Threads And Teams¶
Declared in: |
kernel/OS.h |
Library: |
A thread is a synchronous process that executes a series of program instructions. When you launch an application, an initial thread—the main thread—is automatically created (or spawned) and told to run. From the main thread you can spawn and run additional threads; from each of these threads you can spawn and run more threads, and so on. The collection of threads that are spawned from the main thread—in other words, the threads that comprise an application—is called a team. All the threads in all teams run concurrently and asynchronously with each other.
For more information on threads and teams, see “Threads And Teams Overview”.
Thread and Team Functions¶
estimate_max_scheduling_latency()
Declared in: kernel/scheduler.h
bigtime_t estimate_max_scheduling_latency(thread_id thread = -1)
Returns the scheduling latency, in microseconds, of the specified thread. Specify a thread_id of -1 to return the scheduling latency of the current thread.
exit_thread(), kill_thread(), kill_team()
void exit_thread(status_t return_value)
status_t kill_thread(thread_id thread)
status_t kill_team(team_id team)
status_t on_exit_thread(void (*callback)(void *), void *data)
These functions command one or more threads to halt execution:
Function |
Description |
|---|---|
exit_thread() |
Tells the calling thread to exit with a return value as given by the
argument. Declaring the return value is only useful if some other thread is
sitting in a |
kill_thread() |
Kills the thread given by the argument. The value that the thread will
return to |
kill_team() |
Kills all the threads within the given team. Again, the threads’ return values are random. kill_team() is the same as sending a SIGKILL signal to any thread in the team. Each of the threads in the team is then handed a SIGKILLTHR signal. |
on_exit_thread() |
Sets up the specified callback to be executed when the calling thread exits. The callback will receive the pointer data as an input argument. |
Exiting a thread is a fairly safe thing to do—since a thread can only exit itself, it’s assumed that the thread knows what it’s doing. Killing some other thread or an entire team is a bit more drastic since the death certificate(s) will be delivered at an indeterminate time. In addition, killing a thread can leak memory since resources that were allocated by the thread may not be freed. Killing an entire team, on the other hand, won’t leak since the system reclaims all resources when the team dies.
Keep in mind that threads die automatically (and their resources are reclaimed) if they’re allowed to exit naturally. You should only need to kill a thread if something has gone screwy.
Return Code |
Description |
|---|---|
|
The thread or team was successfully killed. |
|
Invalid thread value. |
|
Invalid team value. |
|
Returned by on_exit_thread() if there’s no memory to construct the internal callback record. |
find_thread()
thread_id find_thread(const char *name)
Finds and returns the thread with the given name. A name argument of NULL returns the calling thread.
A thread’s name is assigned when the thread is spawned. The name can be
changed thereafter through the rename_thread() function. Keep
in mind that thread names needn’t be unique: If two (or more) threads boast
the same name, a find_thread() call on that name returns the first so-named
thread that it finds. There’s no way to iterate through identically-named
threads.
Return Code |
Description |
|---|---|
|
name doesn’t identify a valid thread. |
get_team_info(), get_next_team_info()
status_t get_team_info(team_id team, team_info *info)
status_t get_next_team_info(int32 *cookie, team_info *info)
The functions copy, into the info argument, the
team_info structure for a particular team. The
get_team_info() function retrieves information for the team identified by
team. For information about the kernel, use
B_SYSTEM_TEAM as the team argument.
The get_next_team_info() version lets you step through the list of all
teams. The cookie argument is a placemark; you set it to 0 on
your first call, and let the function do the rest. The function returns
B_BAD_VALUE when there are no more areas to visit:
/* Get the team_info for every team. */
team_info info;
int32 cookie = 0;
while (get_next_team_info(0, &cookie, &info) == B_OK)
...
See team_info for a description of that structure.
Return Code |
Description |
|---|---|
|
The desired team information was found. |
|
team doesn’t identify an existing team, or there are no more areas to visit. |
get_thread_info(), get_next_thread_info()
status_t get_thread_info(thread_id thread, thread_info *info)
status_t get_next_thread_info(team_id team, int32 *cookie, thread_info *info)
These functions copy, into the info argument, the
thread_info structure for a particular thread:
The get_thread_info() function gets the information for the thread identified by thread.
