The Rockbox kernel
General
The Rockbox kernel is an advanced priority aware cooperative kernel. It consists of thin threads with some means of passing messages and locking data structures.
Threads
A thread is simply a normal C function that never returns. It has its own stack, and is executed until it allows the kernel to run another thread. Threads are executed in the order they were created.
The kernel related functions are declared in thread.h and kernel.h.
When the kernel switches the execution to another thread, it is called a context switch. You can force a context switch by calling the yield() function, therefore a forced context switch is often called yielding. A function that may cause a context switch is often called a blocking function, or a function that blocks.
The thread context
The thread context consists of the CPU registers that are normally preserved in normal C calls, plus the stack pointer. The "scratch registers" are not included. Also, no extra registers, such as MAC accumulators or status registers are preserved.
Thread functions
struct thread_entry* create_thread(void* function, void* stack, int stack_size, const char *name IF_PRIO(, int priority) IF_COP(, unsigned int core, bool fallback))
Creates a thread. It is immediately inserted in the list of threads, ready to execute. A thread function must not return. You need to use the remove_thread() function to remove a thread. A pointer to the thread entry is returned by create_thread. It is generally of no interest, but is is needed if you want to remove a thread.
The priority is the priority at which the thread is scheduled, as defined in thread.h. The core is the core which you want the thread to run on on dual processor targets (iPod, iriver H10, Sansa e200.) This will generally be "CPU" (the main CPU), but in certain cases may be "COP" (the coprocessor). Fallback defines the behaviour if a thread cannot be created on the selected core - true to attempt to create the thread on the other core, false to fail immediately. Normally, if a thread can't be created on the coprocessor (for example, early Rockbox bootloaders do not initialise the coprocessor correctly) it is desirable to try to create it on the main CPU, however a thread that has not been written with dual core use in mind should not be created on the coprocessor.
void remove_thread(struct thread_entry *thread)
Removes a thread from the kernel. The thread parameter is the thread entry returned by create_thread().
void yield(void)
Forces a context switch. The next thread in the list will be executed.
void sleep(int ticks)
Sleeps for ticks number of system ticks, allowing other threads to execute.
Complete thread example
You create a thread in three steps:
- Write a C function. The thread function should not take any arguments, and obviously not return anything.
- Reserve some stack. A normal thread should not use that much stack. If in doubt, use the DEFAULT_STACK_SIZE macro, which should be large enough for simple threads. Most CPU's want the stack long-aligned, so it might be a good idea to use the 'long' type.
- Call
create_thread()
%CODE{"cpp"}%
#include "thread.h"
/* My thread function */
void my_thread(void)
{
while(1) {
yield();
}
}
static long my_stack[DEFAULT_STACK_SIZE/sizeof(long)];
void init_my_thread(void)
{
create_thread(my_thread, my_stack, sizeof(my_stack), "my thread", IF_PRIO(, PRIORITY_BACKGROUND) IF_COP(, CPU, false));
}
%ENDCODE%
The system tick
There is a general timing mechanism called the system tick. It is a 32-bit counter that is increased HZ times per second. You can read the current tick count in the current_tick global variable.
Tick helper functions
TIME_AFTER(a, b)
Returns true if the a count is later than b.
Example:
%CODE{"cpp"}%
unsigned long my_delay = current_tick + HZ;
if(TIME_AFTER(current_tick, my_delay)) {
splash(HZ, true, "1 second has passed");
}
%ENDCODE%
TIME_BEFORE(a, b)
Returns true if the a count is earlier than b.
Example:
%CODE{"cpp"}%
unsigned long my_delay = current_tick + HZ;
while(TIME_BEFORE(current_tick, my_delay)) {
yield();
}
splash(HZ, true, "1 second has passed");
%ENDCODE%
Note: This is (almost) how the sleep() function is implemented.
Tick tasks
You can declare a function to be called on every kernel tick. It will be executed in interrupt context and has to be very short and simple.
int tick_add_task(void (*f)(void))
Adds a tick task to the kernel. Returns 0 on success or -1 on failure.
Example:
%CODE{"cpp"}%
void my_tick_task(void)
{
/* Do something fun */
}
void my_init(void)
{
tick_add_task(my_tick_task);
}
%ENDCODE%
int tick_remove_task(void (*f)(void))
Removes a tick task from the kernel. Returns 0 if it was found and removed or -1 if the task wasn't found in the kernel list.
