In the Linux kernel, there is no such thing as a thread. Threads are the same as processes, it's just that they share some resources (most importantly, a virtual address space). Linux processes are fairly lightweight anyway, so this works out well.
fork() and pthread_create() are
implemented via clone() and
sys_clone() calls. Cloning a process involves
specifying which resources are to be shared via a set of flags
that are ORed together. For example, CLONE_VM
means the new process shares virtual memory with the creating
process, CLONE_FS means the new process shares file
system information with its parent, and CLONE_FILES
mean the child and parent share open files.
This means that there is no need for a user-level threads package at all, and no need for a separate scheduler. Threads are scheduled by the kernel just like regular processes, because they are regular processes.
The Linux kernel itself is multi-threaded. All kernel code
runs using physical addresses in Linux, so all kernel processes
share their address space (technically, they don't have an
address space, and the corresponding field in their
task_struct is set to NULL. Thus the
kernel_thread() call invokes a version of
clone() just like fork() and
pthread_create().
Modified: 25 March 2007