per-task window rollover. We don't care about the idle + * task or exiting tasks. + */ + if (new_window && !is_idle_task(p) && !exiting_task(p)) { + u32 curr_window = 0; + + if (!full_window) + curr_window = p->ravg.curr_window; + + p->ravg.prev_window = curr_window; + p->ravg.curr_window = 0; + } + + if (flip_counters) { + u64 curr_sum = *curr_runnable_sum; + u64 nt_curr_sum = *nt_curr_runnable_sum; + + if (prev_sum_reset) + curr_sum = nt_curr_sum = 0; + + *prev_runnable_sum = curr_sum; + *nt_prev_runnable_sum = nt_curr_sum; + + *curr_runnable_sum = 0; + *nt_curr_runnable_sum = 0; + } + + if (!account_busy_for_cpu_time(rq, p, irqtime, event)) { + /* + * account_busy_for_cpu_time() = 0, so no update to the + * task's current window needs to be made. This could be + * for example + * + * - a wakeup event on a task within the current + * window (!new_window below, no action required), + * - switching to a new task from idle (PICK_NEXT_TASK) + * in a new window where irqtime is 0 and we aren't + * waiting on IO + */ + + if (!new_window) + return; + + /* + * A new window has started. The RQ demand must be rolled + * over if p is the current task. + */ + if (p_is_curr_task) { + /* p is idle task */ + BUG_ON(p != rq->idle); + } + + return; + } + + if (!new_window) { + /* + * account_busy_for_cpu_time() = 1 so busy time needs + * to be accounted to the current window. No rollover + * since we didn't start a new window. An example of this is + * when a task starts execution and then sleeps within the + * same window. + */ + + if (!irqtime || !is_idle_task(p) || cpu_is_waiting_on_io(rq)) + delta = wallclock - mark_start; + else + delta = irqtime; + delta = scale_exec_time(delta, rq); + *curr_runnable_sum += delta; + if (new_task) + *nt_curr_runnable_sum += delta; + + if (!is_idle_task(p) && !exiting_task(p)) + p->ravg.curr_window += delta; + + return; + } + + if (!p_is_curr_task) { + /* + * account_busy_for_cpu_time() = 1 so busy time needs + * to be accounted to the current window. A new window + * has also started, but p is not the current task, so the + * window is not rolled over - just split up and account + * as necessary into curr and prev. The window is only + * rolled over when a new window is processed for the current + * task. + * + * Irqtime can't be accounted by a task that isn't the + * currently running task. + */ + + if (!full_window) { + /* + * A full window hasn't elapsed, account partial + * contribution to previous completed window. + */ + delta = scale_exec_time(window_start - mark_start, rq); + if (!exiting_task(p)) + p->ravg.prev_window += delta; + } else { + /* + * Since at least one full window has elapsed, + * the contribution to the previous window is the + * full window (window_size). + */ + delta = scale_exec_time(window_size, rq); + if (!exiting_task(p)) + p->ravg.prev_window = delta; + } + + *prev_runnable_sum += delta; + if (new_task) + *nt_prev_runnable_sum += delta; + + /* Account piece of busy time in the current window. */ + delta = scale_exec_time(wallclock - window_start, rq); + *curr_runnable_sum += delta; + if (new_task) + *nt_curr_runnable_sum += delta; + + if (!exiting_task(p)) + p->ravg.curr_window = delta; + + return; + } + + if (!irqtime || !is_idle_task(p) || cpu_is_waiting_on_io(rq)) { + /* + * account_busy_for_cpu_time() = 1 so busy time needs + * to be accounted to the current window. A new window + * has started and p is the current task so rollover is + * needed. If any of these three above conditions are true + * then this busy time can't be accounted as irqtime. + * + * Busy time for the idle task or exiting tasks need not + * be accounted. + * + * An example of this would be a task that starts execution + * and then sleeps once a new window has begun. + */ + + if (!full_window) { + /* + * A full window hasn't elapsed, account partial + * contribution to previous completed window. + */ + delta = scale_exec_time(window_start - mark_start, rq); + if (!is_idle_task(p) && !exiting_task(p)) + p->ravg.prev_window += delta; + } else { + /* + * Since at least one full window has elapsed, + * the contribution to the previous window is the + * full window (window_size). + */ + delta = scale_exec_time(window_size, rq); + if (!is_idle_task(p) && !exiting_task(p)) + p->ravg.prev_window = delta; + } + + /* + * Rollover is done here by overwriting the values in + * prev_runnable_sum and curr_runnable_sum. + */ + *prev_runnable_sum += delta; + if (new_task) + *nt_prev_runnable_sum += delta; + + /* Account piece of busy time in the current window. */ + delta = scale_exec_time(wallclock - window_start, rq); + *curr_runnable_sum += delta; + if (new_task) + *nt_curr_runnable_sum += delta; + + if (!is_idle_task(p) && !exiting_task(p)) + p->ravg.curr_window = delta; + + return; + } + + if (irqtime) { + /* + * account_busy_for_cpu_time() = 1 so busy time needs + * to be accounted to the current window. A new window + * has started and p is the current task so rollover is + * needed. The current task must be the idle task because + * irqtime is not accounted for any other task. + * + * Irqtime will be accounted each time we process IRQ activity + * after a period of idleness, so we know the IRQ busy time + * started at wallclock - irqtime. + */ + + BUG_ON(!is_idle_task(p)); + mark_start = wallclock - irqtime; + + /* + * Roll window over. If IRQ busy time was just in the current + * window then that is all that need be accounted. + */ + if (mark_start > window_start) { + *curr_runnable_sum = scale_exec_time(irqtime, rq); + return; + } + + /* + * The IRQ busy time spanned multiple windows. Process the + * busy time preceding the current window start first. + */ + delta = window_start - mark_start; + if (delta > window_size) + delta = window_size; + delta = scale_exec_time(delta, rq); + *prev_runnable_sum += delta; + + /* Process the remaining IRQ busy time in the current window. */ + delta = wallclock - window_start; + rq->curr_runnable_sum = scale_exec_time(delta, rq); + + return; + } + + BUG(); +} + +static inline u32 predict_and_update_buckets(struct rq *rq, + struct task_struct *p, u32 runtime) { + + int bidx; + u32 pred_demand; + + bidx = busy_to_bucket(runtime); + pred_demand = get_pred_busy(rq, p, bidx, runtime); + bucket_increase(p->ravg.busy_buckets, bidx); + + return pred_demand; +} + +static void update_task_cpu_cycles(struct task_struct *p, int cpu) +{ + if (use_cycle_counter) + p->cpu_cycles = cpu_cycle_counter_cb.get_cpu_cycle_counter(cpu); +} + +static void +update_task_rq_cpu_cycles(struct task_struct *p, struct rq *rq, int event, + u64 wallclock, u64 irqtime) +{ + u64 cur_cycles; + int cpu = cpu_of(rq); + + lockdep_assert_held(&rq->lock); + + if (!use_cycle_counter) { + rq->cc.cycles = cpu_cur_freq(cpu); + rq->cc.time = 1; + return; + } + + cur_cycles = cpu_cycle_counter_cb.get_cpu_cycle_counter(cpu); + + /* + * If current task is idle task and irqtime == 0 CPU was + * indeed idle and probably its cycle counter was not + * increasing. We still need estimatied CPU frequency + * for IO wait time accounting. Use the previously + * calculated frequency in such a case. + */ + if (!is_idle_task(rq->curr) || irqtime) { + if (unlikely(cur_cycles < p->cpu_cycles)) + rq->cc.cycles = cur_cycles + (U64_MAX - p->cpu_cycles); + else + rq->cc.cycles = cur_cycles - p->cpu_cycles; + rq->cc.cycles = rq->cc.cycles * NSEC_PER_MSEC; + + if (event == IRQ_UPDATE && is_idle_task(p)) + /* + * Time between mark_start of idle task and IRQ handler + * entry time is CPU cycle counter stall period. + * Upon IRQ handler entry sched_account_irqstart() + * replenishes idle task's cpu cycle counter so + * rq->cc.cycles now represents increased cycles during + * IRQ handler rather than time between idle entry and + * IRQ exit. Thus use irqtime as time delta. + */ + rq->cc.time = irqtime; + else + rq->cc.time = wallclock - p->ravg.mark_start; + BUG_ON((s64)rq->cc.time < 0); + } + + p->cpu_cycles = cur_cycles; + + trace_sched_get_task_cpu_cycles(cpu, event, rq->cc.cycles, rq->cc.time); +} + +static int account_busy_for_task_demand(struct task_struct *p, int event) +{ + /* + * No need to bother updating task demand for exiting tasks + * or the idle task. + */ + if (exiting_task(p) || is_idle_task(p)) + return 0; + + /* + * When a task is waking up it is completing a segment of non-busy + * time. Likewise, if wait time is not treated as busy time, then + * when a task begins to run or is migrated, it is not running and + * is completing a segment of non-busy time. + */ + if (event == TASK_WAKE || (!SCHED_ACCOUNT_WAIT_TIME && + (event == PICK_NEXT_TASK || event == TASK_MIGRATE))) + return 0; + + return 1; +} + +/* + * Called when new window is starting for a task, to record cpu usage over + * recently concluded window(s). Normally 'samples' should be 1. It can be > 1 + * when, say, a real-time task runs without preemption for several windows at a + * stretch. + */ +static void update_history(struct rq *rq, struct task_struct *p, + u32 runtime, int samples, int event) +{ + u32 *hist = &p->ravg.sum_history[0]; + int ridx, widx; + u32 max = 0, avg, demand, pred_demand; + u64 sum = 0; + + /* Ignore windows where task had no activity */ + if (!runtime || is_idle_task(p) || exiting_task(p) || !samples) + goto done; + + /* Push new 'runtime' value onto stack */ + widx = sched_ravg_hist_size - 1; + ridx = widx - samples; + for (; ridx >= 0; --widx, --ridx) { + hist[widx] = hist[ridx]; + sum += hist[widx]; + if (hist[widx] > max) + max = hist[widx]; + } + + for (widx = 0; widx < samples && widx < sched_ravg_hist_size; widx++) { + hist[widx] = runtime; + sum += hist[widx]; + if (hist[widx] > max) + max = hist[widx]; + } + + p->ravg.sum = 0; + + if (sched_window_stats_policy == WINDOW_STATS_RECENT) { + demand = runtime; + } else if (sched_window_stats_policy == WINDOW_STATS_MAX) { + demand = max; + } else { + avg = div64_u64(sum, sched_ravg_hist_size); + if (sched_window_stats_policy == WINDOW_STATS_AVG) + demand = avg; + else + demand = max(avg, runtime); + } + pred_demand = predict_and_update_buckets(rq, p, runtime); + + /* + * A throttled deadline sched class task gets dequeued without + * changing p->on_rq. Since the dequeue decrements hmp stats + * avoid decrementing it here again. + */ + if (task_on_rq_queued(p) && (!task_has_dl_policy(p) || + !p->dl.dl_throttled)) + p->sched_class->fixup_hmp_sched_stats(rq, p, demand, + pred_demand); + + p->ravg.demand = demand; + p->ravg.pred_demand = pred_demand; + +done: + trace_sched_update_history(rq, p, runtime, samples, event); +} + +static void add_to_task_demand(struct rq *rq, struct task_struct *p, u64 delta) +{ + delta = scale_exec_time(delta, rq); + p->ravg.sum += delta; + if (unlikely(p->ravg.sum > sched_ravg_window)) + p->ravg.sum = sched_ravg_window; +} + +/* + * Account cpu demand of task and/or update task's cpu demand history + * + * ms = p->ravg.mark_start; + * wc = wallclock + * ws = rq->window_start + * + * Three possibilities: + * + * a) Task event is contained within one window. + * window_start < mark_start < wallclock + * + * ws ms