1 =================
2 Scheduler Domains
3 =================
4
5 Each CPU has a "base" scheduling domain (struc
6 hierarchy is built from these base domains via
7 MUST be NULL terminated, and domain structures
8 locklessly updated.
9
10 Each scheduling domain spans a number of CPUs
11 A domain's span MUST be a superset of it child
12 be relaxed if the need arises), and a base dom
13 i. The top domain for each CPU will generally
14 although strictly it doesn't have to, but this
15 CPUs will never be given tasks to run unless t
16 explicitly set. A sched domain's span means "b
17 CPUs".
18
19 Each scheduling domain must have one or more C
20 which are organised as a circular one way link
21 pointer. The union of cpumasks of these groups
22 domain's span. The group pointed to by the ->g
23 to which the domain belongs. Groups may be sha
24 read only data after they have been set up. Th
25 any two of these groups may be non empty. If t
26 flag is set on the corresponding scheduling do
27 shared between CPUs.
28
29 Balancing within a sched domain occurs between
30 is treated as one entity. The load of a group
31 load of each of its member CPUs, and only when
32 out of balance are tasks moved between groups.
33
34 In kernel/sched/core.c, sched_balance_trigger(
35 through sched_tick(). It raises a softirq afte
36 rebalancing event for the current runqueue has
37 balancing workhorse, sched_balance_softirq()->
38 in softirq context (SCHED_SOFTIRQ).
39
40 The latter function takes two arguments: the r
41 the CPU was idle at the time the sched_tick()
42 sched domains our CPU is on, starting from its
43 chain. While doing that, it checks to see if t
44 rebalance interval. If so, it runs sched_balan
45 the parent sched_domain (if it exists), and th
46 forth.
47
48 Initially, sched_balance_rq() finds the busies
49 If it succeeds, it looks for the busiest runqu
50 that group. If it manages to find such a runqu
51 CPU's runqueue and the newly found busiest one
52 to our runqueue. The exact number of tasks amo
53 computed while iterating over this sched domai
54
55 Implementing sched domains
56 ==========================
57
58 The "base" domain will "span" the first level
59 of SMT, you'll span all siblings of the physic
60 a single virtual CPU.
61
62 In SMP, the parent of the base domain will spa
63 node. Each group being a single physical CPU.
64 of the SMP domain will span the entire machine
65 cpumask of a node. Or, you could do multi-leve
66 might have just one domain covering its one NU
67
68 The implementor should read comments in includ
69 SD_* to get an idea of the specifics and what
70 of a sched_domain.
71
72 Architectures may override the generic domain
73 for a given topology level by creating a sched
74 calling set_sched_topology() with this array a
75
76 The sched-domains debugging infrastructure can
77 CONFIG_SCHED_DEBUG and adding 'sched_verbose'
78 forgot to tweak your cmdline, you can also fli
79 /sys/kernel/debug/sched/verbose knob. This ena
80 the sched domains which should catch most poss
81 also prints out the domain structure in a visu
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.