Skip to content

Latest commit

 

History

History
 
 

concurrency-xv6-threads

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

Kernel Threads

In this project, you'll be adding real kernel threads to xv6. Sound like fun? Well, it should. Because you are on your way to becoming a real kernel hacker. And what could be more fun than that?

Specifically, you'll do three things. First, you'll define a new system call to create a kernel thread, called clone(), as well as one to wait for a thread called join(). Then, you'll use clone() to build a little thread library, with a thread_create() call and lock_acquire() and lock_release() functions. That's it! And now, for some details.

Overview

Your new clone system call should look like this: int clone(void(*fcn)(void *, void *), void *arg1, void *arg2, void *stack). This call creates a new kernel thread which shares the calling process's address space. File descriptors are copied as in fork(). The new process uses stack as its user stack, which is passed two arguments (arg1 and arg2) and uses a fake return PC (0xffffffff); a proper thread will simply call exit() when it is done (and not return). The stack should be one page in size and page-aligned. The new thread starts executing at the address specified by fcn. As with fork(), the PID of the new thread is returned to the parent (for simplicity, threads each have their own process ID).

The other new system call is int join(void **stack). This call waits for a child thread that shares the address space with the calling process to exit. It returns the PID of waited-for child or -1 if none. The location of the child's user stack is copied into the argument stack (which can then be freed).

You also need to think about the semantics of a couple of existing system calls. For example, int wait() should wait for a child process that does not share the address space with this process. It should also free the address space if this is last reference to it. Also, exit() should work as before but for both processes and threads; little change is required here.

Your thread library will be built on top of this, and just have a simple int thread_create(void (*start_routine)(void *, void *), void *arg1, void *arg2) routine. This routine should call malloc() to create a new user stack, use clone() to create the child thread and get it running. It returns the newly created PID to the parent and 0 to the child (if successful), -1 otherwise. An int thread_join() call should also be created, which calls the underlying join() system call, frees the user stack, and then returns. It returns the waited-for PID (when successful), -1 otherwise.

Your thread library should also have a simple ticket lock (read this book chapter for more information on this). There should be a type lock_t that one uses to declare a lock, and two routines void lock_acquire(lock_t *) and void lock_release(lock_t *), which acquire and release the lock. The spin lock should use x86 atomic add to build the lock -- see this wikipedia page for a way to create an atomic fetch-and-add routine using the x86 xaddl instruction. One last routine, void lock_init(lock_t *), is used to initialize the lock as need be (it should only be called by one thread).

The thread library should be available as part of every program that runs in xv6. Thus, you should add prototypes to user/user.h and the actual code to implement the library routines in user/ulib.c.

One thing you need to be careful with is when an address space is grown by a thread in a multi-threaded process (for example, when malloc() is called, it may call sbrk to grow the address space of the process). Trace this code path carefully and see where a new lock is needed and what else needs to be updated to grow an address space in a multi-threaded process correctly.

Building clone() from fork()

To implement clone(), you should study (and mostly copy) the fork() system call. The fork() system call will serve as a template for clone(), with some modifications. For example, in kernel/proc.c, we see the beginning of the fork() implementation:

int
fork(void)
{
  int i, pid;
  struct proc *np;

  // Allocate process.
  if((np = allocproc()) == 0)
    return -1;

  // Copy process state from p.
  if((np->pgdir = copyuvm(proc->pgdir, proc->sz)) == 0){
    kfree(np->kstack);
    np->kstack = 0;
    np->state = UNUSED;
    return -1;
  }
  np->sz = proc->sz;
  np->parent = proc;
  *np->tf = *proc->tf;

This code does some work you need to have done for clone(), for example, calling allocproc() to allocate a slot in the process table, creating a kernel stack for the new thread, etc.

However, as you can see, the next thing fork() does is copy the address space and point the page directory (np->pgdir) to a new page table for that address space. When creating a thread (as clone() does), you'll want the new child thread to be in the same address space as the parent; thus, there is no need to create a copy of the address space, and the new thread's np->pgdir should be the same as the parent's -- they now share the address space, and thus have the same page table.

Once that part is complete, there is a little more effort you'll have to apply inside clone() to make it work. Specifically, you'll have to set up the kernel stack so that when clone() returns in the child (i.e., in the newly created thread), it runs on the user stack passed into clone (stack), that the function fcn is the starting point of the child thread, and that the arguments arg1 and arg2 are available to that function. This will be a little work on your part to figure out; have fun!

x86 Calling Convention

One other thing you'll have to understand to make this all work is the x86 calling convention, and exactly how the stack works when calling a function. This is you can read about in Programming From The Ground Up, a free online book. Specifically, you should understand Chapter 4 (and maybe Chapter 3) and the details of call/return. All of this will be useful in getting clone() above to set things up properly on the user stack of the child thread.