Expand description
The scheduler: tasks.c and queue.c as a state machine over indices.
Every public function here is a FreeRTOS API function, and its body
follows the C body statement for statement — including where the
critical sections open and close, because on the sim that is where
time passes (ORACLES.md, sim contract v1, rule 3), and including where
the trace* macros fire, because the order of those lines is the gate.
The subtlety that costs a diff if you miss it: FreeRTOS yields from
inside the critical section in vTaskResume, vTaskPrioritySet,
xTaskResumeAll and the queue calls, and from outside it in
vTaskSuspend and vTaskDelay. Since portYIELD() on the Posix port
is itself a critical section, an inside yield nests (and counts one
outermost exit) where an outside yield counts two. Each site below says
which it is.
Where the C has a configASSERT, this has an Error: the C kernel
stops the world on a bad handle, and the whole point of the remake is
that we do not.
Structs§
- Kernel
- The scheduler.
- Start
Handles - What
Kernel::start_schedulercreated, so a runner can attach bodies.
Enums§
- Notify
Action eNotifyAction: what a notification does to the value already there.- Notify
State ucNotifyState[]: what a task’s notification slot is doing.- Stall
- A trace line owed by a task that was switched out before it could
emit one. Only the queue failure paths can owe one: they are the only
places FreeRTOS traces after leaving a critical section.
Why
vTaskSwitchContextcould not choose a task to run. - Task
State eTaskState: what a task is doing, derived from the list it is in exactly aseTaskGetStatederives it.
Constants§
- MAX_
NOTIFICATION_ ENTRIES - How many notification slots a task has room for.