Assignment 7: Design and Implement a CPU Profiler (Part 2)

Part 2: Record the stack trace and calculate the task execution time

[40 points + 20 bonus points] In Part 2, you will modify your kernel module, perftop, to: (a) track the time that both user and kernel tasks spend on a CPU and (b) print the 20 most scheduled tasks on the system using /proc.

Part 2.1. Storing kernel stack trace

[20 points] Modify your kernel module from Part 1 to track tasks instead of PIDs. A task is defined as a unique user or kernel stack trace. The hash table key will now be a hash of a task stack trace, and the value will be the number of times the unique kernel stack trace is scheduled on the CPU. Particularly, referring to background reading stack trace and Jenkins Hash in Part 1 of the project (Assignment 6).

Tasks:

• In the Kprobes event handler, get a given task's stack trace

  • Use stack_trace_save function for a kernel task
  • Use stack_trace_save_user function for a user task
  • If the function you want to use is not exported, you can obtain a function pointer with kallsyms_lookup_name()

• Modify the kernel modules hash table to store the stack trace instead of PID as the key (you can pass the stack trace buffer to Jenkins hash)

• Increment the schedule count of each task (stack trace) in the hash table

• Modify the open function of the proc file to print with cat /proc/perftop the stack trace dump and its corresponding number of times it has been scheduled

Deliverables:

• Load perftop module
• Invoke cat /proc/perftop
• Add a screenshot of the output
• Upload the source code tarball

Part 2.2. Store schedule time

[20 points] Calculate the time spent by each kernel task scheduled on a CPU and store it.

Task:

• Modify the Kprobes event handler to measure the time spent by each task on CPU (the time the task is actively running TASK_RUNNING)

  • Modify the hash table value to store task and corresponding time spent on CPU
  • Measure the time using rdtsc counter
  • When the event handler is invoked: 1) A task is scheduled out and a new task is scheduled in; 2) Calculate the time spent on the CPU for task being scheduled out and update the hash table; 3) Start the timer for the new task being scheduled in

• Modify the open function in the /proc file to print the stack trace and the cumulative time spent by the task on the cpu. The time can be in rdtsc ticks; please denote this unit of measurement in your print line.

Deliverables:

• Load perftop module
• Invoke cat /proc/perftop
• Add a screenshot of the output
• Upload the source code tarball

Part 2.3: Use rb-tree to get 20 most scheduled tasks [Optional]

[20 points] Modify the kernel module to print the 20 most scheduled tasks.

Note: This is an optional task. Credits are added to your total homework points, provided they do not cause your total to exceed the maximum points for the homework section.

Tasks:

• Maintain a rb-tree with the key as the cumulative time spent by a task scheduled on the CPU and value as the stack trace;
• Whenever a task is scheduled out, remove the old entry of the task from rb-tree and add the new cumulative time spend and stack trace to the rb-tree;
• Modify the open function of proc file to print the top 20 most scheduled tasks. Specifically: print the rank of the stack trace (rank 1 scheduled task: ..., rank 2 scheduled task: ...) stack trace jenkins hash, the total time spent on CPU, and the stack trace dump with only max depth of 4 frames.

Deliverables:

• Load perftop module
• Invoke cat /proc/perftop
• Add a screenshot of the output
• Upload the source code tarball