]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
locking/lockdep: Untangle xhlock history save/restore from task independence
authorPeter Zijlstra <peterz@infradead.org>
Tue, 29 Aug 2017 08:59:39 +0000 (10:59 +0200)
committerIngo Molnar <mingo@kernel.org>
Tue, 29 Aug 2017 13:14:38 +0000 (15:14 +0200)
Where XHLOCK_{SOFT,HARD} are save/restore points in the xhlocks[] to
ensure the temporal IRQ events don't interact with task state, the
XHLOCK_PROC is a fundament different beast that just happens to share
the interface.

The purpose of XHLOCK_PROC is to annotate independent execution inside
one task. For example workqueues, each work should appear to run in its
own 'pristine' 'task'.

Remove XHLOCK_PROC in favour of its own interface to avoid confusion.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Byungchul Park <byungchul.park@lge.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: david@fromorbit.com
Cc: johannes@sipsolutions.net
Cc: kernel-team@lge.com
Cc: oleg@redhat.com
Cc: tj@kernel.org
Link: http://lkml.kernel.org/r/20170829085939.ggmb6xiohw67micb@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
include/linux/irqflags.h
include/linux/lockdep.h
kernel/locking/lockdep.c
kernel/workqueue.c

index 9bc050bc81b2e6cf4b0caac12a096d30f72dcd27..5fdd93bb93009ccb3bb70404a59f9fd31939bcb4 100644 (file)
@@ -26,7 +26,7 @@
 # define trace_hardirq_enter()                 \
 do {                                           \
        current->hardirq_context++;             \
-       crossrelease_hist_start(XHLOCK_HARD, 0);\
+       crossrelease_hist_start(XHLOCK_HARD);   \
 } while (0)
 # define trace_hardirq_exit()                  \
 do {                                           \
@@ -36,7 +36,7 @@ do {                                          \
 # define lockdep_softirq_enter()               \
 do {                                           \
        current->softirq_context++;             \
-       crossrelease_hist_start(XHLOCK_SOFT, 0);\
+       crossrelease_hist_start(XHLOCK_SOFT);   \
 } while (0)
 # define lockdep_softirq_exit()                        \
 do {                                           \
index 78bb7133abed015d42008fc4aad5422de1c1580f..bfa8e0b0d6f1b26442fbe1f36be54b749dffb7bc 100644 (file)
@@ -551,7 +551,6 @@ struct pin_cookie { };
 enum xhlock_context_t {
        XHLOCK_HARD,
        XHLOCK_SOFT,
-       XHLOCK_PROC,
        XHLOCK_CTX_NR,
 };
 
@@ -580,8 +579,9 @@ extern void lock_commit_crosslock(struct lockdep_map *lock);
 #define STATIC_LOCKDEP_MAP_INIT(_name, _key) \
        { .name = (_name), .key = (void *)(_key), .cross = 0, }
 
-extern void crossrelease_hist_start(enum xhlock_context_t c, bool force);
+extern void crossrelease_hist_start(enum xhlock_context_t c);
 extern void crossrelease_hist_end(enum xhlock_context_t c);
+extern void lockdep_invariant_state(bool force);
 extern void lockdep_init_task(struct task_struct *task);
 extern void lockdep_free_task(struct task_struct *task);
 #else /* !CROSSRELEASE */
@@ -593,8 +593,9 @@ extern void lockdep_free_task(struct task_struct *task);
 #define STATIC_LOCKDEP_MAP_INIT(_name, _key) \
        { .name = (_name), .key = (void *)(_key), }
 
-static inline void crossrelease_hist_start(enum xhlock_context_t c, bool force) {}
+static inline void crossrelease_hist_start(enum xhlock_context_t c) {}
 static inline void crossrelease_hist_end(enum xhlock_context_t c) {}
+static inline void lockdep_invariant_state(bool force) {}
 static inline void lockdep_init_task(struct task_struct *task) {}
 static inline void lockdep_free_task(struct task_struct *task) {}
 #endif /* CROSSRELEASE */
index f73ca595b81ec1e0c8afd54b43b12fbbcd016579..44c8d0d17170af808ab0f11c4e556d49044a6c78 100644 (file)
@@ -4623,13 +4623,8 @@ asmlinkage __visible void lockdep_sys_exit(void)
        /*
         * The lock history for each syscall should be independent. So wipe the
         * slate clean on return to userspace.
-        *
-        * crossrelease_hist_end() works well here even when getting here
-        * without starting (i.e. just after forking), because it rolls back
-        * the index to point to the last entry, which is already invalid.
         */
-       crossrelease_hist_end(XHLOCK_PROC);
-       crossrelease_hist_start(XHLOCK_PROC, false);
+       lockdep_invariant_state(false);
 }
 
 void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
@@ -4723,19 +4718,47 @@ static inline void invalidate_xhlock(struct hist_lock *xhlock)
 }
 
