Header refactoring: unify and de-catchall mutex module

34 files changed, 304 insertions, 287 deletions

diff --git a/include/jemalloc/internal/arena_inlines_b.h b/include/jemalloc/internal/arena_inlines_b.h
index a105718..8db6e9a 100644
--- a/include/jemalloc/internal/arena_inlines_b.h
+++ b/include/jemalloc/internal/arena_inlines_b.h
@@ -2,6 +2,7 @@
 #define JEMALLOC_INTERNAL_ARENA_INLINES_B_H
 
 #include "jemalloc/internal/jemalloc_internal_types.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/size_classes.h"
 #include "jemalloc/internal/ticker.h"
 
diff --git a/include/jemalloc/internal/arena_structs_b.h b/include/jemalloc/internal/arena_structs_b.h
index 95680c0..f98f45c 100644
--- a/include/jemalloc/internal/arena_structs_b.h
+++ b/include/jemalloc/internal/arena_structs_b.h
@@ -4,6 +4,7 @@
 #include "jemalloc/internal/atomic.h"
 #include "jemalloc/internal/bitmap.h"
 #include "jemalloc/internal/jemalloc_internal_types.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/nstime.h"
 #include "jemalloc/internal/ql.h"
 #include "jemalloc/internal/size_classes.h"
diff --git a/include/jemalloc/internal/base_structs.h b/include/jemalloc/internal/base_structs.h
index 1d0a1f3..18e227b 100644
--- a/include/jemalloc/internal/base_structs.h
+++ b/include/jemalloc/internal/base_structs.h
@@ -2,6 +2,7 @@
 #define JEMALLOC_INTERNAL_BASE_STRUCTS_H
 
 #include "jemalloc/internal/jemalloc_internal_types.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/size_classes.h"
 
 /* Embedded at the beginning of every block of base-managed virtual memory. */
diff --git a/include/jemalloc/internal/extent_externs.h b/include/jemalloc/internal/extent_externs.h
index 7a5b38c..9d5daf5 100644
--- a/include/jemalloc/internal/extent_externs.h
+++ b/include/jemalloc/internal/extent_externs.h
@@ -1,8 +1,9 @@
 #ifndef JEMALLOC_INTERNAL_EXTENT_EXTERNS_H
 #define JEMALLOC_INTERNAL_EXTENT_EXTERNS_H
 
-#include "jemalloc/internal/rb.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/ph.h"
+#include "jemalloc/internal/rb.h"
 
 extern rtree_t			extents_rtree;
 extern const extent_hooks_t	extent_hooks_default;
diff --git a/include/jemalloc/internal/extent_inlines.h b/include/jemalloc/internal/extent_inlines.h
index 2ebd945..a99a635 100644
--- a/include/jemalloc/internal/extent_inlines.h
+++ b/include/jemalloc/internal/extent_inlines.h
@@ -1,6 +1,7 @@
 #ifndef JEMALLOC_INTERNAL_EXTENT_INLINES_H
 #define JEMALLOC_INTERNAL_EXTENT_INLINES_H
 
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/mutex_pool_inlines.h"
 #include "jemalloc/internal/pages.h"
 #include "jemalloc/internal/prng.h"
diff --git a/include/jemalloc/internal/extent_structs.h b/include/jemalloc/internal/extent_structs.h
index 62bae39..457891d 100644
--- a/include/jemalloc/internal/extent_structs.h
+++ b/include/jemalloc/internal/extent_structs.h
@@ -3,6 +3,7 @@
 
 #include "jemalloc/internal/atomic.h"
 #include "jemalloc/internal/bitmap.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/ql.h"
 #include "jemalloc/internal/rb.h"
 #include "jemalloc/internal/ph.h"
diff --git a/include/jemalloc/internal/jemalloc_internal_includes.h b/include/jemalloc/internal/jemalloc_internal_includes.h
index 2abc478..b1a6f17 100644
--- a/include/jemalloc/internal/jemalloc_internal_includes.h
+++ b/include/jemalloc/internal/jemalloc_internal_includes.h
@@ -40,7 +40,6 @@
 /* TYPES */
 /******************************************************************************/
 
-#include "jemalloc/internal/mutex_types.h"
 #include "jemalloc/internal/extent_types.h"
 #include "jemalloc/internal/extent_dss_types.h"
 #include "jemalloc/internal/base_types.h"
@@ -53,7 +52,6 @@
 /* STRUCTS */
 /******************************************************************************/
 
-#include "jemalloc/internal/mutex_structs.h"
 #include "jemalloc/internal/mutex_pool_structs.h"
 #include "jemalloc/internal/arena_structs_a.h"
 #include "jemalloc/internal/extent_structs.h"
@@ -70,7 +68,6 @@
 /******************************************************************************/
 
 #include "jemalloc/internal/jemalloc_internal_externs.h"
-#include "jemalloc/internal/mutex_externs.h"
 #include "jemalloc/internal/extent_externs.h"
 #include "jemalloc/internal/extent_dss_externs.h"
 #include "jemalloc/internal/extent_mmap_externs.h"
@@ -86,7 +83,6 @@
 /* INLINES */
 /******************************************************************************/
 
