2011-04-21 01:43:41

by Steven Rostedt

[permalink] [raw]
Subject: [PATCH 4/7] lockdep: Printk nice description for irq inversion bug

From: Steven Rostedt <[email protected]>

The irq inversion and irq dependency bug are only subtly different.
The diffenerence lies where the interrupt occurred.

For irq dependency:

irq_disable
lock(A)
lock(B)
unlock(B)
unlock(A)
irq_enable

lock(B)
unlock(B)

<interrupt>
lock(A)

The interrupt comes in after it has been established that lock A
can be held when taking an irq unsafe lock. Lockdep detects the problem
when taking lock A in interrupt context.

<interrupt>
lock(A)

irq_disable
lock(A)
lock(B)
unlock(B)
unlock(A)
irq_enable

lock(B)
unlock(B)

With the irq_inversion the irq happens before it is established
and lockdep detects the problem with the taking of lock B.

Since the problem with the locking for both of these issues are
in actuality the same, they both should report the same scenario.

other info that might help us debug this:

Chain exists of:
&rq->lock --> lockA --> lockC

Possible interrupt unsafe locking scenario:

CPU0 CPU1
---- ----
lock(lockC);
local_irq_disable();
lock(&rq->lock);
lock(lockA);
<Interrupt>
lock(&rq->lock);

*** DEADLOCK ***

Acked-by: Peter Zijlstra <[email protected]>
Signed-off-by: Steven Rostedt <[email protected]>
---
kernel/lockdep.c | 34 +++++++++++++++++++++++++++++-----
1 files changed, 29 insertions(+), 5 deletions(-)

diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index a0f253b..8646b8c 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -1395,15 +1395,15 @@ print_shortest_lock_dependencies(struct lock_list *leaf,
static void
print_irq_lock_scenario(struct lock_list *safe_entry,
struct lock_list *unsafe_entry,
- struct held_lock *prev,
- struct held_lock *next)
+ struct lock_class *prev_class,
+ struct lock_class *next_class)
{
struct lock_class *safe_class = safe_entry->class;
struct lock_class *unsafe_class = unsafe_entry->class;
- struct lock_class *middle_class = hlock_class(prev);
+ struct lock_class *middle_class = prev_class;

if (middle_class == safe_class)
- middle_class = hlock_class(next);
+ middle_class = next_class;

/*
* A direct locking problem where unsafe_class lock is taken
@@ -1499,7 +1499,8 @@ print_bad_irq_dependency(struct task_struct *curr,
print_stack_trace(forwards_entry->class->usage_traces + bit2, 1);

printk("\nother info that might help us debug this:\n\n");
- print_irq_lock_scenario(backwards_entry, forwards_entry, prev, next);
+ print_irq_lock_scenario(backwards_entry, forwards_entry,
+ hlock_class(prev), hlock_class(next));

lockdep_print_held_locks(curr);

@@ -2219,6 +2220,10 @@ print_irq_inversion_bug(struct task_struct *curr,
struct held_lock *this, int forwards,
const char *irqclass)
{
+ struct lock_list *entry = other;
+ struct lock_list *middle = NULL;
+ int depth;
+
if (!debug_locks_off_graph_unlock() || debug_locks_silent)
return 0;

@@ -2237,6 +2242,25 @@ print_irq_inversion_bug(struct task_struct *curr,
printk("\n\nand interrupts could create inverse lock ordering between them.\n\n");

printk("\nother info that might help us debug this:\n");
+
+ /* Find a middle lock (if one exists) */
+ depth = get_lock_depth(other);
+ do {
+ if (depth == 0 && (entry != root)) {
+ printk("lockdep:%s bad path found in chain graph\n", __func__);
+ break;
+ }
+ middle = entry;
+ entry = get_lock_parent(entry);
+ depth--;
+ } while (entry && entry != root && (depth >= 0));
+ if (forwards)
+ print_irq_lock_scenario(root, other,
+ middle ? middle->class : root->class, other->class);
+ else
+ print_irq_lock_scenario(other, root,
+ middle ? middle->class : other->class, root->class);
+
lockdep_print_held_locks(curr);

printk("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
--
1.7.2.3


2011-04-22 12:21:20

by Steven Rostedt

[permalink] [raw]
Subject: [tip:core/locking] lockdep: Print a nicer description for irq inversion bugs

Commit-ID: dad3d7435e1d8c254d6877dc06852dc00c5da812
Gitweb: http://git.kernel.org/tip/dad3d7435e1d8c254d6877dc06852dc00c5da812
Author: Steven Rostedt <[email protected]>
AuthorDate: Wed, 20 Apr 2011 21:41:57 -0400
Committer: Ingo Molnar <[email protected]>
CommitDate: Fri, 22 Apr 2011 11:06:58 +0200

lockdep: Print a nicer description for irq inversion bugs

Irq inversion and irq dependency bugs are only subtly
different. The diffenerence lies where the interrupt occurred.

