lockdep: annotate reclaim context (__GFP_NOFS)
authorNick Piggin <npiggin@suse.de>
Wed, 21 Jan 2009 07:12:39 +0000 (08:12 +0100)
committerIngo Molnar <mingo@elte.hu>
Sat, 14 Feb 2009 22:27:49 +0000 (23:27 +0100)
Here is another version, with the incremental patch rolled up, and
added reclaim context annotation to kswapd, and allocation tracing
to slab allocators (which may only ever reach the page allocator
in rare cases, so it is good to put annotations here too).

Haven't tested this version as such, but it should be getting closer
to merge worthy ;)

--
After noticing some code in mm/filemap.c accidentally perform a __GFP_FS
allocation when it should not have been, I thought it might be a good idea to
try to catch this kind of thing with lockdep.

I coded up a little idea that seems to work. Unfortunately the system has to
actually be in __GFP_FS page reclaim, then take the lock, before it will mark
it. But at least that might still be some orders of magnitude more common
(and more debuggable) than an actual deadlock condition, so we have some
improvement I hope (the concept is no less complete than discovery of a lock's
interrupt contexts).

I guess we could even do the same thing with __GFP_IO (normal reclaim), and
even GFP_NOIO locks too... but filesystems will have the most locks and fiddly
code paths, so let's start there and see how it goes.

It *seems* to work. I did a quick test.

=================================
[ INFO: inconsistent lock state ]
2.6.28-rc6-00007-ged31348-dirty #26
---------------------------------
inconsistent {in-reclaim-W} -> {ov-reclaim-W} usage.
modprobe/8526 [HC0[0]:SC0[0]:HE1:SE1] takes:
 (testlock){--..}, at: [<ffffffffa0020055>] brd_init+0x55/0x216 [brd]
{in-reclaim-W} state was registered at:
  [<ffffffff80267bdb>] __lock_acquire+0x75b/0x1a60
  [<ffffffff80268f71>] lock_acquire+0x91/0xc0
  [<ffffffff8070f0e1>] mutex_lock_nested+0xb1/0x310
  [<ffffffffa002002b>] brd_init+0x2b/0x216 [brd]
  [<ffffffff8020903b>] _stext+0x3b/0x170
  [<ffffffff80272ebf>] sys_init_module+0xaf/0x1e0
  [<ffffffff8020c3fb>] system_call_fastpath+0x16/0x1b
  [<ffffffffffffffff>] 0xffffffffffffffff
irq event stamp: 3929
hardirqs last  enabled at (3929): [<ffffffff8070f2b5>] mutex_lock_nested+0x285/0x310
hardirqs last disabled at (3928): [<ffffffff8070f089>] mutex_lock_nested+0x59/0x310
softirqs last  enabled at (3732): [<ffffffff8061f623>] sk_filter+0x83/0xe0
softirqs last disabled at (3730): [<ffffffff8061f5b6>] sk_filter+0x16/0xe0

other info that might help us debug this:
1 lock held by modprobe/8526:
 #0:  (testlock){--..}, at: [<ffffffffa0020055>] brd_init+0x55/0x216 [brd]

stack backtrace:
Pid: 8526, comm: modprobe Not tainted 2.6.28-rc6-00007-ged31348-dirty #26
Call Trace:
 [<ffffffff80265483>] print_usage_bug+0x193/0x1d0
 [<ffffffff80266530>] mark_lock+0xaf0/0xca0
 [<ffffffff80266735>] mark_held_locks+0x55/0xc0
 [<ffffffffa0020000>] ? brd_init+0x0/0x216 [brd]
 [<ffffffff802667ca>] trace_reclaim_fs+0x2a/0x60
 [<ffffffff80285005>] __alloc_pages_internal+0x475/0x580
 [<ffffffff8070f29e>] ? mutex_lock_nested+0x26e/0x310
 [<ffffffffa0020000>] ? brd_init+0x0/0x216 [brd]
 [<ffffffffa002006a>] brd_init+0x6a/0x216 [brd]
 [<ffffffffa0020000>] ? brd_init+0x0/0x216 [brd]
 [<ffffffff8020903b>] _stext+0x3b/0x170
 [<ffffffff8070f8b9>] ? mutex_unlock+0x9/0x10
 [<ffffffff8070f83d>] ? __mutex_unlock_slowpath+0x10d/0x180
 [<ffffffff802669ec>] ? trace_hardirqs_on_caller+0x12c/0x190
 [<ffffffff80272ebf>] sys_init_module+0xaf/0x1e0
 [<ffffffff8020c3fb>] system_call_fastpath+0x16/0x1b

