Students: This lecture involved looking at the details of a lot of code. Please see the slides for those implementations: yield(), wait(), yield_wait().
- Introduction
- Today: get rid of assumption that we only have one program per CPU.
- Sharing CPU is a problem because one program can block another.
- Threads
- thread = virtual processor
- API: suspend(), resume()
- Need to capture program’s state: value of all registers, all of its memory.
- Big question: when to suspend/resume a thread?
- yield()
- Command to tell kernel that thread is waiting for an event.
- Implementation does three things: Suspends running thread, chooses new thread to run, resumes new thread.
- Data structures: Threads table, CPUs table, t_lock.
- Suspending current thread: Save stack pointer and page-table register.
- Choosing a new thread: Round-robin fashion until we hit a RUNNABLE thread (perhaps the one that just called yield).
- Resuming new thread: Reload state.
- All of this happens as an atomic action.
- Condition Variables
- Allow kernel to notify threads instead of having threads constantly make checks.
- “Lost notify” problem:
- T1 has lock on buffer, finds it full, releases lock.
- Prior to T1 calling wait, T2 acquires lock, reads message, notifies waiting threads that the buffer is not full.
- ..but T1 is not yet waiting; it was interrupted before it could call wait.
- Solution: API is wait(cv, lock), not wait(cv)..
- when a thread calls wait, it goes to sleep and releases the lock.
- Wait implementation:
- requires a different version of yield() — yield_wait() — to prevent deadlock.
- yield_wait() releases and re-acquires t_lock in the middle, and must point to a special stack to prevent stack corruption.
- Preemption
- If a thread never calls yield or wait, it’s okay; special hardware will periodically generate an interrupt and forcibly call yield.
- ..But what if this interrupt occurs while the CPU is running yield()? Deadlock.
- Solution: Hardware mechanism to disable interrupts.
- Reflection/Summary
- We’ve enforced modularity on a single machine, assuming that the OS itself is indeed correct.
- Locks and threads are interesting: we needed them to get bounded buffers to work, but they bring up modularity issues. We had to reason globally about locks.
- To truly enforce modularity, we needed kernel and/or hardware support.