Threads

Creating a new thread

In SVR4.2MP and UnixWare 2.0 processes are single-threaded when they start running new programs (that is, on entry to main, following a successful call to exec(2)). The thread created to execute main is known as the ``initial thread''. If no additional threads are created, a process can continue to execute with the same semantics as the traditional UNIX process.

New threads can be created via the thr_create(3thread) routine

   int thr_create(
   	void     *stack_address,
   	size_t   stack_size,
   	void     *(*start_routine)(void *arg),
   	void     *arg,
   	long     flags,
   	thread_t *new_thread
   );

which takes the following parameters:

stack_address and stack_size

These define the stack space for the new thread. (This space is used for function call transactions and for automatic variables in functions called by the thread.)

The stack of the traditional UNIX process has ``autogrow'' support by the operating system. That is, if the stack grows beyond its initial size the operating system automatically increases its size as needed (or until it runs into some other defined segment). However, threads (other than the initial thread) use stacks that do not have autogrow support; consequently, the stacks should be allocated to meet the maximum needs of the thread.

As a convenience, the Threads Library will implicitly allocate a reasonably-sized stack (twice the page size or 16K bytes, whichever is greater) if stack_address and stack_size are set to NULL and 0, respectively,

The programmer can specify other sizes if needed. The value must not be less than that returned by thr_minstack(3thread). Note that stack_address should point to the base address (lowest) of the allocated space.
In this implementation, the Threads Library manages the process address space so that stack overflows will result in an addressing error (SIGSEGV). For most applications, this is a desirable behavior. It is better to discover such errors as soon as they occur rather than have one stack corrupt another.

start_routine and arg

These parameters define the starting condition of the newly created thread. start_routine is the function address where the new thread's execution will begin and arg is the argument that start_routine will receive.

start_routine takes a single parameter of type (void *) and returns a value of the same type. These values can be used (with type casts) to pass values (or aggregations of values in structures) of any type.

NOTE: For portability, do not cast an int to (void *), and then cast it back to int. These values should only be used as pointers; otherwise, information can be lost.

Of course, a thread need not be entirely defined by a single function. That initial function will typically call other functions.

flags

These flags will be discussed as their respective topics arise later in this section. These flags are not mutually exclusive; they can be combined with a bitwise inclusive OR. For each flag, the relevant section is shown:

THR_SUSPENDED: ``Managing thread scheduling''
THR_BOUND: ``Managing threads concurrency''
THR_DETACHED: ``Waiting for thread termination''
THR_INCR_CONC: ``Managing threads concurrency''
THR_DAEMON: ``Terminating a thread''

new_thread

The ``thread ID'' of the newly created thread is delivered at this address.

This value can be used in other functions to influence that thread.
The scope of the value is limited to the enclosing process; it is not relevant to threads in other processes.
A thread can learn its own thread ID number by the thr_self(3thread) function.

NOTE: thr_create(3thread) and most other functions in the Threads Library return 0 on success. On failure, instead of setting the errno global variable, they return the error code as the function's value.

The creation of one thread by another is conceptually similar but not identical to the creation of a new process by another process via the fork(2) system call. Some differences are:

After a fork(2) system call both the creator (parent) and created (child) processes resume from the same point of computation -- the return from fork(2).
In contrast, a new thread starts execution at the start_function specified by the creator (in some respects similar to the exec(2) system call), while the creating thread returns from thr_create(3thread).
The operating system maintains a parent/child relationship between creating and created processes that affects later interactions at process termination (for example, wait(2) semantics). In contrast, there is no innate hierarchy among threads. Each is a ``sibling'' of the other. Thus, the creator might wait for the newly created thread to terminate or, just as easily, the new thread can wait for its creator to terminate. {See discussion of thr_join(3thread) below.)