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
include/linux/lockdep.h
include/linux/sched.h
kernel/lockdep.c
kernel/lockdep_internals.h
kernel/lockdep_proc.c
mm/page_alloc.c
mm/slab.c
mm/slob.c
mm/slub.c
mm/vmscan.c

index 23bf02f..cc97bdb 100644 (file)
@@ -27,12 +27,16 @@ enum lock_usage_bit
        LOCK_USED = 0,
        LOCK_USED_IN_HARDIRQ,
        LOCK_USED_IN_SOFTIRQ,
+       LOCK_USED_IN_RECLAIM_FS,
        LOCK_ENABLED_SOFTIRQS,
        LOCK_ENABLED_HARDIRQS,
+       LOCK_HELD_OVER_RECLAIM_FS,
        LOCK_USED_IN_HARDIRQ_READ,
        LOCK_USED_IN_SOFTIRQ_READ,
+       LOCK_USED_IN_RECLAIM_FS_READ,
        LOCK_ENABLED_SOFTIRQS_READ,
        LOCK_ENABLED_HARDIRQS_READ,
+       LOCK_HELD_OVER_RECLAIM_FS_READ,
        LOCK_USAGE_STATES
 };
 
@@ -42,16 +46,20 @@ enum lock_usage_bit
 #define LOCKF_USED                     (1 << LOCK_USED)
 #define LOCKF_USED_IN_HARDIRQ          (1 << LOCK_USED_IN_HARDIRQ)
 #define LOCKF_USED_IN_SOFTIRQ          (1 << LOCK_USED_IN_SOFTIRQ)
+#define LOCKF_USED_IN_RECLAIM_FS       (1 << LOCK_USED_IN_RECLAIM_FS)
 #define LOCKF_ENABLED_HARDIRQS         (1 << LOCK_ENABLED_HARDIRQS)
 #define LOCKF_ENABLED_SOFTIRQS         (1 << LOCK_ENABLED_SOFTIRQS)
+#define LOCKF_HELD_OVER_RECLAIM_FS     (1 << LOCK_HELD_OVER_RECLAIM_FS)
 
 #define LOCKF_ENABLED_IRQS (LOCKF_ENABLED_HARDIRQS | LOCKF_ENABLED_SOFTIRQS)
 #define LOCKF_USED_IN_IRQ (LOCKF_USED_IN_HARDIRQ | LOCKF_USED_IN_SOFTIRQ)
 
 #define LOCKF_USED_IN_HARDIRQ_READ     (1 << LOCK_USED_IN_HARDIRQ_READ)
 #define LOCKF_USED_IN_SOFTIRQ_READ     (1 << LOCK_USED_IN_SOFTIRQ_READ)
+#define LOCKF_USED_IN_RECLAIM_FS_READ  (1 << LOCK_USED_IN_RECLAIM_FS_READ)
 #define LOCKF_ENABLED_HARDIRQS_READ    (1 << LOCK_ENABLED_HARDIRQS_READ)
 #define LOCKF_ENABLED_SOFTIRQS_READ    (1 << LOCK_ENABLED_SOFTIRQS_READ)
+#define LOCKF_HELD_OVER_RECLAIM_FS_READ        (1 << LOCK_HELD_OVER_RECLAIM_FS_READ)
 
 #define LOCKF_ENABLED_IRQS_READ \
                (LOCKF_ENABLED_HARDIRQS_READ | LOCKF_ENABLED_SOFTIRQS_READ)
@@ -324,7 +332,11 @@ static inline void lock_set_subclass(struct lockdep_map *lock,
        lock_set_class(lock, lock->name, lock->key, subclass, ip);
 }
 