The get_next_thread_info() function lets you step through the list of a
team’s threads through iterated calls. The team argument
identifies the team you want to look at; a team value of 0 means
the team of the calling thread. The cookie argument is a
placemark; you set it to 0 on your first call, and let the function do the
rest. The function returns B_BAD_VALUE when there are no
more threads to visit:
/* Get the thread_info for every thread in this team. */
thread_info info;
int32 cookie = 0;
while (get_next_thread_info(0, &cookie, &info) == B_OK)
...
The value of the priority field describes the thread’s “urgency”;
the higher the value, the more urgent the thread. The more urgent the
thread, the more attention it gets from the CPU. Expected priority values
fall between 0 and 120. See “Thread Priorities” for the full story.
Return Code |
Description |
|---|---|
|
The thread was found; info contains valid information. |
|
thread doesn’t identify an existing thread, team doesn’t identify an existing team, or there are no more threads to visit. |
rename_thread()
status_t rename_thread(thread_id thread, const char *name)
Changes the name of the given thread to name. The name can be no
longer than B_OS_NAME_LENGTH (32 characters).
Return Code |
Description |
|---|---|
|
The thread was successfully named. |
|
thread argument isn’t a valid thread_id number. |
resume_thread()
status_t resume_thread(thread_id thread)
Tells a new or suspended thread to begin executing instructions. If the
thread has just been spawned, it enters and executes the thread function
declared in spawn_thread(). If the thread was previously
suspended(through suspend_thread()), it continues from where it
was suspended.
You can’t use this function to wake up a sleeping thread, or to unblock a
thread that’s waiting to acquire a semaphore or waiting in a
receive_data() call. However, you can unblock any of these
threads by suspending and then resuming. Blocked threads that are resumed
return B_INTERRUPTED.
resume_thread() is the same as sending a SIGCONT signal to the thread.
Return Code |
Description |
|---|---|
|
The thread was successfully resumed. |
|
thread argument isn’t a valid thread_id number. |
|
The thread isn’t suspended. |
receive_data()
int32 receive_data(thread_id sender, void *buffer, size_t buffer_size)
Retrieves a message from the thread’s message cache. The message will have
been placed there through a previous send_data() function call.
If the cache is empty, receive_data() blocks until one shows up—it never
returns empty-handed.
The thread_id of the thread that called send_data() is
returned by reference in the sender argument. Note that there’s no
guarantee that the sender will still be alive by the time you get its ID.
Also, the value of sender going into the function is ignored—you can’t ask
for a message from a particular sender.
The send_data() function copies two pieces of data into a
thread’s message cache:
A single four-byte code that’s delivered as receive_data()’s return value,
and an arbitrarily long data buffer that’s copied into receive_data()’s buffer argument (you must allocate and free buffer yourself). The buffer_size argument tells the function how many bytes of data to copy. If you don’t need the data buffer—if the code value returned directly by the function is sufficient—you set buffer to NULL and buffer_size to 0.
Unfortunately, there’s no way to tell how much data is in the cache before you call receive_data():
If there’s more data than buffer can accommodate, the unaccommodated portion is discarded—a second receive_data() call will not read the rest of the message.
Conversely, if receive_data() asks for more data than was sent, the function returns with the excess portion of buffer unmodified—receive_data() doesn’t wait for another
send_data()call to provide more data with which to fill up the buffer.
Each receive_data() corresponds to exactly one send_data().
Lacking a previous invocation of its mate, receive_data() will block until
send_data() is called. If you don’t want to block, you should
call has_data() before calling receive_data() (and proceed to
receive_data() only if has_data() returns true).
Return Code |
Description |
|---|---|
If successful |
Returns the message’s four-byte code. |
|
A blocked receive_data() call was interrupted by a signal. |
send_data()
status_t send_data(thread_id sender, void *buffer, size_t buffer_size)
send_data() copies a message into thread’s message cache. The target thread
retrieves the message (and empties the cache) by calling
receive_data().
There are two parts to the message:
A single four-byte code passed as an argument to send_data() and returned directly by
receive_data().A buffer of data that’s buffer_size bytes long (buffer can be NULL, in which case buffer_size should be 0). The data is copied into the target thread’s cache, and then copied into
receive_data()’s buffer (which must be allocated). The calling threads retain responsibility for freeing their buffers.
