diff --git a/Doc/library/threading.rst b/Doc/library/threading.rst
index eca3e76d84a1cfd..5d9a7b6314b1668 100644
--- a/Doc/library/threading.rst
+++ b/Doc/library/threading.rst
@@ -515,7 +515,7 @@ since it is impossible to detect the termination of alien threads.
    This constructor should always be called with keyword arguments.  Arguments
    are:
 
-   *group* should be ``None``; reserved for future extension when a
+   *group* must be ``None`` as it is reserved for future extension when a
    :class:`!ThreadGroup` class is implemented.
 
    *target* is the callable object to be invoked by the :meth:`run` method.
diff --git a/Include/internal/pycore_sliceobject.h b/Include/internal/pycore_sliceobject.h
index ba8b1f1cb27dee3..b6c821764886c36 100644
--- a/Include/internal/pycore_sliceobject.h
+++ b/Include/internal/pycore_sliceobject.h
@@ -12,7 +12,7 @@ extern "C" {
 /* runtime lifecycle */
 
 PyAPI_FUNC(PyObject *)
-_PyBuildSlice_ConsumeRefs(PyObject *start, PyObject *stop);
+_PyBuildSlice_ConsumeRefs(PyObject *start, PyObject *stop, PyObject *step);
 
 #ifdef __cplusplus
 }
diff --git a/Lib/asyncio/locks.py b/Lib/asyncio/locks.py
index fa3a94764b507ae..0d5e362188ab21d 100644
--- a/Lib/asyncio/locks.py
+++ b/Lib/asyncio/locks.py
@@ -145,10 +145,7 @@ def _wake_up_first(self):
         """Ensure that the first waiter will wake up."""
         if not self._waiters:
             return
-        try:
-            fut = next(iter(self._waiters))
-        except StopIteration:
-            return
+        fut = next(iter(self._waiters))
 
         # .done() means that the waiter is already set to wake up.
         if not fut.done():
diff --git a/Lib/profiling/sampling/_flamegraph_assets/flamegraph.js b/Lib/profiling/sampling/_flamegraph_assets/flamegraph.js
index 1611bf754424c13..840acf2c27d1201 100644
--- a/Lib/profiling/sampling/_flamegraph_assets/flamegraph.js
+++ b/Lib/profiling/sampling/_flamegraph_assets/flamegraph.js
@@ -7,6 +7,7 @@ let invertedData = null;
 let currentThreadFilter = 'all';
 let isInverted = false;
 let useModuleNames = true;
+let zoomedNodeValue = null;
 
 // Heat colors are now defined in CSS variables (--heat-1 through --heat-8)
 // and automatically switch with theme changes - no JS color arrays needed!
@@ -316,6 +317,7 @@ function createPythonTooltip(data) {
     const selfSamples = d.data.self || 0;
     const selfMs = (selfSamples / 1000).toFixed(2);
     const percentage = ((d.data.value / data.value) * 100).toFixed(2);
+    const relativePercentage = Math.min(100, ((d.data.value / (zoomedNodeValue ?? data.value)) * 100)).toFixed(2);
     const calls = d.data.calls || 0;
     const childCount = d.children ? d.children.length : 0;
     const source = d.data.source;
@@ -439,6 +441,11 @@ function createPythonTooltip(data) {
         <span class="tooltip-stat-label">Percentage:</span>
         <span class="tooltip-stat-value accent">${percentage}%</span>
 
+        ${relativePercentage != percentage && relativePercentage != "100.00" ? `
+          <span class="tooltip-stat-label">Relative Percentage:</span>
+          <span class="tooltip-stat-value accent">${relativePercentage}%</span>
+        ` : ''}
+
         ${calls > 0 ? `
           <span class="tooltip-stat-label">Function Calls:</span>
           <span class="tooltip-stat-value">${calls.toLocaleString()}</span>
@@ -620,6 +627,9 @@ function createFlamegraph(tooltip, rootValue, data) {
       const percentage = d.data.value / rootValue;
       const level = getHeatLevel(percentage);
       return heatColors[level];
+    })
+    .onClick(function (d) {
+      zoomedNodeValue = d.data.value;
     });
 
   return chart;
@@ -629,6 +639,7 @@ function renderFlamegraph(chart, data) {
   d3.select("#chart").datum(data).call(chart);
   window.flamegraphChart = chart;
   window.flamegraphData = data;
+  zoomedNodeValue = null;
   populateStats(data);
 }
 
@@ -1269,6 +1280,7 @@ function filterDataByThread(data, threadId) {
 
 function resetZoom() {
   if (window.flamegraphChart) {
+    zoomedNodeValue = null;
     window.flamegraphChart.resetZoom();
   }
 }
diff --git a/Lib/test/test_type_cache.py b/Lib/test/test_type_cache.py
index 22ad9f6243eda91..849a2afd8ed7986 100644
--- a/Lib/test/test_type_cache.py
+++ b/Lib/test/test_type_cache.py
@@ -1,4 +1,5 @@
 """ Tests for the internal type cache in CPython. """
+import collections.abc
 import dis
 import unittest
 import warnings
@@ -114,6 +115,25 @@ class HolderSub(Holder):
             Holder.set_value()
             HolderSub.value
 