-#include "jemalloc/internal/mutex_inlines.h"
 #include "jemalloc/internal/mutex_pool_inlines.h"
 #include "jemalloc/internal/jemalloc_internal_inlines_a.h"
 #include "jemalloc/internal/rtree_inlines.h"
diff --git a/include/jemalloc/internal/mutex.h b/include/jemalloc/internal/mutex.h
new file mode 100644
index 0000000..6520c25
--- /dev/null
+++ b/include/jemalloc/internal/mutex.h
@@ -0,0 +1,248 @@
+#ifndef JEMALLOC_INTERNAL_MUTEX_H
+#define JEMALLOC_INTERNAL_MUTEX_H
+
+#include "jemalloc/internal/atomic.h"
+#include "jemalloc/internal/mutex_prof.h"
+#include "jemalloc/internal/tsd.h"
+#include "jemalloc/internal/witness.h"
+
+typedef enum {
+	/* Can only acquire one mutex of a given witness rank at a time. */
+	malloc_mutex_rank_exclusive,
+	/*
+	 * Can acquire multiple mutexes of the same witness rank, but in
+	 * address-ascending order only.
+	 */
+	malloc_mutex_address_ordered
+} malloc_mutex_lock_order_t;
+
+typedef struct malloc_mutex_s malloc_mutex_t;
+struct malloc_mutex_s {
+	union {
+		struct {
+			/*
+			 * prof_data is defined first to reduce cacheline
+			 * bouncing: the data is not touched by the mutex holder
+			 * during unlocking, while might be modified by
+			 * contenders.  Having it before the mutex itself could
+			 * avoid prefetching a modified cacheline (for the
+			 * unlocking thread).
+			 */
+			mutex_prof_data_t	prof_data;
+#ifdef _WIN32
+#  if _WIN32_WINNT >= 0x0600
+			SRWLOCK         	lock;
+#  else
+			CRITICAL_SECTION	lock;
+#  endif
+#elif (defined(JEMALLOC_OS_UNFAIR_LOCK))
+			os_unfair_lock		lock;
+#elif (defined(JEMALLOC_OSSPIN))
+			OSSpinLock		lock;
+#elif (defined(JEMALLOC_MUTEX_INIT_CB))
+			pthread_mutex_t		lock;
+			malloc_mutex_t		*postponed_next;
+#else
+			pthread_mutex_t		lock;
+#endif
+		};
+		/*
+		 * We only touch witness when configured w/ debug.  However we
+		 * keep the field in a union when !debug so that we don't have
+		 * to pollute the code base with #ifdefs, while avoid paying the
+		 * memory cost.
+		 */
+#if !defined(JEMALLOC_DEBUG)
+		witness_t			witness;
+		malloc_mutex_lock_order_t	lock_order;
+#endif
+	};
+
+#if defined(JEMALLOC_DEBUG)
+	witness_t			witness;
+	malloc_mutex_lock_order_t	lock_order;
+#endif
+};
+
+/*
+ * Based on benchmark results, a fixed spin with this amount of retries works
+ * well for our critical sections.
+ */
+#define MALLOC_MUTEX_MAX_SPIN 250
+
+#ifdef _WIN32
+#  if _WIN32_WINNT >= 0x0600
+#    define MALLOC_MUTEX_LOCK(m)    AcquireSRWLockExclusive(&(m)->lock)
+#    define MALLOC_MUTEX_UNLOCK(m)  ReleaseSRWLockExclusive(&(m)->lock)
+#    define MALLOC_MUTEX_TRYLOCK(m) (!TryAcquireSRWLockExclusive(&(m)->lock))
+#  else
+#    define MALLOC_MUTEX_LOCK(m)    EnterCriticalSection(&(m)->lock)
+#    define MALLOC_MUTEX_UNLOCK(m)  LeaveCriticalSection(&(m)->lock)
+#    define MALLOC_MUTEX_TRYLOCK(m) (!TryEnterCriticalSection(&(m)->lock))
+#  endif
+#elif (defined(JEMALLOC_OS_UNFAIR_LOCK))
+#    define MALLOC_MUTEX_LOCK(m)    os_unfair_lock_lock(&(m)->lock)
+#    define MALLOC_MUTEX_UNLOCK(m)  os_unfair_lock_unlock(&(m)->lock)
+#    define MALLOC_MUTEX_TRYLOCK(m) (!os_unfair_lock_trylock(&(m)->lock))
+#elif (defined(JEMALLOC_OSSPIN))
+#    define MALLOC_MUTEX_LOCK(m)    OSSpinLockLock(&(m)->lock)
+#    define MALLOC_MUTEX_UNLOCK(m)  OSSpinLockUnlock(&(m)->lock)
+#    define MALLOC_MUTEX_TRYLOCK(m) (!OSSpinLockTry(&(m)->lock))
+#else
+#    define MALLOC_MUTEX_LOCK(m)    pthread_mutex_lock(&(m)->lock)
+#    define MALLOC_MUTEX_UNLOCK(m)  pthread_mutex_unlock(&(m)->lock)
+#    define MALLOC_MUTEX_TRYLOCK(m) (pthread_mutex_trylock(&(m)->lock) != 0)
+#endif
+
+#define LOCK_PROF_DATA_INITIALIZER					\
+    {NSTIME_ZERO_INITIALIZER, NSTIME_ZERO_INITIALIZER, 0, 0, 0,		\
+	    ATOMIC_INIT(0), 0, NULL, 0}
+
+#ifdef _WIN32
+#  define MALLOC_MUTEX_INITIALIZER
+#elif (defined(JEMALLOC_OS_UNFAIR_LOCK))
+#  define MALLOC_MUTEX_INITIALIZER					\
+     {{{LOCK_PROF_DATA_INITIALIZER, OS_UNFAIR_LOCK_INIT}},		\
+      WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
+#elif (defined(JEMALLOC_OSSPIN))
+#  define MALLOC_MUTEX_INITIALIZER					\
+     {{{LOCK_PROF_DATA_INITIALIZER, 0}},				\
+      WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
+#elif (defined(JEMALLOC_MUTEX_INIT_CB))
+#  define MALLOC_MUTEX_INITIALIZER					\
+     {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, NULL}},	\
+      WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
+#else
+#    define MALLOC_MUTEX_TYPE PTHREAD_MUTEX_DEFAULT
+#    define MALLOC_MUTEX_INITIALIZER					\
+       {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER}},	\
+        WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
+#endif
+
+#ifdef JEMALLOC_LAZY_LOCK