wc + * | | | + * V V V + * |---------------| + * + * In this case, p->ravg.sum is updated *iff* event is appropriate + * (ex: event == PUT_PREV_TASK) + * + * b) Task event spans two windows. + * mark_start < window_start < wallclock + * + * ms ws wc + * | | | + * V V V + * -----|------------------- + * + * In this case, p->ravg.sum is updated with (ws - ms) *iff* event + * is appropriate, then a new window sample is recorded followed + * by p->ravg.sum being set to (wc - ws) *iff* event is appropriate. + * + * c) Task event spans more than two windows. + * + * ms ws_tmp ws wc + * | | | | + * V V V V + * ---|-------|-------|-------|-------|------ + * | | + * |<------ nr_full_windows ------>| + * + * In this case, p->ravg.sum is updated with (ws_tmp - ms) first *iff* + * event is appropriate, window sample of p->ravg.sum is recorded, + * 'nr_full_window' samples of window_size is also recorded *iff* + * event is appropriate and finally p->ravg.sum is set to (wc - ws) + * *iff* event is appropriate. + * + * IMPORTANT : Leave p->ravg.mark_start unchanged, as update_cpu_busy_time() + * depends on it! + */ +static void update_task_demand(struct task_struct *p, struct rq *rq, + int event, u64 wallclock) +{ + u64 mark_start = p->ravg.mark_start; + u64 delta, window_start = rq->window_start; + int new_window, nr_full_windows; + u32 window_size = sched_ravg_window; + + new_window = mark_start < window_start; + if (!account_busy_for_task_demand(p, event)) { + if (new_window) + /* + * If the time accounted isn't being accounted as + * busy time, and a new window started, only the + * previous window need be closed out with the + * pre-existing demand. Multiple windows may have + * elapsed, but since empty windows are dropped, + * it is not necessary to account those. + */ + update_history(rq, p, p->ravg.sum, 1, event); + return; + } + + if (!new_window) { + /* + * The simple case - busy time contained within the existing + * window. + */ + add_to_task_demand(rq, p, wallclock - mark_start); + return; + } + + /* + * Busy time spans at least two windows. Temporarily rewind + * window_start to first window boundary after mark_start. + */ + delta = window_start - mark_start; + nr_full_windows = div64_u64(delta, window_size); + window_start -= (u64)nr_full_windows * (u64)window_size; + + /* Process (window_start - mark_start) first */ + add_to_task_demand(rq, p, window_start - mark_start); + + /* Push new sample(s) into task's demand history */ + update_history(rq, p, p->ravg.sum, 1, event); + if (nr_full_windows) + update_history(rq, p, scale_exec_time(window_size, rq), + nr_full_windows, event); + + /* + * Roll window_start back to current to process any remainder + * in current window. + */ + window_start += (u64)nr_full_windows * (u64)window_size; + + /* Process (wallclock - window_start) next */ + mark_start = window_start; + add_to_task_demand(rq, p, wallclock - mark_start); +} + +/* Reflect task activity on its demand and cpu's busy time statistics */ +void update_task_ravg(struct task_struct *p, struct rq *rq, int event, + u64 wallclock, u64 irqtime) +{ + if (!rq->window_start || sched_disable_window_stats) + return; + + lockdep_assert_held(&rq->lock); + + update_window_start(rq, wallclock); + + if (!p->ravg.mark_start) { + update_task_cpu_cycles(p, cpu_of(rq)); + goto done; + } + + update_task_rq_cpu_cycles(p, rq, event, wallclock, irqtime); + update_task_demand(p, rq, event, wallclock); + update_cpu_busy_time(p, rq, event, wallclock, irqtime); + update_task_pred_demand(rq, p, event); +done: + trace_sched_update_task_ravg(p, rq, event, wallclock, irqtime, + rq->cc.cycles, rq->cc.time, + _group_cpu_time(p->grp, cpu_of(rq))); + + p->ravg.mark_start = wallclock; +} + +void sched_account_irqtime(int cpu, struct task_struct *curr, + u64 delta, u64 wallclock) +{ + struct rq *rq = cpu_rq(cpu); + unsigned long flags, nr_windows; + u64 cur_jiffies_ts; + + raw_spin_lock_irqsave(&rq->lock, flags); + + /* + * cputime (wallclock) uses sched_clock so use the same here for + * consistency. + */ + delta += sched_clock() - wallclock; + cur_jiffies_ts = get_jiffies_64(); + + if (is_idle_task(curr)) + update_task_ravg(curr, rq, IRQ_UPDATE, sched_ktime_clock(), + delta); + + nr_windows = cur_jiffies_ts - rq->irqload_ts; + + if (nr_windows) { + if (nr_windows < 10) { + /* Decay CPU's irqload by 3/4 for each window. */ + rq->avg_irqload *= (3 * nr_windows); + rq->avg_irqload = div64_u64(rq->avg_irqload, + 4 * nr_windows); + } else { + rq->avg_irqload = 0; + } + rq->avg_irqload += rq->cur_irqload; + rq->cur_irqload = 0; + } + + rq->cur_irqload += delta; + rq->irqload_ts = cur_jiffies_ts; + raw_spin_unlock_irqrestore(&rq->lock, flags); +} + +void sched_account_irqstart(int cpu, struct task_struct *curr, u64 wallclock) +{ + struct rq *rq = cpu_rq(cpu); + + if (!rq->window_start || sched_disable_window_stats) + return; + + if (is_idle_task(curr)) { + /* We're here without rq->lock held, IRQ disabled */ + raw_spin_lock(&rq->lock); + update_task_cpu_cycles(curr, cpu); + raw_spin_unlock(&rq->lock); + } +} + +void reset_task_stats(struct task_struct *p) +{ + u32 sum = 0; + + if (exiting_task(p)) + sum = EXITING_TASK_MARKER; + + memset(&p->ravg, 0, sizeof(struct ravg)); + /* Retain EXITING_TASK marker */ + p->ravg.sum_history[0] = sum; +} + +void mark_task_starting(struct task_struct *p) +{ + u64 wallclock; + struct rq *rq = task_rq(p); + + if (!rq->window_start || sched_disable_window_stats) { + reset_task_stats(p); + return; + } + + wallclock = sched_ktime_clock(); + p->ravg.mark_start = p->last_wake_ts = wallclock; + p->last_cpu_selected_ts = wallclock; + p->last_switch_out_ts = 0; + update_task_cpu_cycles(p, cpu_of(rq)); +} + +void set_window_start(struct rq *rq) +{ + int cpu = cpu_of(rq); + struct rq *sync_rq = cpu_rq(sync_cpu); + + if (rq->window_start) + return; + + if (cpu == sync_cpu) { + rq->window_start = sched_ktime_clock(); + } else { + raw_spin_unlock(&rq->lock); + double_rq_lock(rq, sync_rq); + rq->window_start = cpu_rq(sync_cpu)->window_start; + rq->curr_runnable_sum = rq->prev_runnable_sum = 0; + rq->nt_curr_runnable_sum = rq->nt_prev_runnable_sum = 0; + raw_spin_unlock(&sync_rq->lock); + } + + rq->curr->ravg.mark_start = rq->window_start; +} + +void migrate_sync_cpu(int cpu) +{ + if (cpu == sync_cpu) + sync_cpu = smp_processor_id(); +} + +static void reset_all_task_stats(void) +{ + struct task_struct *g, *p; + + read_lock(&tasklist_lock); + do_each_thread(g, p) { + reset_task_stats(p); + } while_each_thread(g, p); + read_unlock(&tasklist_lock); +} + +static void disable_window_stats(void) +{ + unsigned long flags; + int i; + + local_irq_save(flags); + for_each_possible_cpu(i) + raw_spin_lock(&cpu_rq(i)->lock); + + sched_disable_window_stats = 1; + + for_each_possible_cpu(i) + raw_spin_unlock(&cpu_rq(i)->lock); + + local_irq_restore(flags); +} + +/* Called with all cpu's rq->lock held */ +static void enable_window_stats(void) +{ + sched_disable_window_stats = 0; + +} + +enum reset_reason_code { + WINDOW_CHANGE, + POLICY_CHANGE, + HIST_SIZE_CHANGE, + FREQ_AGGREGATE_CHANGE, +}; + +const char *sched_window_reset_reasons[] = { + "WINDOW_CHANGE", + "POLICY_CHANGE", + "HIST_SIZE_CHANGE", +}; + +/* Called with IRQs enabled */ +void reset_all_window_stats(u64 window_start, unsigned int window_size) +{ + int cpu; + unsigned long flags; + u64 start_ts = sched_ktime_clock(); + int reason = WINDOW_CHANGE; + unsigned int old = 0, new = 0; + struct related_thread_group *grp; + + disable_window_stats(); + + reset_all_task_stats(); + + local_irq_save(flags); + + read_lock(&related_thread_group_lock); + + for_each_possible_cpu(cpu) + raw_spin_lock(&cpu_rq(cpu)->lock); + + list_for_each_entry(grp, &related_thread_groups, list) { + int j; + + for_each_possible_cpu(j) { + struct group_cpu_time *cpu_time; + /* Protected by rq lock */ + cpu_time = _group_cpu_time(grp, j); + memset(cpu_time, 0, sizeof(struct group_cpu_time)); + if (window_start) + cpu_time->window_start = window_start; + } + } + + if (window_size) { + sched_ravg_window = window_size * TICK_NSEC; + set_hmp_defaults(); + } + + enable_window_stats(); + + for_each_possible_cpu(cpu) { + struct rq *rq = cpu_rq(cpu); + + if (window_start) + rq->window_start = window_start; + rq->curr_runnable_sum = rq->prev_runnable_sum = 0; + rq->nt_curr_runnable_sum = rq->nt_prev_runnable_sum = 0; + reset_cpu_hmp_stats(cpu, 1); + } + + if (sched_window_stats_policy != sysctl_sched_window_stats_policy) { + reason = POLICY_CHANGE; + old = sched_window_stats_policy; + new = sysctl_sched_window_stats_policy; + sched_window_stats_policy = sysctl_sched_window_stats_policy; + } else if (sched_ravg_hist_size != sysctl_sched_ravg_hist_size) { + reason = HIST_SIZE_CHANGE; + old = sched_ravg_hist_size; + new = sysctl_sched_ravg_hist_size; + sched_ravg_hist_size = sysctl_sched_ravg_hist_size; + } else if (sched_freq_aggregate != + sysctl_sched_freq_aggregate) { + reason = FREQ_AGGREGATE_CHANGE; + old = sched_freq_aggregate; + new = sysctl_sched_freq_aggregate; + sched_freq_aggregate = sysctl_sched_freq_aggregate; + } + + for_each_possible_cpu(cpu) + raw_spin_unlock(&cpu_rq(cpu)->lock); + + read_unlock(&related_thread_group_lock); + + local_irq_restore(flags); + + trace_sched_reset_all_window_stats(window_start, window_size, + sched_ktime_clock() - start_ts, reason, old, new); +} + +static inline void +sync_window_start(struct rq *rq, struct group_cpu_time *cpu_time); + +void sched_get_cpus_busy(struct sched_load *busy, + const struct cpumask *query_cpus) +{ + unsigned long flags; + struct rq *rq; + const int cpus = cpumask_weight(query_cpus); + u64 load[cpus], group_load[cpus]; + u64 nload[cpus], ngload[cpus]; + u64 pload[cpus]; + unsigned int cur_freq[cpus], max_freq[cpus]; + int notifier_sent = 0; + int early_detection[cpus]; + int cpu, i = 0; + unsigned int window_size; + u64 max_prev_sum = 0; + int max_busy_cpu = cpumask_first(query_cpus); + struct related_thread_group *grp; + u64 total_group_load = 0, total_ngload = 0; + bool aggregate_load = false; + + if (unlikely(cpus == 0)) + return; + + /* + * This function could be called in timer context, and the + * current task may have been executing for a long time. Ensure + * that the window stats are current by doing an update. + */ + read_lock(&related_thread_group_lock); + + local_irq_save(flags); + for_each_cpu(cpu, query_cpus) + raw_spin_lock(&cpu_rq(cpu)->lock); + + window_size = sched_ravg_window; + + for_each_cpu(cpu, query_cpus) { + rq = cpu_rq(cpu); + + update_task_ravg(rq->curr, rq, TASK_UPDATE, sched_ktime_clock(), + 0); + cur_freq[i] = cpu_cycles_to_freq(rq->cc.cycles, rq->cc.time); + + load[i] = rq->old_busy_time = rq->prev_runnable_sum; + nload[i] = rq->nt_prev_runnable_sum; + pload[i] = rq->hmp_stats.pred_demands_sum; + rq->old_estimated_time = pload[i]; + + if (load[i] > max_prev_sum) { + max_prev_sum = load[i]; + max_busy_cpu = cpu; + } + + /* + * sched_get_cpus_busy() is called for all CPUs in a + * frequency domain. So the notifier_sent flag per + * cluster works even when a frequency domain spans + * more than 1 cluster. + */ + if (rq->cluster->notifier_sent) { + notifier_sent = 1; + rq->cluster->notifier_sent = 0; + } + early_detection[i] = (rq->ed_task != NULL); + cur_freq[i] = cpu_cur_freq(cpu); + max_freq[i] = cpu_max_freq(cpu); + i++; + } + + for_each_related_thread_group(grp) { + for_each_cpu(cpu, query_cpus) { + /* Protected by rq_lock */ + struct group_cpu_time *cpu_time = + _group_cpu_time(grp, cpu); + sync_window_start(cpu_rq(cpu), cpu_time); + } + } + + group_load_in_freq_domain( + &cpu_rq(max_busy_cpu)->freq_domain_cpumask, + &total_group_load, &total_ngload); + aggregate_load = !!