libata: reorder ata_queued_cmd to remove alignment padding on 64 bit builds
[pandora-kernel.git] / mm / oom_kill.c
index 5014e50..4029583 100644 (file)
@@ -121,8 +121,8 @@ struct task_struct *find_lock_task_mm(struct task_struct *p)
 }
 
 /* return true if the task is not adequate as candidate victim task. */
-static bool oom_unkillable_task(struct task_struct *p, struct mem_cgroup *mem,
-                          const nodemask_t *nodemask)
+static bool oom_unkillable_task(struct task_struct *p,
+               const struct mem_cgroup *mem, const nodemask_t *nodemask)
 {
        if (is_global_init(p))
                return true;
@@ -208,8 +208,13 @@ unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
         */
        points += p->signal->oom_score_adj;
 
-       if (points < 0)
-               return 0;
+       /*
+        * Never return 0 for an eligible task that may be killed since it's
+        * possible that no single user task uses more than 0.1% of memory and
+        * no single admin tasks uses more than 3.0%.
+        */
+       if (points <= 0)
+               return 1;
        return (points < 1000) ? points : 1000;
 }
 
@@ -339,26 +344,24 @@ static struct task_struct *select_bad_process(unsigned int *ppoints,
 /**
  * dump_tasks - dump current memory state of all system tasks
  * @mem: current's memory controller, if constrained
+ * @nodemask: nodemask passed to page allocator for mempolicy ooms
  *
- * Dumps the current memory state of all system tasks, excluding kernel threads.
+ * Dumps the current memory state of all eligible tasks.  Tasks not in the same
+ * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
+ * are not shown.
  * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
  * value, oom_score_adj value, and name.
  *
- * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are
- * shown.
- *
  * Call with tasklist_lock read-locked.
  */
-static void dump_tasks(const struct mem_cgroup *mem)
+static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
 {
        struct task_struct *p;
        struct task_struct *task;
 
        pr_info("[ pid ]   uid  tgid total_vm      rss cpu oom_adj oom_score_adj name\n");
        for_each_process(p) {
-               if (p->flags & PF_KTHREAD)
-                       continue;
-               if (mem && !task_in_mem_cgroup(p, mem))
+               if (oom_unkillable_task(p, mem, nodemask))
                        continue;
 
                task = find_lock_task_mm(p);
@@ -372,7 +375,7 @@ static void dump_tasks(const struct mem_cgroup *mem)
                }
 
                pr_info("[%5d] %5d %5d %8lu %8lu %3u     %3d         %5d %s\n",
-                       task->pid, __task_cred(task)->uid, task->tgid,
+                       task->pid, task_uid(task), task->tgid,
                        task->mm->total_vm, get_mm_rss(task->mm),
                        task_cpu(task), task->signal->oom_adj,
                        task->signal->oom_score_adj, task->comm);
@@ -381,7 +384,7 @@ static void dump_tasks(const struct mem_cgroup *mem)
 }
 
 static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
-                                                       struct mem_cgroup *mem)
+                       struct mem_cgroup *mem, const nodemask_t *nodemask)
 {
        task_lock(current);
        pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
@@ -394,17 +397,16 @@ static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
        mem_cgroup_print_oom_info(mem, p);
        show_mem();
        if (sysctl_oom_dump_tasks)
-               dump_tasks(mem);
+               dump_tasks(mem, nodemask);
 }
 
 #define K(x) ((x) << (PAGE_SHIFT-10))
 static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
 {
        p = find_lock_task_mm(p);
-       if (!p) {
-               task_unlock(p);
+       if (!p)
                return 1;
-       }
+
        pr_err("Killed process %d (%s) total-vm:%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, MM_ANONPAGES)),
@@ -437,7 +439,7 @@ static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
        unsigned int victim_points = 0;
 
        if (printk_ratelimit())
-               dump_header(p, gfp_mask, order, mem);
+               dump_header(p, gfp_mask, order, mem, nodemask);
 
        /*
         * If the task is already exiting, don't alarm the sysadmin or kill
@@ -483,7 +485,7 @@ static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
  * Determines whether the kernel must panic because of the panic_on_oom sysctl.
  */
 static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
-                               int order)
+                               int order, const nodemask_t *nodemask)
 {
        if (likely(!sysctl_panic_on_oom))
                return;
@@ -497,7 +499,7 @@ static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
                        return;
        }
        read_lock(&tasklist_lock);
-       dump_header(NULL, gfp_mask, order, NULL);
+       dump_header(NULL, gfp_mask, order, NULL, nodemask);
        read_unlock(&tasklist_lock);
        panic("Out of memory: %s panic_on_oom is enabled\n",
                sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
@@ -510,7 +512,7 @@ void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
        unsigned int points = 0;
        struct task_struct *p;
 
-       check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0);
+       check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
        limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
        read_lock(&tasklist_lock);
 retry:
@@ -642,11 +644,13 @@ static void clear_system_oom(void)
 void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
                int order, nodemask_t *nodemask)
 {
+       const nodemask_t *mpol_mask;
        struct task_struct *p;
        unsigned long totalpages;
        unsigned long freed = 0;
        unsigned int points;
        enum oom_constraint constraint = CONSTRAINT_NONE;
+       int killed = 0;
 
        blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
        if (freed > 0)
@@ -670,7 +674,8 @@ void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
         */
        constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
                                                &totalpages);
-       check_panic_on_oom(constraint, gfp_mask, order);
+       mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
+       check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
 
        read_lock(&tasklist_lock);
        if (sysctl_oom_kill_allocating_task &&
@@ -684,19 +689,17 @@ void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
                if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
                                NULL, nodemask,
                                "Out of memory (oom_kill_allocating_task)"))
-                       return;
+                       goto out;
        }
 
 retry:
-       p = select_bad_process(&points, totalpages, NULL,
-                       constraint == CONSTRAINT_MEMORY_POLICY ? nodemask :
-                                                                NULL);
+       p = select_bad_process(&points, totalpages, NULL, mpol_mask);
        if (PTR_ERR(p) == -1UL)
-               return;
+               goto out;
 
        /* Found nothing?!?! Either we hang forever, or we panic. */
        if (!p) {
-               dump_header(NULL, gfp_mask, order, NULL);
+               dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
                read_unlock(&tasklist_lock);
                panic("Out of memory and no killable processes...\n");
        }
@@ -704,13 +707,15 @@ retry:
        if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
                                nodemask, "Out of memory"))
                goto retry;
+       killed = 1;
+out:
        read_unlock(&tasklist_lock);
 
        /*
         * Give "p" a good chance of killing itself before we
         * retry to allocate memory unless "p" is current
         */
-       if (!test_thread_flag(TIF_MEMDIE))
+       if (killed && !test_thread_flag(TIF_MEMDIE))
                schedule_timeout_uninterruptible(1);
 }