cksystemsteaching / AOS-Winter-2015

Advanced Operating Systems 2015

Home Work Flow

• Step 0: form a team of 2-3 members
• Step 1: get a github account (for each member)
• Step 2: one person per group forks the AOS-Winter-2015 repository by clicking here and adds the other team members as collaborators
• Step 3: check out the branch named selfie-master in your fork of AOS-Winter-2015
• Step 4: implement the first assignment (see below)
• Step 5: add your names to the AUTHORS file
• Step 6: send a pull request containing your solution via github.com to cksystemsteaching/AOS-Winter-2015/tree/selfie-master

Assignment 0: Basic data structures

Review linked lists and implement a simple program using a singly linked list in C*. The minimal requirements are as follows:

• must be implemented in C*
• must compile with selfie
• must run on selfie
• the list must be dynamically allocated
• every node must be dynamically allocated
• inserting nodes to the list and removing nodes from the list
• list iteration
• Bonus: sort the list. Any way you like
• Deadline: Oct 15, end of day

Assignment 1: Loading, scheduling, switching, execution

Implement basic concurrent execution of n processes in mipster. n >= 2

• understand how mipster interprets and executes binary instructions. Tipp: add your own comments to the code
• mipster maintains a local state for a process (running executable), e.g., pc, registers, memory
• understand the purpose of each variable and data structure
• duplicate the process state n times
• running mipster like: ./selfie -m 32 yourbinary should generate n instances of yourbinary in a single instance of mipster
• implement preemptive multitasking, i.e., switching between the n instances of yourbinary is determined by mipster
• switch processes every m instructions. 1 <= m <= number of instructions in yourbinary
• implement round-robin scheduling
• add some output in yourbinary to demonstrate context switching
• Deadline: Oct 22, end of day

Assignment 2: Memory segmentation, yield system call

This assignment deals with cooperative multitasking of n processes in mipster using a single instance of physical memory.

• again, duplicate the process state n times

• but, do not duplicate the whole main memory

• instead, split the main memory into segments by implementing a segment table in mipster

• each process has an entry in the segment table for the segment start address and segment size

• design the segment table for constant time access

• translate the addresses of read and write operations to memory

• implement cooperative multitasking through a yield system call, i.e., a user process calling sched_yield() will cause the OS to re-schedule

• implement a simple user program that demonstrates yielding, e.g, yield each time after printing a counter to the console

• Deadline: Oct 29, end of day

Assignment 3: bootstrapping the kernel

At the end of this assignment you will have the operating system running on top if mipster along with other processes.

• implement the operating system in selfie.c and use the provided flag (-k) to execute the kernel code.

• whenever a trap (e.g. a syscall instruction) or an interrupt (e.g. scheduling timer) happens, the operating is invoked instead of handling the trap or interrupt by the emulator. However, the OS cannot modify the machine state directly, i.e., modifying the memory pointer and registers array is not possible. Therefore:

• provide a special system call, e.g., switch(int previous_process, int next_process) in the emulator that is invoked by the operating system only and modifies the machine state. One issue remains: after the OS invokes switch, the OS process must be reset to interrupt-trap handling mode. You can rely on the following convention: mipster starts executing a binay at address 0x0, the main method of selfie.c. Resetting the PC of the OS process to 0x0 after switch will reset the OS but not its heap and globals. The OS stack must be reset as well. Important: selfie -k must start with interrupt/trap handling. If no interrupt or trap is to be handled, the OS switches to the first ready process. If not ready process exists, the OS loads some_program.mips or terminates.

• Deadline: Nov 5, end of day

Assignment 4: Mutual Exclusion

• implement a single global lock through mipster syscalls, e.g., a lock() and unlock() call.

• implement a simple user program that demonstrates mutual exclusion, e.g, show that one process inside the critical section makes progress, processes not taking the lock make progress, and processes waiting for the lock do not make progress. Hint: you can implement the getpid system call to identify processes.

• experiment with and demonstrate different interleavings: using locks, no locks, different time slices

• Deadline: Nov 12, end of day

• Bonus: implement basic multi-threading support

• Idea: threads share one address space, processes don't

• when duplicating processes, create threads instead, i.e., shared code, heap, globals, but private call stacks, private PC, private registers

Assignment 5: Virtual Memory

• implement on-demand paging instead of memory segmentation in the kernel.

• partition the emulated memory (like physical memory on a real machine) into 4KB frames.

• each process gets a 4MB virtual address space, i.e., each process may access any address between 0x0 to 0x3FFFFFF. Virtual address space is organised in 4KB pages

• whenever a process actually accesses an address, allocate a frame and provide a mapping between pages and frames, i.e., a page table for each process

• modify the tlb function accordingly

• provide a demo program that allocates the whole virtual memory space but only accesses, for example, 128 different addresses evenly distributed within that region. Make sure that in this case only 128 frames are actually allocated, not the whole 4MB.

• page replacement and swapping is not required in this assignment

• Deadline: November 26, end of day

Assignment 6 + 7: Towards Microkernels

1. In the emulator implement a microkernel that supports context and page table programming. You will need at least create_context, switch_context, delete_context, map_page_in_context, and flush_page_in_context. We distinguish these five calls into the microkernel from all other syscalls and call them hypercalls. create_context returns a unique ID to identify the new context, switch_context switches contexts, delete_context deletes contexts and recycles IDs, map_page_in_context creates a mapping for a virtual page, and flush_page_in_context unmaps it again. The key simplification is that the microkernel maintains contexts and page tables in the emulator. All processes are now in virtual memory including the OS.

Deadline: December 17, end of day

Until January 14 implement also:

2. On top of the emulator your OS in a single process that manages user processes as before but uses the microkernel in the emulator for switching. This process should have all of its virtual memory mapped 1-to-1 to physical memory. For paging the OS process may maintain its own copies of user page tables (in addition to the page tables in the microkernel) and also the free-list of page frames (most of the heap part of the virtual pages of the OS process become the page frames of user processes). This way the OS can read and write all user process memory, e.g. for loading code.
3. Similarly, all other events (timer interrupts, page faults) should also lead to "traps" into the microkernel which then make the microkernel also switch contexts to the OS process to handle them. Make sure that the OS process always runs to completion and is not interrupted by anything. This means the OS process sees something to do whenever it returns from context switching to a user process.

Assignment 8: Concurrent Data Structures

Implement the Micheal-Scott queue as discussed in class. For that you need to implement a Compare and Swap instruction (CAS) and shared memory through a thread_fork system call.

Demonstrate your data structure with a meaningful test program, e.g., the Producer–consumer problem

Deadline: January 28, 2016, end of day