+extern bool isthreaded;
+#else
+#  undef isthreaded /* Undo private_namespace.h definition. */
+#  define isthreaded true
+#endif
+
+bool malloc_mutex_init(malloc_mutex_t *mutex, const char *name,
+    witness_rank_t rank, malloc_mutex_lock_order_t lock_order);
+void malloc_mutex_prefork(tsdn_t *tsdn, malloc_mutex_t *mutex);
+void malloc_mutex_postfork_parent(tsdn_t *tsdn, malloc_mutex_t *mutex);
+void malloc_mutex_postfork_child(tsdn_t *tsdn, malloc_mutex_t *mutex);
+bool malloc_mutex_boot(void);
+void malloc_mutex_prof_data_reset(tsdn_t *tsdn, malloc_mutex_t *mutex);
+
+void malloc_mutex_lock_slow(malloc_mutex_t *mutex);
+
+static inline void
+malloc_mutex_lock_final(malloc_mutex_t *mutex) {
+	MALLOC_MUTEX_LOCK(mutex);
+}
+
+static inline bool
+malloc_mutex_trylock_final(malloc_mutex_t *mutex) {
+	return MALLOC_MUTEX_TRYLOCK(mutex);
+}
+
+static inline void
+mutex_owner_stats_update(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	if (config_stats) {
+		mutex_prof_data_t *data = &mutex->prof_data;
+		data->n_lock_ops++;
+		if (data->prev_owner != tsdn) {
+			data->prev_owner = tsdn;
+			data->n_owner_switches++;
+		}
+	}
+}
+
+/* Trylock: return false if the lock is successfully acquired. */
+static inline bool
+malloc_mutex_trylock(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	witness_assert_not_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
+	if (isthreaded) {
+		if (malloc_mutex_trylock_final(mutex)) {
+			return true;
+		}
+		mutex_owner_stats_update(tsdn, mutex);
+	}
+	witness_lock(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
+
+	return false;
+}
+
+/* Aggregate lock prof data. */
+static inline void
+malloc_mutex_prof_merge(mutex_prof_data_t *sum, mutex_prof_data_t *data) {
+	nstime_add(&sum->tot_wait_time, &data->tot_wait_time);
+	if (nstime_compare(&sum->max_wait_time, &data->max_wait_time) < 0) {
+		nstime_copy(&sum->max_wait_time, &data->max_wait_time);
+	}
+
+	sum->n_wait_times += data->n_wait_times;
+	sum->n_spin_acquired += data->n_spin_acquired;
+
+	if (sum->max_n_thds < data->max_n_thds) {
+		sum->max_n_thds = data->max_n_thds;
+	}
+	uint32_t cur_n_waiting_thds = atomic_load_u32(&sum->n_waiting_thds,
+	    ATOMIC_RELAXED);
+	uint32_t new_n_waiting_thds = cur_n_waiting_thds + atomic_load_u32(
+	    &data->n_waiting_thds, ATOMIC_RELAXED);
+	atomic_store_u32(&sum->n_waiting_thds, new_n_waiting_thds,
+	    ATOMIC_RELAXED);
+	sum->n_owner_switches += data->n_owner_switches;
+	sum->n_lock_ops += data->n_lock_ops;
+}
+
+static inline void
+malloc_mutex_lock(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	witness_assert_not_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
+	if (isthreaded) {
+		if (malloc_mutex_trylock_final(mutex)) {
+			malloc_mutex_lock_slow(mutex);
+		}
+		mutex_owner_stats_update(tsdn, mutex);
+	}
+	witness_lock(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
+}
+
+static inline void
+malloc_mutex_unlock(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	witness_unlock(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
+	if (isthreaded) {
+		MALLOC_MUTEX_UNLOCK(mutex);
+	}
+}
+
+static inline void
+malloc_mutex_assert_owner(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	witness_assert_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
+}
+
+static inline void
+malloc_mutex_assert_not_owner(tsdn_t *tsdn, malloc_mutex_t *mutex) {
+	witness_assert_not_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
+}
+
+/* Copy the prof data from mutex for processing. */
+static inline void
+malloc_mutex_prof_read(tsdn_t *tsdn, mutex_prof_data_t *data,
+    malloc_mutex_t *mutex) {
+	mutex_prof_data_t *source = &mutex->prof_data;
+	/* Can only read holding the mutex. */
+	malloc_mutex_assert_owner(tsdn, mutex);
+
+	/*
+	 * Not *really* allowed (we shouldn't be doing non-atomic loads of
+	 * atomic data), but the mutex protection makes this safe, and writing
+	 * a member-for-member copy is tedious for this situation.
+	 */
+	*data = *source;
+	/* n_wait_thds is not reported (modified w/o locking). */
+	atomic_store_u32(&data->n_waiting_thds, 0, ATOMIC_RELAXED);
+}
+
+#endif /* JEMALLOC_INTERNAL_MUTEX_H */
diff --git a/include/jemalloc/internal/mutex_externs.h b/include/jemalloc/internal/mutex_externs.h
deleted file mode 100644
index d0139f2..0000000
--- a/include/jemalloc/internal/mutex_externs.h
+++ /dev/null
@@ -1,21 +0,0 @@
-#ifndef JEMALLOC_INTERNAL_MUTEX_EXTERNS_H
-#define JEMALLOC_INTERNAL_MUTEX_EXTERNS_H
-
-#include "jemalloc/internal/tsd_types.h"
-
-#ifdef JEMALLOC_LAZY_LOCK
-extern bool isthreaded;
-#else
-#  undef isthreaded /* Undo private_namespace.h definition. */
-#  define isthreaded true
-#endif
-
