[PATCH 1/7] lockdep: Print a nice description of an irq locking issue

[Steven Rostedt <rostedt@goodmis.org>]

[-- Attachment #1: 0001-lockdep-Print-a-nice-description-of-an-irq-locking-i.patch --]
[-- Type: text/plain, Size: 5045 bytes --]

From: Steven Rostedt <srostedt@redhat.com>

Locking order inversion due to interrupts is a subtle problem.
When a locking inversion due to interrupts is discovered by lockdep,
it currently reports something like this:

[ INFO: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected ]

And then writes the locks that are involved as well as back traces.
But several developers are confused by what a HARDIRQ->safe to unsafe
issue is all about, and sometimes even blow it off as a bug in lockdep.
As it is not obvious when lockdep describes this about a lock that
is never taken in interrupt context.

After explaining the problems that lockdep is reporting, I decided
to add a description of the problem in visual form. Now the following
is shown:

 ---
other info that might help us debug this:

 Possible interrupt unsafe locking scenario:

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

 *** DEADLOCK ***

 ---

The above is the case when the unsafe lock is taken while holding
a lock taken in irq context. But when a lock is taken that also
grabs a unsafe lock, the call chain is shown:

 ---
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 <a.p.zijlstra@chello.nl>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
---
 kernel/lockdep.c |   70 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 70 insertions(+), 0 deletions(-)

diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index 0d2058d..bb77c030 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -490,6 +490,18 @@ void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
 	usage[i] = '\0';
 }
 
+static int __print_lock_name(struct lock_class *class)
+{
+	char str[KSYM_NAME_LEN];
+	const char *name;
+
+	name = class->name;
+	if (!name)
+		name = __get_key_name(class->key, str);
+
+	return printk("%s", name);
+}
+
 static void print_lock_name(struct lock_class *class)
 {
 	char str[KSYM_NAME_LEN], usage[LOCK_USAGE_CHARS];
@@ -1325,6 +1337,62 @@ print_shortest_lock_dependencies(struct lock_list *leaf,
 	return;
 }
 
+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 *safe_class = safe_entry->class;
+	struct lock_class *unsafe_class = unsafe_entry->class;
+	struct lock_class *middle_class = hlock_class(prev);
+
+	if (middle_class == safe_class)
+		middle_class = hlock_class(next);
+
+	/*
+	 * A direct locking problem where unsafe_class lock is taken
+	 * directly by safe_class lock, then all we need to show
+	 * is the deadlock scenario, as it is obvious that the
+	 * unsafe lock is taken under the safe lock.
+	 *
+	 * But if there is a chain instead, where the safe lock takes
+	 * an intermediate lock (middle_class) where this lock is
+	 * not the same as the safe lock, then the lock chain is
+	 * used to describe the problem. Otherwise we would need
+	 * to show a different CPU case for each link in the chain
+	 * from the safe_class lock to the unsafe_class lock.
+	 */
+	if (middle_class != unsafe_class) {
+		printk("Chain exists of:\n  ");
+		__print_lock_name(safe_class);
+		printk(" --> ");
+		__print_lock_name(middle_class);
+		printk(" --> ");
+		__print_lock_name(unsafe_class);
+		printk("\n\n");
+	}
+
+	printk(" Possible interrupt unsafe locking scenario:\n\n");
+	printk("       CPU0                    CPU1\n");
+	printk("       ----                    ----\n");
+	printk("  lock(");
+	__print_lock_name(unsafe_class);
+	printk(");\n");
+	printk("                               local_irq_disable();\n");
+	printk("                               lock(");
+	__print_lock_name(safe_class);
+	printk(");\n");
+	printk("                               lock(");
+	__print_lock_name(middle_class);
+	printk(");\n");
+	printk("  <Interrupt>\n");
+	printk("    lock(");
+	__print_lock_name(safe_class);
+	printk(");\n");
+	printk("\n *** DEADLOCK ***\n\n");
+}
+
 static int
 print_bad_irq_dependency(struct task_struct *curr,
 			 struct lock_list *prev_root,
@@ -1376,6 +1444,8 @@ 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);
+
 	lockdep_print_held_locks(curr);
 
 	printk("\nthe dependencies between %s-irq-safe lock", irqclass);
-- 
1.7.2.3