6.10 Exercises

Comment out the calls to acquire and release in kalloc (kernel/kalloc.c:69). This seems like it should cause problems for kernel code that calls kalloc; what symptoms do you expect to see? When you run xv6, do you see these symptoms? How about when running usertests? If you don’t see a problem, why not? See if you can provoke a problem by inserting dummy loops into the critical section of kalloc.

Suppose that you instead commented out the locking in kfree (after restoring locking in kalloc). What might now go wrong? Is lack of locks in kfree less harmful than in kalloc?

If two CPUs call kalloc at the same time, one will have to wait for the other, which is bad for performance. Modify kalloc.c to have more parallelism, so that simultaneous calls to kalloc from different CPUs can proceed without waiting for each other.

Write a parallel program using POSIX threads, which is supported on most operating systems. For example, implement a parallel hash table and measure if the number of puts/gets scales with increasing number of CPUs.

Implement a subset of Pthreads in xv6. That is, implement a user-level thread library so that a user process can have more than 1 thread and arrange that these threads can run in parallel on different CPUs. Come up with a design that correctly handles a thread making a blocking system call and changing its shared address space.