diff options
Diffstat (limited to 'Documentation/scheduler')
| -rw-r--r-- | Documentation/scheduler/index.rst | 2 | ||||
| -rw-r--r-- | Documentation/scheduler/sched-debug.rst | 54 | ||||
| -rw-r--r-- | Documentation/scheduler/sched-domains.rst | 8 | ||||
| -rw-r--r-- | Documentation/scheduler/schedutil.rst (renamed from Documentation/scheduler/schedutil.txt) | 30 |
4 files changed, 77 insertions, 17 deletions
diff --git a/Documentation/scheduler/index.rst b/Documentation/scheduler/index.rst index 88900aabdbf7..b430d856056a 100644 --- a/Documentation/scheduler/index.rst +++ b/Documentation/scheduler/index.rst @@ -14,9 +14,11 @@ Linux Scheduler sched-domains sched-capacity sched-energy + schedutil sched-nice-design sched-rt-group sched-stats + sched-debug text_files diff --git a/Documentation/scheduler/sched-debug.rst b/Documentation/scheduler/sched-debug.rst new file mode 100644 index 000000000000..4d3d24f2a439 --- /dev/null +++ b/Documentation/scheduler/sched-debug.rst @@ -0,0 +1,54 @@ +================= +Scheduler debugfs +================= + +Booting a kernel with CONFIG_SCHED_DEBUG=y will give access to +scheduler specific debug files under /sys/kernel/debug/sched. Some of +those files are described below. + +numa_balancing +============== + +`numa_balancing` directory is used to hold files to control NUMA +balancing feature. If the system overhead from the feature is too +high then the rate the kernel samples for NUMA hinting faults may be +controlled by the `scan_period_min_ms, scan_delay_ms, +scan_period_max_ms, scan_size_mb` files. + + +scan_period_min_ms, scan_delay_ms, scan_period_max_ms, scan_size_mb +------------------------------------------------------------------- + +Automatic NUMA balancing scans tasks address space and unmaps pages to +detect if pages are properly placed or if the data should be migrated to a +memory node local to where the task is running. Every "scan delay" the task +scans the next "scan size" number of pages in its address space. When the +end of the address space is reached the scanner restarts from the beginning. + +In combination, the "scan delay" and "scan size" determine the scan rate. +When "scan delay" decreases, the scan rate increases. The scan delay and +hence the scan rate of every task is adaptive and depends on historical +behaviour. If pages are properly placed then the scan delay increases, +otherwise the scan delay decreases. The "scan size" is not adaptive but +the higher the "scan size", the higher the scan rate. + +Higher scan rates incur higher system overhead as page faults must be +trapped and potentially data must be migrated. However, the higher the scan +rate, the more quickly a tasks memory is migrated to a local node if the +workload pattern changes and minimises performance impact due to remote +memory accesses. These files control the thresholds for scan delays and +the number of pages scanned. + +``scan_period_min_ms`` is the minimum time in milliseconds to scan a +tasks virtual memory. It effectively controls the maximum scanning +rate for each task. + +``scan_delay_ms`` is the starting "scan delay" used for a task when it +initially forks. + +``scan_period_max_ms`` is the maximum time in milliseconds to scan a +tasks virtual memory. It effectively controls the minimum scanning +rate for each task. + +``scan_size_mb`` is how many megabytes worth of pages are scanned for +a given scan. diff --git a/Documentation/scheduler/sched-domains.rst b/Documentation/scheduler/sched-domains.rst index 84dcdcd2911c..e57ad28301bd 100644 --- a/Documentation/scheduler/sched-domains.rst +++ b/Documentation/scheduler/sched-domains.rst @@ -37,10 +37,10 @@ rebalancing event for the current runqueue has arrived. The actual load balancing workhorse, run_rebalance_domains()->rebalance_domains(), is then run in softirq context (SCHED_SOFTIRQ). -The latter function takes two arguments: the current CPU and whether it was idle -at the time the scheduler_tick() happened and iterates over all sched domains -our CPU is on, starting from its base domain and going up