Remove _Py_tss_stats thread local.

Add pystats pointer to PyThreadState and use from there instead.  This
is slightly slower but shouldn't matter in practice.  This simplifies
the attach/detach logic as well.

Author: nascheme

Include/cpython/pystate.h

@@ -208,6 +208,10 @@ struct _ts {
     */
     PyObject *threading_local_sentinel;
     _PyRemoteDebuggerSupport remote_debugger_support;
+
+#ifdef Py_STATS
+    PyStats *pystats;  // pointer PyStats structure, NULL if recording is off
+#endif
 };
 
 /* other API */
@@ -230,6 +234,19 @@ PyAPI_FUNC(void) PyThreadState_EnterTracing(PyThreadState *tstate);
 // function is set, otherwise disable them.
 PyAPI_FUNC(void) PyThreadState_LeaveTracing(PyThreadState *tstate);
 
+#if defined(HAVE_THREAD_LOCAL) && !defined(Py_BUILD_CORE_MODULE)
+extern _Py_thread_local PyThreadState* _Py_tss_tstate;
+
+static inline PyStats*
+_PyThreadState_GetStatsFast(void)
+{
+    if (_Py_tss_tstate == NULL) {
+        return NULL; // no attached thread state
+    }
+    return _Py_tss_tstate->pystats;
+}
+#endif
+
 /* PyGILState */
 
 /* Helper/diagnostic function - return 1 if the current thread

Include/cpython/pystats.h

@@ -14,7 +14,7 @@
 // - sys._stats_dump()
 //
 // Python must be built with ./configure --enable-pystats to define the
-// _PyStats_GET() function.
+// _PyStats_GET() macro.
 //
 // Define _PY_INTERPRETER macro to increment interpreter_increfs and
 // interpreter_decrefs. Otherwise, increment increfs and decrefs.
@@ -112,8 +112,11 @@ typedef struct _gc_stats {
 #ifdef Py_GIL_DISABLED
 // stats specific to free-threaded build
 typedef struct _ft_stats {
+    // number of times interpreter had to spin or park when trying to acquire a mutex
     uint64_t mutex_sleeps;
+    // number of times that the QSBR mechanism polled (compute read sequence value)
     uint64_t qsbr_polls;
+    // number of times stop-the-world mechanism was used
     uint64_t world_stops;
 } FTStats;
 #endif
@@ -189,25 +192,16 @@ typedef struct _stats {
     GCStats gc_stats[3]; // must match NUM_GENERATIONS
 } PyStats;
 
-
-#if defined(HAVE_THREAD_LOCAL) && !defined(Py_BUILD_CORE_MODULE)
-extern _Py_thread_local PyStats *_Py_tss_stats;
-#endif
-
-// Export for most shared extensions, used via _PyStats_GET() static
-// inline function.
+// Export for most shared extensions
 PyAPI_FUNC(PyStats *) _PyStats_GetLocal(void);
 
-// Return pointer to the PyStats structure, NULL if recording is off.
-static inline PyStats*
-_PyStats_GET(void)
-{
 #if defined(HAVE_THREAD_LOCAL) && !defined(Py_BUILD_CORE_MODULE)
-    return _Py_tss_stats;
+// use inline function version defined in cpython/pystate.h
+static inline PyStats* _PyThreadState_GetStatsFast(void);
+#define _PyStats_GET _PyThreadState_GetStatsFast
 #else
-    return _PyStats_GetLocal();
+#define _PyStats_GET _PyStats_GetLocal
 #endif
-}
 
 #define _Py_STATS_EXPR(expr) \
     do { \

Include/internal/pycore_tstate.h

@@ -71,16 +71,13 @@ 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
-#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
+     // per-thread stats, will be merged into interp->pystats_struct
+     PyStats *pystats_struct; // allocated by _PyStats_ThreadInit()
 #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