Event queues
Threads can communicate using event queues. An event queue is a circular buffer of events. A thread can wait for messages to arrive in the queue while other threads are executing. When an event arrives, the thread will receive the event the next time it is scheduled.
An event looks like this:
%CODE{"cpp"}%
struct event
{
long id;
void *data;
};
%ENDCODE%
The id is what identifies this particular event. It can be any 32-bit positive number. Negative numbers are reserved for system events. The data can be anything you want to pass along with the event.
If you want to receive events in your thread, declare a global event queue and you're set.
%CODE{"cpp"}%
struct event_queue my_queue;
%ENDCODE%
Before using the queue, you have to initialize it.
%CODE{"cpp"}%
queue_init(&my_queue);
%ENDCODE%
Queue functions
Here are the functions for using event queues.
void queue_init(struct event_queue *queue)
Initializes a queue. This must be done before using it.
Example:
%CODE{"cpp"}%
struct event_queue my_queue;
queue_init(&my_queue)
%ENDCODE%
void queue_delete(struct event_queue *q)
Removes q from the queue list. You should use this function if the queue could be destroyed (like in plugins), otherwise calling of queue_broadcast() or sending an event to this queue could/will do a very bad thing (tm).
void queue_wait(struct event_queue *q, struct event *ev)
Waits for an event to arrive, removes the event from the queue and copies it to the event structure pointed to by ev.
Example:
%CODE{"cpp"}%
struct event ev;
queue_wait(&my_queue, &ev);
if(ev.id == 1) {
splash(HZ, true, "Got event 1");
}
%ENDCODE%
void queue_wait_w_tmo(struct event_queue *q, struct event *ev, int ticks)
Waits for an event to arrive, removes the event from the queue and copies it to the event structure pointed to by ev. If ticks system ticks have passed, it returns anyway with ev.id set to SYS_TIMEOUT.
Example:
%CODE{"cpp"}%
struct event ev;
queue_wait_w_tmo(&my_queue, &ev, HZ);
switch(ev.id) {
case 1:
splash(HZ, true, "Got event 1");
break;
case SYS_TIMEOUT:
splash(HZ, true, "Timeout");
break;
}
%ENDCODE%
bool queue_empty(const struct event_queue* q)
Returns true if a queue is empty
Example:
%CODE{"cpp"}%
while(queue_empty(&my_queue)) {
calculate();
yield();
}
queue_wait(&my_queue, &ev);
if(ev.id == 1) {
splash(HZ, true, "Got event 1");
}
%ENDCODE%
void queue_clear(struct event_queue* q)
Empties a queue.
void queue_post(struct event_queue *q, long id, void *data)
Sends an event to a queue. This can be called from another thread or from an interrupt handler.
Example:
%CODE{"cpp"}%
extern struct event_queue your_queue;
queue_post(&your_queue, 1, NULL);
%ENDCODE%
int queue_broadcast(long id, void *data)
Sends the event to all threads in the system, including the sending thread.
Protecting your shared data
To protect your shared data, there is a mechanism called a mutex. It is a simple locking mechanism to allow exclusive access to a resource. Mutexes can only be used in a thread context and not in an interrupt handler.
Mutex functions
void mutex_init(struct mutex *m)
Initializes a mutex before it can be used.
void mutex_lock(struct mutex *m)
Attempts to lock a mutex. If it is already locked, it blocks until the mutex is unlocked.
void mutex_unlock(struct mutex *m)
Unlocks a mutex.
Mutex example
%CODE{"cpp"}%
struct mutex my_mutex;
void my_protected_function(void)
{
mutex_lock(&my_mutex);
/* Do the stuff that requires exclusive access */
mutex_unlock(&my_mutex);
}
void init_my_protected_function(void)
{
mutex_init(&my_mutex);
}
%ENDCODE%
Guidelines for developers
When you develop Rockbox code, you have to be nice to the other threads in the system.
- Don't spend time in an interrupt handler (or tick task)
- Yield or sleep every so often. Many threads in the system rely on being executed regularly. If your code performs lengthy operations, like Mandelbrot calculations, you must call yield() once in a while, so the other threads can run.
- Sleeping is often better than yielding, because the kernel can execute a SLEEP instruction (architecture specific) if all threads are asleep, which will reduce the power consumption.
|