Fix threads-related profiling bugs.

Initialize bt2cnt_tsd so that cleanup at thread exit actually happens.

Associate (prof_ctx_t *) with allocated objects, rather than
(prof_thr_cnt_t *).  Each thread must always operate on its own
(prof_thr_cnt_t *), and an object may outlive the thread that allocated it.

jemalloc/include/jemalloc/internal/arena.h

@@ -98,7 +98,7 @@ struct arena_chunk_map_s {
 
 #ifdef JEMALLOC_PROF
 	/* Profile counters, used for large object runs. */
-	prof_thr_cnt_t			*prof_cnt;
+	prof_ctx_t			*prof_ctx;
 #endif
 
 	/*
@@ -246,10 +246,10 @@ struct arena_bin_s {
 
 #ifdef JEMALLOC_PROF
 	/*
-	 * Offset of first (prof_cnt_t *) in a run header for this bin's size
+	 * Offset of first (prof_ctx_t *) in a run header for this bin's size
 	 * class, or 0 if (opt_prof == false).
 	 */
-	uint32_t	cnt0_offset;
+	uint32_t	ctx0_offset;
 #endif
 
 	/* Offset of first region in a run for this bin's size class. */
@@ -438,8 +438,8 @@ size_t	arena_salloc(const void *ptr);
 #ifdef JEMALLOC_PROF
 void	arena_prof_promoted(const void *ptr, size_t size);
 size_t	arena_salloc_demote(const void *ptr);
-prof_thr_cnt_t	*arena_prof_cnt_get(const void *ptr);
-void	arena_prof_cnt_set(const void *ptr, prof_thr_cnt_t *cnt);
+prof_ctx_t	*arena_prof_ctx_get(const void *ptr);
+void	arena_prof_ctx_set(const void *ptr, prof_ctx_t *ctx);
 #endif
 void	arena_dalloc_bin(arena_t *arena, arena_chunk_t *chunk, void *ptr,
     arena_chunk_map_t *mapelm);

jemalloc/include/jemalloc/internal/extent.h

@@ -19,7 +19,7 @@ struct extent_node_s {
 
 #ifdef JEMALLOC_PROF
 	/* Profile counters, used for huge objects. */
-	prof_thr_cnt_t		*prof_cnt;
+	prof_ctx_t		*prof_ctx;
 #endif
 
 	/* Pointer to the extent that this tree node is responsible for. */

jemalloc/include/jemalloc/internal/huge.h

@@ -25,8 +25,8 @@ void	*huge_ralloc(void *ptr, size_t size, size_t oldsize);
 void	huge_dalloc(void *ptr);
 size_t	huge_salloc(const void *ptr);
 #ifdef JEMALLOC_PROF
-prof_thr_cnt_t	*huge_prof_cnt_get(const void *ptr);
-void	huge_prof_cnt_set(const void *ptr, prof_thr_cnt_t *cnt);
+prof_ctx_t	*huge_prof_ctx_get(const void *ptr);
+void	huge_prof_ctx_set(const void *ptr, prof_ctx_t *ctx);
 #endif
 bool	huge_boot(void);
 

jemalloc/include/jemalloc/internal/prof.h

@@ -98,6 +98,9 @@ struct prof_thr_cnt_s {
 };
 
 struct prof_ctx_s {
+	/* Associated backtrace. */
+	prof_bt_t		*bt;
+
 	/* Protects cnt_merged and sets_ql. */
 	malloc_mutex_t		lock;
 
@@ -151,10 +154,10 @@ bool	prof_init(prof_t *prof, bool master);
 void	prof_destroy(prof_t *prof);
 
 prof_thr_cnt_t	*prof_alloc_prep(size_t size);
-prof_thr_cnt_t	*prof_cnt_get(const void *ptr);
+prof_ctx_t	*prof_ctx_get(const void *ptr);
 void	prof_malloc(const void *ptr, prof_thr_cnt_t *cnt);
 void	prof_realloc(const void *ptr, prof_thr_cnt_t *cnt, const void *old_ptr,
-    size_t old_size, prof_thr_cnt_t *old_cnt);
+    size_t old_size, prof_ctx_t *old_ctx);
 void	prof_free(const void *ptr);
 void	prof_idump(void);
 bool	prof_mdump(const char *filename);