Linux Kernel Architecture

Chapter 2: Process Management and Scheduling

class->

move_one_task

No migration

requests?

schedule

Get migration

request

Iterate over all

scheduler classes

1

1

migration_thread

rq->active_balance set?

active_load_balance

move_one_task

__migrate_task

Figure 2-27: Code flow diagram formigration_thread.

migration_threadruns an infinite loop and sleeps when there is nothing to do. First of all, the function

checks if active balancing is required, and if this is the case,active_load_balanceis called to satisfy this

request. The function tries to move one task from the current run queue to the run queue of the CPU that

initiated the request for active balancing. It usesmove_one_taskforthispurpose,which,inturn,ends

up calling the scheduler-class specificmove_one_taskfunctions of all scheduler classes until one of them

succeeds. Note that these functions try to move processes more aggressively thanload_balance.For

instance, they do not perform the previously mentioned priority comparison, so they are more likely to

succeed.

Once the active load balancing is finished, the migration thread checks if any migration requests from the

scheduler are pending in themigrate_reqlist. If none is available, the thread can reschedule. Otherwise,

the request is fulfilled with__migrate_task, which performs the desired process movement directly

without further interaction with the scheduler classes.

Core SchedulerChanges

Besides the additions discussed above, some changes to the existing methods are required in the core

scheduler on SMP systems. While numerous small details change all over the place, the most important

differences as compared to uniprocessor systems are the following:

❑ When a new process is started with theexecsystem call, a good opportunity for the sched-

uler to move the task across CPUs arises. Naturally, it has not been running yet, so there can-

not be any negative effects on the CPU cache by moving the task to another CPU.sched_exec

is the hook function invoked by theexecsystem call, and the code flow diagram is shown in

Figure 2-28.

sched_balance_selfpicks the CPU that is currently least loaded (and on which the process

is also allowed to run). If this is not the current CPU, thensched_migrate_taskforwards an

according migration request to the migration thread usingsched_migrate_task.

❑ The scheduling granularity of the completely fair scheduler scales with the number of CPUs.

The more processors present in the system, the larger the granularities that can be employed.

Bothsysctl_sched_min_granularityandsysctl_sched_latencyforsysctl_sched_min_