In addition to returning the code directly, and copying the
message data into its buffer argument, receive_data()
sets sender to the id of the thread that sent the message.
send_data() blocks if there’s an unread message in the target thread’s
cache; otherwise it returns immediately (i.e. it doesn’t wait for the
target to call receive_data(). Analogously,
receive_data() blocks until there’s a message to retrieve.
In the following example, the main thread spawns a thread, sends it a message, and then tells the thread to run:
main()
{
thread_id other_thread;
int32 code = 63;
char *buf = "Hello";
other_thread = spawn_thread(thread_func, ...);
send_data(other_thread, code, (void *)buf, strlen(buf));
resume_thread(other_thread);
...
}
To retrieve the message, the target thread calls
receive_data():
int32 thread_func(void *data)
{
thread_id sender;
int32 code;
char buf[512];
code = receive_data(&sender, (void *)buf, sizeof(buf));
...
}
Keep in mind that the message data is copied into the buffer; you must allocate adequate storage for the data. If the buffer isn’t big enough to accommodate all the datain the message, the left-over portion is thrown away. Note, however, that there isn’t any way for a thread to determine how much data has been copied into its message cache.
Return Code |
Description |
|---|---|
|
The data was successfuly sent. |
|
thread doesn’t identify a valid thread. |
|
The target couldn’t allocate enough memory for its copy of buffer. |
|
The function blocked, but a signal unblocked it. |
has_data()
bool has_data(thread_id thread)
has_data() returns true if thread has a message in
its message cache. Ostensibly, you use this function before calling
send_data() or receive_data() to avoid blocking:
if (!has_data(target_thread))
err = send_data(target_thread, ...);
/* or */
if (has_data(find_thread(NULL))
code = receive_data(...);
This works for receive_data(), but notice that there’s a race
condition between the has_data() and send_data() calls. Another
thread could send a message to the target in the interim.
set_thread_priority(), suggest_thread_priority()
status_t set_thread_priority(thread_id thread, int32 new_priority)
int32 suggest_thread_priority(uint32 what = B_DEFAULT_MEDIA_PRIORITY, int32 period = 0, bigtime_t jitter = 0, bigtime_t length = 0)
Declared in: kernel/scheduler.h
set_thread_priority() resets the given thread’s priority to new_priority. The priority is expected to be between 0 and 120. See “Thread Priorities” for a description of the priority scheme, and “Thread Priority Values” for a list of pre-defined priority constants.
suggest_thread_priority() takes information about a thread and returns a
suggested priority that you can pass to set_thread_priority() (or, more
likely, to spawn_thread()).
The what value is a bit mask that indicates the type of activities the thread will be used for. The possible values are listed in Suggested Thread Priorities.
period is the number of times per second the thread needs to be run (specify 0 if it needs to run continuously). jitter is an estimate, in microseconds, of how much the period can vary as long as the average stays at period times per second.
length is an approximation of the amount of time, in microseconds, the thread will typically run per invocation (i.e., the amount of time that will pass between the moment it receives a message, through processing it, until it’s again waiting for another message).
For example, if you’re spawning a thread to handle video refresh for a computer game, and you want the display to update 30 times per second, you might use code similar to the following:
int32 priority;
priority = suggest_thread_priority(B_LIVE_3D_RENDERING, 30, 1000, 150);
th = spawn_thread(func, "render_thread", priority, NULL)
This spawns the rendering thread with a priority appropriate for a thread for live 3D rendering which wants to be run 30 times per second, with a variation of only 1000 microseconds. Each invocation of the thread’s code is estimated to take 150 microseconds. Obviously the jitter and length values would have to be tuned to the particular application.
set_thread_priority() returns…
Return Code |
Description |
|---|---|
Positive integers. |
If the function is successful, the previous priority is returned. |
|
thread doesn’t identify a valid thread. |
snooze(), snooze_until()
status_t snooze(bigtime_t microseconds, int32 new_priority)
status_t snooze_until(bigtime_t microseconds, int timebase)
snooze() blocks the calling thread for the given number of microseconds.
snooze_until() blocks until an absolute time measured in the given
timebase. Currently, the only allowed value for
timebase is B_SYSTEM_TIMEBASE, which measures
time against the system clock (as reported by system_time()).