(total_group_load > sched_freq_aggregate_threshold); + + i = 0; + for_each_cpu(cpu, query_cpus) { + group_load[i] = 0; + ngload[i] = 0; + + if (early_detection[i]) + goto skip_early; + + rq = cpu_rq(cpu); + if (aggregate_load) { + if (cpu == max_busy_cpu) { + group_load[i] = total_group_load; + ngload[i] = total_ngload; + } + } else { + _group_load_in_cpu(cpu, &group_load[i], &ngload[i]); + } + + load[i] += group_load[i]; + nload[i] += ngload[i]; + /* + * Scale load in reference to cluster max_possible_freq. + * + * Note that scale_load_to_cpu() scales load in reference to + * the cluster max_freq. + */ + load[i] = scale_load_to_cpu(load[i], cpu); + nload[i] = scale_load_to_cpu(nload[i], cpu); + pload[i] = scale_load_to_cpu(pload[i], cpu); +skip_early: + i++; + } + + for_each_cpu(cpu, query_cpus) + raw_spin_unlock(&(cpu_rq(cpu))->lock); + local_irq_restore(flags); + + read_unlock(&related_thread_group_lock); + + i = 0; + for_each_cpu(cpu, query_cpus) { + rq = cpu_rq(cpu); + + if (early_detection[i]) { + busy[i].prev_load = div64_u64(sched_ravg_window, + NSEC_PER_USEC); + busy[i].new_task_load = 0; + goto exit_early; + } + + /* + * When the load aggregation is controlled by + * sched_freq_aggregate_threshold, allow reporting loads + * greater than 100 @ Fcur to ramp up the frequency + * faster. + */ + if (notifier_sent || (aggregate_load && + sched_freq_aggregate_threshold)) { + load[i] = scale_load_to_freq(load[i], max_freq[i], + cpu_max_possible_freq(cpu)); + nload[i] = scale_load_to_freq(nload[i], max_freq[i], + cpu_max_possible_freq(cpu)); + } else { + load[i] = scale_load_to_freq(load[i], max_freq[i], + cur_freq[i]); + nload[i] = scale_load_to_freq(nload[i], max_freq[i], + cur_freq[i]); + if (load[i] > window_size) + load[i] = window_size; + if (nload[i] > window_size) + nload[i] = window_size; + + load[i] = scale_load_to_freq(load[i], cur_freq[i], + cpu_max_possible_freq(cpu)); + nload[i] = scale_load_to_freq(nload[i], cur_freq[i], + cpu_max_possible_freq(cpu)); + } + pload[i] = scale_load_to_freq(pload[i], max_freq[i], + rq->cluster->max_possible_freq); + + busy[i].prev_load = div64_u64(load[i], NSEC_PER_USEC); + busy[i].new_task_load = div64_u64(nload[i], NSEC_PER_USEC); + busy[i].predicted_load = div64_u64(pload[i], NSEC_PER_USEC); + +exit_early: + trace_sched_get_busy(cpu, busy[i].prev_load, + busy[i].new_task_load, + busy[i].predicted_load, + early_detection[i]); + i++; + } +} + +void sched_set_io_is_busy(int val) +{ + sched_io_is_busy = val; +} + +int sched_set_window(u64 window_start, unsigned int window_size) +{ + u64 now, cur_jiffies, jiffy_ktime_ns; + s64 ws; + unsigned long flags; + + if (window_size * TICK_NSEC < MIN_SCHED_RAVG_WINDOW) + return -EINVAL; + + mutex_lock(&policy_mutex); + + /* + * Get a consistent view of ktime, jiffies, and the time + * since the last jiffy (based on last_jiffies_update). + */ + local_irq_save(flags); + cur_jiffies = jiffy_to_ktime_ns(&now, &jiffy_ktime_ns); + local_irq_restore(flags); + + /* translate window_start from jiffies to nanoseconds */ + ws = (window_start - cur_jiffies); /* jiffy difference */ + ws *= TICK_NSEC; + ws += jiffy_ktime_ns; + + /* + * Roll back calculated window start so that it is in + * the past (window stats must have a current window). + */ + while (ws > now) + ws -= (window_size * TICK_NSEC); + + BUG_ON(sched_ktime_clock() < ws); + + reset_all_window_stats(ws, window_size); + + sched_update_freq_max_load(cpu_possible_mask); + + mutex_unlock(&policy_mutex); + + return 0; +} + +void fixup_busy_time(struct task_struct *p, int new_cpu) +{ + struct rq *src_rq = task_rq(p); + struct rq *dest_rq = cpu_rq(new_cpu); + u64 wallclock; + u64 *src_curr_runnable_sum, *dst_curr_runnable_sum; + u64 *src_prev_runnable_sum, *dst_prev_runnable_sum; + u64 *src_nt_curr_runnable_sum, *dst_nt_curr_runnable_sum; + u64 *src_nt_prev_runnable_sum, *dst_nt_prev_runnable_sum; + int migrate_type; + struct migration_sum_data d; + bool new_task; + struct related_thread_group *grp; + + if (!p->on_rq && p->state != TASK_WAKING) + return; + + if (exiting_task(p)) { + clear_ed_task(p, src_rq); + return; + } + + if (p->state == TASK_WAKING) + double_rq_lock(src_rq, dest_rq); + + if (sched_disable_window_stats) + goto done; + + wallclock = sched_ktime_clock(); + + update_task_ravg(task_rq(p)->curr, task_rq(p), + TASK_UPDATE, + wallclock, 0); + update_task_ravg(dest_rq->curr, dest_rq, + TASK_UPDATE, wallclock, 0); + + update_task_ravg(p, task_rq(p), TASK_MIGRATE, + wallclock, 0); + + update_task_cpu_cycles(p, new_cpu); + + new_task = is_new_task(p); + /* Protected by rq_lock */ + grp = p->grp; + if (grp && sched_freq_aggregate) { + struct group_cpu_time *cpu_time; + + migrate_type = GROUP_TO_GROUP; + /* Protected by rq_lock */ + cpu_time = _group_cpu_time(grp, cpu_of(src_rq)); + d.src_rq = NULL; + d.src_cpu_time = cpu_time; + src_curr_runnable_sum = &cpu_time->curr_runnable_sum; + src_prev_runnable_sum = &cpu_time->prev_runnable_sum; + src_nt_curr_runnable_sum = &cpu_time->nt_curr_runnable_sum; + src_nt_prev_runnable_sum = &cpu_time->nt_prev_runnable_sum; + + /* Protected by rq_lock */ + cpu_time = _group_cpu_time(grp, cpu_of(dest_rq)); + d.dst_rq = NULL; + d.dst_cpu_time = cpu_time; + dst_curr_runnable_sum = &cpu_time->curr_runnable_sum; + dst_prev_runnable_sum = &cpu_time->prev_runnable_sum; + dst_nt_curr_runnable_sum = &cpu_time->nt_curr_runnable_sum; + dst_nt_prev_runnable_sum = &cpu_time->nt_prev_runnable_sum; + sync_window_start(dest_rq, cpu_time); + } else { + migrate_type = RQ_TO_RQ; + d.src_rq = src_rq; + d.src_cpu_time = NULL; + d.dst_rq = dest_rq; + d.dst_cpu_time = NULL; + src_curr_runnable_sum = &src_rq->curr_runnable_sum; + src_prev_runnable_sum = &src_rq->prev_runnable_sum; + src_nt_curr_runnable_sum = &src_rq->nt_curr_runnable_sum; + src_nt_prev_runnable_sum = &src_rq->nt_prev_runnable_sum; + + dst_curr_runnable_sum = &dest_rq->curr_runnable_sum; + dst_prev_runnable_sum = &dest_rq->prev_runnable_sum; + dst_nt_curr_runnable_sum = &dest_rq->nt_curr_runnable_sum; + dst_nt_prev_runnable_sum = &dest_rq->nt_prev_runnable_sum; + } + + if (p->ravg.curr_window) { + *src_curr_runnable_sum -= p->ravg.curr_window; + *dst_curr_runnable_sum += p->ravg.curr_window; + if (new_task) { + *src_nt_curr_runnable_sum -= p->ravg.curr_window; + *dst_nt_curr_runnable_sum += p->ravg.curr_window; + } + } + + if (p->ravg.prev_window) { + *src_prev_runnable_sum -= p->ravg.prev_window; + *dst_prev_runnable_sum += p->ravg.prev_window; + if (new_task) { + *src_nt_prev_runnable_sum -= p->ravg.prev_window; + *dst_nt_prev_runnable_sum += p->ravg.prev_window; + } + } + + if (p == src_rq->ed_task) { + src_rq->ed_task = NULL; + if (!dest_rq->ed_task) + dest_rq->ed_task = p; + } + + trace_sched_migration_update_sum(p, migrate_type, &d); + BUG_ON((s64)*src_prev_runnable_sum < 0); + BUG_ON((s64)*src_curr_runnable_sum < 0); + BUG_ON((s64)*src_nt_prev_runnable_sum < 0); + BUG_ON((s64)*src_nt_curr_runnable_sum < 0); + +done: + if (p->state == TASK_WAKING) + double_rq_unlock(src_rq, dest_rq); +} + +#define sched_up_down_migrate_auto_update 1 +static void check_for_up_down_migrate_update(const struct cpumask *cpus) +{ + int i = cpumask_first(cpus); + + if (!sched_up_down_migrate_auto_update) + return; + + if (cpu_max_possible_capacity(i) == max_possible_capacity) + return; + + if (cpu_max_possible_freq(i) == cpu_max_freq(i)) + up_down_migrate_scale_factor = 1024; + else + up_down_migrate_scale_factor = (1024 * + cpu_max_possible_freq(i)) / cpu_max_freq(i); + + update_up_down_migrate(); +} + +/* Return cluster which can offer required capacity for group */ +static struct sched_cluster * +best_cluster(struct related_thread_group *grp, u64 total_demand) +{ + struct sched_cluster *cluster = NULL; + + for_each_sched_cluster(cluster) { + if (group_will_fit(cluster, grp, total_demand)) + return cluster; + } + + return NULL; +} + +static void _set_preferred_cluster(struct related_thread_group *grp) +{ + struct task_struct *p; + u64 combined_demand = 0; + + if (!sysctl_sched_enable_colocation) { + grp->last_update = sched_ktime_clock(); + grp->preferred_cluster = NULL; + return; + } + + /* + * wakeup of two or more related tasks could race with each other and + * could result in multiple calls to _set_preferred_cluster being issued + * at same time. Avoid overhead in such cases of rechecking preferred + * cluster + */ + if (sched_ktime_clock() - grp->last_update < sched_ravg_window / 10) + return; + + list_for_each_entry(p, &grp->tasks, grp_list) + combined_demand += p->ravg.demand; + + grp->preferred_cluster = best_cluster(grp, combined_demand); + grp->last_update = sched_ktime_clock(); + trace_sched_set_preferred_cluster(grp, combined_demand); +} + +void set_preferred_cluster(struct related_thread_group *grp) +{ + raw_spin_lock(&grp->lock); + _set_preferred_cluster(grp); + raw_spin_unlock(&grp->lock); +} + +#define ADD_TASK 0 +#define REM_TASK 1 + +static inline void free_group_cputime(struct related_thread_group *grp) +{ + free_percpu(grp->cpu_time); +} + +static int alloc_group_cputime(struct related_thread_group *grp) +{ + int i; + struct group_cpu_time *cpu_time; + int cpu = raw_smp_processor_id(); + struct rq *rq = cpu_rq(cpu); + u64 window_start = rq->window_start; + + grp->cpu_time = alloc_percpu(struct group_cpu_time); + if (!grp->cpu_time) + return -ENOMEM; + + for_each_possible_cpu(i) { + cpu_time = per_cpu_ptr(grp->cpu_time, i); + memset(cpu_time, 0, sizeof(struct group_cpu_time)); + cpu_time->window_start = window_start; + } + + return 0; +} + +/* + * A group's window_start may be behind. When moving it forward, flip prev/curr + * counters. When moving forward > 1 window, prev counter is set to 0 + */ +static inline void +sync_window_start(struct rq *rq, struct group_cpu_time *cpu_time) +{ + u64 delta; + int nr_windows; + u64 curr_sum = cpu_time->curr_runnable_sum; + u64 nt_curr_sum = cpu_time->nt_curr_runnable_sum; + + delta = rq->window_start - cpu_time->window_start; + if (!delta) + return; + + nr_windows = div64_u64(delta, sched_ravg_window); + if (nr_windows > 1) + curr_sum = nt_curr_sum = 0; + + cpu_time->prev_runnable_sum = curr_sum; + cpu_time->curr_runnable_sum = 0; + + cpu_time->nt_prev_runnable_sum = nt_curr_sum; + cpu_time->nt_curr_runnable_sum = 0; + + cpu_time->window_start = rq->window_start; +} + +/* + * Task's cpu usage is accounted in: + * rq->curr/prev_runnable_sum, when its ->grp is NULL + * grp->cpu_time[cpu]->curr/prev_runnable_sum, when its ->grp is !NULL + * + * Transfer task's