Commit | Line | Data |
---|---|---|
e2495b57 | 1 | Each CPU has a "base" scheduling domain (struct sched_domain). The domain |
1da177e4 | 2 | hierarchy is built from these base domains via the ->parent pointer. ->parent |
e2495b57 BP |
3 | MUST be NULL terminated, and domain structures should be per-CPU as they are |
4 | locklessly updated. | |
1da177e4 LT |
5 | |
6 | Each scheduling domain spans a number of CPUs (stored in the ->span field). | |
7 | A domain's span MUST be a superset of it child's span (this restriction could | |
8 | be relaxed if the need arises), and a base domain for CPU i MUST span at least | |
9 | i. The top domain for each CPU will generally span all CPUs in the system | |
10 | although strictly it doesn't have to, but this could lead to a case where some | |
11 | CPUs will never be given tasks to run unless the CPUs allowed mask is | |
12 | explicitly set. A sched domain's span means "balance process load among these | |
13 | CPUs". | |
14 | ||
15 | Each scheduling domain must have one or more CPU groups (struct sched_group) | |
16 | which are organised as a circular one way linked list from the ->groups | |
17 | pointer. The union of cpumasks of these groups MUST be the same as the | |
18 | domain's span. The intersection of cpumasks from any two of these groups | |
19 | MUST be the empty set. The group pointed to by the ->groups pointer MUST | |
20 | contain the CPU to which the domain belongs. Groups may be shared among | |
21 | CPUs as they contain read only data after they have been set up. | |
22 | ||
23 | Balancing within a sched domain occurs between groups. That is, each group | |
24 | is treated as one entity. The load of a group is defined as the sum of the | |
25 | load of each of its member CPUs, and only when the load of a group becomes | |
26 | out of balance are tasks moved between groups. | |
27 | ||
e2495b57 BP |
28 | In kernel/sched.c, trigger_load_balance() is run periodically on each CPU |
29 | through scheduler_tick(). It raises a softirq after the next regularly scheduled | |
30 | rebalancing event for the current runqueue has arrived. The actual load | |
31 | balancing workhorse, run_rebalance_domains()->rebalance_domains(), is then run | |
32 | in softirq context (SCHED_SOFTIRQ). | |
33 | ||
34 | The latter function takes two arguments: the current CPU and whether it was idle | |
35 | at the time the scheduler_tick() happened and iterates over all sched domains | |
36 | our CPU is on, starting from its base domain and going up the ->parent chain. | |
37 | While doing that, it checks to see if the current domain has exhausted its | |
38 | rebalance interval. If so, it runs load_balance() on that domain. It then checks | |
39 | the parent sched_domain (if it exists), and the parent of the parent and so | |
40 | forth. | |
41 | ||
42 | Initially, load_balance() finds the busiest group in the current sched domain. | |
43 | If it succeeds, it looks for the busiest runqueue of all the CPUs' runqueues in | |
44 | that group. If it manages to find such a runqueue, it locks both our initial | |
45 | CPU's runqueue and the newly found busiest one and starts moving tasks from it | |
46 | to our runqueue. The exact number of tasks amounts to an imbalance previously | |
47 | computed while iterating over this sched domain's groups. | |
1da177e4 LT |
48 | |
49 | *** Implementing sched domains *** | |
50 | The "base" domain will "span" the first level of the hierarchy. In the case | |
51 | of SMT, you'll span all siblings of the physical CPU, with each group being | |
52 | a single virtual CPU. | |
53 | ||
54 | In SMP, the parent of the base domain will span all physical CPUs in the | |
55 | node. Each group being a single physical CPU. Then with NUMA, the parent | |
56 | of the SMP domain will span the entire machine, with each group having the | |
57 | cpumask of a node. Or, you could do multi-level NUMA or Opteron, for example, | |
58 | might have just one domain covering its one NUMA level. | |
59 | ||
60 | The implementor should read comments in include/linux/sched.h: | |
61 | struct sched_domain fields, SD_FLAG_*, SD_*_INIT to get an idea of | |
62 | the specifics and what to tune. | |
63 | ||
1da177e4 LT |
64 | Architectures may retain the regular override the default SD_*_INIT flags |
65 | while using the generic domain builder in kernel/sched.c if they wish to | |
66 | retain the traditional SMT->SMP->NUMA topology (or some subset of that). This | |
67 | can be done by #define'ing ARCH_HASH_SCHED_TUNE. | |
68 | ||
69 | Alternatively, the architecture may completely override the generic domain | |
70 | builder by #define'ing ARCH_HASH_SCHED_DOMAIN, and exporting your | |
71 | arch_init_sched_domains function. This function will attach domains to all | |
72 | CPUs using cpu_attach_domain. | |
73 | ||
e29c98d1 GS |
74 | The sched-domains debugging infrastructure can be enabled by enabling |
75 | CONFIG_SCHED_DEBUG. This enables an error checking parse of the sched domains | |
1da177e4 LT |
76 | which should catch most possible errors (described above). It also prints out |
77 | the domain structure in a visual format. |