-bool malloc_mutex_init(malloc_mutex_t *mutex, const char *name,
-    witness_rank_t rank, malloc_mutex_lock_order_t lock_order);
-void malloc_mutex_prefork(tsdn_t *tsdn, malloc_mutex_t *mutex);
-void malloc_mutex_postfork_parent(tsdn_t *tsdn, malloc_mutex_t *mutex);
-void malloc_mutex_postfork_child(tsdn_t *tsdn, malloc_mutex_t *mutex);
-bool malloc_mutex_boot(void);
-void malloc_mutex_prof_data_reset(tsdn_t *tsdn, malloc_mutex_t *mutex);
-
-#endif /* JEMALLOC_INTERNAL_MUTEX_EXTERNS_H */
diff --git a/include/jemalloc/internal/mutex_inlines.h b/include/jemalloc/internal/mutex_inlines.h
deleted file mode 100644
index b86a4ad..0000000
--- a/include/jemalloc/internal/mutex_inlines.h
+++ /dev/null
@@ -1,118 +0,0 @@
-#ifndef JEMALLOC_INTERNAL_MUTEX_INLINES_H
-#define JEMALLOC_INTERNAL_MUTEX_INLINES_H
-
-#include "jemalloc/internal/nstime.h"
-#include "jemalloc/internal/tsd_types.h"
-
-void	malloc_mutex_lock_slow(malloc_mutex_t *mutex);
-
-static inline void
-malloc_mutex_lock_final(malloc_mutex_t *mutex) {
-	MALLOC_MUTEX_LOCK(mutex);
-}
-
-static inline bool
-malloc_mutex_trylock_final(malloc_mutex_t *mutex) {
-	return MALLOC_MUTEX_TRYLOCK(mutex);
-}
-
-static inline void
-mutex_owner_stats_update(tsdn_t *tsdn, malloc_mutex_t *mutex) {
-	if (config_stats) {
-		mutex_prof_data_t *data = &mutex->prof_data;
-		data->n_lock_ops++;
-		if (data->prev_owner != tsdn) {
-			data->prev_owner = tsdn;
-			data->n_owner_switches++;
-		}
-	}
-}
-
-/* Trylock: return false if the lock is successfully acquired. */
-static inline bool
-malloc_mutex_trylock(tsdn_t *tsdn, malloc_mutex_t *mutex) {
-	witness_assert_not_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
-	if (isthreaded) {
-		if (malloc_mutex_trylock_final(mutex)) {
-			return true;
-		}
-		mutex_owner_stats_update(tsdn, mutex);
-	}
-	witness_lock(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
-
-	return false;
-}
-
-/* Aggregate lock prof data. */
-static inline void
-malloc_mutex_prof_merge(mutex_prof_data_t *sum, mutex_prof_data_t *data) {
-	nstime_add(&sum->tot_wait_time, &data->tot_wait_time);
-	if (nstime_compare(&sum->max_wait_time, &data->max_wait_time) < 0) {
-		nstime_copy(&sum->max_wait_time, &data->max_wait_time);
-	}
-
-	sum->n_wait_times += data->n_wait_times;
-	sum->n_spin_acquired += data->n_spin_acquired;
-
-	if (sum->max_n_thds < data->max_n_thds) {
-		sum->max_n_thds = data->max_n_thds;
-	}
-	uint32_t cur_n_waiting_thds = atomic_load_u32(&sum->n_waiting_thds,
-	    ATOMIC_RELAXED);
-	uint32_t new_n_waiting_thds = cur_n_waiting_thds + atomic_load_u32(
-	    &data->n_waiting_thds, ATOMIC_RELAXED);
-	atomic_store_u32(&sum->n_waiting_thds, new_n_waiting_thds,
-	    ATOMIC_RELAXED);
-	sum->n_owner_switches += data->n_owner_switches;
-	sum->n_lock_ops += data->n_lock_ops;
-}
-
-static inline void
-malloc_mutex_lock(tsdn_t *tsdn, malloc_mutex_t *mutex) {
-	witness_assert_not_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
-	if (isthreaded) {
-		if (malloc_mutex_trylock_final(mutex)) {
-			malloc_mutex_lock_slow(mutex);
-		}
-		mutex_owner_stats_update(tsdn, mutex);
-	}
-	witness_lock(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
-}
-
-static inline void
-malloc_mutex_unlock(tsdn_t *tsdn, malloc_mutex_t *mutex) {
-	witness_unlock(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
-	if (isthreaded) {
-		MALLOC_MUTEX_UNLOCK(mutex);
-	}
-}
-
-static inline void
-malloc_mutex_assert_owner(tsdn_t *tsdn, malloc_mutex_t *mutex) {
-	witness_assert_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
-}
-
-static inline void
-malloc_mutex_assert_not_owner(tsdn_t *tsdn, malloc_mutex_t *mutex) {
-	witness_assert_not_owner(tsdn_witness_tsdp_get(tsdn), &mutex->witness);
-}
-
-/* Copy the prof data from mutex for processing. */
-static inline void
-malloc_mutex_prof_read(tsdn_t *tsdn, mutex_prof_data_t *data,
-    malloc_mutex_t *mutex) {
-	mutex_prof_data_t *source = &mutex->prof_data;
-	/* Can only read holding the mutex. */
-	malloc_mutex_assert_owner(tsdn, mutex);
-
-	/*
-	 * Not *really* allowed (we shouldn't be doing non-atomic loads of
-	 * atomic data), but the mutex protection makes this safe, and writing
-	 * a member-for-member copy is tedious for this situation.
-	 */
-	*data = *source;
-	/* n_wait_thds is not reported (modified w/o locking). */
-	atomic_store_u32(&data->n_waiting_thds, 0, ATOMIC_RELAXED);
-}
-
-#endif /* JEMALLOC_INTERNAL_MUTEX_INLINES_H */
diff --git a/include/jemalloc/internal/mutex_pool_inlines.h b/include/jemalloc/internal/mutex_pool_inlines.h
index bc257ea..19b5ab4 100644
--- a/include/jemalloc/internal/mutex_pool_inlines.h
+++ b/include/jemalloc/internal/mutex_pool_inlines.h
@@ -2,7 +2,7 @@
 #define JEMALLOC_INTERNAL_MUTEX_POOL_INLINES_H
 
