On Sun, 23 Mar 2014, Thomas Gleixner wrote:
> On Sat, 22 Mar 2014, Thomas Gleixner wrote:
> > On Sat, 22 Mar 2014, Thomas Gleixner wrote:
> > > On Fri, 21 Mar 2014, Vince Weaver wrote:
> > >
> > > > On Fri, 21 Mar 2014, Thomas Gleixner wrote:
> > > > >
> > > > > I'm a complete idiot. I was staring at oaddr and did not notice that
> > > > > descr->name is the real culprit. Sorry. Delta patch below.
> > > >
> > > > OK. The log was much longer this time, attached.
> > >
> > > Hmmm.
> > >
> > > [ 2.739858] NULL pointer dereference at (null)
> > > [ 2.747390] IP: [< (null)>] (null)
> > > [ 2.752970] PGD 0
> > > [ 2.755287] Oops: 0010 [#1] SMP
> > >
> > > So this time the CPU branched to NULL. So let me recap.
> > >
> > > First you had the explosion in the hrtimer code. After enabling debug
> > > stuff it went to the timer_list and now it looks different again.
> > >
> > > So that looks more like a random memory corruption.
> > >
> > > Nasty to debug. And of course it does not reproduce here. I'll throw
> > > your config at more machines in the hope that something will trigger
> > > it.
> >
> > I've refined the trace_printk stuff in the hope to get a bit more info
> > out of it.
>
> You might also try with the trace_printks removed. That wont give us
> the history, but maybe then the bug happens again at some decodable
> place.
>
I tried to reproduce on a few machines, but no luck.
But I tooks some time to tweak the debugobjects patch further in the
hope that it helps.
I wonder whether it might be helpful to instrument memset/memcpy with
debugobjects. Those are pretty much the same as a free from an active
object POV. But that's a non trivial thing to do.
I'm traveling tomorrow, so I wont be able to respond before Tuesday.
Thanks,
tglx
--------------->
include/linux/debugobjects.h | 4 ++
kernel/timer.c | 7 +++
lib/debugobjects.c | 83 +++++++++++++++++++++++++++++++++++++++++--
3 files changed, 90 insertions(+), 4 deletions(-)
Index: linux-2.6/include/linux/debugobjects.h
===================================================================
--- linux-2.6.orig/include/linux/debugobjects.h
+++ linux-2.6/include/linux/debugobjects.h
@@ -30,6 +30,7 @@ struct debug_obj {
unsigned int astate;
void *object;
struct debug_obj_descr *descr;
+ void *hint;
};
/**
@@ -68,6 +69,7 @@ extern void debug_object_deactivate(void
extern void debug_object_destroy (void *addr, struct debug_obj_descr *descr);
extern void debug_object_free (void *addr, struct debug_obj_descr *descr);
extern void debug_object_assert_init(void *addr, struct debug_obj_descr
*descr);
+extern void debug_object_info(void *addr, struct debug_obj_descr *descr);
/*
* Active state:
@@ -95,6 +97,8 @@ static inline void
debug_object_free (void *addr, struct debug_obj_descr *descr) { }
static inline void
debug_object_assert_init(void *addr, struct debug_obj_descr *descr) { }
+static inline void
+debug_object_info(void *addr, struct debug_obj_descr *descr) { }
static inline void debug_objects_early_init(void) { }
static inline void debug_objects_mem_init(void) { }
Index: linux-2.6/kernel/timer.c
===================================================================
--- linux-2.6.orig/kernel/timer.c
+++ linux-2.6/kernel/timer.c
@@ -1081,7 +1081,12 @@ static int cascade(struct tvec_base *bas
* don't have to detach them individually.