-# define INIT_LOCKDEP                          .lockdep_recursion = 0,
+extern void lockdep_set_current_reclaim_state(gfp_t gfp_mask);
+extern void lockdep_clear_current_reclaim_state(void);
+extern void lockdep_trace_alloc(gfp_t mask);
+
+# define INIT_LOCKDEP                          .lockdep_recursion = 0, .lockdep_reclaim_gfp = 0,
 
 #define lockdep_depth(tsk)     (debug_locks ? (tsk)->lockdep_depth : 0)
 
@@ -342,6 +354,9 @@ static inline void lockdep_on(void)
 # define lock_release(l, n, i)                 do { } while (0)
 # define lock_set_class(l, n, k, s, i)         do { } while (0)
 # define lock_set_subclass(l, s, i)            do { } while (0)
+# define lockdep_set_current_reclaim_state(g)  do { } while (0)
+# define lockdep_clear_current_reclaim_state() do { } while (0)
+# define lockdep_trace_alloc(g)                        do { } while (0)
 # define lockdep_init()                                do { } while (0)
 # define lockdep_info()                                do { } while (0)
 # define lockdep_init_map(lock, name, key, sub) \
index 4efb552..b00a77f 100644 (file)
@@ -1313,6 +1313,7 @@ struct task_struct {
        int lockdep_depth;
        unsigned int lockdep_recursion;
        struct held_lock held_locks[MAX_LOCK_DEPTH];
+       gfp_t lockdep_reclaim_gfp;
 #endif
 
 /* journalling filesystem info */
index 06b0c35..977f940 100644 (file)
@@ -310,12 +310,14 @@ EXPORT_SYMBOL(lockdep_on);
 #if VERBOSE
 # define HARDIRQ_VERBOSE       1
 # define SOFTIRQ_VERBOSE       1
+# define RECLAIM_VERBOSE       1
 #else
 # define HARDIRQ_VERBOSE       0
 # define SOFTIRQ_VERBOSE       0
+# define RECLAIM_VERBOSE       0
 #endif
 
-#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE
+#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE || RECLAIM_VERBOSE
 /*
  * Quick filtering for interesting events:
  */
@@ -454,6 +456,10 @@ static const char *usage_str[] =
        [LOCK_USED_IN_SOFTIRQ_READ] =   "in-softirq-R",
        [LOCK_ENABLED_SOFTIRQS_READ] =  "softirq-on-R",
        [LOCK_ENABLED_HARDIRQS_READ] =  "hardirq-on-R",
+       [LOCK_USED_IN_RECLAIM_FS] =     "in-reclaim-W",
+       [LOCK_USED_IN_RECLAIM_FS_READ] = "in-reclaim-R",
+       [LOCK_HELD_OVER_RECLAIM_FS] =   "ov-reclaim-W",
+       [LOCK_HELD_OVER_RECLAIM_FS_READ] = "ov-reclaim-R",
 };
 
 const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
@@ -462,9 +468,10 @@ const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
 }
 
 void
-get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3, char *c4)
+get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3,
+                                       char *c4, char *c5, char *c6)
 {
-       *c1 = '.', *c2 = '.', *c3 = '.', *c4 = '.';
+       *c1 = '.', *c2 = '.', *c3 = '.', *c4 = '.', *c5 = '.', *c6 = '.';
 
        if (class->usage_mask & LOCKF_USED_IN_HARDIRQ)
                *c1 = '+';
@@ -493,14 +500,29 @@ get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3, char *c4
                if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS_READ)
                        *c4 = '?';
        }
+
+       if (class->usage_mask & LOCKF_USED_IN_RECLAIM_FS)
+               *c5 = '+';
+       else
+               if (class->usage_mask & LOCKF_HELD_OVER_RECLAIM_FS)
+                       *c5 = '-';
+
+       if (class->usage_mask & LOCKF_HELD_OVER_RECLAIM_FS_READ)
+               *c6 = '-';
+       if (class->usage_mask & LOCKF_USED_IN_RECLAIM_FS_READ) {
+               *c6 = '+';
+               if (class->usage_mask & LOCKF_HELD_OVER_RECLAIM_FS_READ)
+                       *c6 = '?';
+       }
+
 }
 
 static void print_lock_name(struct lock_class *class)
 {
-       char str[KSYM_NAME_LEN], c1, c2, c3, c4;
+       char str[KSYM_NAME_LEN], c1, c2, c3, c4, c5, c6;
        const char *name;
 