 #include "jemalloc/internal/hash.h"
-#include "jemalloc/internal/mutex_inlines.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/mutex_pool_structs.h"
 #include "jemalloc/internal/witness.h"
 
diff --git a/include/jemalloc/internal/mutex_pool_structs.h b/include/jemalloc/internal/mutex_pool_structs.h
index a662166..b32fb5a 100644
--- a/include/jemalloc/internal/mutex_pool_structs.h
+++ b/include/jemalloc/internal/mutex_pool_structs.h
@@ -1,6 +1,8 @@
 #ifndef JEMALLOC_INTERNAL_MUTEX_POOL_STRUCTS_H
 #define JEMALLOC_INTERNAL_MUTEX_POOL_STRUCTS_H
 
+#include "jemalloc/internal/mutex.h"
+
 /* This file really combines "structs" and "types", but only transitionally. */
 
 /* We do mod reductions by this value, so it should be kept a power of 2. */
diff --git a/include/jemalloc/internal/mutex_structs.h b/include/jemalloc/internal/mutex_structs.h
deleted file mode 100644
index c1b6552..0000000
--- a/include/jemalloc/internal/mutex_structs.h
+++ /dev/null
@@ -1,55 +0,0 @@
-#ifndef JEMALLOC_INTERNAL_MUTEX_STRUCTS_H
-#define JEMALLOC_INTERNAL_MUTEX_STRUCTS_H
-
-#include "jemalloc/internal/atomic.h"
-#include "jemalloc/internal/mutex_prof.h"
-#include "jemalloc/internal/witness.h"
-
-struct malloc_mutex_s {
-	union {
-		struct {
-			/*
-			 * prof_data is defined first to reduce cacheline
-			 * bouncing: the data is not touched by the mutex holder
-			 * during unlocking, while might be modified by
-			 * contenders.  Having it before the mutex itself could
-			 * avoid prefetching a modified cacheline (for the
-			 * unlocking thread).
-			 */
-			mutex_prof_data_t	prof_data;
-#ifdef _WIN32
-#  if _WIN32_WINNT >= 0x0600
-			SRWLOCK         	lock;
-#  else
-			CRITICAL_SECTION	lock;
-#  endif
-#elif (defined(JEMALLOC_OS_UNFAIR_LOCK))
-			os_unfair_lock		lock;
-#elif (defined(JEMALLOC_OSSPIN))
-			OSSpinLock		lock;
-#elif (defined(JEMALLOC_MUTEX_INIT_CB))
-			pthread_mutex_t		lock;
-			malloc_mutex_t		*postponed_next;
-#else
-			pthread_mutex_t		lock;
-#endif
-		};
-		/*
-		 * We only touch witness when configured w/ debug.  However we
-		 * keep the field in a union when !debug so that we don't have
-		 * to pollute the code base with #ifdefs, while avoid paying the
-		 * memory cost.
-		 */
-#if !defined(JEMALLOC_DEBUG)
-		witness_t			witness;
-		malloc_mutex_lock_order_t	lock_order;
-#endif
-	};
-
-#if defined(JEMALLOC_DEBUG)
-	witness_t			witness;
-	malloc_mutex_lock_order_t	lock_order;
-#endif
-};
-
-#endif /* JEMALLOC_INTERNAL_MUTEX_STRUCTS_H */
diff --git a/include/jemalloc/internal/mutex_types.h b/include/jemalloc/internal/mutex_types.h
deleted file mode 100644
index 65a9938..0000000
--- a/include/jemalloc/internal/mutex_types.h
+++ /dev/null
@@ -1,71 +0,0 @@
-#ifndef JEMALLOC_INTERNAL_MUTEX_TYPES_H
-#define JEMALLOC_INTERNAL_MUTEX_TYPES_H
-
-typedef struct malloc_mutex_s malloc_mutex_t;
-
-typedef enum {
-	/* Can only acquire one mutex of a given witness rank at a time. */
-	malloc_mutex_rank_exclusive,
-	/*
