Thread Switch

This lecture talks about xv6’s mechanism of thread switching.

Background knowledge

Thread : one serial execution, it should have its own pc, registers and stack. The scheduling system focus on the interleave multiple threads. There are two main methods to get this done, the first is using multi-cores and the second is to switch.

To deal with a computation-bound program which takes a long run to finish the job, the scheduling system uses timer interrupt to pre-empt the kernel then switch to another process.

  • In xv6, the scheduling does is first pre-empt the kernel then voluntarily yields cpu to other process.
    ![](Lec9– Thread Switch/Lec9-pic1.png)

  • Recall the workflow of interrupt. Trampoline.S stores all registers and pc in trapframe.

  • Sometimes, in kernel, kernel thread just does a simple syscall and returns back to user thread by restore pc.

  • But sometimes, kernel decides to swtich to other thread.

    • The kernel thread should save its own state, we call it kernel context, includes kstack and kernel registers.

  • Two time to make context switch – time interrupt and IO waiting.

What xv6 does

Actually, xv6’s the kernel context switch is conducted by scheduler thread, means kernel thread first switch to a middle thread then it chooses which process to run next.
The stored states includes kstack pointer and pc. Once the scheduler switch next process, it also has its previously stored states to be reloaded.

  • The scheduler is pin to each CPU, so it exactly chooses which runs on the CPU
  • Kernel context is stored in p->context.
  • Scheduler’s context is stored on the structure CPU for that core.
  • Important, one core could just run one thread at one time and one thread can only run on one core at one time.

scheduler

swtch function store the current thread’s context, includes ra, sp... etc. then load the target thread context.
one process kernel thread -> sched() -> swtch to scheduler() -> find a runable thread() -> swtch() -> other process kernel thread

Some thoughts

  • Each child process are set a fake ra in allocproc(), because it needs a ra & sp to be switch to and execute to set p->trapframe->epc = 0 in forkret()
    1
    2
    3
    4
    5
    // Set up new context to start executing at forkret,
    // which returns to user space.
    memset(&p->context, 0, sizeof(p->context));
    p->context.ra = (uint64)forkret;
    p->context.sp = p->kstack + PGSIZE;
  • In xv6, even in the scheduler, the kernel turns on the interrupt to provent kernel from deadlock.