CS533 Concepts of Operating Systems Class 3 Integrated Task and Stack Management

CS533 - Concepts of Operating Systems 2 Questions  What do these terms mean? o process, thread, kernel-level thread, user-level thread, fiber? o user-level thread stack, user stack, kernel stack? o Address space?  In what way are user-level threads “faster” than kernel-level threads? o Aren’t we just layering abstractions on abstractions?

CS533 - Concepts of Operating Systems 3 Questions  How are synchronization operations such as spin- locks and semaphores implemented, i.e., how do you spin and how do you block? o For user level threads? o For kernel level threads?  Why can’t you do preemptive user-level thread scheduling without special kernel (or language runtime) support?

CS533 - Concepts of Operating Systems 4 Questions  How might a user-level thread library support synchronous I/O calls for its threads without losing control of the CPU due to blocking? o Does this constitute stack ripping?  Why can’t you handle page-faults this way?  Why are preemptions of kernel threads a problem for user level thread libraries?

CS533 - Concepts of Operating Systems 5 Questions  What stack management model do most OS kernels export? o How are I/O and page faults events handled?  What differences are there between virtual CPUs (threads) and real CPUs (hardware)? o How are I/O and page fault events handled by code on a real CPU vs on a virtual CPU?  What similarities do kernel-provided scheduler activations have to hardware-provided interrupts?

CS533 - Concepts of Operating Systems 6 Questions  What synchronization-related issues do you have to worry about when writing an interrupt handler? o Do these occur with scheduler activation handlers?  Why must the user-level thread library be reentrant? o How is its pattern of entry points and synchronization concerns similar to or different from kernel-level scheduler code? With scheduler activations and without scheduler activations

CS533 - Concepts of Operating Systems 7 Questions  Why does the kernel delay delivery of a second page-fault SA until the first page-fault SA on the same address has completed?  Why might a user level thread scheduler choose to delay context switching a user level thread that is holding a spin lock? o How can it do this in the presence of kernel preemption and notification via scheduler activations?

CS533 - Concepts of Operating Systems 8 Questions  In what ways do scheduler activations mix manual and automatic stack management models?

CS533 - Concepts of Operating Systems 9 Duality  What really happens underneath the abstraction? o For synchronous procedure calling and message passing o For asynchronous procedure calling and message passing  Fundamental tasks: execute new instruction stream exchange data  What is the cost of pushing state onto the stack and popping it off, vs the cost of allocating copying then freeing message memory?

CS533 - Concepts of Operating Systems 10 Optimizations Discussed in Future Classes  How could you optimize context switch costs for local message passing?  How could you optimize data movement costs for local message passing? o What conventions would you need to follow to avoid needing synchronization? o How does this approach compare to local state management in procedure calling?  How might you further optimize pointer passing? o … and what does this remind you of? Hint: SP.

CS533 - Concepts of Operating Systems 11 Duality  Is threading just an automated “pattern” of context switch and data exchange that is done manually in event models?  Do the two models have equal memory consumption? o Is over-allocation of memory inevitable in thread-based systems? If not, how can you avoid it … and at what cost?