 /*
- * Lock history stacks; we have 3 nested lock history stacks:
+ * Lock history stacks; we have 2 nested lock history stacks:
  *
  *   HARD(IRQ)
  *   SOFT(IRQ)
- *   PROC(ess)
  *
  * The thing is that once we complete a HARD/SOFT IRQ the future task locks
  * should not depend on any of the locks observed while running the IRQ.  So
  * what we do is rewind the history buffer and erase all our knowledge of that
  * temporal event.
- *
- * The PROCess one is special though; it is used to annotate independence
- * inside a task.
+ */
+
+void crossrelease_hist_start(enum xhlock_context_t c)
+{
+       struct task_struct *cur = current;
+
+       if (!cur->xhlocks)
+               return;
+
+       cur->xhlock_idx_hist[c] = cur->xhlock_idx;
+       cur->hist_id_save[c]    = cur->hist_id;
+}
+
+void crossrelease_hist_end(enum xhlock_context_t c)
+{
+       struct task_struct *cur = current;
+
+       if (cur->xhlocks) {
+               unsigned int idx = cur->xhlock_idx_hist[c];
+               struct hist_lock *h = &xhlock(idx);
+
+               cur->xhlock_idx = idx;
+
+               /* Check if the ring was overwritten. */
+               if (h->hist_id != cur->hist_id_save[c])
+                       invalidate_xhlock(h);
+       }
+}
+
+/*
+ * lockdep_invariant_state() is used to annotate independence inside a task, to
+ * make one task look like multiple independent 'tasks'.
  *
  * Take for instance workqueues; each work is independent of the last. The
  * completion of a future work does not depend on the completion of a past work
@@ -4758,40 +4781,14 @@ static inline void invalidate_xhlock(struct hist_lock *xhlock)
  * entry. Similarly, independence per-definition means it does not depend on
  * prior state.
  */
-void crossrelease_hist_start(enum xhlock_context_t c, bool force)
+void lockdep_invariant_state(bool force)
 {
-       struct task_struct *cur = current;
-
-       if (!cur->xhlocks)
-               return;
-
        /*
         * We call this at an invariant point, no current state, no history.
+        * Verify the former, enforce the latter.
         */
-       if (c == XHLOCK_PROC) {
-               /* verified the former, ensure the latter */
-               WARN_ON_ONCE(!force && cur->lockdep_depth);
-               invalidate_xhlock(&xhlock(cur->xhlock_idx));
-       }
-
-       cur->xhlock_idx_hist[c] = cur->xhlock_idx;
-       cur->hist_id_save[c]    = cur->hist_id;
-}
-
-void crossrelease_hist_end(enum xhlock_context_t c)
-{
-       struct task_struct *cur = current;
-
-       if (cur->xhlocks) {
-               unsigned int idx = cur->xhlock_idx_hist[c];
-               struct hist_lock *h = &xhlock(idx);
-
-               cur->xhlock_idx = idx;
-
-               /* Check if the ring was overwritten. */
-               if (h->hist_id != cur->hist_id_save[c])
-                       invalidate_xhlock(h);
-       }
+       WARN_ON_ONCE(!force && current->lockdep_depth);
+       invalidate_xhlock(&xhlock(current->xhlock_idx));
 }
 
 static int cross_lock(struct lockdep_map *lock)
index c0331891dec1d11ea1400f4837d1c7af0c0c8460..ab3c0dc8c7ed6e966fa073b9102be04967e748b9 100644 (file)
@@ -2094,8 +2094,8 @@ __acquires(&pool->lock)
        lock_map_acquire(&pwq->wq->lockdep_map);
        lock_map_acquire(&lockdep_map);
        /*
-        * Strictly speaking we should do start(PROC) without holding any
-        * locks, that is, before these two lock_map_acquire()'s.
+        * Strictly speaking we should mark the invariant state without holding
+        * any locks, that is, before these two lock_map_acquire()'s.
         *
         * However, that would result in:
         *
@@ -2107,14 +2107,14 @@ __acquires(&pool->lock)
         * Which would create W1->C->W1 dependencies, even though there is no
         * actual deadlock possible. There are two solutions, using a
         * read-recursive acquire on the work(queue) 'locks', but this will then
-        * hit the lockdep limitation on recursive locks, or simly discard
+        * hit the lockdep limitation on recursive locks, or simply discard
         * these locks.
         *
         * AFAICT there is no possible deadlock scenario between the
         * flush_work() and complete() primitives (except for single-threaded
         * workqueues), so hiding them isn't a problem.
         */
-       crossrelease_hist_start(XHLOCK_PROC, true);
+       lockdep_invariant_state(true);
        trace_workqueue_execute_start(work);
        worker->current_func(work);
        /*
@@ -2122,7 +2122,6 @@ __acquires(&pool->lock)
         * point will only record its address.
         */
        trace_workqueue_execute_end(work);
-       crossrelease_hist_end(XHLOCK_PROC);
        lock_map_release(&lockdep_map);
        lock_map_release(&pwq->wq->lockdep_map);