For irq dependency:

irq_disable
lock(A)
lock(B)
unlock(B)
unlock(A)
irq_enable

lock(B)
unlock(B)

<interrupt>
lock(A)

The interrupt comes in after it has been established that lock A
can be held when taking an irq unsafe lock. Lockdep detects the
problem when taking lock A in interrupt context.

With the irq_inversion the irq happens before it is established
and lockdep detects the problem with the taking of lock B:

<interrupt>
lock(A)

irq_disable
lock(A)
lock(B)
unlock(B)
unlock(A)
irq_enable

lock(B)
unlock(B)

Since the problem with the locking logic for both of these issues
is in actuality the same, they both should report the same scenario.
This patch implements that and prints this:

other info that might help us debug this:

Chain exists of:
&rq->lock --> lockA --> lockC

Possible interrupt unsafe locking scenario:

CPU0 CPU1
---- ----
lock(lockC);
local_irq_disable();
lock(&rq->lock);
lock(lockA);
<Interrupt>
lock(&rq->lock);

*** DEADLOCK ***

Signed-off-by: Steven Rostedt <[email protected]>
Acked-by: Peter Zijlstra <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
---
kernel/lockdep.c | 34 +++++++++++++++++++++++++++++-----
1 files changed, 29 insertions(+), 5 deletions(-)

diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index c4cc5d1..0b497dd 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -1395,15 +1395,15 @@ print_shortest_lock_dependencies(struct lock_list *leaf,
static void
print_irq_lock_scenario(struct lock_list *safe_entry,
struct lock_list *unsafe_entry,
- struct held_lock *prev,
- struct held_lock *next)
+ struct lock_class *prev_class,
+ struct lock_class *next_class)
{
struct lock_class *safe_class = safe_entry->class;
struct lock_class *unsafe_class = unsafe_entry->class;
- struct lock_class *middle_class = hlock_class(prev);
+ struct lock_class *middle_class = prev_class;

if (middle_class == safe_class)
- middle_class = hlock_class(next);
+ middle_class = next_class;

/*
* A direct locking problem where unsafe_class lock is taken
@@ -1499,7 +1499,8 @@ print_bad_irq_dependency(struct task_struct *curr,
print_stack_trace(forwards_entry->class->usage_traces + bit2, 1);

printk("\nother info that might help us debug this:\n\n");
- print_irq_lock_scenario(backwards_entry, forwards_entry, prev, next);
+ print_irq_lock_scenario(backwards_entry, forwards_entry,
+ hlock_class(prev), hlock_class(next));

lockdep_print_held_locks(curr);

@@ -2219,6 +2220,10 @@ print_irq_inversion_bug(struct task_struct *curr,
struct held_lock *this, int forwards,
const char *irqclass)
{
+ struct lock_list *entry = other;
+ struct lock_list *middle = NULL;
+ int depth;
+
if (!debug_locks_off_graph_unlock() || debug_locks_silent)
return 0;

@@ -2237,6 +2242,25 @@ print_irq_inversion_bug(struct task_struct *curr,
printk("\n\nand interrupts could create inverse lock ordering between them.\n\n");

printk("\nother info that might help us debug this:\n");
+
+ /* Find a middle lock (if one exists) */
+ depth = get_lock_depth(other);
+ do {
+ if (depth == 0 && (entry != root)) {
+ printk("lockdep:%s bad path found in chain graph\n", __func__);
+ break;
+ }
+ middle = entry;
+ entry = get_lock_parent(entry);
+ depth--;
+ } while (entry && entry != root && (depth >= 0));
+ if (forwards)
+ print_irq_lock_scenario(root, other,
+ middle ? middle->class : root->class, other->class);
+ else
+ print_irq_lock_scenario(other, root,
+ middle ? middle->class : other->class, root->class);
+
lockdep_print_held_locks(curr);

printk("\nthe shortest dependencies between 2nd lock and 1st lock:\n");