-       get_usage_chars(class, &c1, &c2, &c3, &c4);
+       get_usage_chars(class, &c1, &c2, &c3, &c4, &c5, &c6);
 
        name = class->name;
        if (!name) {
@@ -513,7 +535,7 @@ static void print_lock_name(struct lock_class *class)
                if (class->subclass)
                        printk("/%d", class->subclass);
        }
-       printk("){%c%c%c%c}", c1, c2, c3, c4);
+       printk("){%c%c%c%c%c%c}", c1, c2, c3, c4, c5, c6);
 }
 
 static void print_lockdep_cache(struct lockdep_map *lock)
@@ -1306,6 +1328,26 @@ check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
                                        LOCK_ENABLED_SOFTIRQS, "soft"))
                return 0;
 
+       /*
+        * Prove that the new dependency does not connect a reclaim-fs-safe
+        * lock with a reclaim-fs-unsafe lock - to achieve this we search
+        * the backwards-subgraph starting at <prev>, and the
+        * forwards-subgraph starting at <next>:
+        */
+       if (!check_usage(curr, prev, next, LOCK_USED_IN_RECLAIM_FS,
+                                       LOCK_HELD_OVER_RECLAIM_FS, "reclaim-fs"))
+               return 0;
+
+       /*
+        * Prove that the new dependency does not connect a reclaim-fs-safe-read
+        * lock with a reclaim-fs-unsafe lock - to achieve this we search
+        * the backwards-subgraph starting at <prev>, and the
+        * forwards-subgraph starting at <next>:
+        */
+       if (!check_usage(curr, prev, next, LOCK_USED_IN_RECLAIM_FS_READ,
+                                       LOCK_HELD_OVER_RECLAIM_FS, "reclaim-fs-read"))
+               return 0;
+
        return 1;
 }
 
@@ -1949,6 +1991,14 @@ static int softirq_verbose(struct lock_class *class)
        return 0;
 }
 
+static int reclaim_verbose(struct lock_class *class)
+{
+#if RECLAIM_VERBOSE
+       return class_filter(class);
+#endif
+       return 0;
+}
+
 #define STRICT_READ_CHECKS     1
 