*/
list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
- BUG_ON(tbase_get_base(timer->base) != base);
+ if (tbase_get_base(timer->base) != base) {
+ pr_err("Invalid timer base: tmr %p tmr->base %p base
%p\n",
+ timer, timer->base, base);
+ debug_object_info(timer, &timer_debug_descr);
+ BUG();
+ }
/* No accounting, while moving them */
__internal_add_timer(base, timer);
}
Index: linux-2.6/lib/debugobjects.c
===================================================================
--- linux-2.6.orig/lib/debugobjects.c
+++ linux-2.6/lib/debugobjects.c
@@ -70,13 +70,14 @@ static int __init disable_object_debug(c
early_param("debug_objects", enable_object_debug);
early_param("no_debug_objects", disable_object_debug);
-static const char *obj_states[ODEBUG_STATE_MAX] = {
+static const char *obj_states[ODEBUG_STATE_MAX + 1] = {
[ODEBUG_STATE_NONE] = "none",
[ODEBUG_STATE_INIT] = "initialized",
[ODEBUG_STATE_INACTIVE] = "inactive",
[ODEBUG_STATE_ACTIVE] = "active",
[ODEBUG_STATE_DESTROYED] = "destroyed",
[ODEBUG_STATE_NOTAVAILABLE] = "not available",
+ [ODEBUG_STATE_MAX] = "free",
};
static void fill_pool(void)
@@ -140,6 +141,7 @@ alloc_object(void *addr, struct debug_bu
obj->descr = descr;
obj->state = ODEBUG_STATE_NONE;
obj->astate = 0;
+ obj->hint = NULL;
hlist_del(&obj->node);
hlist_add_head(&obj->node, &b->list);
@@ -245,6 +247,47 @@ static struct debug_bucket *get_bucket(u
return &obj_hash[hash];
}
+static void debug_object_update_hint(struct debug_obj *obj, int newstate)
+{
+ struct debug_obj_descr *descr = obj->descr;
+ bool has_hint = descr->debug_hint != NULL;
+ void *hint;
+
+ static int limit;
+
+ if (!has_hint || limit > 10)
+ return;
+
+ hint = descr->debug_hint(obj->object);
+#if 1
+ trace_printk("obj: %p %s --> %s (active state %u) type: %s hint: %pS ->
%pS\n",
+ obj->object, obj_states[obj->state], obj_states[newstate],
obj->astate,
+ descr->name, obj->hint, hint);
+#endif
+ /* This does not care about improper state transitions! */
+ switch (newstate) {
+ case ODEBUG_STATE_INIT:
+ return;
+ case ODEBUG_STATE_ACTIVE:
+ if (!hint || !kernel_text_address((unsigned long) hint))
+ break;
+ obj->hint = hint;
+ return;
+ case ODEBUG_STATE_INACTIVE:
+ if (!hint || !kernel_text_address((unsigned long) hint))
+ break;
+ /* Someone changed the callback of an active object */
+ if (obj->hint != hint && obj->state == ODEBUG_STATE_ACTIVE)
+ break;
+ default:
+ return;
+ }
+ WARN(1, KERN_ERR "ODEBUG: %s -> %s (active state %u) object type: %s
hint: %pS -> %pS\n",
+ obj_states[obj->state], obj_states[newstate], obj->astate,
descr->name,
+ obj->hint, hint);
+ limit++;
+}
+
static void debug_print_object(struct debug_obj *obj, char *msg)
{
struct debug_obj_descr *descr = obj->descr;
@@ -255,9 +298,9 @@ static void debug_print_object(struct de
descr->debug_hint(obj->object) : NULL;
limit++;
WARN(1, KERN_ERR "ODEBUG: %s %s (active state %u) "
- "object type: %s hint: %pS\n",
+ "object type: %s hint: %pS %pS\n",
msg, obj_states[obj->state], obj->astate,
- descr->name, hint);
+ descr->name, hint, obj->hint);
}
debug_objects_warnings++;
}
@@ -326,6 +369,8 @@ __debug_object_init(void *addr, struct d
debug_object_is_on_stack(addr, onstack);
}
+ debug_object_update_hint(obj, ODEBUG_STATE_INIT);
+
switch (obj->state) {
case ODEBUG_STATE_NONE:
case ODEBUG_STATE_INIT:
@@ -377,6 +422,28 @@ void debug_object_init_on_stack(void *ad
__debug_object_init(addr, descr, 1);
}
+void debug_object_info(void *addr, struct debug_obj_descr *descr)
+{
+ struct debug_bucket *db;
+ struct debug_obj *obj;
+ unsigned long flags;
+
+ if (!debug_objects_enabled)
+ return;
+
+ db = get_bucket((unsigned long) addr);
+
+ raw_spin_lock_irqsave(&db->lock, flags);
+
+ obj = lookup_object(addr, db);
+ if (!obj)
+ pr_err("Object unknown %p\n", addr);
+ else
+ debug_print_object(obj, "Info");
+
+ raw_spin_unlock_irqrestore(&db->lock, flags);
+}
+
/**
* debug_object_activate - debug checks when an object is activated
* @addr: address of the object
@@ -403,6 +470,9 @@ int debug_object_activate(void *addr, st
obj = lookup_object(addr, db);
if (obj) {
+
+ debug_object_update_hint(obj, ODEBUG_STATE_ACTIVE);
+
switch (obj->state) {
case ODEBUG_STATE_INIT:
case ODEBUG_STATE_INACTIVE:
@@ -425,6 +495,7 @@ int debug_object_activate(void *addr, st
ret = 0;
break;
}
+
raw_spin_unlock_irqrestore(&db->lock, flags);
return ret;
}
@@ -463,6 +534,8 @@ void debug_object_deactivate(void *addr,
obj = lookup_object(addr, db);
if (obj) {
+ debug_object_update_hint(obj, ODEBUG_STATE_INACTIVE);
+
switch (obj->state) {
case ODEBUG_STATE_INIT:
case ODEBUG_STATE_INACTIVE:
@@ -513,6 +586,8 @@ void debug_object_destroy(void *addr, st
if (!obj)
goto out_unlock;
+ debug_object_update_hint(obj, ODEBUG_STATE_DESTROYED);
+
switch (obj->state) {
case ODEBUG_STATE_NONE:
case ODEBUG_STATE_INIT:
@@ -559,6 +634,8 @@ void debug_object_free(void *addr, struc
if (!obj)
goto out_unlock;
+ debug_object_update_hint(obj, ODEBUG_STATE_MAX);
+
switch (obj->state) {
case ODEBUG_STATE_ACTIVE:
debug_print_object(obj, "free");
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
