xref: /linux/Documentation/scheduler/sched-domains.rst (revision 3f41368fbfe1b3d5922d317fe1a0a0cab6846802)
1=================
2Scheduler Domains
3=================
4
5Each CPU has a "base" scheduling domain (struct sched_domain). The domain
6hierarchy is built from these base domains via the ->parent pointer. ->parent
7MUST be NULL terminated, and domain structures should be per-CPU as they are
8locklessly updated.
9
10Each scheduling domain spans a number of CPUs (stored in the ->span field).
11A domain's span MUST be a superset of it child's span (this restriction could
12be relaxed if the need arises), and a base domain for CPU i MUST span at least
13i. The top domain for each CPU will generally span all CPUs in the system
14although strictly it doesn't have to, but this could lead to a case where some
15CPUs will never be given tasks to run unless the CPUs allowed mask is
16explicitly set. A sched domain's span means "balance process load among these
17CPUs".
18
19Each scheduling domain must have one or more CPU groups (struct sched_group)
20which are organised as a circular one way linked list from the ->groups
21pointer. The union of cpumasks of these groups MUST be the same as the
22domain's span. The group pointed to by the ->groups pointer MUST contain the CPU
23to which the domain belongs. Groups may be shared among CPUs as they contain
24read only data after they have been set up. The intersection of cpumasks from
25any two of these groups may be non empty. If this is the case the SD_OVERLAP
26flag is set on the corresponding scheduling domain and its groups may not be
27shared between CPUs.
28
29Balancing within a sched domain occurs between groups. That is, each group
30is treated as one entity. The load of a group is defined as the sum of the
31load of each of its member CPUs, and only when the load of a group becomes
32out of balance are tasks moved between groups.
33
34In kernel/sched/core.c, sched_balance_trigger() is run periodically on each CPU
35through sched_tick(). It raises a softirq after the next regularly scheduled
36rebalancing event for the current runqueue has arrived. The actual load
37balancing workhorse, sched_balance_softirq()->sched_balance_domains(), is then run
38in softirq context (SCHED_SOFTIRQ).
39
40The latter function takes two arguments: the runqueue of current CPU and whether
41the CPU was idle at the time the sched_tick() happened and iterates over all
42sched domains our CPU is on, starting from its base domain and going up the ->parent
43chain. While doing that, it checks to see if the current domain has exhausted its
44rebalance interval. If so, it runs sched_balance_rq() on that domain. It then checks
45the parent sched_domain (if it exists), and the parent of the parent and so
46forth.
47
48Initially, sched_balance_rq() finds the busiest group in the current sched domain.
49If it succeeds, it looks for the busiest runqueue of all the CPUs' runqueues in
50that group. If it manages to find such a runqueue, it locks both our initial
51CPU's runqueue and the newly found busiest one and starts moving tasks from it
52to our runqueue. The exact number of tasks amounts to an imbalance previously
53computed while iterating over this sched domain's groups.
54
55Implementing sched domains
56==========================
57
58The "base" domain will "span" the first level of the hierarchy. In the case
59of SMT, you'll span all siblings of the physical CPU, with each group being
60a single virtual CPU.
61
62In SMP, the parent of the base domain will span all physical CPUs in the
63node. Each group being a single physical CPU. Then with NUMA, the parent
64of the SMP domain will span the entire machine, with each group having the
65cpumask of a node. Or, you could do multi-level NUMA or Opteron, for example,
66might have just one domain covering its one NUMA level.
67
68The implementor should read comments in include/linux/sched/sd_flags.h:
69SD_* to get an idea of the specifics and what to tune for the SD flags
70of a sched_domain.
71
72Architectures may override the generic domain builder and the default SD flags
73for a given topology level by creating a sched_domain_topology_level array and
74calling set_sched_topology() with this array as the parameter.
75
76The sched-domains debugging infrastructure can be enabled by enabling
77CONFIG_SCHED_DEBUG and adding 'sched_verbose' to your cmdline. If you
78forgot to tweak your cmdline, you can also flip the
79/sys/kernel/debug/sched/verbose knob. This enables an error checking parse of
80the sched domains which should catch most possible errors (described above). It
81also prints out the domain structure in a visual format.
82