-	 * Can acquire multiple mutexes of the same witness rank, but in
-	 * address-ascending order only.
-	 */
-	malloc_mutex_address_ordered
-} malloc_mutex_lock_order_t;
-
-/*
- * Based on benchmark results, a fixed spin with this amount of retries works
- * well for our critical sections.
- */
-#define MALLOC_MUTEX_MAX_SPIN 250
-
-#ifdef _WIN32
-#  if _WIN32_WINNT >= 0x0600
-#    define MALLOC_MUTEX_LOCK(m)    AcquireSRWLockExclusive(&(m)->lock)
-#    define MALLOC_MUTEX_UNLOCK(m)  ReleaseSRWLockExclusive(&(m)->lock)
-#    define MALLOC_MUTEX_TRYLOCK(m) (!TryAcquireSRWLockExclusive(&(m)->lock))
-#  else
-#    define MALLOC_MUTEX_LOCK(m)    EnterCriticalSection(&(m)->lock)
-#    define MALLOC_MUTEX_UNLOCK(m)  LeaveCriticalSection(&(m)->lock)
-#    define MALLOC_MUTEX_TRYLOCK(m) (!TryEnterCriticalSection(&(m)->lock))
-#  endif
-#elif (defined(JEMALLOC_OS_UNFAIR_LOCK))
-#    define MALLOC_MUTEX_LOCK(m)    os_unfair_lock_lock(&(m)->lock)
-#    define MALLOC_MUTEX_UNLOCK(m)  os_unfair_lock_unlock(&(m)->lock)
-#    define MALLOC_MUTEX_TRYLOCK(m) (!os_unfair_lock_trylock(&(m)->lock))
-#elif (defined(JEMALLOC_OSSPIN))
-#    define MALLOC_MUTEX_LOCK(m)    OSSpinLockLock(&(m)->lock)
-#    define MALLOC_MUTEX_UNLOCK(m)  OSSpinLockUnlock(&(m)->lock)
-#    define MALLOC_MUTEX_TRYLOCK(m) (!OSSpinLockTry(&(m)->lock))
-#else
-#    define MALLOC_MUTEX_LOCK(m)    pthread_mutex_lock(&(m)->lock)
-#    define MALLOC_MUTEX_UNLOCK(m)  pthread_mutex_unlock(&(m)->lock)
-#    define MALLOC_MUTEX_TRYLOCK(m) (pthread_mutex_trylock(&(m)->lock) != 0)
-#endif
-
-#define LOCK_PROF_DATA_INITIALIZER					\
-    {NSTIME_ZERO_INITIALIZER, NSTIME_ZERO_INITIALIZER, 0, 0, 0,		\
-	    ATOMIC_INIT(0), 0, NULL, 0}
-
-#ifdef _WIN32
-#  define MALLOC_MUTEX_INITIALIZER
-#elif (defined(JEMALLOC_OS_UNFAIR_LOCK))
-#  define MALLOC_MUTEX_INITIALIZER					\
-     {{{LOCK_PROF_DATA_INITIALIZER, OS_UNFAIR_LOCK_INIT}},		\
-      WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
-#elif (defined(JEMALLOC_OSSPIN))
-#  define MALLOC_MUTEX_INITIALIZER					\
-     {{{LOCK_PROF_DATA_INITIALIZER, 0}},				\
-      WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
-#elif (defined(JEMALLOC_MUTEX_INIT_CB))
-#  define MALLOC_MUTEX_INITIALIZER					\
-     {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, NULL}},	\
-      WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
-#else
-#    define MALLOC_MUTEX_TYPE PTHREAD_MUTEX_DEFAULT
-#    define MALLOC_MUTEX_INITIALIZER					\
-       {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER}},	\
-        WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
-#endif
-
-#endif /* JEMALLOC_INTERNAL_MUTEX_TYPES_H */
diff --git a/include/jemalloc/internal/prof_externs.h b/include/jemalloc/internal/prof_externs.h
index 2891b8b..0434869 100644
--- a/include/jemalloc/internal/prof_externs.h
+++ b/include/jemalloc/internal/prof_externs.h
@@ -1,6 +1,8 @@
 #ifndef JEMALLOC_INTERNAL_PROF_EXTERNS_H
 #define JEMALLOC_INTERNAL_PROF_EXTERNS_H
 
