Struct Scheduler

pub struct Scheduler<P>
where
    P: Platform,
{
    pub tasklet_scheduler: TaskletScheduler,
    /* private fields */
}

The global Scheduler coordinates the main ‘run loop’ of the kernel, allocating CPU time to userspace tasks. There is one global Scheduler instance, which then holds a CpuScheduler for each running processor to coordinate tasks running on that processor.

It is also responsible for managing spawned kernel asynchronous tasklets (which are somewhat confusingly also often called Tasks) - this involves tracking tasks that have been ‘woken’ (are ready to make progress) and making sure they are polled regularly. The forward progress of both userspace tasks and kernel tasklets are intertwined, and so are managed together.

Fields

tasklet_scheduler: TaskletScheduler

Implementations

impl<P> Scheduler<P>

where P: Platform,

pub fn new() -> Scheduler<P>

pub fn add_task(&self, task: Arc<Task<P>>)

pub fn for_this_cpu(&self) -> SpinlockGuard<'_, CpuScheduler<P>>

pub fn start_scheduling(&self) -> !

Start scheduling! This should be called after a platform has finished initializing, and is diverging. It gives kernel tasklets an initial poll while we’re here in the kernel, and then drops down into userspace.

pub fn schedule(&self, new_state: TaskState)

Called when a userspace task yields or is pre-empted. This is responsible for the ‘scheduling’ part of the scheduler - it polls kernel tasklets as they need attention, and shares CPU time between userspace tasks.

On each call to schedule, the kernel can choose to:

• Give CPU time to the kernel-space tasklet scheduler
• Switch to another userspace task
• Steal work from another CPU’s scheduler
• Idle the CPU, if there is nothing to be done
• Nothing

If the current task is switched away from, it will be placed in the state new_state. This allows the caller to block the current task on a dependency. If a task has been pre-empted or yields, it should be placed into TaskState::Ready.