Lab 5: CoW fork

Pitfalls I've fallen into:

1. When adding reference count to kalloc.c, the kfree function should behave like this:

void kfree(void *pa) {
    acquire(&kmem.lock);
    --refcount[pa];
    if refcount[pa] == 0 {
        memset to 1
        add the page to freelist
    }
    release(&kmem.lock);
}

but I'm afraid that memset is time-consuming, not good for doing while holding the lock. So I did it like this:

void kfree(void *pa) {
    int is_released = 0;
    acquire(&kmem.lock);
    --refcount[pa];
    if refcount[pa] == 0 {
        is_released = 1;
        add the page to freelist
    }
    release(&kmem.lock);
    if is_released {
        memset to 1
    }
}

which is wrong because the freed page need to hold the pointer to the next free page, but memset will overwrite it. So the correct way is to memset before adding to the freelist.

2. The original uvmcopy function is:

int
uvmcopy(pagetable_t old, pagetable_t new, uint64 sz)
{
  pte_t *pte;
  uint64 pa, i;
  uint flags;
  char *mem;

  for(i = 0; i < sz; i += PGSIZE){
    if((pte = walk(old, i, 0)) == 0)
      continue;   // page table entry hasn't been allocated
    if((*pte & PTE_V) == 0)
      continue;   // physical page hasn't been allocated
    pa = PTE2PA(*pte);
    flags = PTE_FLAGS(*pte);
    if((mem = kalloc()) == 0)
      goto err;
    memmove(mem, (char*)pa, PGSIZE);
    if(mappages(new, i, PGSIZE, (uint64)mem, flags) != 0){
      kfree(mem);
      goto err;
    }
  }
  return 0;

 err:
  uvmunmap(new, 0, i / PGSIZE, 1);
  return -1;
}

I modified *pte to *pte & ~PTE_W | PTE_COW, thinking the flag will pass to the new page table's page, but no, I mistakenly added the CoW logic under flags = PTE_FLAGS(*pte), so the new page will still use the original flag.