[PATCH v3 12/30] locking/lockdep: Treat every lock dependency as in a new lock chain

[Yuyang Du <duyuyang@gmail.com>]

For a lock chain that has a lock on top of a trylock or a multiple of
consecutive trylocks, if they are in the same context, we check each
prev trylock -> next and the first prev non-trylock -> next lock
dependency, illustrated as:

    (The top is the latest lock.)

    Lock1
    Trylock2
    Trylock3
    Lock4
    ...

If the prev lock is not the direct previous lock to next (e.g., Trylock3
and Lock4), this dependency may not have a lock chain associated with
it. IOW, we may never make a lock chain, but the chain is actually
checked in such circumstances. This patch fixes this by treating each
such depdnency as if it is from a new lock chain. If the chain already
exists, then this is a chain hit and the check is actually not needed.

After this, it is guarantteed that each dependency has at least a lock
chain associated with it.

Signed-off-by: Yuyang Du <duyuyang@gmail.com>
---
 kernel/locking/lockdep.c | 223 +++++++++++++++++++++++------------------------
 1 file changed, 108 insertions(+), 115 deletions(-)

diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
index 5d19dc6..6f457ef 100644
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -1207,6 +1207,11 @@ static bool is_dynamic_key(const struct lock_class_key *key)
 }
 
 #ifdef CONFIG_PROVE_LOCKING
+static inline struct
+lock_chain *lookup_chain_cache_add(struct task_struct *curr,
+				   struct held_lock *hlock,
+				   u64 chain_key, int depth);
+
 /*
  * Allocate a lockdep entry. (assumes the graph_lock held, returns
  * with NULL on failure)
@@ -2432,87 +2437,6 @@ static inline void inc_chains(void)
 	return 2;
 }
 
-/*
- * Add the dependency to all directly-previous locks that are 'relevant'.
- * The ones that are relevant are (in increasing distance from curr):
- * all consecutive trylock entries and the final non-trylock entry - or
- * the end of this context's lock-chain - whichever comes first.
- */
-static int
-check_prevs_add(struct task_struct *curr, struct held_lock *next,
-		struct lock_chain *chain)
-{
-	struct lock_trace trace = { .nr_entries = 0 };
-	int depth = curr->lockdep_depth;
-	struct held_lock *hlock;
-
-	/*
-	 * Debugging checks.
-	 *
-	 * Depth must not be zero for a non-head lock:
-	 */
-	if (!depth)
-		goto out_bug;
-	/*
-	 * At least two relevant locks must exist for this
-	 * to be a head:
-	 */
-	if (curr->held_locks[depth].irq_context !=
-			curr->held_locks[depth-1].irq_context)
-		goto out_bug;
-
-	for (;;) {
-		int distance = curr->lockdep_depth - depth + 1;
-		hlock = curr->held_locks + depth - 1;
-
-		/*
-		 * Only non-recursive-read entries get new dependencies
-		 * added:
-		 */
-		if (hlock->read != 2 && hlock->check) {
-			int ret = check_prev_add(curr, hlock, next, distance,
-						 &trace, chain);
-			if (!ret)
-				return 0;
-
-			/*
-			 * Stop after the first non-trylock entry,
-			 * as non-trylock entries have added their
-			 * own direct dependencies already, so this
-			 * lock is connected to them indirectly:
-			 */
-			if (!hlock->trylock)
-				break;
-		}
-
-		depth--;
-		/*
-		 * End of lock-stack?
-		 */
-		if (!depth)
-			break;
-		/*
-		 * Stop the search if we cross into another context:
-		 */
-		if (curr->held_locks[depth].irq_context !=
-				curr->held_locks[depth-1].irq_context)
-			break;
-	}
-	return 1;
-out_bug:
-	if (!debug_locks_off_graph_unlock())
-		return 0;
-
-	/*
-	 * Clearly we all shouldn't be here, but since we made it we
-	 * can reliable say we messed up our state. See the above two
-	 * gotos for reasons why we could possibly end up here.
-	 */
-	WARN_ON(1);
-
-	return 0;
-}
-
 struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
 static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS);
 int nr_chain_hlocks;
@@ -2810,66 +2734,135 @@ static inline struct lock_chain *lookup_chain_cache(u64 chain_key)
 	return add_chain_cache(curr, hlock, chain_key, depth);
 }
 