 static int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
@@ -2007,6 +2057,31 @@ static int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
                if (softirq_verbose(hlock_class(this)))
                        ret = 2;
                break;
+       case LOCK_USED_IN_RECLAIM_FS:
+               if (!valid_state(curr, this, new_bit, LOCK_HELD_OVER_RECLAIM_FS))
+                       return 0;
+               if (!valid_state(curr, this, new_bit,
+                                LOCK_HELD_OVER_RECLAIM_FS_READ))
+                       return 0;
+               /*
+                * just marked it reclaim-fs-safe, check that this lock
+                * took no reclaim-fs-unsafe lock in the past:
+                */
+               if (!check_usage_forwards(curr, this,
+                                         LOCK_HELD_OVER_RECLAIM_FS, "reclaim-fs"))
+                       return 0;
+#if STRICT_READ_CHECKS
+               /*
+                * just marked it reclaim-fs-safe, check that this lock
+                * took no reclaim-fs-unsafe-read lock in the past:
+                */
+               if (!check_usage_forwards(curr, this,
+                               LOCK_HELD_OVER_RECLAIM_FS_READ, "reclaim-fs-read"))
+                       return 0;
+#endif
+               if (reclaim_verbose(hlock_class(this)))
+                       ret = 2;
+               break;
        case LOCK_USED_IN_HARDIRQ_READ:
                if (!valid_state(curr, this, new_bit, LOCK_ENABLED_HARDIRQS))
                        return 0;
@@ -2033,6 +2108,19 @@ static int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
                if (softirq_verbose(hlock_class(this)))
                        ret = 2;
                break;
+       case LOCK_USED_IN_RECLAIM_FS_READ:
+               if (!valid_state(curr, this, new_bit, LOCK_HELD_OVER_RECLAIM_FS))
+                       return 0;
+               /*
+                * just marked it reclaim-fs-read-safe, check that this lock
+                * took no reclaim-fs-unsafe lock in the past:
+                */
+               if (!check_usage_forwards(curr, this,
+                                         LOCK_HELD_OVER_RECLAIM_FS, "reclaim-fs"))
+                       return 0;
+               if (reclaim_verbose(hlock_class(this)))
+                       ret = 2;
+               break;
        case LOCK_ENABLED_HARDIRQS:
                if (!valid_state(curr, this, new_bit, LOCK_USED_IN_HARDIRQ))
                        return 0;
@@ -2085,6 +2173,32 @@ static int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
                if (softirq_verbose(hlock_class(this)))
                        ret = 2;
                break;
+       case LOCK_HELD_OVER_RECLAIM_FS:
+               if (!valid_state(curr, this, new_bit, LOCK_USED_IN_RECLAIM_FS))
+                       return 0;
+               if (!valid_state(curr, this, new_bit,
+                                LOCK_USED_IN_RECLAIM_FS_READ))
+                       return 0;
+               /*
+                * just marked it reclaim-fs-unsafe, check that no reclaim-fs-safe
+                * lock in the system ever took it in the past:
+                */
+               if (!check_usage_backwards(curr, this,
+                                          LOCK_USED_IN_RECLAIM_FS, "reclaim-fs"))
+                       return 0;
+#if STRICT_READ_CHECKS
+               /*
+                * just marked it softirq-unsafe, check that no
+                * softirq-safe-read lock in the system ever took
+                * it in the past:
+                */
+               if (!check_usage_backwards(curr, this,
+                                  LOCK_USED_IN_RECLAIM_FS_READ, "reclaim-fs-read"))
+                       return 0;
+#endif
+               if (reclaim_verbose(hlock_class(this)))
+                       ret = 2;
+               break;
        case LOCK_ENABLED_HARDIRQS_READ:
                if (!valid_state(curr, this, new_bit, LOCK_USED_IN_HARDIRQ))
                        return 0;
@@ -2115,6 +2229,21 @@ static int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
                if (softirq_verbose(hlock_class(this)))
                        ret = 2;
                break;
+       case LOCK_HELD_OVER_RECLAIM_FS_READ:
+               if (!valid_state(curr, this, new_bit, LOCK_USED_IN_RECLAIM_FS))
+                       return 0;
+#if STRICT_READ_CHECKS
+               /*
+                * just marked it reclaim-fs-read-unsafe, check that no
+                * reclaim-fs-safe lock in the system ever took it in the past:
+                */
+               if (!check_usage_backwards(curr, this,
+                                          LOCK_USED_IN_RECLAIM_FS, "reclaim-fs"))
+                       return 0;
+#endif
+               if (reclaim_verbose(hlock_class(this)))
+                       ret = 2;
+               break;
        default:
                WARN_ON(1);
                break;
@@ -2123,11 +2252,17 @@ static int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
        return ret;
 }
 
+enum mark_type {
+       HARDIRQ,
+       SOFTIRQ,
+       RECLAIM_FS,
+};
+
 /*
  * Mark all held locks with a usage bit:
  */
 static int
-mark_held_locks(struct task_struct *curr, int hardirq)
+mark_held_locks(struct task_struct *curr, enum mark_type mark)
 {
        enum lock_usage_bit usage_bit;
        struct held_lock *hlock;
@@ -2136,17 +2271,32 @@ mark_held_locks(struct task_struct *curr, int hardirq)
        for (i = 0; i < curr->lockdep_depth; i++) {
                hlock = curr->held_locks + i;
 
