Department of Computer Science
University of São Paulo
São Paulo, Brazil

description

functions

---

• static unsigned int int_sqrt(unsigned int x)

  ---

• static int badness (struct task_struct *p)

  description

  It basically calculates a numeric value for how "bad" this task has been. This is the way we measure which process is the best to kill. So along the function, the more "bad" points it acquires, more likely it'll be the chosen for the kill. This is called from select_bad_process() to select the highest ranked (worst) process.

  The idea of good here is that is (taken from original oom_kill.c):

  • we lose the minimum amount of work done
  • we recover a large amount of memory
  • we don't kill anything innocent of eating tons of memory
  • we want to kill the minimum amount of processes (one)
  • we try to kill the process the user expects us to kill, this algorithm has been meticulously tuned to meet the priniciple of least surprise.

  Here are some heurists rules for setting the points to the processes:

  • The processes' points start with the memory size of that process (p->mm->total_vm).
  • The points are divided by the square root of their cpu_time and run_time, this way we give less points to a process that has been running for a long time.
  • If a process is niced, then it is liked not very important, therefore it's points are doubled.
  • If a process is owned by root, then we probably don't want to kill (it's prbably important for the whole system), so it's points are divided by four.
  • If a process has direct hardware access, then we probably don't want to kill it, so also divide it's points by four.
  • Finished. Return the accumulated points.

  parameters

  • p - task struct of the task (process) we should calculate.

  return value

  • returns the points this task has accumulated.

  ---

• static struct task_struct select_bad_process(void)

  description

  Loops all tasks (using for_each_task in sched.h), looking for the task highest number of points. All it does is call badness() for each task.

  parameters

  • none

  return value

  • returns the chosen task (to be killed in oom_kill()).

  ---

• void oom_kill(void)

  description

  Kills the "best" process so that we can free memory in the system. This is only called if the system has no more free memory. This is called from kswapd(), only if out_of_memory() returns true.
  • Call select_bad_process() to get the process we will kill
  • Give this process high priority and access to memory, so it can exit quickly
  • If the process has hardware access throw a SIGTERM signal, otherwise throw a SIGKILL
  • Yield kswapd so that this process has a chance to killing itself before another process wants some more memory.

  parameters

  • none

  return value

  • none

  ---

• int out_of_memory(void)

  description

  • Check if the system is really out of free memory
  • If nr_free_pages() is above the minimum (freepages.min) then return 0 (we still have memory)
  • If nr_free_pages() + nr_inactive_clean_pages() is above freepages.low, then we still have some memory (still return 0)
  • Check from si_swapinfo() if there is still space in swap, if so return 0.
  • If we've still not returned, then all free memory is gone! Return 1.

  parameters

  • none

  return value

  • returns 0 if there is still some memory left
  • returns 1 otherwise (no more memory)

  ---

Page last updated on "Mon Apr 30 08:44:25 2001"
Send feedback to Rodrigo S. de Castro<[email protected]>