oom: only oom kill exiting tasks with attached memory

When a task is chosen for oom kill and is found to be PF_EXITING,
__oom_kill_task() is called to elevate the task's timeslice and give it
access to memory reserves so that it may quickly exit.

This privilege is unnecessary, however, if the task has already detached
its mm.  Although its possible for the mm to become detached later since
task_lock() is not held, __oom_kill_task() will simply be a no-op in such
circumstances.

Subsequently, it is no longer necessary to warn about killing mm-less
tasks since it is a no-op.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/oom_kill.c

@@ -325,11 +325,8 @@ static void __oom_kill_task(struct task_struct *p, int verbose)
 		return;
 	}
 
-	if (!p->mm) {
-		WARN_ON(1);
-		printk(KERN_WARNING "tried to kill an mm-less task!\n");
+	if (!p->mm)
 		return;
-	}
 
 	if (verbose)
 		printk(KERN_ERR "Killed process %d (%s)\n",
@@ -397,8 +394,9 @@ static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
 	/*
 	 * If the task is already exiting, don't alarm the sysadmin or kill
 	 * its children or threads, just set TIF_MEMDIE so it can die quickly
+	 * if its mm is still attached.
 	 */
-	if (p->flags & PF_EXITING) {
+	if (p->mm && (p->flags & PF_EXITING)) {
 		__oom_kill_task(p, 0);
 		return 0;
 	}