-               if (hardirq) {
+               switch (mark) {
+               case HARDIRQ:
                        if (hlock->read)
                                usage_bit = LOCK_ENABLED_HARDIRQS_READ;
                        else
                                usage_bit = LOCK_ENABLED_HARDIRQS;
-               } else {
+                       break;
+
+               case SOFTIRQ:
                        if (hlock->read)
                                usage_bit = LOCK_ENABLED_SOFTIRQS_READ;
                        else
                                usage_bit = LOCK_ENABLED_SOFTIRQS;
+                       break;
+
+               case RECLAIM_FS:
+                       if (hlock->read)
+                               usage_bit = LOCK_HELD_OVER_RECLAIM_FS_READ;
+                       else
+                               usage_bit = LOCK_HELD_OVER_RECLAIM_FS;
+                       break;
+
+               default:
+                       BUG();
                }
+
                if (!mark_lock(curr, hlock, usage_bit))
                        return 0;
        }
@@ -2200,7 +2350,7 @@ void trace_hardirqs_on_caller(unsigned long ip)
         * We are going to turn hardirqs on, so set the
         * usage bit for all held locks:
         */
-       if (!mark_held_locks(curr, 1))
+       if (!mark_held_locks(curr, HARDIRQ))
                return;
        /*
         * If we have softirqs enabled, then set the usage
@@ -2208,7 +2358,7 @@ void trace_hardirqs_on_caller(unsigned long ip)
         * this bit from being set before)
         */
        if (curr->softirqs_enabled)
-               if (!mark_held_locks(curr, 0))
+               if (!mark_held_locks(curr, SOFTIRQ))
                        return;
 
        curr->hardirq_enable_ip = ip;
@@ -2288,7 +2438,7 @@ void trace_softirqs_on(unsigned long ip)
         * enabled too:
         */
        if (curr->hardirqs_enabled)
-               mark_held_locks(curr, 0);
+               mark_held_locks(curr, SOFTIRQ);
 }
 
 /*
@@ -2317,6 +2467,31 @@ void trace_softirqs_off(unsigned long ip)
                debug_atomic_inc(&redundant_softirqs_off);
 }
 
+void lockdep_trace_alloc(gfp_t gfp_mask)
+{
+       struct task_struct *curr = current;
+
+       if (unlikely(!debug_locks))
+               return;
+
+       /* no reclaim without waiting on it */
+       if (!(gfp_mask & __GFP_WAIT))
+               return;
+
+       /* this guy won't enter reclaim */
+       if ((curr->flags & PF_MEMALLOC) && !(gfp_mask & __GFP_NOMEMALLOC))
+               return;
+
+       /* We're only interested __GFP_FS allocations for now */
+       if (!(gfp_mask & __GFP_FS))
+               return;
+
+       if (DEBUG_LOCKS_WARN_ON(irqs_disabled()))
+               return;
+
+       mark_held_locks(curr, RECLAIM_FS);
+}
+
 static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
 {
        /*
@@ -2362,6 +2537,22 @@ static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
                }
        }
 
+       /*
+        * We reuse the irq context infrastructure more broadly as a general
+        * context checking code. This tests GFP_FS recursion (a lock taken
+        * during reclaim for a GFP_FS allocation is held over a GFP_FS
+        * allocation).
+        */
+       if (!hlock->trylock && (curr->lockdep_reclaim_gfp & __GFP_FS)) {
+               if (hlock->read) {
+                       if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS_READ))
+                                       return 0;
+               } else {
+                       if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS))
+                                       return 0;
+               }
+       }
+
        return 1;
 }
 
@@ -2453,6 +2644,10 @@ static int mark_lock(struct task_struct *curr, struct held_lock *this,
        case LOCK_ENABLED_SOFTIRQS:
        case LOCK_ENABLED_HARDIRQS_READ:
        case LOCK_ENABLED_SOFTIRQS_READ:
+       case LOCK_USED_IN_RECLAIM_FS:
+       case LOCK_USED_IN_RECLAIM_FS_READ:
+       case LOCK_HELD_OVER_RECLAIM_FS:
+       case LOCK_HELD_OVER_RECLAIM_FS_READ:
                ret = mark_lock_irq(curr, this, new_bit);
                if (!ret)
                        return 0;
@@ -2966,6 +3161,16 @@ void lock_release(struct lockdep_map *lock, int nested,
 }
 EXPORT_SYMBOL_GPL(lock_release);
 