cpu usage between those counters when transitioning between + * groups + */ +static void transfer_busy_time(struct rq *rq, struct related_thread_group *grp, + struct task_struct *p, int event) +{ + u64 wallclock; + struct group_cpu_time *cpu_time; + u64 *src_curr_runnable_sum, *dst_curr_runnable_sum; + u64 *src_prev_runnable_sum, *dst_prev_runnable_sum; + u64 *src_nt_curr_runnable_sum, *dst_nt_curr_runnable_sum; + u64 *src_nt_prev_runnable_sum, *dst_nt_prev_runnable_sum; + struct migration_sum_data d; + int migrate_type; + + if (!sched_freq_aggregate) + return; + + wallclock = sched_ktime_clock(); + + update_task_ravg(rq->curr, rq, TASK_UPDATE, wallclock, 0); + update_task_ravg(p, rq, TASK_UPDATE, wallclock, 0); + + /* cpu_time protected by related_thread_group_lock, grp->lock rq_lock */ + cpu_time = _group_cpu_time(grp, cpu_of(rq)); + if (event == ADD_TASK) { + sync_window_start(rq, cpu_time); + migrate_type = RQ_TO_GROUP; + d.src_rq = rq; + d.src_cpu_time = NULL; + d.dst_rq = NULL; + d.dst_cpu_time = cpu_time; + src_curr_runnable_sum = &rq->curr_runnable_sum; + dst_curr_runnable_sum = &cpu_time->curr_runnable_sum; + src_prev_runnable_sum = &rq->prev_runnable_sum; + dst_prev_runnable_sum = &cpu_time->prev_runnable_sum; + + src_nt_curr_runnable_sum = &rq->nt_curr_runnable_sum; + dst_nt_curr_runnable_sum = &cpu_time->nt_curr_runnable_sum; + src_nt_prev_runnable_sum = &rq->nt_prev_runnable_sum; + dst_nt_prev_runnable_sum = &cpu_time->nt_prev_runnable_sum; + } else { + migrate_type = GROUP_TO_RQ; + d.src_rq = NULL; + d.src_cpu_time = cpu_time; + d.dst_rq = rq; + d.dst_cpu_time = NULL; + + /* + * In case of REM_TASK, cpu_time->window_start would be + * uptodate, because of the update_task_ravg() we called + * above on the moving task. Hence no need for + * sync_window_start() + */ + src_curr_runnable_sum = &cpu_time->curr_runnable_sum; + dst_curr_runnable_sum = &rq->curr_runnable_sum; + src_prev_runnable_sum = &cpu_time->prev_runnable_sum; + dst_prev_runnable_sum = &rq->prev_runnable_sum; + + src_nt_curr_runnable_sum = &cpu_time->nt_curr_runnable_sum; + dst_nt_curr_runnable_sum = &rq->nt_curr_runnable_sum; + src_nt_prev_runnable_sum = &cpu_time->nt_prev_runnable_sum; + dst_nt_prev_runnable_sum = &rq->nt_prev_runnable_sum; + } + + *src_curr_runnable_sum -= p->ravg.curr_window; + *dst_curr_runnable_sum += p->ravg.curr_window; + + *src_prev_runnable_sum -= p->ravg.prev_window; + *dst_prev_runnable_sum += p->ravg.prev_window; + + if (is_new_task(p)) { + *src_nt_curr_runnable_sum -= p->ravg.curr_window; + *dst_nt_curr_runnable_sum += p->ravg.curr_window; + *src_nt_prev_runnable_sum -= p->ravg.prev_window; + *dst_nt_prev_runnable_sum += p->ravg.prev_window; + } + + trace_sched_migration_update_sum(p, migrate_type, &d); + + BUG_ON((s64)*src_curr_runnable_sum < 0); + BUG_ON((s64)*src_prev_runnable_sum < 0); +} + +static inline struct group_cpu_time * +task_group_cpu_time(struct task_struct *p, int cpu) +{ + return _group_cpu_time(rcu_dereference(p->grp), cpu); +} + +static inline struct group_cpu_time * +_group_cpu_time(struct related_thread_group *grp, int cpu) +{ + return grp ? per_cpu_ptr(grp->cpu_time, cpu) : NULL; +} + +struct related_thread_group *alloc_related_thread_group(int group_id) +{ + struct related_thread_group *grp; + + grp = kzalloc(sizeof(*grp), GFP_KERNEL); + if (!grp) + return ERR_PTR(-ENOMEM); + + if (alloc_group_cputime(grp)) { + kfree(grp); + return ERR_PTR(-ENOMEM); + } + + grp->id = group_id; + INIT_LIST_HEAD(&grp->tasks); + INIT_LIST_HEAD(&grp->list); + raw_spin_lock_init(&grp->lock); + + return grp; +} + +struct related_thread_group *lookup_related_thread_group(unsigned int group_id) +{ + struct related_thread_group *grp; + + list_for_each_entry(grp, &related_thread_groups, list) { + if (grp->id == group_id) + return grp; + } + + return NULL; +} + +/* See comments before preferred_cluster() */ +static void free_related_thread_group(struct rcu_head *rcu) +{ + struct related_thread_group *grp = container_of(rcu, struct + related_thread_group, rcu); + + free_group_cputime(grp); + kfree(grp); +} + +static void remove_task_from_group(struct task_struct *p) +{ + struct related_thread_group *grp = p->grp; + struct rq *rq; + int empty_group = 1; + struct rq_flags rf; + + raw_spin_lock(&grp->lock); + + rq = __task_rq_lock(p, &rf); + transfer_busy_time(rq, p->grp, p, REM_TASK); + list_del_init(&p->grp_list); + rcu_assign_pointer(p->grp, NULL); + __task_rq_unlock(rq, &rf); + + if (!list_empty(&grp->tasks)) { + empty_group = 0; + _set_preferred_cluster(grp); + } + + raw_spin_unlock(&grp->lock); + + if (empty_group) { + list_del(&grp->list); + call_rcu(&grp->rcu, free_related_thread_group); + } +} + +static int +add_task_to_group(struct task_struct *p, struct related_thread_group *grp) +{ + struct rq *rq; + struct rq_flags rf; + + raw_spin_lock(&grp->lock); + + /* + * Change p->grp under rq->lock. Will prevent races with read-side + * reference of p->grp in various hot-paths + */ + rq = __task_rq_lock(p, &rf); + transfer_busy_time(rq, grp, p, ADD_TASK); + list_add(&p->grp_list, &grp->tasks); + rcu_assign_pointer(p->grp, grp); + __task_rq_unlock(rq, &rf); + + _set_preferred_cluster(grp); + + raw_spin_unlock(&grp->lock); + + return 0; +} + +void add_new_task_to_grp(struct task_struct *new) +{ + unsigned long flags; + struct related_thread_group *grp; + struct task_struct *parent; + + if (!sysctl_sched_enable_thread_grouping) + return; + + if (thread_group_leader(new)) + return; + + parent = new->group_leader; + + /* + * The parent's pi_lock is required here to protect race + * against the parent task being removed from the + * group. + */ + raw_spin_lock_irqsave(&parent->pi_lock, flags); + + /* protected by pi_lock. */ + grp = task_related_thread_group(parent); + if (!grp) { + raw_spin_unlock_irqrestore(&parent->pi_lock, flags); + return; + } + raw_spin_lock(&grp->lock); + + rcu_assign_pointer(new->grp, grp); + list_add(&new->grp_list, &grp->tasks); + + raw_spin_unlock(&grp->lock); + raw_spin_unlock_irqrestore(&parent->pi_lock, flags); +} + +int sched_set_group_id(struct task_struct *p, unsigned int group_id) +{ + int rc = 0, destroy = 0; + unsigned long flags; + struct related_thread_group *grp = NULL, *new = NULL; + +redo: + raw_spin_lock_irqsave(&p->pi_lock, flags); + + if ((current != p && p->flags & PF_EXITING) || + (!p->grp && !group_id) || + (p->grp && p->grp->id == group_id)) + goto done; + + write_lock(&related_thread_group_lock); + + if (!group_id) { + remove_task_from_group(p); + write_unlock(&related_thread_group_lock); + goto done; + } + + if (p->grp && p->grp->id != group_id) + remove_task_from_group(p); + + grp = lookup_related_thread_group(group_id); + if (!grp && !new) { + /* New group */ + write_unlock(&related_thread_group_lock); + raw_spin_unlock_irqrestore(&p->pi_lock, flags); + new = alloc_related_thread_group(group_id); + if (IS_ERR(new)) + return -ENOMEM; + destroy = 1; + /* Rerun checks (like task exiting), since we dropped pi_lock */ + goto redo; + } else if (!grp && new) { + /* New group - use object allocated before */ + destroy = 0; + list_add(&new->list, &related_thread_groups); + grp = new; + } + + BUG_ON(!grp); + rc = add_task_to_group(p, grp); + write_unlock(&related_thread_group_lock); +done: + raw_spin_unlock_irqrestore(&p->pi_lock, flags); + + if (new && destroy) { + free_group_cputime(new); + kfree(new); + } + + return rc; +} + +unsigned int sched_get_group_id(struct task_struct *p) +{ + unsigned int group_id; + struct related_thread_group *grp; + + rcu_read_lock(); + grp = task_related_thread_group(p); + group_id = grp ? grp->id : 0; + rcu_read_unlock(); + + return group_id; +} + +static void update_cpu_cluster_capacity(const cpumask_t *cpus) +{ + int i; + struct sched_cluster *cluster; + struct cpumask cpumask; + + cpumask_copy(&cpumask, cpus); + pre_big_task_count_change(cpu_possible_mask); + + for_each_cpu(i, &cpumask) { + cluster = cpu_rq(i)->cluster; + cpumask_andnot(&cpumask, &cpumask, &cluster->cpus); + + cluster->capacity = compute_capacity(cluster); + cluster->load_scale_factor = compute_load_scale_factor(cluster); + + /* 'cpus' can contain cpumask more than one cluster */ + check_for_up_down_migrate_update(&cluster->cpus); + } + + __update_min_max_capacity(); + + post_big_task_count_change(cpu_possible_mask); +} + +static DEFINE_SPINLOCK(cpu_freq_min_max_lock); +void sched_update_cpu_freq_min_max(const cpumask_t *cpus, u32 fmin, u32 fmax) +{ + struct cpumask cpumask; + struct sched_cluster *cluster; + int i, update_capacity = 0; + unsigned long flags; + + spin_lock_irqsave(&cpu_freq_min_max_lock, flags); + cpumask_copy(&cpumask, cpus); + for_each_cpu(i, &cpumask) { + cluster = cpu_rq(i)->cluster; + cpumask_andnot(&cpumask, &cpumask, &cluster->cpus); + + update_capacity += (cluster->max_mitigated_freq != fmax); + cluster->max_mitigated_freq = fmax; + } + spin_unlock_irqrestore(&cpu_freq_min_max_lock, flags); + + if (update_capacity) + update_cpu_cluster_capacity(cpus); +} + +static int cpufreq_notifier_policy(struct notifier_block *nb, + unsigned long val, void *data) +{ + struct cpufreq_policy *policy = (struct cpufreq_policy *)data; + struct sched_cluster *cluster = NULL; + struct cpumask policy_cluster = *policy->related_cpus; + unsigned int orig_max_freq = 0; + int i, j, update_capacity = 0; + + if (val != CPUFREQ_NOTIFY && val != CPUFREQ_REMOVE_POLICY && + val != CPUFREQ_CREATE_POLICY) + return 0; + + if (val == CPUFREQ_REMOVE_POLICY || val == CPUFREQ_CREATE_POLICY) { + update_min_max_capacity(); + return 0; + } + + max_possible_freq = max(max_possible_freq, policy->cpuinfo.max_freq); + if (min_max_freq == 1) + min_max_freq = UINT_MAX; + min_max_freq = min(min_max_freq, policy->cpuinfo.max_freq); + BUG_ON(!min_max_freq); + BUG_ON(!policy->max); + + for_each_cpu(i, &policy_cluster) { + cluster = cpu_rq(i)->cluster; + cpumask_andnot(&policy_cluster, &policy_cluster, + &cluster->cpus); + + orig_max_freq = cluster->max_freq; + cluster->min_freq = policy->min; + cluster->max_freq = policy->max; + cluster->cur_freq = policy->cur; + + if (!cluster->freq_init_done) { + mutex_lock(&cluster_lock); + for_each_cpu(j, &cluster->cpus) + cpumask_copy(&cpu_rq(j)->freq_domain_cpumask, + policy->related_cpus); + cluster->max_possible_freq = policy->cpuinfo.max_freq; + cluster->max_possible_capacity = + compute_max_possible_capacity(cluster); + cluster->freq_init_done = true; + + sort_clusters(); + update_all_clusters_stats(); + mutex_unlock(&cluster_lock); + continue; + } + + update_capacity += (orig_max_freq != cluster->max_freq); + } + + if (update_capacity) + update_cpu_cluster_capacity(policy->related_cpus); + + return 0; +} + +static int cpufreq_notifier_trans(struct notifier_block *nb, + unsigned long val, void *data) +{ + struct cpufreq_freqs *freq = (struct cpufreq_freqs *)data; + unsigned int cpu = freq->cpu, new_freq = freq->new; + unsigned long flags; + struct sched_cluster *cluster; + struct cpumask policy_cpus = cpu_rq(cpu)->freq_domain_cpumask; + int i, j; + + if (val != CPUFREQ_POSTCHANGE) + return 0; + + BUG_ON(!new_freq); + + if (cpu_cur_freq(cpu) == new_freq) + return 0; + + for_each_cpu(i, &policy_cpus) { + cluster = cpu_rq(i)->cluster; + + for_each_cpu(j, &cluster->cpus) { + struct rq *rq = cpu_rq(j); + + raw_spin_lock_irqsave(&rq->lock, flags); + update_task_ravg(rq->curr, rq, TASK_UPDATE, + sched_ktime_clock(), 0); + raw_spin_unlock_irqrestore(&rq->lock, flags); + } + + cluster->cur_freq = new_freq; + cpumask_andnot(&policy_cpus, &policy_cpus, &cluster->cpus); + } + + return 0; +} + +static int pwr_stats_ready_notifier(struct notifier_block *nb, + unsigned long cpu, void *data) +{ + cpumask_t mask = CPU_MASK_NONE; + + cpumask_set_cpu(cpu, &mask); + sched_update_freq_max_load(&mask); + + mutex_lock(&cluster_lock); + sort_clusters(); + mutex_unlock(&cluster_lock); + + return 0; +} + +static struct notifier_block notifier_policy_block = { + .notifier_call = cpufreq_notifier_policy +}; + +static struct notifier_block notifier_trans_block = { + .notifier_call = cpufreq_notifier_trans +}; + +static struct notifier_block notifier_pwr_stats_ready = { + .notifier_call = pwr_stats_ready_notifier +}; + +int __weak register_cpu_pwr_stats_ready_notifier(struct notifier_block *nb) +{ + return -EINVAL; +} + +static int register_sched_callback(void) +{ + int ret; + + ret = cpufreq_register_notifier(¬ifier_policy_block, + CPUFREQ_POLICY_NOTIFIER); + + if (!ret) + ret = cpufreq_register_notifier(¬ifier_trans_block, + CPUFREQ_TRANSITION_NOTIFIER); + + register_cpu_pwr_stats_ready_notifier(¬ifier_pwr_stats_ready); + + return 0; +} + +/* + * cpufreq callbacks can be registered at core_initcall or later time. + * Any registration done prior to that is "forgotten" by cpufreq. See + * initialization of variable init_cpufreq_transition_notifier_list_called + * for further information. + */ +core_initcall(register_sched_callback); + +int update_preferred_cluster(struct related_thread_group *grp, + struct task_struct *p, u32 old_load) +{ + u32 new_load = task_load(p); + + if (!grp) + return 0; + + /* + * Update if task's load has changed significantly or a complete window + * has passed since we last updated preference + */ + if (abs(new_load - old_load) > sched_ravg_window / 4 || + sched_ktime_clock() - grp->last_update > sched_ravg_window) + return 1; + + return 0; +} + +bool early_detection_notify(struct rq *rq, u64 wallclock) +{ + struct task_struct *p; + int loop_max = 10; + + if (!sched_boost() || !rq->cfs.h_nr_running) + return 0; + + rq->ed_task = NULL; + list_for_each_entry(p, &rq->cfs_tasks, se.group_node) { + if (!loop_max) + break; + + if (wallclock - p->last_wake_ts >= EARLY_DETECTION_DURATION) { + rq->ed_task = p; + return 1; + } + + loop_max--; + } + + return 0; +} + +#ifdef CONFIG_CGROUP_SCHED +u64 cpu_upmigrate_discourage_read_u64(struct cgroup_subsys_state *css, + struct cftype *cft) +{ + struct task_group *tg = css_tg(css); + + return tg->upmigrate_discouraged; +} + +int cpu_upmigrate_discourage_write_u64(struct cgroup_subsys_state *css, + struct cftype *cft, u64 upmigrate_discourage) +{ + struct task_group *tg = css_tg(css); + int discourage = upmigrate_discourage > 0; + + if (tg->upmigrate_discouraged == discourage) + return 0; + + /* + * Revisit big-task classification for tasks of this cgroup. It would + * have been efficient to walk tasks of just this cgroup in running + * state, but we don't have easy means to do that. Walk all tasks in + * running state on all cpus instead and re-visit their big task + * classification. + */ + get_online_cpus(); + pre_big_task_count_change(cpu_online_mask); + + tg->upmigrate_discouraged = discourage; + + post_big_task_count_change(cpu_online_mask); + put_online_cpus(); + + return 0; +} +#endif /* CONFIG_CGROUP_SCHED */ diff --git a/kernel/sched/idle_task.c b/kernel/sched/idle_task.c index 5405d3feb112..13c881832f6d 100644 --- a/kernel/sched/idle_task.c +++ b/kernel/sched/idle_task.c @@ -78,6 +78,14 @@ static void update_curr_idle(struct rq *rq) { } +#ifdef CONFIG_SCHED_HMP +static void +fixup_hmp_sched_stats_idle(struct rq *rq, struct task_struct *p, + u32 new_task_load, u32 new_pred_demand) +{ +} +#endif + /* * Simple, special scheduling class for the per-CPU idle tasks: */ @@ -106,4 +114,7 @@ const struct sched_class idle_sched_class = { .prio_changed = prio_changed_idle, .switched_to = switched_to_idle, .update_curr = update_curr_idle, +#ifdef CONFIG_SCHED_HMP + .fixup_hmp_sched_stats = fixup_hmp_sched_stats_idle, +#endif }; diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 2516b8df6dbb..a791486cdb49 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -7,6 +7,7 @@ #include #include +#include int sched_rr_timeslice = RR_TIMESLICE; @@ -897,6 +898,51 @@ static inline int rt_se_prio(struct sched_rt_entity *rt_se) return rt_task_of(rt_se)->prio; } +static void dump_throttled_rt_tasks(struct rt_rq *rt_rq) +{ + struct rt_prio_array *array = &rt_rq->active; + struct sched_rt_entity *rt_se; + char buf[500]; + char *pos = buf; + char *end = buf + sizeof(buf); + int idx; + + pos += snprintf(pos, sizeof(buf), + "sched: RT throttling activated for rt_rq %p (cpu %d)\n", + rt_rq, cpu_of(rq_of_rt_rq(rt_rq))); + + if (bitmap_empty(array->bitmap, MAX_RT_PRIO)) + goto out; + + pos += snprintf(pos, end - pos, "potential CPU hogs:\n"); + idx = sched_find_first_bit(array->bitmap); + while (idx < MAX_RT_PRIO) { + list_for_each_entry(rt_se, array->queue + idx, run_list) { + struct task_struct *p; + + if (!rt_entity_is_task(rt_se)) + continue; + + p = rt_task_of(rt_se); + if (pos < end) + pos += snprintf(pos, end - pos, "\t%s (%d)\n", + p->comm, p->pid); + } + idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx + 1); + } +out: +#ifdef CONFIG_PANIC_ON_RT_THROTTLING + /* + * Use pr_err() in the BUG() case since printk_sched() will + * not get flushed and deadlock is not a concern. + */ + pr_err("%s", buf); + BUG(); +#else + printk_deferred("%s", buf); +#endif +} + static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) { u64 runtime = sched_rt_runtime(rt_rq); @@ -920,8 +966,14 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) * but accrue some time due to boosting. */ if (likely(rt_b->rt_runtime)) { + static bool once; + rt_rq->rt_throttled = 1; - printk_deferred_once("sched: RT throttling activated\n"); + + if (!once) { + once = true; + dump_throttled_rt_tasks(rt_rq); + } } else { /* * In case we did anyway, make it go away, @@ -1141,6 +1193,41 @@ void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} #endif /* CONFIG_RT_GROUP_SCHED */ +#ifdef CONFIG_SCHED_HMP + +static void +inc_hmp_sched_stats_rt(struct rq *rq, struct task_struct *p) +{ + inc_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +static void +dec_hmp_sched_stats_rt(struct rq *rq, struct task_struct *p) +{ + dec_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +static void +fixup_hmp_sched_stats_rt(struct rq *rq, struct task_struct *p, + u32 new_task_load, u32 new_pred_demand) +{ + s64 task_load_delta = (s64)new_task_load - task_load(p); + s64 pred_demand_delta = PRED_DEMAND_DELTA; + + fixup_cumulative_runnable_avg(&rq->hmp_stats, p, task_load_delta, + pred_demand_delta); +} + +#else /* CONFIG_SCHED_HMP */ + +static inline void +inc_hmp_sched_stats_rt(struct rq *rq, struct task_struct *p) { } + +static inline void +dec_hmp_sched_stats_rt(struct rq *rq, struct task_struct *p) { } + +#endif /* CONFIG_SCHED_HMP */ + static inline unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se) { @@ -1322,6 +1409,7 @@ enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) rt_se->timeout = 0; enqueue_rt_entity(rt_se, flags); + inc_hmp_sched_stats_rt(rq, p); if (!task_current(rq, p) && tsk_nr_cpus_allowed(p) > 1) enqueue_pushable_task(rq, p); @@ -1333,6 +1421,7 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) update_curr_rt(rq); dequeue_rt_entity(rt_se, flags); + dec_hmp_sched_stats_rt(rq, p); dequeue_pushable_task(rq, p); } @@ -1374,12 +1463,32 @@ static void yield_task_rt(struct rq *rq) #ifdef CONFIG_SMP static int find_lowest_rq(struct task_struct *task); +#ifdef CONFIG_SCHED_HMP +static int +select_task_rq_rt_hmp(struct task_struct *p, int cpu, int sd_flag, int flags) +{ + int target; + + rcu_read_lock(); + target = find_lowest_rq(p); + if (target != -1) + cpu = target; + rcu_read_unlock(); + + return cpu; +} +#endif /* CONFIG_SCHED_HMP */ + static int select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags) { struct task_struct *curr; struct rq *rq; +#ifdef CONFIG_SCHED_HMP + return select_task_rq_rt_hmp(p, cpu, sd_flag, flags); +#endif + /* For anything but wake ups, just return the task_cpu */ if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK) goto out; @@ -1617,6 +1726,65 @@ static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu) static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); +#ifdef CONFIG_SCHED_HMP +static int find_lowest_rq_hmp(struct task_struct *task) +{ + struct cpumask *lowest_mask = *this_cpu_ptr(&local_cpu_mask); + struct cpumask candidate_mask = CPU_MASK_NONE; + struct sched_cluster *cluster; + int best_cpu = -1; + int prev_cpu = task_cpu(task); + u64 cpu_load, min_load = ULLONG_MAX; + int i; + int restrict_cluster = sched_boost() ? 0 : + sysctl_sched_restrict_cluster_spill; + + /* Make sure the mask is initialized first */ + if (unlikely(!lowest_mask)) + return best_cpu; + + if (task->nr_cpus_allowed == 1) + return best_cpu; /* No other targets possible */ + + if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) + return best_cpu; /* No targets found */ + + /* + * At this point we have built a mask of cpus representing the + * lowest priority tasks in the system. Now we want to elect + * the best one based on our affinity and topology. + */ + + for_each_sched_cluster(cluster) { + cpumask_and(&candidate_mask, &cluster->cpus, lowest_mask); + + if (cpumask_empty(&candidate_mask)) + continue; + + for_each_cpu(i, &candidate_mask) { + if (sched_cpu_high_irqload(i)) + continue; + + cpu_load = cpu_rq(i)->hmp_stats.cumulative_runnable_avg; + if (!restrict_cluster) + cpu_load = scale_load_to_cpu(cpu_load, i); + + if (cpu_load < min_load || + (cpu_load == min_load && + (i == prev_cpu || (best_cpu != prev_cpu && + cpus_share_cache(prev_cpu, i))))) { + min_load = cpu_load; + best_cpu = i; + } + } + if (restrict_cluster && best_cpu != -1) + break; + } + + return best_cpu; +} +#endif /* CONFIG_SCHED_HMP */ + static int find_lowest_rq(struct task_struct *task) { struct sched_domain *sd; @@ -1624,6 +1792,10 @@ static int find_lowest_rq(struct task_struct *task) int this_cpu = smp_processor_id(); int cpu = task_cpu(task); +#ifdef CONFIG_SCHED_HMP + return find_lowest_rq_hmp(task); +#endif + /* Make sure the mask is initialized first */ if (unlikely(!lowest_mask)) return -1; @@ -1841,9 +2013,11 @@ retry: goto retry; } + next_task->on_rq = TASK_ON_RQ_MIGRATING; deactivate_task(rq, next_task, 0); set_task_cpu(next_task, lowest_rq->cpu); activate_task(lowest_rq, next_task, 0); + next_task->on_rq = TASK_ON_RQ_QUEUED; ret = 1; resched_curr(lowest_rq); @@ -2095,9 +2269,11 @@ static void pull_rt_task(struct rq *this_rq) resched = true; + p->on_rq = TASK_ON_RQ_MIGRATING; deactivate_task(src_rq, p, 0); set_task_cpu(p, this_cpu); activate_task(this_rq, p, 0); + p->on_rq = TASK_ON_RQ_QUEUED; /* * We continue with the search, just in * case there's an even higher prio task @@ -2352,6 +2528,9 @@ const struct sched_class rt_sched_class = { .switched_to = switched_to_rt, .update_curr = update_curr_rt, +#ifdef CONFIG_SCHED_HMP + .fixup_hmp_sched_stats = fixup_hmp_sched_stats_rt, +#endif }; #ifdef CONFIG_SCHED_DEBUG diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 055f935d4421..f6e2bf132d08 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -255,6 +255,10 @@ struct cfs_bandwidth { struct task_group { struct cgroup_subsys_state css; +#ifdef CONFIG_SCHED_HMP + bool upmigrate_discouraged; +#endif + #ifdef CONFIG_FAIR_GROUP_SCHED /* schedulable entities of this group on each cpu */ struct sched_entity **se; @@ -365,12 +369,83 @@ static inline void set_task_rq_fair(struct sched_entity *se, #endif /* CONFIG_SMP */ #endif /* CONFIG_FAIR_GROUP_SCHED */ +extern struct task_group *css_tg(struct cgroup_subsys_state *css); + #else /* CONFIG_CGROUP_SCHED */ struct cfs_bandwidth { }; #endif /* CONFIG_CGROUP_SCHED */ +#ifdef CONFIG_SCHED_HMP + +struct hmp_sched_stats { + int nr_big_tasks; + u64 cumulative_runnable_avg; + u64 pred_demands_sum; +}; + +struct sched_cluster { + struct list_head list; + struct cpumask cpus; + int id; + int max_power_cost; + int min_power_cost; + int max_possible_capacity; + int capacity; + int efficiency; /* Differentiate cpus with different IPC capability */ + int load_scale_factor; + unsigned int exec_scale_factor; + /* + * max_freq = user maximum + * max_mitigated_freq = thermal defined maximum + * max_possible_freq = maximum supported by hardware + */ + unsigned int cur_freq, max_freq, max_mitigated_freq, min_freq; + unsigned int max_possible_freq; + bool freq_init_done; + int dstate, dstate_wakeup_latency, dstate_wakeup_energy; + unsigned int static_cluster_pwr_cost; + int notifier_sent; +}; + +extern unsigned long all_cluster_ids[]; + +static inline int cluster_first_cpu(struct sched_cluster *cluster) +{ + return cpumask_first(&cluster->cpus); +} + +struct related_thread_group { + int id; + raw_spinlock_t lock; + struct list_head tasks; + struct list_head list; + struct sched_cluster *preferred_cluster; + struct rcu_head rcu; + u64 last_update; + struct group_cpu_time __percpu *cpu_time; /* one per cluster */ +}; + +struct migration_sum_data { + struct rq *src_rq, *dst_rq; + struct group_cpu_time *src_cpu_time, *dst_cpu_time; +}; + +extern struct list_head cluster_head; +extern int num_clusters; +extern struct sched_cluster *sched_cluster[NR_CPUS]; + +struct cpu_cycle { + u64 cycles; + u64 time; +}; + +#define for_each_sched_cluster(cluster) \ + list_for_each_entry_rcu(cluster, &cluster_head, list) + +#endif /* CONFIG_SCHED_HMP */ + /* CFS-related fields in a runqueue */ struct cfs_rq { struct load_weight load; @@ -439,6 +514,10 @@ struct cfs_rq { struct task_group *tg; /* group that "owns" this runqueue */ #ifdef CONFIG_CFS_BANDWIDTH +#ifdef CONFIG_SCHED_HMP + struct hmp_sched_stats hmp_stats; +#endif + int runtime_enabled; u64 runtime_expires; s64 runtime_remaining; @@ -656,6 +735,7 @@ struct rq { /* For active balancing */ int active_balance; int push_cpu; + struct task_struct *push_task; struct cpu_stop_work active_balance_work; /* cpu of this runqueue: */ int cpu; @@ -672,6 +752,29 @@ struct rq { u64 max_idle_balance_cost; #endif +#ifdef CONFIG_SCHED_HMP + struct sched_cluster *cluster; + struct cpumask freq_domain_cpumask; + struct hmp_sched_stats hmp_stats; + + int cstate, wakeup_latency, wakeup_energy; + u64 window_start; + unsigned long hmp_flags; + + u64 cur_irqload; + u64 avg_irqload; + u64 irqload_ts; + unsigned int static_cpu_pwr_cost; + struct task_struct *ed_task; + struct cpu_cycle cc; + u64 old_busy_time, old_busy_time_group; + u64 old_estimated_time; + u64 curr_runnable_sum; + u64 prev_runnable_sum; + u64 nt_curr_runnable_sum; + u64 nt_prev_runnable_sum; +#endif + #ifdef CONFIG_IRQ_TIME_ACCOUNTING u64 prev_irq_time; #endif @@ -962,6 +1065,583 @@ static inline void sched_ttwu_pending(void) { } #include "stats.h" #include "auto_group.h" +#ifdef CONFIG_SCHED_HMP + +#define WINDOW_STATS_RECENT 0 +#define WINDOW_STATS_MAX 1 +#define WINDOW_STATS_MAX_RECENT_AVG 2 +#define WINDOW_STATS_AVG 3 +#define WINDOW_STATS_INVALID_POLICY 4 + +#define MAJOR_TASK_PCT 85 +#define SCHED_UPMIGRATE_MIN_NICE 15 +#define EXITING_TASK_MARKER 0xdeaddead + +#define UP_MIGRATION 1 +#define DOWN_MIGRATION 2 +#define IRQLOAD_MIGRATION 3 + +extern struct mutex policy_mutex; +extern unsigned int sched_ravg_window; +extern unsigned int sched_disable_window_stats; +extern unsigned int max_possible_freq; +extern unsigned int min_max_freq; +extern unsigned int pct_task_load(struct task_struct *p); +extern unsigned int max_possible_efficiency; +extern unsigned int min_possible_efficiency; +extern unsigned int max_capacity; +extern unsigned int min_capacity; +extern unsigned int max_load_scale_factor; +extern unsigned int max_possible_capacity; +extern unsigned int min_max_possible_capacity; +extern unsigned int sched_upmigrate; +extern unsigned int sched_downmigrate; +extern unsigned int sched_init_task_load_windows; +extern unsigned int up_down_migrate_scale_factor; +extern unsigned int sysctl_sched_restrict_cluster_spill; +extern unsigned int sched_pred_alert_load; +extern unsigned int sched_major_task_runtime; +extern struct sched_cluster init_cluster; +extern unsigned int __read_mostly sched_short_sleep_task_threshold; +extern unsigned int __read_mostly sched_long_cpu_selection_threshold; +extern unsigned int __read_mostly sched_big_waker_task_load; +extern unsigned int __read_mostly sched_small_wakee_task_load; +extern unsigned int __read_mostly sched_spill_load; +extern unsigned int __read_mostly sched_upmigrate; +extern unsigned int __read_mostly sched_downmigrate; +extern unsigned int __read_mostly sysctl_sched_spill_nr_run; + +extern void init_new_task_load(struct task_struct *p); +extern u64 sched_ktime_clock(void); +extern int got_boost_kick(void); +extern int register_cpu_cycle_counter_cb(struct cpu_cycle_counter_cb *cb); +extern void update_task_ravg(struct task_struct *p, struct rq *rq, int event, + u64 wallclock, u64 irqtime); +extern bool early_detection_notify(struct rq *rq, u64 wallclock); +extern void clear_ed_task(struct task_struct *p, struct rq *rq); +extern void fixup_busy_time(struct task_struct *p, int new_cpu); +extern void clear_boost_kick(int cpu); +extern void clear_hmp_request(int cpu); +extern void mark_task_starting(struct task_struct *p); +extern void set_window_start(struct rq *rq); +extern void migrate_sync_cpu(int cpu); +extern void update_cluster_topology(void); +extern void set_task_last_wake(struct task_struct *p, u64 wallclock); +extern void set_task_last_switch_out(struct task_struct *p, u64 wallclock); +extern void init_clusters(void); +extern void reset_cpu_hmp_stats(int cpu, int reset_cra); +extern unsigned int max_task_load(void); +extern void sched_account_irqtime(int cpu, struct task_struct *curr, + u64 delta, u64 wallclock); +extern void sched_account_irqstart(int cpu, struct task_struct *curr, + u64 wallclock); +extern unsigned int cpu_temp(int cpu); +extern unsigned int nr_eligible_big_tasks(int cpu); +extern void update_up_down_migrate(void); +extern int update_preferred_cluster(struct related_thread_group *grp, + struct task_struct *p, u32 old_load); +extern void set_preferred_cluster(struct related_thread_group *grp); +extern void add_new_task_to_grp(struct task_struct *new); + +enum sched_boost_type { + SCHED_BOOST_NONE, + SCHED_BOOST_ON_BIG, + SCHED_BOOST_ON_ALL, +}; + +static inline struct sched_cluster *cpu_cluster(int cpu) +{ + return cpu_rq(cpu)->cluster; +} + +static inline int cpu_capacity(int cpu) +{ + return cpu_rq(cpu)->cluster->capacity; +} + +static inline int cpu_max_possible_capacity(int cpu) +{ + return cpu_rq(cpu)->cluster->max_possible_capacity; +} + +static inline int cpu_load_scale_factor(int cpu) +{ + return cpu_rq(cpu)->cluster->load_scale_factor; +} + +static inline int cpu_efficiency(int cpu) +{ + return cpu_rq(cpu)->cluster->efficiency; +} + +static inline unsigned int cpu_cur_freq(int cpu) +{ + return cpu_rq(cpu)->cluster->cur_freq; +} + +static inline unsigned int cpu_min_freq(int cpu) +{ + return cpu_rq(cpu)->cluster->min_freq; +} + +static inline unsigned int cluster_max_freq(struct sched_cluster *cluster) +{ + /* + * Governor and thermal driver don't know the other party's mitigation + * voting. So struct cluster saves both and return min() for current + * cluster fmax. + */ + return min(cluster->max_mitigated_freq, cluster->max_freq); +} + +static inline unsigned int cpu_max_freq(int cpu) +{ + return cluster_max_freq(cpu_rq(cpu)->cluster); +} + +static inline unsigned int cpu_max_possible_freq(int cpu) +{ + return cpu_rq(cpu)->cluster->max_possible_freq; +} + +static inline int same_cluster(int src_cpu, int dst_cpu) +{ + return cpu_rq(src_cpu)->cluster == cpu_rq(dst_cpu)->cluster; +} + +static inline int cpu_max_power_cost(int cpu) +{ + return cpu_rq(cpu)->cluster->max_power_cost; +} + +static inline u32 cpu_cycles_to_freq(u64 cycles, u32 period) +{ + return div64_u64(cycles, period); +} + +static inline bool hmp_capable(void) +{ + return max_possible_capacity != min_max_possible_capacity; +} + +/* + * 'load' is in reference to "best cpu" at its best frequency. + * Scale that in reference to a given cpu, accounting for how bad it is + * in reference to "best cpu". + */ +static inline u64 scale_load_to_cpu(u64 task_load, int cpu) +{ + u64 lsf = cpu_load_scale_factor(cpu); + + if (lsf != 1024) { + task_load *= lsf; + task_load /= 1024; + } + + return task_load; +} + +static inline unsigned int task_load(struct task_struct *p) +{ + return p->ravg.demand; +} + +static inline void +inc_cumulative_runnable_avg(struct hmp_sched_stats *stats, + struct task_struct *p) +{ + u32 task_load; + + if (sched_disable_window_stats) + return; + + task_load = sched_disable_window_stats ? 0 : p->ravg.demand; + + stats->cumulative_runnable_avg += task_load; + stats->pred_demands_sum += p->ravg.pred_demand; +} + +static inline void +dec_cumulative_runnable_avg(struct hmp_sched_stats *stats, + struct task_struct *p) +{ + u32 task_load; + + if (sched_disable_window_stats) + return; + + task_load = sched_disable_window_stats ? 