Return Code |
Description |
|---|---|
|
The thread went to sleep and is now awake. |
|
The thread received a signal while it was sleeping. |
spawn_thread()
thread_id spawn_thread(thread_func func, const char *name, int32 priority, void *data)
Creates a new thread and returns its thread_id identifier (a positive integer). The arguments are:
Parameter |
Description |
|---|---|
func |
Is a pointer to a thread function. This is the function that the thread will execute when it’s told to run. See “The Thread Function” for details. |
name |
Is the name that you wish to give the thread. It can be, at most,
|
priority |
Is the CPU priority level of the thread. This value should be between 0 and 120; he higher the priority, the more attention the thread gets. See Thread Priorities for a description of the priorities, and Thread Priority Values for a list of priority constants. |
data |
Is forwarded as the argument to the thread function. |
A newly spawned thread is in a suspended state
(B_THREAD_SUSPENDED). To tell the thread to run, you pass
its thread_id to the resume_thread() function. The
thread will continue to run until the thread function exits, or until the
thread is explicitly killed (through a signal or a call to
exit_thread(), kill_thread(), or
kill_team()).
Return Code |
Description |
|---|---|
|
All thread_id numbers are currently in use. |
|
Not enough memory to allocate the resources for another thread. |
suspend_thread()
status_t suspend_thread(thread_id thread)
Halts the execution of the given thread, but doesn’t kill the thread
entirely. The thread remains suspended (suspend_thread() blocks) until it’s
told to run through the resume_thread() function. Nothing
prevents you from suspending your own thread, i.e.:
suspend_thread(find_thread(NULL));
Of course, this is only smart if you have some other thread that will resume you later.
You can suspend any thread, regardless of its current state. But be
careful: If the thread is blocked on a semaphore (for example), the
subsequent resume_thread() call will “hop over” the semaphore
acquisition.
Suspensions don’t nest. A single resume_thread() unsuspends a
thread regardless of the number of suspend_thread() calls it has received.
suspend_thread() is the same as sending a SIGSTOP signal to the thread.
Return Code |
Description |
|---|---|
|
The thread is now suspended. |
|
thread isn’t a valid thread_id number. |
wait_for_thread()
status_t wait_for_thread(thread_id thread, status_t *exit_value)
This function causes the calling thread to wait until thread (the “target thread”) has died. If thread is suspended (or freshly spawned), wait_for_thread() will resume it.
When the target thread is dead, the value that was returned by its thread
function (or imposed by exit_thread()) is returned in
exit_value. If the target thread was killed (by
kill_thread() or kill_team()), or if the thread
function doesn’t return a value, the value returned in exit_value
will be unreliable.
You must pass a valid pointer as the second argument to wait_for_thread(). You mustn’t pass NULL even if you’re not interested in the return value.
Return Code |
Description |
|---|---|
|
The target is now dead. |
|
thread isn’t a valid thread_id number. |
|
The target was killed by a signal. This includes |
Thread and Team Structures and Types¶
team_id, thread_id¶
typedef int32 team_id ;
typedef int32 thread_id ;
These id numbers uniquely identify teams and threads, respecitvely.
team_info¶
typedef struct {
team_id team;
int32 thread_count;
int32 image_count;
int32 area_count;
thread_id debugger_nub_thread;
port_id debugger_nub_port;
int32 argc;
char args[64];
uid_t uid;
gid_t gid;
} team_info;
The team_info structure returns information about a team. To
retrieve one of these structures, use get_team_info() or
get_next_team_info().
The first field is obvious; the next three reasonably so: They give the number of threads that have been spawned, images that have been loaded, and areas that have been created or cloned within this team.
The debugger fields are used by the, uhm, the…debugger?
The argc field is the number of command line arguments that were used to launch the team; args is a copy of the first 64 characters from the command line invocation. If this team is an application that was launched through the user interface (by double-clicking, or by accepting a dropped icon), then argc is 1 and args is the name of the application’s executable file.
uid and gid identify the user and group that “owns” the team. You can use these values to play permission games.
thread_func¶
typedef int32 (*thread_func)(void *data);
thread_func is the prototype for a thread’s thread function. You
specify a thread function by passing a thread_func as the first
argument to spawn_thread(); the last argument to
spawn_thread() is forwarded as the thread function’s data
argument. When the thread function exits, the spawned thread is
automatically killed. To retrieve a thread_func’s return value,
some other thread must be waiting in a wait_for_thread() call.