+#include "jemalloc/internal/mutex.h"
+
 extern malloc_mutex_t	bt2gctx_mtx;
 
 extern bool	opt_prof;
diff --git a/include/jemalloc/internal/prof_inlines_a.h b/include/jemalloc/internal/prof_inlines_a.h
index 6203cbd..eda6839 100644
--- a/include/jemalloc/internal/prof_inlines_a.h
+++ b/include/jemalloc/internal/prof_inlines_a.h
@@ -1,6 +1,8 @@
 #ifndef JEMALLOC_INTERNAL_PROF_INLINES_A_H
 #define JEMALLOC_INTERNAL_PROF_INLINES_A_H
 
+#include "jemalloc/internal/mutex.h"
+
 static inline bool
 prof_accum_add(tsdn_t *tsdn, prof_accum_t *prof_accum, uint64_t accumbytes) {
 	cassert(config_prof);
diff --git a/include/jemalloc/internal/prof_structs.h b/include/jemalloc/internal/prof_structs.h
index a26a042..0d58ae1 100644
--- a/include/jemalloc/internal/prof_structs.h
+++ b/include/jemalloc/internal/prof_structs.h
@@ -2,6 +2,7 @@
 #define JEMALLOC_INTERNAL_PROF_STRUCTS_H
 
 #include "jemalloc/internal/ckh.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/prng.h"
 #include "jemalloc/internal/rb.h"
 
diff --git a/include/jemalloc/internal/rtree_structs.h b/include/jemalloc/internal/rtree_structs.h
index ba0f96d..a02a1f6 100644
--- a/include/jemalloc/internal/rtree_structs.h
+++ b/include/jemalloc/internal/rtree_structs.h
@@ -2,7 +2,7 @@
 #define JEMALLOC_INTERNAL_RTREE_STRUCTS_H
 
 #include "jemalloc/internal/atomic.h"
-#include "jemalloc/internal/mutex_pool_structs.h"
+#include "jemalloc/internal/mutex.h"
 
 struct rtree_node_elm_s {
 	atomic_p_t	child; /* (rtree_{node,leaf}_elm_t *) */
diff --git a/include/jemalloc/internal/stats.h b/include/jemalloc/internal/stats.h
index 3f5c20c..47ca4f9 100644
--- a/include/jemalloc/internal/stats.h
+++ b/include/jemalloc/internal/stats.h
@@ -3,9 +3,9 @@
 
 #include "jemalloc/internal/atomic.h"
 #include "jemalloc/internal/mutex_prof.h"
-#include "jemalloc/internal/mutex_types.h"
-#include "jemalloc/internal/mutex_structs.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/size_classes.h"
+#include "jemalloc/internal/stats_tsd.h"
 
 /* The opt.stats_print storage. */
 extern bool opt_stats_print;
@@ -26,14 +26,6 @@ typedef atomic_u64_t arena_stats_u64_t;
 typedef uint64_t arena_stats_u64_t;
 #endif
 
-typedef struct tcache_bin_stats_s {
-	/*
-	 * Number of allocation requests that corresponded to the size of this
-	 * bin.
-	 */
-	uint64_t	nrequests;
-} tcache_bin_stats_t;
-
 typedef struct malloc_bin_stats_s {
 	/*
 	 * Total number of allocation/deallocation requests served directly by
diff --git a/include/jemalloc/internal/stats_tsd.h b/include/jemalloc/internal/stats_tsd.h
new file mode 100644
index 0000000..d0c3bbe
--- /dev/null
+++ b/include/jemalloc/internal/stats_tsd.h
@@ -0,0 +1,12 @@
+#ifndef JEMALLOC_INTERNAL_STATS_TSD_H
+#define JEMALLOC_INTERNAL_STATS_TSD_H
+
+typedef struct tcache_bin_stats_s {
+	/*
+	 * Number of allocation requests that corresponded to the size of this
+	 * bin.
+	 */
+	uint64_t	nrequests;
+} tcache_bin_stats_t;
+
+#endif /* JEMALLOC_INTERNAL_STATS_TSD_H */
diff --git a/include/jemalloc/internal/tcache_structs.h b/include/jemalloc/internal/tcache_structs.h
index 7c0afb0..7eb516f 100644
--- a/include/jemalloc/internal/tcache_structs.h
+++ b/include/jemalloc/internal/tcache_structs.h
@@ -3,7 +3,7 @@
 
 #include "jemalloc/internal/ql.h"
 #include "jemalloc/internal/size_classes.h"
-#include "jemalloc/internal/stats.h"
+#include "jemalloc/internal/stats_tsd.h"
 #include "jemalloc/internal/ticker.h"
 
 /*
diff --git a/include/jemalloc/internal/tsd_generic.h b/include/jemalloc/internal/tsd_generic.h
index d59cb74..1e52ef7 100644
--- a/include/jemalloc/internal/tsd_generic.h
+++ b/include/jemalloc/internal/tsd_generic.h
@@ -10,11 +10,8 @@ struct tsd_init_block_s {
 	void *data;
 };
 
+/* Defined in tsd.c, to allow the mutex headers to have tsd dependencies. */
 typedef struct tsd_init_head_s tsd_init_head_t;
-struct tsd_init_head_s {
-	ql_head(tsd_init_block_t) blocks;
-	malloc_mutex_t lock;
-};
 
 typedef struct {
 	bool initialized;
diff --git a/src/arena.c b/src/arena.c
index 35b4e5a..bc8fd28 100644
--- a/src/arena.c
+++ b/src/arena.c
@@ -3,6 +3,7 @@
 #include "jemalloc/internal/jemalloc_internal_includes.h"
 