0 : p->ravg.demand; + + stats->cumulative_runnable_avg -= task_load; + + BUG_ON((s64)stats->cumulative_runnable_avg < 0); + + stats->pred_demands_sum -= p->ravg.pred_demand; + BUG_ON((s64)stats->pred_demands_sum < 0); +} + +static inline void +fixup_cumulative_runnable_avg(struct hmp_sched_stats *stats, + struct task_struct *p, s64 task_load_delta, + s64 pred_demand_delta) +{ + if (sched_disable_window_stats) + return; + + stats->cumulative_runnable_avg += task_load_delta; + BUG_ON((s64)stats->cumulative_runnable_avg < 0); + + stats->pred_demands_sum += pred_demand_delta; + BUG_ON((s64)stats->pred_demands_sum < 0); +} + +#define pct_to_real(tunable) \ + (div64_u64((u64)tunable * (u64)max_task_load(), 100)) + +#define real_to_pct(tunable) \ + (div64_u64((u64)tunable * (u64)100, (u64)max_task_load())) + +#define SCHED_HIGH_IRQ_TIMEOUT 3 +static inline u64 sched_irqload(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + s64 delta; + + delta = get_jiffies_64() - rq->irqload_ts; + /* + * Current context can be preempted by irq and rq->irqload_ts can be + * updated by irq context so that delta can be negative. + * But this is okay and we can safely return as this means there + * was recent irq occurrence. + */ + + if (delta < SCHED_HIGH_IRQ_TIMEOUT) + return rq->avg_irqload; + else + return 0; +} + +static inline int sched_cpu_high_irqload(int cpu) +{ + return sched_irqload(cpu) >= sysctl_sched_cpu_high_irqload; +} + +static inline bool task_in_related_thread_group(struct task_struct *p) +{ + return !!(rcu_access_pointer(p->grp) != NULL); +} + +static inline +struct related_thread_group *task_related_thread_group(struct task_struct *p) +{ + return rcu_dereference(p->grp); +} + +#define PRED_DEMAND_DELTA ((s64)new_pred_demand - p->ravg.pred_demand) + +extern void +check_for_freq_change(struct rq *rq, bool check_pred, bool check_groups); + +extern void notify_migration(int src_cpu, int dest_cpu, + bool src_cpu_dead, struct task_struct *p); + +struct group_cpu_time { + u64 curr_runnable_sum; + u64 prev_runnable_sum; + u64 nt_curr_runnable_sum; + u64 nt_prev_runnable_sum; + u64 window_start; +}; + +/* Is frequency of two cpus synchronized with each other? */ +static inline int same_freq_domain(int src_cpu, int dst_cpu) +{ + struct rq *rq = cpu_rq(src_cpu); + + if (src_cpu == dst_cpu) + return 1; + + return cpumask_test_cpu(dst_cpu, &rq->freq_domain_cpumask); +} + +#define BOOST_KICK 0 +#define CPU_RESERVED 1 + +static inline int is_reserved(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + + return test_bit(CPU_RESERVED, &rq->hmp_flags); +} + +static inline int mark_reserved(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + + /* Name boost_flags as hmp_flags? */ + return test_and_set_bit(CPU_RESERVED, &rq->hmp_flags); +} + +static inline void clear_reserved(int cpu) +{ + struct rq *rq = cpu_rq(cpu); + + clear_bit(CPU_RESERVED, &rq->hmp_flags); +} + +static inline u64 cpu_cravg_sync(int cpu, int sync) +{ + struct rq *rq = cpu_rq(cpu); + u64 load; + + load = rq->hmp_stats.cumulative_runnable_avg; + + /* + * If load is being checked in a sync wakeup environment, + * we may want to discount the load of the currently running + * task. + */ + if (sync && cpu == smp_processor_id()) { + if (load > rq->curr->ravg.demand) + load -= rq->curr->ravg.demand; + else + load = 0; + } + + return load; +} + +extern void check_for_migration(struct rq *rq, struct task_struct *p); +extern void pre_big_task_count_change(const struct cpumask *cpus); +extern void post_big_task_count_change(const struct cpumask *cpus); +extern void set_hmp_defaults(void); +extern int power_delta_exceeded(unsigned int cpu_cost, unsigned int base_cost); +extern unsigned int power_cost(int cpu, u64 demand); +extern void reset_all_window_stats(u64 window_start, unsigned int window_size); +extern void boost_kick(int cpu); +extern int sched_boost(void); +extern int task_load_will_fit(struct task_struct *p, u64 task_load, int cpu, + enum sched_boost_type boost_type); +extern enum sched_boost_type sched_boost_type(void); +extern int task_will_fit(struct task_struct *p, int cpu); +extern int group_will_fit(struct sched_cluster *cluster, + struct related_thread_group *grp, u64 demand); +extern u64 cpu_load(int cpu); +extern u64 cpu_load_sync(int cpu, int sync); +extern int preferred_cluster(struct sched_cluster *cluster, + struct task_struct *p); +extern void inc_nr_big_task(struct hmp_sched_stats *stats, + struct task_struct *p); +extern void dec_nr_big_task(struct hmp_sched_stats *stats, + struct task_struct *p); +extern void inc_rq_hmp_stats(struct rq *rq, + struct task_struct *p, int change_cra); +extern void dec_rq_hmp_stats(struct rq *rq, + struct task_struct *p, int change_cra); +extern int is_big_task(struct task_struct *p); +extern int upmigrate_discouraged(struct task_struct *p); +extern struct sched_cluster *rq_cluster(struct rq *rq); +extern int nr_big_tasks(struct rq *rq); +extern void fixup_nr_big_tasks(struct hmp_sched_stats *stats, + struct task_struct *p, s64 delta); +extern void reset_task_stats(struct task_struct *p); +extern void reset_cfs_rq_hmp_stats(int cpu, int reset_cra); +extern void inc_hmp_sched_stats_fair(struct rq *rq, + struct task_struct *p, int change_cra); +extern u64 cpu_upmigrate_discourage_read_u64(struct cgroup_subsys_state *css, + struct cftype *cft); +extern int cpu_upmigrate_discourage_write_u64(struct cgroup_subsys_state *css, + struct cftype *cft, u64 upmigrate_discourage); + +#else /* CONFIG_SCHED_HMP */ + +struct hmp_sched_stats; +struct related_thread_group; +struct sched_cluster; + +static inline int got_boost_kick(void) +{ + return 0; +} + +static inline void update_task_ravg(struct task_struct *p, struct rq *rq, + int event, u64 wallclock, u64 irqtime) { } + +static inline bool early_detection_notify(struct rq *rq, u64 wallclock) +{ + return 0; +} + +static inline void clear_ed_task(struct task_struct *p, struct rq *rq) { } +static inline void fixup_busy_time(struct task_struct *p, int new_cpu) { } +static inline void clear_boost_kick(int cpu) { } +static inline void clear_hmp_request(int cpu) { } +static inline void mark_task_starting(struct task_struct *p) { } +static inline void set_window_start(struct rq *rq) { } +static inline void migrate_sync_cpu(int cpu) { } +static inline void update_cluster_topology(void) { } +static inline void set_task_last_wake(struct task_struct *p, u64 wallclock) { } +static inline void set_task_last_switch_out(struct task_struct *p, + u64 wallclock) { } + +static inline int task_will_fit(struct task_struct *p, int cpu) +{ + return 1; +} + +static inline unsigned int power_cost(int cpu, u64 demand) +{ + return SCHED_CAPACITY_SCALE; +} + +static inline int sched_boost(void) +{ + return 0; +} + +static inline int is_big_task(struct task_struct *p) +{ + return 0; +} + +static inline int nr_big_tasks(struct rq *rq) +{ + return 0; +} + +static inline int is_cpu_throttling_imminent(int cpu) +{ + return 0; +} + +static inline int is_task_migration_throttled(struct task_struct *p) +{ + return 0; +} + +static inline unsigned int cpu_temp(int cpu) +{ + return 0; +} + +static inline void +inc_rq_hmp_stats(struct rq *rq, struct task_struct *p, int change_cra) { } + +static inline void +dec_rq_hmp_stats(struct rq *rq, struct task_struct *p, int change_cra) { } + +static inline int +preferred_cluster(struct sched_cluster *cluster, struct task_struct *p) +{ + return 1; +} + +static inline struct sched_cluster *rq_cluster(struct rq *rq) +{ + return NULL; +} + +static inline void init_new_task_load(struct task_struct *p) { } + +static inline u64 scale_load_to_cpu(u64 load, int cpu) +{ + return load; +} + +static inline unsigned int nr_eligible_big_tasks(int cpu) +{ + return 0; +} + +static inline int pct_task_load(struct task_struct *p) { return 0; } + +static inline int cpu_capacity(int cpu) +{ + return SCHED_CAPACITY_SCALE; +} + +static inline int same_cluster(int src_cpu, int dst_cpu) { return 1; } + +static inline void inc_cumulative_runnable_avg(struct hmp_sched_stats *stats, + struct task_struct *p) +{ +} + +static inline void dec_cumulative_runnable_avg(struct hmp_sched_stats *stats, + struct task_struct *p) +{ +} + +static inline void sched_account_irqtime(int cpu, struct task_struct *curr, + u64 delta, u64 wallclock) +{ +} + +static inline void sched_account_irqstart(int cpu, struct task_struct *curr, + u64 wallclock) +{ +} + +static inline int sched_cpu_high_irqload(int cpu) { return 0; } + +static inline void set_preferred_cluster(struct related_thread_group *grp) { } + +static inline bool task_in_related_thread_group(struct task_struct *p) +{ + return false; +} + +static inline +struct related_thread_group *task_related_thread_group(struct task_struct *p) +{ + return NULL; +} + +static inline u32 task_load(struct task_struct *p) { return 0; } + +static inline int update_preferred_cluster(struct related_thread_group *grp, + struct task_struct *p, u32 old_load) +{ + return 0; +} + +static inline void add_new_task_to_grp(struct task_struct *new) {} + +#define sched_freq_legacy_mode 1 +#define sched_migration_fixup 0 +#define PRED_DEMAND_DELTA (0) + +static inline void +check_for_freq_change(struct rq *rq, bool check_pred, bool check_groups) { } + +static inline void notify_migration(int src_cpu, int dest_cpu, + bool src_cpu_dead, struct task_struct *p) { } + +static inline int same_freq_domain(int src_cpu, int dst_cpu) +{ + return 1; +} + +static inline void check_for_migration(struct rq *rq, struct task_struct *p) { } +static inline void pre_big_task_count_change(void) { } +static inline void post_big_task_count_change(void) { } +static inline void set_hmp_defaults(void) { } + +static inline void clear_reserved(int cpu) { } + +#define trace_sched_cpu_load(...) +#define trace_sched_cpu_load_lb(...) +#define trace_sched_cpu_load_cgroup(...) +#define trace_sched_cpu_load_wakeup(...) + +#endif /* CONFIG_SCHED_HMP */ + +/* + * Returns the rq capacity of any rq in a group. This does not play + * well with groups where rq capacity can change independently. + */ +#define group_rq_capacity(group) cpu_capacity(group_first_cpu(group)) + #ifdef CONFIG_CGROUP_SCHED /* @@ -1163,6 +1843,7 @@ static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ #define WF_FORK 0x02 /* child wakeup after fork */ #define WF_MIGRATED 0x4 /* internal use, task got migrated */ +#define WF_NO_NOTIFIER 0x08 /* do not notify governor */ /* * To aid in avoiding the subversion of "niceness" due to uneven distribution @@ -1278,6 +1959,10 @@ struct sched_class { #ifdef CONFIG_FAIR_GROUP_SCHED void (*task_change_group) (struct task_struct *p, int type); #endif +#ifdef CONFIG_SCHED_HMP + void (*fixup_hmp_sched_stats)(struct rq *rq, struct task_struct *p, + u32 new_task_load, u32 new_pred_demand); +#endif }; static inline void put_prev_task(struct rq *rq, struct task_struct *prev) @@ -1391,6 +2076,7 @@ static inline void add_nr_running(struct rq *rq, unsigned count) { unsigned prev_nr = rq->nr_running; + sched_update_nr_prod(cpu_of(rq), count, true); rq->nr_running = prev_nr + count; if (prev_nr < 2 && rq->nr_running >= 2) { @@ -1405,6 +2091,7 @@ static inline void add_nr_running(struct rq *rq, unsigned count) static inline void sub_nr_running(struct rq *rq, unsigned count) { + sched_update_nr_prod(cpu_of(rq), count, false); rq->nr_running -= count; /* Check if we still need preemption */ sched_update_tick_dependency(rq); @@ -1732,6 +2419,9 @@ enum rq_nohz_flag_bits { NOHZ_BALANCE_KICK, }; +#define NOHZ_KICK_ANY 0 +#define NOHZ_KICK_RESTRICT 1 + #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) extern void nohz_balance_exit_idle(unsigned int cpu); diff --git a/kernel/sched/sched_avg.c b/kernel/sched/sched_avg.c new file mode 100644 index 000000000000..c70e0466c36c --- /dev/null +++ b/kernel/sched/sched_avg.c @@ -0,0 +1,128 @@ +/* Copyright (c) 2012, 2015, The Linux Foundation. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 and + * only version 2 as published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + */ + +/* + * Scheduler hook for average runqueue determination + */ +#include +#include +#include +#include +#include + +#include "sched.h" +#include + +static DEFINE_PER_CPU(u64, nr_prod_sum); +static DEFINE_PER_CPU(u64, last_time); +static DEFINE_PER_CPU(u64, nr_big_prod_sum); +static DEFINE_PER_CPU(u64, nr); + +static DEFINE_PER_CPU(unsigned long, iowait_prod_sum); +static DEFINE_PER_CPU(spinlock_t, nr_lock) = __SPIN_LOCK_UNLOCKED(nr_lock); +static s64 last_get_time; + +/** + * sched_get_nr_running_avg + * @return: Average nr_running, iowait and nr_big_tasks value since last poll. + * Returns the avg * 100 to return up to two decimal points + * of accuracy. + * + * Obtains the average nr_running value since the last poll. + * This function may not be called concurrently with itself + */ +void sched_get_nr_running_avg(int *avg, int *iowait_avg, int *big_avg) +{ + int cpu; + u64 curr_time = sched_clock(); + u64 diff = curr_time - last_get_time; + u64 tmp_avg = 0, tmp_iowait = 0, tmp_big_avg = 0; + + *avg = 0; + *iowait_avg = 0; + *big_avg = 0; + + if (!diff) + return; + + /* read and reset nr_running counts */ + for_each_possible_cpu(cpu) { + unsigned long flags; + + spin_lock_irqsave(&per_cpu(nr_lock, cpu), flags); + curr_time = sched_clock(); + tmp_avg += per_cpu(nr_prod_sum, cpu); + tmp_avg += per_cpu(nr, cpu) * + (curr_time - per_cpu(last_time, cpu)); + + tmp_big_avg += per_cpu(nr_big_prod_sum, cpu); + tmp_big_avg += nr_eligible_big_tasks(cpu) * + (curr_time - per_cpu(last_time, cpu)); + + tmp_iowait += per_cpu(iowait_prod_sum, cpu); + tmp_iowait += nr_iowait_cpu(cpu) * + (curr_time - per_cpu(last_time, cpu)); + + per_cpu(last_time, cpu) = curr_time; + + per_cpu(nr_prod_sum, cpu) = 0; + per_cpu(nr_big_prod_sum, cpu) = 0; + per_cpu(iowait_prod_sum, cpu) = 0; + + spin_unlock_irqrestore(&per_cpu(nr_lock, cpu), flags); + } + + diff = curr_time - last_get_time; + last_get_time = curr_time; + + *avg = (int)div64_u64(tmp_avg * 100, diff); + *big_avg = (int)div64_u64(tmp_big_avg * 100, diff); + *iowait_avg = (int)div64_u64(tmp_iowait * 100, diff); + + trace_sched_get_nr_running_avg(*avg, *big_avg, *iowait_avg); + + BUG_ON(*avg < 0 || *big_avg < 0 || *iowait_avg < 0); + pr_debug("%s - avg:%d big_avg:%d iowait_avg:%d\n", + __func__, *avg, *big_avg, *iowait_avg); +} +EXPORT_SYMBOL(sched_get_nr_running_avg); + +/** + * sched_update_nr_prod + * @cpu: The core id of the nr running driver. + * @delta: Adjust nr by 'delta' amount + * @inc: Whether we are increasing or decreasing the count + * @return: N/A + * + * Update average with latest nr_running value for CPU + */ +void sched_update_nr_prod(int cpu, long delta, bool inc) +{ + int diff; + s64 curr_time; + unsigned long flags, nr_running; + + spin_lock_irqsave(&per_cpu(nr_lock, cpu), flags); + nr_running = per_cpu(nr, cpu); + curr_time = sched_clock(); + diff = curr_time - per_cpu(last_time, cpu); + per_cpu(last_time, cpu) = curr_time; + per_cpu(nr, cpu) = nr_running + (inc ? delta : -delta); + + BUG_ON((s64)per_cpu(nr, cpu) < 0); + + per_cpu(nr_prod_sum, cpu) += nr_running * diff; + per_cpu(nr_big_prod_sum, cpu) += nr_eligible_big_tasks(cpu) * diff; + per_cpu(iowait_prod_sum, cpu) += nr_iowait_cpu(cpu) * diff; + spin_unlock_irqrestore(&per_cpu(nr_lock, cpu), flags); +} +EXPORT_SYMBOL(sched_update_nr_prod); diff --git a/kernel/sched/stop_task.c b/kernel/sched/stop_task.c index 604297a08b3a..427bc8f8b88c 100644 --- a/kernel/sched/stop_task.c +++ b/kernel/sched/stop_task.c @@ -17,6 +17,41 @@ select_task_rq_stop(struct task_struct *p, int cpu, int sd_flag, int flags) } #endif /* CONFIG_SMP */ +#ifdef CONFIG_SCHED_HMP + +static void +inc_hmp_sched_stats_stop(struct rq *rq, struct task_struct *p) +{ + inc_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +static void +dec_hmp_sched_stats_stop(struct rq *rq, struct task_struct *p) +{ + dec_cumulative_runnable_avg(&rq->hmp_stats, p); +} + +static void +fixup_hmp_sched_stats_stop(struct rq *rq, struct task_struct *p, + u32 new_task_load, u32 new_pred_demand) +{ + s64 task_load_delta = (s64)new_task_load - task_load(p); + s64 pred_demand_delta = PRED_DEMAND_DELTA; + + fixup_cumulative_runnable_avg(&rq->hmp_stats, p, task_load_delta, + pred_demand_delta); +} + +#else /* CONFIG_SCHED_HMP */ + +static inline void +inc_hmp_sched_stats_stop(struct rq *rq, struct task_struct *p) { } + +static inline void +dec_hmp_sched_stats_stop(struct rq *rq, struct task_struct *p) { } + +#endif /* CONFIG_SCHED_HMP */ + static void check_preempt_curr_stop(struct rq *rq, struct task_struct *p, int flags) { @@ -42,12 +77,14 @@ static void enqueue_task_stop(struct rq *rq, struct task_struct *p, int flags) { add_nr_running(rq, 1); + inc_hmp_sched_stats_stop(rq, p); } static void dequeue_task_stop(struct rq *rq, struct task_struct *p, int flags) { sub_nr_running(rq, 1); + dec_hmp_sched_stats_stop(rq, p); } static void yield_task_stop(struct rq *rq) @@ -134,4 +171,7 @@ const struct sched_class stop_sched_class = { .prio_changed = prio_changed_stop, .switched_to = switched_to_stop, .update_curr = update_curr_stop, +#ifdef CONFIG_SCHED_HMP + .fixup_hmp_sched_stats = fixup_hmp_sched_stats_stop, +#endif }; diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 706309f9ed84..fe6537188085 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -288,6 +288,190 @@ static struct ctl_table kern_table[] = { .mode = 0644, .proc_handler = proc_dointvec, }, + { + .procname = "sched_wake_to_idle", + .data = &sysctl_sched_wake_to_idle, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec, + }, +#ifdef CONFIG_SCHED_HMP + { + .procname = "sched_freq_inc_notify", + .data = &sysctl_sched_freq_inc_notify, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &zero, + }, + { + .procname = "sched_freq_dec_notify", + .data = &sysctl_sched_freq_dec_notify, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &zero, + }, + { + .procname = "sched_cpu_high_irqload", + .data = &sysctl_sched_cpu_high_irqload, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec, + }, + { + .procname = "sched_ravg_hist_size", + .data = &sysctl_sched_ravg_hist_size, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_window_update_handler, + }, + { + .procname = "sched_window_stats_policy", + .data = &sysctl_sched_window_stats_policy, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_window_update_handler, + }, + { + .procname = "sched_spill_load", + .data = &sysctl_sched_spill_load_pct, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + .extra2 = &one_hundred, + }, + { + .procname = "sched_spill_nr_run", + .data = &sysctl_sched_spill_nr_run, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &zero, + }, + { + .procname = "sched_upmigrate", + .data = &sysctl_sched_upmigrate_pct, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + .extra2 = &one_hundred, + }, + { + .procname = "sched_downmigrate", + .data = &sysctl_sched_downmigrate_pct, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + .extra2 = &one_hundred, + }, + { + .procname = "sched_init_task_load", + .data = &sysctl_sched_init_task_load_pct, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + .extra2 = &one_hundred, + }, + { + .procname = "sched_select_prev_cpu_us", + .data = &sysctl_sched_select_prev_cpu_us, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + }, + { + .procname = "sched_enable_colocation", + .data = &sysctl_sched_enable_colocation, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec, + .extra1 = &zero, + .extra2 = &one, + }, + { + .procname = "sched_restrict_cluster_spill", + .data = &sysctl_sched_restrict_cluster_spill, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &zero, + .extra2 = &one, + }, + { + .procname = "sched_small_wakee_task_load", + .data = &sysctl_sched_small_wakee_task_load_pct, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + .extra2 = &one_hundred, + }, + { + .procname = "sched_big_waker_task_load", + .data = &sysctl_sched_big_waker_task_load_pct, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + .extra2 = &one_hundred, + }, + { + .procname = "sched_enable_thread_grouping", + .data = &sysctl_sched_enable_thread_grouping, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec, + }, + { + .procname = "sched_new_task_windows", + .data = &sysctl_sched_new_task_windows, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_window_update_handler, + }, + { + .procname = "sched_pred_alert_freq", + .data = &sysctl_sched_pred_alert_freq, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = &zero, + }, + { + .procname = "sched_freq_aggregate", + .data = &sysctl_sched_freq_aggregate, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_window_update_handler, + }, + { + .procname = "sched_freq_aggregate_threshold", + .data = &sysctl_sched_freq_aggregate_threshold_pct, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_hmp_proc_update_handler, + .extra1 = &zero, + /* + * Special handling for sched_freq_aggregate_threshold_pct + * which can be greater than 100. Use 1000 as an upper bound + * value which works for all practical use cases. + */ + .extra2 = &one_thousand, + }, + { + .procname = "sched_boost", + .data = &sysctl_sched_boost, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_boost_handler, + }, +#endif /* CONFIG_SCHED_HMP */ #ifdef CONFIG_SCHED_DEBUG { .procname = "sched_min_granularity_ns", diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c index 11ce935633f4..189e77b4bb25 100644 --- a/kernel/time/tick-sched.c +++ b/kernel/time/tick-sched.c @@ -51,6 +51,21 @@ struct tick_sched *tick_get_tick_sched(int cpu) */ static ktime_t last_jiffies_update; +u64 jiffy_to_ktime_ns(u64 *now, u64 *jiffy_ktime_ns) +{ + u64 cur_jiffies; + unsigned long seq; + + do { + seq = read_seqbegin(&jiffies_lock); + *now = ktime_get_ns(); + *jiffy_ktime_ns = ktime_to_ns(last_jiffies_update); + cur_jiffies = get_jiffies_64(); + } while (read_seqretry(&jiffies_lock, seq)); + + return cur_jiffies; +} + /* * Must be called with interrupts disabled ! */ diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug index 5d663b322a3e..9747e380bf47 100644 --- a/lib/Kconfig.debug +++ b/lib/Kconfig.debug @@ -935,6 +935,25 @@ config SCHED_INFO bool default n +config PANIC_ON_SCHED_BUG + bool "Panic on all bugs encountered by the scheduler" + help + Say Y here to panic on all 'BUG:' conditions encountered by the + scheduler, even potentially-recoverable ones such as scheduling + while atomic, sleeping from invalid context, and detection of + broken arch topologies. + + Say N if unsure. + +config PANIC_ON_RT_THROTTLING + bool "Panic on RT throttling" + help + Say Y here to enable the kernel to panic when a realtime + runqueue is throttled. This may be useful for detecting + and debugging RT throttling issues. + + Say N if unsure. + config SCHEDSTATS bool "Collect scheduler statistics" depends on DEBUG_KERNEL && PROC_FS