-static int validate_chain(struct task_struct *curr, struct held_lock *hlock,
+/*
+ * Check whether last held lock:
+ *
+ * - is irq-safe, if this lock is irq-unsafe
+ * - is softirq-safe, if this lock is hardirq-unsafe
+ *
+ * And check whether the new lock's dependency graph could lead back to the
+ * previous lock:
+ *
+ * - within the current held-lock stack or
+ * - across our accumulated lock dependency graph
+ *
+ * any of these scenarios could lead to a deadlock.
+ */
+static int validate_chain(struct task_struct *curr, struct held_lock *next,
 			  u64 chain_key)
 {
 	struct lock_chain *chain;
+	struct lock_trace trace = { .nr_entries = 0 };
+	struct held_lock *hlock;
+	int depth = curr->lockdep_depth;
+
 	/*
 	 * Trylock needs to maintain the stack of held locks, but it
-	 * does not add new dependencies, because trylock can be done
-	 * in any order.
-	 *
+	 * does not add new dependencies unless it is taken, because
+	 * attempting to acquire a trylock does not block.
+	 */
+	if (next->trylock || !next->check)
+		return 1;
+
+	/*
+	 * Add the dependency to all previous locks that are 'relevant'. The
+	 * ones that are relevant are (in increasing distance from next lock
+	 * to acquire): all consecutive trylock entries and the final
+	 * non-trylock entry - or the end of this context's lock-chain
+	 * - whichever comes first.
+	 */
+chain_again:
+	hlock = curr->held_locks + depth - 1;
+
+	/*
 	 * We look up the chain_key and do the O(N^2) check and update of
-	 * the dependencies only if this is a new dependency chain.
-	 * (If lookup_chain_cache_add() return with 1 it acquires
-	 * graph_lock for us)
+	 * the dependencies only if this is a new dependency chain. (If
+	 * lookup_chain_cache_add() return with 1 it acquires graph_lock for
+	 * us.)
 	 */
-	if (!hlock->trylock && hlock->check &&
-	    (chain = lookup_chain_cache_add(curr, hlock, chain_key,
-					    curr->lockdep_depth))) {
-		/*
-		 * Check whether last held lock:
-		 *
-		 * - is irq-safe, if this lock is irq-unsafe
-		 * - is softirq-safe, if this lock is hardirq-unsafe
-		 *
-		 * And check whether the new lock's dependency graph
-		 * could lead back to the previous lock:
-		 *
-		 * - within the current held-lock stack
-		 * - across our accumulated lock dependency records
-		 *
-		 * any of these scenarios could lead to a deadlock.
-		 */
+	chain = lookup_chain_cache_add(curr, next, chain_key, depth);
+	if (depth == curr->lockdep_depth) {
+		int ret;
+
+		if (!chain)
+			return 1;
 		/*
 		 * The simple case: does the current hold the same lock
 		 * already?
 		 */
-		int ret = check_deadlock_current(curr, hlock);
+		ret = check_deadlock_current(curr, next);
 
 		if (!ret)
 			return 0;
 		/*
-		 * Add dependency only if this lock is not the head
-		 * of the chain, and if it's not a secondary read-lock:
+		 * Add dependency only if this lock is not the head of the
+		 * chain, and if it's not a second recursive-read lock. If
+		 * not, there is no need to check further.
 		 */
-		if (chain->depth > 1 && ret != 2) {
-			if (!check_prevs_add(curr, hlock, chain))
+		if (!(chain->depth > 1 && ret != 2))
+			goto out_unlock;
+	}
+
+	/*
+	 * Only non-recursive-read entries get new dependencies
+	 * added:
+	 */
+	if (chain) {
+		if (hlock->read != 2 && hlock->check) {
+			int distance = curr->lockdep_depth - depth + 1;
+
+			if (!check_prev_add(curr, hlock, next, distance,
+					    &trace, chain))
 				return 0;
 		}
 
 		graph_unlock();
-	} else {
-		/* after lookup_chain_cache_add(): */
-		if (unlikely(!debug_locks))
-			return 0;
 	}
 
+	/*
+	 * Stop after the first non-trylock entry, as non-trylock entries
+	 * have added their own direct dependencies already, so this lock is
+	 * connected to them indirectly:
+	 */
+	if (!hlock->trylock)
+		goto out;
+
+	depth--;
+	/*
+	 * End of lock-stack?
+	 */
+	if (!depth)
+		goto out;
+	/*
+	 * Stop the search if we cross into another context:
+	 */
+	if (curr->held_locks[depth].irq_context !=
+			curr->held_locks[depth-1].irq_context)
+		goto out;
+
+	/*
+	 * This is another direct dependency with a further previous lock
+	 * that is separated by a trylock. We compose a lock chain out of
+	 * this, then calculate the chain key, and look it up in the
+	 * lock_chains. If it exists the check is actually not needed.
+	 */
+	chain_key = iterate_chain_key(hlock->prev_chain_key,
+				      hlock_class(next) - lock_classes);
+
+	goto chain_again;
+
+out_unlock:
+	graph_unlock();
+out:
+	/* after lookup_chain_cache_add(): */
+	if (unlikely(!debug_locks))
+		return 0;
+
 	return 1;
 }
 #else
 static inline int validate_chain(struct task_struct *curr,
-				 struct held_lock *hlock,
+				 struct held_lock *next,
 				 u64 chain_key)
 {
 	return 1;
-- 
1.8.3.1