the ->parent chain. -While doing that, it checks to see if the current domain has exhausted its +The latter function takes two arguments: the runqueue of current CPU and whether +the CPU was idle at the time the scheduler_tick() happened and iterates over all +sched domains our CPU is on, starting from its base domain and going up the ->parent +chain. While doing that, it checks to see if the current domain has exhausted its rebalance interval. If so, it runs load_balance() on that domain. It then checks the parent sched_domain (if it exists), and the parent of the parent and so forth. diff --git a/Documentation/scheduler/schedutil.txt b/Documentation/scheduler/schedutil.rst index 78f6b91e2291..32c7d69fc86c 100644 --- a/Documentation/scheduler/schedutil.txt +++ b/Documentation/scheduler/schedutil.rst @@ -1,11 +1,15 @@ +========= +Schedutil +========= +.. note:: -NOTE; all this assumes a linear relation between frequency and work capacity, -we know this is flawed, but it is the best workable approximation. + All this assumes a linear relation between frequency and work capacity, + we know this is flawed, but it is the best workable approximation. PELT (Per Entity Load Tracking) -------------------------------- +=============================== With PELT we track some metrics across the various scheduler entities, from individual tasks to task-group slices to CPU runqueues. As the basis for this @@ -38,8 +42,8 @@ while 'runnable' will increase to reflect the amount of contention. For more detail see: kernel/sched/pelt.c -Frequency- / CPU Invariance ---------------------------- +Frequency / CPU Invariance +========================== Because consuming the CPU for 50% at 1GHz is not the same as consuming the CPU for 50% at 2GHz, nor is running 50% on a LITTLE CPU the same as running 50% on @@ -47,7 +51,7 @@ a big CPU, we allow architectures to scale the time delta with two ratios, one Dynamic Voltage and Frequency Scaling (DVFS) ratio and one microarch ratio. For simple DVFS architectures (where software is in full control) we trivially -compute the ratio as: +compute the ratio as:: f_cur r_dvfs := ----- @@ -55,7 +59,7 @@ compute the ratio as: For more dynamic systems where the hardware is in control of DVFS we use hardware counters (Intel APERF/MPERF, ARMv8.4-AMU) to provide us this ratio. -For Intel specifically, we use: +For Intel specifically, we use:: APERF f_cur := ----- * P0 @@ -87,7 +91,7 @@ For more detail see: UTIL_EST / UTIL_EST_FASTUP --------------------------- +========================== Because periodic tasks have their averages decayed while they sleep, even though when running their expected utilization will be the same, they suffer a @@ -106,7 +110,7 @@ For more detail see: kernel/sched/fair.c:util_est_dequeue() UCLAMP ------- +====== It is possible to set effective u_min and u_max clamps on each CFS or RT task; the runqueue keeps an max aggregate of these clamps for all running tasks. @@ -115,7 +119,7 @@ For more detail see: include/uapi/linux/sched/types.h Schedutil / DVFS ----------------- +================ Every time the scheduler load tracking is updated (task wakeup, task migration, time progression) we call out to schedutil to update the hardware @@ -123,7 +127,7 @@ DVFS state. The basis is the CPU runqueue's 'running' metric, which per the above it is the frequency invariant utilization estimate of the CPU. From this we compute -a desired frequency like: +a desired frequency like:: max( running, util_est ); if UTIL_EST u_cfs := { running; otherwise @@ -135,7 +139,7 @@ a desired frequency like: f_des := min( f_max, 1.25 u * f_max ) -XXX IO-wait; when the update is due to a task wakeup from IO-completion we +XXX IO-wait: when the update is due to a task wakeup from IO-completion we boost 'u' above. This frequency is then used to select a P-state/OPP or directly munged into a @@ -153,7 +157,7 @@ For more information see: kernel/sched/cpufreq_schedutil.c NOTES ------ +===== - On low-load scenarios, where DVFS is most relevant, the 'running' numbers will closely reflect utilization. |