 #include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/size_classes.h"
 #include "jemalloc/internal/util.h"
 
diff --git a/src/base.c b/src/base.c
index dd4b109..892c28d 100644
--- a/src/base.c
+++ b/src/base.c
@@ -3,6 +3,7 @@
 #include "jemalloc/internal/jemalloc_internal_includes.h"
 
 #include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
 
 /******************************************************************************/
 /* Data. */
diff --git a/src/ctl.c b/src/ctl.c
index da5e171..30704ed 100644
--- a/src/ctl.c
+++ b/src/ctl.c
@@ -4,6 +4,7 @@
 
 #include "jemalloc/internal/assert.h"
 #include "jemalloc/internal/ctl.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/nstime.h"
 #include "jemalloc/internal/size_classes.h"
 #include "jemalloc/internal/util.h"
diff --git a/src/extent.c b/src/extent.c
index ff09f7f..fee8198 100644
--- a/src/extent.c
+++ b/src/extent.c
@@ -4,6 +4,7 @@
 
 #include "jemalloc/internal/assert.h"
 #include "jemalloc/internal/ph.h"
+#include "jemalloc/internal/mutex.h"
 
 /******************************************************************************/
 /* Data. */
diff --git a/src/jemalloc.c b/src/jemalloc.c
index f083adc..517fbb9 100644
--- a/src/jemalloc.c
+++ b/src/jemalloc.c
@@ -7,6 +7,7 @@
 #include "jemalloc/internal/ctl.h"
 #include "jemalloc/internal/jemalloc_internal_types.h"
 #include "jemalloc/internal/malloc_io.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/size_classes.h"
 #include "jemalloc/internal/spin.h"
 #include "jemalloc/internal/ticker.h"
diff --git a/src/large.c b/src/large.c
index ed73dc2..55ee352 100644
--- a/src/large.c
+++ b/src/large.c
@@ -3,6 +3,7 @@
 #include "jemalloc/internal/jemalloc_internal_includes.h"
 
 #include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/util.h"
 
 /******************************************************************************/
diff --git a/src/mutex_pool.c b/src/mutex_pool.c
index 004d6d0..95a4573 100644
--- a/src/mutex_pool.c
+++ b/src/mutex_pool.c
@@ -3,6 +3,8 @@
 #include "jemalloc/internal/jemalloc_preamble.h"
 #include "jemalloc/internal/jemalloc_internal_includes.h"
 
+#include "jemalloc/internal/mutex.h"
+
 bool
 mutex_pool_init(mutex_pool_t *pool, const char *name, witness_rank_t rank) {
 	for (int i = 0; i < MUTEX_POOL_SIZE; ++i) {
diff --git a/src/prof.c b/src/prof.c
index 1897881..639e598 100644
--- a/src/prof.c
+++ b/src/prof.c
@@ -6,6 +6,7 @@
 #include "jemalloc/internal/ckh.h"
 #include "jemalloc/internal/hash.h"
 #include "jemalloc/internal/malloc_io.h"
+#include "jemalloc/internal/mutex.h"
 
 /******************************************************************************/
 
diff --git a/src/rtree.c b/src/rtree.c
index 637853c..53702cf 100644
--- a/src/rtree.c
+++ b/src/rtree.c
@@ -3,6 +3,7 @@
 #include "jemalloc/internal/jemalloc_internal_includes.h"
 
 #include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
 
 /*
  * Only the most significant bits of keys passed to rtree_{read,write}() are
diff --git a/src/stats.c b/src/stats.c
index fd10816..b67d46d 100644
--- a/src/stats.c
+++ b/src/stats.c
@@ -4,6 +4,7 @@
 
 #include "jemalloc/internal/assert.h"
 #include "jemalloc/internal/ctl.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/mutex_prof.h"
 
 const char *global_mutex_names[mutex_prof_num_global_mutexes] = {
diff --git a/src/tcache.c b/src/tcache.c
index 4bb2fb8..96ebe67 100644
--- a/src/tcache.c
+++ b/src/tcache.c
@@ -3,6 +3,7 @@
 #include "jemalloc/internal/jemalloc_internal_includes.h"
 
 #include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
 #include "jemalloc/internal/size_classes.h"
 
 /******************************************************************************/
diff --git a/src/tsd.c b/src/tsd.c
index 801d812..525432b 100644
--- a/src/tsd.c
+++ b/src/tsd.c
@@ -3,6 +3,7 @@
 #include "jemalloc/internal/jemalloc_internal_includes.h"
 
 #include "jemalloc/internal/assert.h"
+#include "jemalloc/internal/mutex.h"
 
 /******************************************************************************/
 /* Data. */
@@ -23,6 +24,17 @@ DWORD tsd_tsd;
 tsd_wrapper_t tsd_boot_wrapper = {false, TSD_INITIALIZER};
 bool tsd_booted = false;
 #else
+
+/*
+ * This contains a mutex, but it's pretty convenient to allow the mutex code to
+ * have a dependency on tsd.  So we define the struct here, and only refer to it
+ * by pointer in the header.
+ */
+struct tsd_init_head_s {
+	ql_head(tsd_init_block_t) blocks;
+	malloc_mutex_t lock;
+};
+
 pthread_key_t tsd_tsd;
 tsd_init_head_t	tsd_init_head = {
 	ql_head_initializer(blocks),