oom-kill: show virtual size and rss information of the killed process
authorKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Wed, 16 Dec 2009 00:45:32 +0000 (16:45 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 16 Dec 2009 15:19:57 +0000 (07:19 -0800)
In a typical oom analysis scenario, we frequently want to know whether the
killed process has a memory leak or not at the first step.  This patch
adds vsz and rss information to the oom log to help this analysis.  To
save time for the debugging.

example:
===================================================================
rsyslogd invoked oom-killer: gfp_mask=0x201da, order=0, oom_adj=0
Pid: 1308, comm: rsyslogd Not tainted 2.6.32-rc6 #24
Call Trace:
[<ffffffff8132e35b>] ?_spin_unlock+0x2b/0x40
[<ffffffff810f186e>] oom_kill_process+0xbe/0x2b0

(snip)

492283 pages non-shared
Out of memory: kill process 2341 (memhog) score 527276 or a child
Killed process 2341 (memhog) vsz:1054552kB, anon-rss:970588kB, file-rss:4kB
===========================================================================
                             ^
                             |
                            here

[rientjes@google.com: fix race, add pid & comm to message]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/oom_kill.c

index 492c98624fc19bfd4b6fbb012b9c5661fc7fdb91..6bb8a7a7ec9ae0daa61871d456cb6736ec86c7b8 100644 (file)
@@ -352,6 +352,8 @@ static void dump_header(gfp_t gfp_mask, int order, struct mem_cgroup *mem)
                dump_tasks(mem);
 }
 
+#define K(x) ((x) << (PAGE_SHIFT-10))
+
 /*
  * Send SIGKILL to the selected  process irrespective of  CAP_SYS_RAW_IO
  * flag though it's unlikely that  we select a process with CAP_SYS_RAW_IO
@@ -365,15 +367,23 @@ static void __oom_kill_task(struct task_struct *p, int verbose)
                return;
        }
 
+       task_lock(p);
        if (!p->mm) {
                WARN_ON(1);
-               printk(KERN_WARNING "tried to kill an mm-less task!\n");
+               printk(KERN_WARNING "tried to kill an mm-less task %d (%s)!\n",
+                       task_pid_nr(p), p->comm);
+               task_unlock(p);
                return;
        }
 
        if (verbose)
-               printk(KERN_ERR "Killed process %d (%s)\n",
-                               task_pid_nr(p), p->comm);
+               printk(KERN_ERR "Killed process %d (%s) "
+                      "vsz:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
+                      task_pid_nr(p), p->comm,
+                      K(p->mm->total_vm),
+                      K(get_mm_counter(p->mm, anon_rss)),
+                      K(get_mm_counter(p->mm, file_rss)));
+       task_unlock(p);
 
        /*
         * We give our sacrificial lamb high priority and access to