percpu_counter: add debugobj support
authorTejun Heo <tj@kernel.org>
Tue, 26 Oct 2010 21:23:05 +0000 (14:23 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 26 Oct 2010 23:52:18 +0000 (16:52 -0700)
All percpu counters are linked to a global list on initialization and
removed from it on destruction.  The list is walked during CPU up/down.
If a percpu counter is freed without being properly destroyed, the system
will oops only on the next CPU up/down making it pretty nasty to track
down.  This patch adds debugobj support for percpu counters so that such
problems can be found easily.

As percpu counters don't make sense on stack and can't be statically
initialized, debugobj support is pretty simple.  It's initialized and
activated on counter initialization, and deactivatd and destroyed on
counter destruction.  With this patch applied, the bug fixed by commit
602586a83b719df0fbd94196a1359ed35aeb2df3 (shmem: put_super must
percpu_counter_destroy) triggers the following warning on tmpfs unmount
and the system won't oops on the next cpu up/down operation.

 ------------[ cut here ]------------
 WARNING: at lib/debugobjects.c:259 debug_print_object+0x5c/0x70()
 Hardware name: Bochs
 ODEBUG: free active (active state 0) object type: percpu_counter
 Modules linked in:
 Pid: 3999, comm: umount Not tainted 2.6.36-rc2-work+ #5
 Call Trace:
  [<ffffffff81083f7f>] warn_slowpath_common+0x7f/0xc0
  [<ffffffff81084076>] warn_slowpath_fmt+0x46/0x50
  [<ffffffff813b45cc>] debug_print_object+0x5c/0x70
  [<ffffffff813b50e5>] debug_check_no_obj_freed+0x125/0x210
  [<ffffffff811577d3>] kfree+0xb3/0x2f0
  [<ffffffff81132edd>] shmem_put_super+0x1d/0x30
  [<ffffffff81162e96>] generic_shutdown_super+0x56/0xe0
  [<ffffffff81162f86>] kill_anon_super+0x16/0x60
  [<ffffffff81162ff7>] kill_litter_super+0x27/0x30
  [<ffffffff81163295>] deactivate_locked_super+0x45/0x60
  [<ffffffff81163cfa>] deactivate_super+0x4a/0x70
  [<ffffffff8117d446>] mntput_no_expire+0x86/0xe0
  [<ffffffff8117df7f>] sys_umount+0x6f/0x360
  [<ffffffff8103f01b>] system_call_fastpath+0x16/0x1b
 ---[ end trace cce2a341ba3611a7 ]---

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Thomas Gleixner <tglxlinutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
lib/Kconfig.debug
lib/percpu_counter.c

index 69a32664c289692576677a48d5daacca69212752..0d5c762532a554744762870a26ddc4236686b6c5 100644 (file)
@@ -317,6 +317,14 @@ config DEBUG_OBJECTS_RCU_HEAD
        help
          Enable this to turn on debugging of RCU list heads (call_rcu() usage).
 
+config DEBUG_OBJECTS_PERCPU_COUNTER
+       bool "Debug percpu counter objects"
+       depends on DEBUG_OBJECTS
+       help
+         If you say Y here, additional code will be inserted into the
+         percpu counter routines to track the life time of percpu counter
+         objects and validate the percpu counter operations.
+
 config DEBUG_OBJECTS_ENABLE_DEFAULT
        int "debug_objects bootup default value (0-1)"
         range 0 1
index 209448e1d2b984b2cb8755b3184263994d01493b..1d954ea72331499326b14c9564b4f90f64defec7 100644 (file)
@@ -8,10 +8,53 @@
 #include <linux/init.h>
 #include <linux/cpu.h>
 #include <linux/module.h>
+#include <linux/debugobjects.h>
 
 static LIST_HEAD(percpu_counters);
 static DEFINE_MUTEX(percpu_counters_lock);
 
+#ifdef CONFIG_DEBUG_OBJECTS_PERCPU_COUNTER
+
+static struct debug_obj_descr percpu_counter_debug_descr;
+
+static int percpu_counter_fixup_free(void *addr, enum debug_obj_state state)
+{
+       struct percpu_counter *fbc = addr;
+
+       switch (state) {
+       case ODEBUG_STATE_ACTIVE:
+               percpu_counter_destroy(fbc);
+               debug_object_free(fbc, &percpu_counter_debug_descr);
+               return 1;
+       default:
+               return 0;
+       }
+}
+
+static struct debug_obj_descr percpu_counter_debug_descr = {
+       .name           = "percpu_counter",
+       .fixup_free     = percpu_counter_fixup_free,
+};
+
+static inline void debug_percpu_counter_activate(struct percpu_counter *fbc)
+{
+       debug_object_init(fbc, &percpu_counter_debug_descr);
+       debug_object_activate(fbc, &percpu_counter_debug_descr);
+}
+
+static inline void debug_percpu_counter_deactivate(struct percpu_counter *fbc)
+{
+       debug_object_deactivate(fbc, &percpu_counter_debug_descr);
+       debug_object_free(fbc, &percpu_counter_debug_descr);
+}
+
+#else  /* CONFIG_DEBUG_OBJECTS_PERCPU_COUNTER */
+static inline void debug_percpu_counter_activate(struct percpu_counter *fbc)
+{ }
+static inline void debug_percpu_counter_deactivate(struct percpu_counter *fbc)
+{ }
+#endif /* CONFIG_DEBUG_OBJECTS_PERCPU_COUNTER */
+
 void percpu_counter_set(struct percpu_counter *fbc, s64 amount)
 {
        int cpu;
@@ -75,6 +118,9 @@ int __percpu_counter_init(struct percpu_counter *fbc, s64 amount,
        fbc->counters = alloc_percpu(s32);
        if (!fbc->counters)
                return -ENOMEM;
+
+       debug_percpu_counter_activate(fbc);
+
 #ifdef CONFIG_HOTPLUG_CPU
        INIT_LIST_HEAD(&fbc->list);
        mutex_lock(&percpu_counters_lock);
@@ -90,6 +136,8 @@ void percpu_counter_destroy(struct percpu_counter *fbc)
        if (!fbc->counters)
                return;
 
+       debug_percpu_counter_deactivate(fbc);
+
 #ifdef CONFIG_HOTPLUG_CPU
        mutex_lock(&percpu_counters_lock);
        list_del(&fbc->list);