On Tue, Jul 20, 2004 at 02:29:33 -0400, RSudharsanan@xxxxxxxxxxxxxx wrote: > hi, > What is the simplest way to create kernel threads in 2.6.x kernel.I got > one for 2.4.x > from http://www.scs.ch/~frey/linux/kernelthreads.html8 Hell, that mixes so many things that don't have anything to do with each other. > but the task queues are not implemented here.I suppose i have to go for Task queues and kernel threads are one of them. They are completely distinct things. Unfortunately it's not really obvious from that example. > Work queue interface > Is it necessary first??? just for creating /stopping and killing threads. kernel_thread() behaves exactly like clone() in userspace (but the flags are like for fork). In fact, it calls the same code internaly. You pass it a function that should run in it, it sets up a new task_struct + stack for it and runs it. When it ends, the thread exists. The thread is a process, so it is scheduled like one (you must call schedule() because kernel is not preemptive (2.6 can be, but may be compiled without)). You must call kernel_thread in a process context, which generaly means module_init(), syscall handler or another kernel thread. Task queues are lists of functions that should be called from somewhere. Work queues are similar, generalized. You can use kernel threads without task/work queues and generaly it's done so. Many modules simply spawn a kernel thread in init and the thread then sits in a loop waiting on a wait_queue (which is a synchronization primitive -- nothing to do with task/work queues) for some task to do. You can use task/work queues without kernel threads. There are many places in kernel that run various task/work queues at different times. There is also a generic work-queue running thread, whose services you may use. ------------------------------------------------------------------------------- Jan 'Bulb' Hudec <bulb@xxxxxx>
Attachment:
signature.asc
Description: Digital signature