+void lockdep_set_current_reclaim_state(gfp_t gfp_mask)
+{
+       current->lockdep_reclaim_gfp = gfp_mask;
+}
+
+void lockdep_clear_current_reclaim_state(void)
+{
+       current->lockdep_reclaim_gfp = 0;
+}
+
 #ifdef CONFIG_LOCK_STAT
 static int
 print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock,
index 56b1969..e887b78 100644 (file)
@@ -32,7 +32,8 @@ extern struct list_head all_lock_classes;
 extern struct lock_chain lock_chains[];
 
 extern void
-get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3, char *c4);
+get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3,
+                                       char *c4, char *c5, char *c6);
 
 extern const char * __get_key_name(struct lockdep_subclass_key *key, char *str);
 
index 13716b8..b84a1df 100644 (file)
@@ -84,7 +84,7 @@ static int l_show(struct seq_file *m, void *v)
 {
        struct lock_class *class = v;
        struct lock_list *entry;
-       char c1, c2, c3, c4;
+       char c1, c2, c3, c4, c5, c6;
 
        if (v == SEQ_START_TOKEN) {
                seq_printf(m, "all lock classes:\n");
@@ -100,8 +100,8 @@ static int l_show(struct seq_file *m, void *v)
        seq_printf(m, " BD:%5ld", lockdep_count_backward_deps(class));
 #endif
 
-       get_usage_chars(class, &c1, &c2, &c3, &c4);
-       seq_printf(m, " %c%c%c%c", c1, c2, c3, c4);
+       get_usage_chars(class, &c1, &c2, &c3, &c4, &c5, &c6);
+       seq_printf(m, " %c%c%c%c%c%c", c1, c2, c3, c4, c5, c6);
 
        seq_printf(m, ": ");
        print_name(m, class);
index 5675b30..22b15a4 100644 (file)
@@ -1479,6 +1479,8 @@ __alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
        unsigned long did_some_progress;
        unsigned long pages_reclaimed = 0;
 
+       lockdep_trace_alloc(gfp_mask);
+
        might_sleep_if(wait);
 
        if (should_fail_alloc_page(gfp_mask, order))
@@ -1578,12 +1580,15 @@ nofail_alloc:
         */
        cpuset_update_task_memory_state();
        p->flags |= PF_MEMALLOC;
+
+       lockdep_set_current_reclaim_state(gfp_mask);
        reclaim_state.reclaimed_slab = 0;
        p->reclaim_state = &reclaim_state;
 
        did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
 
        p->reclaim_state = NULL;
+       lockdep_clear_current_reclaim_state();
        p->flags &= ~PF_MEMALLOC;
 
        cond_resched();
index ddc41f3..6b61de8 100644 (file)
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -3318,6 +3318,8 @@ __cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid,
        unsigned long save_flags;
        void *ptr;
 
+       lockdep_trace_alloc(flags);
+
        if (slab_should_failslab(cachep, flags))
                return NULL;
 
@@ -3394,6 +3396,8 @@ __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller)
        unsigned long save_flags;
        void *objp;
 
+       lockdep_trace_alloc(flags);
+
        if (slab_should_failslab(cachep, flags))
                return NULL;
 
index bf7e8fc..1264799 100644 (file)
--- a/mm/slob.c
+++ b/mm/slob.c
@@ -464,6 +464,8 @@ void *__kmalloc_node(size_t size, gfp_t gfp, int node)
        unsigned int *m;
        int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
 
+       lockdep_trace_alloc(flags);
+
        if (size < PAGE_SIZE - align) {
                if (!size)
                        return ZERO_SIZE_PTR;
index bdc9abb..214eb20 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1596,6 +1596,7 @@ static __always_inline void *slab_alloc(struct kmem_cache *s,
        unsigned long flags;
        unsigned int objsize;
 
+       lockdep_trace_alloc(gfpflags);
        might_sleep_if(gfpflags & __GFP_WAIT);
 
        if (should_failslab(s->objsize, gfpflags))
index 9a27c44..303eb65 100644 (file)
@@ -1963,6 +1963,9 @@ static int kswapd(void *p)
        struct reclaim_state reclaim_state = {
                .reclaimed_slab = 0,
        };
+
+       lockdep_set_current_reclaim_state(GFP_KERNEL);
+
        node_to_cpumask_ptr(cpumask, pgdat->node_id);
 
        if (!cpumask_empty(cpumask))