Make pystats global state be per-interpreter.

Author: nascheme

Include/cpython/pystats.h

@@ -186,12 +186,10 @@ typedef struct _stats {
     FTStats ft_stats;
 #endif
     RareEventStats rare_event_stats;
-    GCStats *gc_stats;
+    GCStats gc_stats[3]; // must match NUM_GENERATIONS
 } PyStats;
 
 
-#ifdef Py_GIL_DISABLED
-
 #if defined(HAVE_THREAD_LOCAL) && !defined(Py_BUILD_CORE_MODULE)
 extern _Py_thread_local PyStats *_Py_tss_stats;
 #endif
@@ -200,30 +198,15 @@ extern _Py_thread_local PyStats *_Py_tss_stats;
 // inline function.
 PyAPI_FUNC(PyStats *) _PyStats_GetLocal(void);
 
-#else // !Py_GIL_DISABLED
-
-// Export for shared extensions like 'math'
-PyAPI_DATA(PyStats*) _Py_stats;
-
-#endif
-
 // Return pointer to the PyStats structure, NULL if recording is off.
 static inline PyStats*
 _PyStats_GET(void)
 {
-#ifdef Py_GIL_DISABLED
-
 #if defined(HAVE_THREAD_LOCAL) && !defined(Py_BUILD_CORE_MODULE)
     return _Py_tss_stats;
 #else
     return _PyStats_GetLocal();
 #endif
-
-#else // !Py_GIL_DISABLED
-
-    return _Py_stats;
-
-#endif
 }
 
 #define _Py_STATS_EXPR(expr) \

Include/internal/pycore_interp_structs.h

@@ -198,7 +198,7 @@ enum _GCPhase {
 };
 
 /* If we change this, we need to change the default value in the
-   signature of gc.collect. */
+   signature of gc.collect and change the size of PyStats.gc_stats */
 #define NUM_GENERATIONS 3
 
 struct _gc_runtime_state {
@@ -968,6 +968,18 @@ struct _is {
 #  ifdef Py_STACKREF_CLOSE_DEBUG
     _Py_hashtable_t *closed_stackrefs_table;
 #  endif
+#endif
+
+#ifdef Py_STATS
+    // true if recording of pystats is on, this is used when new threads
+    // are created to decide if recording should be on for them
+    bool pystats_enabled;
+    // allocated when (and if) stats are first enabled
+    PyStats *pystats_struct;
+#ifdef Py_GIL_DISABLED
+    // held when pystats related interpreter state is being updated
+    PyMutex pystats_mutex;
+#endif
 #endif
 
     /* the initial PyInterpreterState.threads.head */

Include/internal/pycore_pystats.h

@@ -9,7 +9,7 @@ extern "C" {
 #endif
 
 #ifdef Py_STATS
-extern void _Py_StatsOn(void);
+extern int _Py_StatsOn(void);
 extern void _Py_StatsOff(void);
 extern void _Py_StatsClear(void);
 extern int _Py_PrintSpecializationStats(int to_file);

Include/internal/pycore_stats.h

@@ -110,7 +110,7 @@ PyAPI_FUNC(PyObject*) _Py_GetSpecializationStats(void);
         RARE_EVENT_INTERP_INC(interp, name); \
     } while (0); \
 
-bool _PyStats_ThreadInit(_PyThreadStateImpl *);
+bool _PyStats_ThreadInit(PyInterpreterState *, _PyThreadStateImpl *);
 void _PyStats_ThreadFini(_PyThreadStateImpl *);
 void _PyStats_Attach(_PyThreadStateImpl *);
 void _PyStats_Detach(_PyThreadStateImpl *);

Include/internal/pycore_tstate.h

@@ -71,14 +71,16 @@ typedef struct _PyThreadStateImpl {
     // When >1, code objects do not immortalize their non-string constants.
     int suppress_co_const_immortalization;
 
+#endif // Py_GIL_DISABLED
+
 #ifdef Py_STATS
-    // per-thread stats, will be merged into the _Py_stats_struct global
+#ifdef Py_GIL_DISABLED
+    // per-thread stats, will be merged into interp->pystats_struct
     PyStats *pystats_struct; // allocated by _PyStats_ThreadInit()
+#endif
     PyStats **pystats_tss;  // pointer to tss variable
 #endif
 
-#endif // Py_GIL_DISABLED
-
 #if defined(Py_REF_DEBUG) && defined(Py_GIL_DISABLED)
     Py_ssize_t reftotal;  // this thread's total refcount operations
 #endif

Objects/object.c

@@ -2418,6 +2418,13 @@ _PyObject_InitState(PyInterpreterState *interp)
     if (refchain_init(interp) < 0) {
         return _PyStatus_NO_MEMORY();
     }
+#endif
+#ifdef Py_STATS
+    if (interp->config._pystats) {
+        // start with pystats enabled, can be disabled via sys._stats_off()
+        // this needs to be set before the first tstate is created
+        interp->pystats_enabled = true;
+    }
 #endif
     return _PyStatus_OK();
 }

Python/initconfig.c

@@ -2807,12 +2807,6 @@ _PyConfig_Write(const PyConfig *config, _PyRuntimeState *runtime)
         return _PyStatus_NO_MEMORY();
     }
 
-#ifdef Py_STATS
-    if (config->_pystats) {
-        _Py_StatsOn();
-    }
-#endif
-
     return _PyStatus_OK();
 }
 

Python/pystate.c

@@ -22,10 +22,9 @@
 #include "pycore_runtime.h"       // _PyRuntime
 #include "pycore_runtime_init.h"  // _PyRuntimeState_INIT
 #include "pycore_stackref.h"      // Py_STACKREF_DEBUG
+#include "pycore_stats.h"         // FT_STAT_WORLD_STOP_INC()
 #include "pycore_time.h"          // _PyTime_Init()
 #include "pycore_uniqueid.h"      // _PyObject_FinalizePerThreadRefcounts()
-#include "pycore_stats.h"         // FT_STAT_WORLD_STOP_INC()
-
 
 
 /* --------------------------------------------------------------------------
@@ -480,6 +479,12 @@ alloc_interpreter(void)
 static void
 free_interpreter(PyInterpreterState *interp)
 {
+#ifdef Py_STATS
+    if (interp->pystats_struct) {
+        PyMem_RawFree(interp->pystats_struct);
+        interp->pystats_struct = NULL;
+    }
+#endif
     // The main interpreter is statically allocated so
     // should not be freed.
     if (interp != &_PyRuntime._main_interpreter) {
@@ -1532,7 +1537,7 @@ new_threadstate(PyInterpreterState *interp, int whence)
 #endif
 #ifdef Py_STATS
     // The PyStats structure is quite large and is allocated separated from tstate.
-    if (!_PyStats_ThreadInit(tstate)) {
+    if (!_PyStats_ThreadInit(interp, tstate)) {
         free_threadstate(tstate);
         return NULL;
     }
@@ -2017,9 +2022,6 @@ tstate_activate(PyThreadState *tstate)
     if (!tstate->_status.bound_gilstate) {
         bind_gilstate_tstate(tstate);
     }
-#ifdef Py_STATS
-    _PyStats_Attach((_PyThreadStateImpl *)tstate);
-#endif
 
     tstate->_status.active = 1;
 }
@@ -2139,6 +2141,10 @@ _PyThreadState_Attach(PyThreadState *tstate)
         _PyCriticalSection_Resume(tstate);
     }
 
+#ifdef Py_STATS
+    _PyStats_Attach((_PyThreadStateImpl *)tstate);
+#endif
+
 #if defined(Py_DEBUG)
     errno = err;
 #endif