VM - Virtual Memory Allocation: brk() and mmap()

Posted on | In | Comments: | Views:

Credit goes to:
https://blog.csdn.net/gfgdsg/article/details/42709943

Notice: page-granularity

All memory in virtual memory address space are page-based. It is the smallest unit of granularity, including the Heap segment in the following.

Virtual Memory Allocation (2 ways):

  1. brk() on Heap segment
  2. mmap() mapped_file segment, between Heap and Stack

Usually user process calls malloc() to invoke kernel
system calls brk() or mmap() to get a virtual memory. Which one will be executed depends on requesting memory size(128KB is the line)

  • size<128KB, kernel calls brk() to assign virtual memory on Heap to user process
    • BAD: only be sequentially assign and free, like Stack
    • GOOD: can be used as cache ()
  • size>128KB, kerel calls mmap() to get a piece of memory between Heap and Stack to user process
    • GOOD: can independently assign and free (flexible)
    • BAD: repeately causing page_fault to involve physical memory (CPU costly)

System API：do_brk( )

1
2
3
4
5
6
7
8
int do_brk(void *addr, void **ret)

设置 process 地址空间中Heap Segment的break，即Heap的“堆顶”。  
- malloc()、free()均会调用该API，增加、减小heap break
- malloc()时向高地址增加; free()时向低地址减小

输入： addr 要更新的break地址； ret 更新后的break地址
输出：0-成功
  1. 如果 addr == NULL，返回当前 brk
  2. 判断 addr 的合法性：与 proc_start_brk比较，不能比他还小
  3. 处理 addr 与 proc_brk 在 page_not_aligned 的情况；并获得old proc_brk所在 virtual area 的最后一页 end_page
    • addr : newvfn (new virtual frame number)
    • curproc->p_brk: oldvma->vma_end
    • if(newvfn > oldvma->vma_end): 对应于malloc()
    • if(newvfn < oldvma->vma_end): 对应于free()
  4. 最后更新两个量：
    • oldvma->vma_end = newvfn;
    • curproc->p_brk = addr;