Note that spawn_thread() doesn’t copy the data that data points
to. It simply passes the pointer through literally. Never pass a pointer
that’s allocated locally (on the stack).
thread_info¶
typedef struct {
thread_id thread;
team_id team;
char name[B_OS_NAME_LENGTH];
thread_state state;
sem_id sem;
int32 priority;
bigtime_t user_time;
bigtime_t kernel_time;
void * stack_base;
void * stack_end;
} thread_info
The thread_info structure contains information about a thread. To
retrieve one of these structure, use get_thread_info() or
get_next_thread_info().
The thread, team, and name fields contain the indicated information.
state describes what the thread is currently doing (see
thread_state for the list of states). If the
thread is waiting to acquire a semaphore, sem is that semaphore.
priority is a value that indicates the level of attention the thread gets (see Thread Priority).
user_time and kernel_time are the amounts of time, in microseconds, the thread has spent executing user code and the amount of time the kernel has run on the thread’s behalf, respectively.
stack_base and stack_end are pointers to the first byte and last bytes in the thread’s execution stack. Currently, the stack size is fixed at around 256k.
Warning
The two stack pointers are currently inverted such that stack_base is less than stack_end. (In a stack-grows-down world, the base should be greater than the end.)
Thread and Team Constants¶
B_SYSTEM_TEAM
#define B_SYSTEM_TEAM ...
Use this constant as the first argument to get_team_info() to
get team information about the kernel).
B_SYSTEM_TIMEBASE
#define B_SYSTEM_TIMEBASE ...
The system timebase constant is used as a basis for time measurement in the
snooze_until() function. (Currently, it’s the only timebase
available.)
be_task_flags¶
enum be_task_flags {
B_DEFAULT_MEDIA_PRIORITY,
B_OFFLINE_PROCESSING,
B_STATUS_RENDERING,
B_USER_INPUT_HANDLING,
B_LIVE_VIDEO_MANIPULATION,
B_VIDEO_PLAYBACK,
B_VIDEO_RECORDING,
B_LIVE_AUDIO_MANIPULATION,
B_AUDIO_PLAYBACK,
B_AUDIO_RECORDING,
B_LIVE_3D_RENDERING,
B_NUMBER_CRUNCHING
};
Declared in: kernel/scheduler.h
Constant |
Description |
|---|---|
|
The thread isn’t doing anything specialized. |
|
The thread is doing non-real-time computations. |
|
The thread is rendering a status or preview display. |
|
The thread is handling user input. |
|
The thread is processing live video (filtering, compression, decompression, etc.). |
|
The thread is playing back video from a hardware device. |
|
The thread is recording video from a hardware device. |
|
The thread is doing real-time manipulation of live audio data (filtering, compression, decompression, etc.). |
|
The thread is playing back audio from a hardware device. |
|
The thread is recording audio from a hardware device. |
|
The thread is performing live 3D rendering. |
|
The thread is doing data processing. |
These constants describe what the thread is designed to do. You use these
constants when asking for a suggested priority (see
suggest_thread_priority()).
Warning
These constants may not be used as actual thread priority values—do not
pass one of these values as the priority argument to
spawn_thread().
Thread Priority Values¶
Time-Sharing Priority |
Value |
|---|---|
|
5 |
|
10 |
|
15 |
|
20 |
Real-Time Priority |
Value |
|---|---|
|
100 |
|
110 |
|
120 |
The thread priority values are used to set the “urgency” of a thread.
Although you can reset a thread’s priority through
set_thread_priority(), the priority is initially—and almost
always permanently—set in spawn_thread().
thread_state¶
enum { ... } thread_state
Constant |
Description |
|---|---|
|
The thread is currently receiving attention from a CPU. |
|
The thread is waiting for its turn to receive attention. |
|
The thread has been suspended or is freshly-spawned and is waiting to start. |
|
The thread is waiting to acquire a semaphore. The sem field of
the thread’s |
|
The thread is sitting in a |
|
The thread is sitting in a |
A thread’s state tells you what the thread is currently doing. To get the
state, look in the state field of the thread_info structure (retrieved through
get_thread_info()).