Python/pystats.c

@@ -3,6 +3,7 @@
 #include "pycore_opcode_metadata.h" // _PyOpcode_Caches
 #include "pycore_pyatomic_ft_wrappers.h"
 #include "pycore_pylifecycle.h"     // _PyOS_URandomNonblock()
+#include "pycore_tstate.h"
 #include "pycore_uop_metadata.h"    // _PyOpcode_uop_name
 #include "pycore_uop_ids.h"         // MAX_UOP_ID
 #include "pycore_pystate.h"         // _PyThreadState_GET()
@@ -12,13 +13,14 @@
 
 #ifdef Py_STATS
 
-// Pointer to Thread-local stats structure, null if recording is off.
-_Py_thread_local PyStats *_Py_tss_stats;
-
 PyStats *
 _PyStats_GetLocal(void)
 {
-    return _Py_tss_stats;
+    PyThreadState *tstate = _PyThreadState_GET();
+    if (tstate) {
+        return tstate->pystats;
+    }
+    return NULL;
 }
 
 #ifdef Py_GIL_DISABLED
@@ -582,20 +584,18 @@ stats_toggle_on_off(PyThreadState *tstate, int on)
         if (!ts->_status.active) {
             continue;
         }
-        _PyThreadStateImpl *ts_impl = (_PyThreadStateImpl *)ts;
         PyStats *s;
         if (interp->pystats_enabled) {
 #ifdef Py_GIL_DISABLED
-            s = ts_impl->pystats_struct;
+            s = ((_PyThreadStateImpl *)ts)->pystats_struct;
 #else
             s = ((PyThreadState *)tstate)->interp->pystats_struct;
 #endif
         }
         else {
             s = NULL;
         }
-        // write to the tss variable for the 'ts' thread
-        *ts_impl->pystats_tss = s;
+        ts->pystats = s;
     }
     _PyEval_StartTheWorld(interp);
     return 0;
@@ -759,31 +759,28 @@ _PyStats_ThreadFini(_PyThreadStateImpl *tstate)
 }
 
 void
-_PyStats_Attach(_PyThreadStateImpl *tstate)
+_PyStats_Attach(_PyThreadStateImpl *tstate_impl)
 {
     PyStats *s;
-    PyInterpreterState *interp = ((PyThreadState *)tstate)->interp;
+    PyThreadState *tstate = (PyThreadState *)tstate_impl;
+    PyInterpreterState *interp = tstate->interp;
     if (FT_ATOMIC_LOAD_INT_RELAXED(interp->pystats_enabled)) {
 #ifdef Py_GIL_DISABLED
-        s = tstate->pystats_struct;
+        s = ((_PyThreadStateImpl *)tstate)->pystats_struct;
 #else
-        s = ((PyThreadState *)tstate)->interp->pystats_struct;
+        s = tstate->interp->pystats_struct;
 #endif
     }
     else {
         s = NULL;
     }
-    // use correct TSS variable for thread
-    tstate->pystats_tss = &_Py_tss_stats;
-    // write to the tss variable for the 'ts' thread
-    _Py_tss_stats = s;
+    tstate->pystats = s;
 }
 
 void
-_PyStats_Detach(_PyThreadStateImpl *tstate)
+_PyStats_Detach(_PyThreadStateImpl *tstate_impl)
 {
-    tstate->pystats_tss = NULL;
-    _Py_tss_stats = NULL;
+    ((PyThreadState *)tstate_impl)->pystats = NULL;
 }
 
 #endif // Py_STATS

Tools/c-analyzer/cpython/ignored.tsv

@@ -194,7 +194,6 @@ Python/pyfpe.c	-	PyFPE_counter	-
 Python/import.c	-	pkgcontext	-
 Python/pystate.c	-	_Py_tss_tstate	-
 Python/pystate.c	-	_Py_tss_gilstate	-
-Python/pystats.c	-	_Py_tss_stats	-
 
 ##-----------------------
 ## should be const