+    def test_abc_register_invalidates_subclass_versions(self):
+        class Parent:
+            pass
+
+        class Child(Parent):
+            pass
+
+        type_assign_version(Parent)
+        type_assign_version(Child)
+        parent_version = type_get_version(Parent)
+        child_version = type_get_version(Child)
+        if parent_version == 0 or child_version == 0:
+            self.skipTest("Could not assign valid type versions")
+
+        collections.abc.Mapping.register(Parent)
+
+        self.assertEqual(type_get_version(Parent), 0)
+        self.assertEqual(type_get_version(Child), 0)
+
 @support.cpython_only
 class TypeCacheWithSpecializationTests(unittest.TestCase):
     def tearDown(self):
diff --git a/Makefile.pre.in b/Makefile.pre.in
index 9435bf534fb5121..765c72290e46ee4 100644
--- a/Makefile.pre.in
+++ b/Makefile.pre.in
@@ -3111,9 +3111,18 @@ config.status:	$(srcdir)/configure
 
 .PRECIOUS: config.status $(BUILDPYTHON) Makefile Makefile.pre
 
-Python/asm_trampoline.o: $(srcdir)/Python/asm_trampoline.S
+Python/asm_trampoline_x86_64.o: $(srcdir)/Python/asm_trampoline_x86_64.S
 	$(CC) -c $(PY_CORE_CFLAGS) -o $@ $<
 
+Python/asm_trampoline_aarch64.o: $(srcdir)/Python/asm_trampoline_aarch64.S
+	$(CC) -c $(PY_CORE_CFLAGS) -o $@ $<
+
+Python/asm_trampoline_riscv64.o: $(srcdir)/Python/asm_trampoline_riscv64.S
+	$(CC) -c $(PY_CORE_CFLAGS) -o $@ $<
+
+Python/asm_trampoline_universal2.o: Python/asm_trampoline_aarch64.o Python/asm_trampoline_x86_64.o
+	lipo -create -output $@ Python/asm_trampoline_aarch64.o Python/asm_trampoline_x86_64.o
+
 Python/emscripten_trampoline_inner.wasm: $(srcdir)/Python/emscripten_trampoline_inner.c
 	# emcc has a path that ends with emsdk/upstream/emscripten/emcc, we're looking for emsdk/upstream/bin/clang.
 	$$(dirname $$(dirname $(CC)))/bin/clang -o $@ $< -mgc -O2 -Wl,--no-entry -Wl,--import-table -Wl,--import-memory -target wasm32-unknown-unknown -nostdlib
diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2026-04-15-15-48-04.gh-issue-148450.2MEVqH.rst b/Misc/NEWS.d/next/Core_and_Builtins/2026-04-15-15-48-04.gh-issue-148450.2MEVqH.rst
new file mode 100644
index 000000000000000..2a7d0d9bb3a7f7e
--- /dev/null
+++ b/Misc/NEWS.d/next/Core_and_Builtins/2026-04-15-15-48-04.gh-issue-148450.2MEVqH.rst
@@ -0,0 +1 @@
+Fix ``abc.register()`` so it invalidates type version tags for registered classes.
diff --git a/Misc/NEWS.d/next/Tools-Demos/2026-05-22-18-51-09.gh-issue-150258.dh8GVK.rst b/Misc/NEWS.d/next/Tools-Demos/2026-05-22-18-51-09.gh-issue-150258.dh8GVK.rst
new file mode 100644
index 000000000000000..02cad6c4f53d928
--- /dev/null
+++ b/Misc/NEWS.d/next/Tools-Demos/2026-05-22-18-51-09.gh-issue-150258.dh8GVK.rst
@@ -0,0 +1 @@
+Update the tooltip on the Tachyon flame graph to show both absolute and relative percentages.
diff --git a/Modules/_io/bufferedio.c b/Modules/_io/bufferedio.c
index 0fdae7b2d210040..5537947f6a51c11 100644
--- a/Modules/_io/bufferedio.c
+++ b/Modules/_io/bufferedio.c
@@ -1509,7 +1509,9 @@ buffered_iternext(PyObject *op)
         tp == state->PyBufferedRandom_Type)
     {
         /* Skip method call overhead for speed */
+        Py_BEGIN_CRITICAL_SECTION(self);
         line = _buffered_readline(self, -1);
+        Py_END_CRITICAL_SECTION();
     }
     else {
         line = PyObject_CallMethodNoArgs((PyObject *)self,
diff --git a/Modules/_testinternalcapi/test_cases.c.h b/Modules/_testinternalcapi/test_cases.c.h
index a2506524f0bb6dc..b463bb18b160564 100644
--- a/Modules/_testinternalcapi/test_cases.c.h
+++ b/Modules/_testinternalcapi/test_cases.c.h
@@ -12007,7 +12007,8 @@
                 v = stack_pointer[-4];
                 _PyFrame_SetStackPointer(frame, stack_pointer);
                 PyObject *slice = _PyBuildSlice_ConsumeRefs(PyStackRef_AsPyObjectSteal(start),
-                    PyStackRef_AsPyObjectSteal(stop));
+                    PyStackRef_AsPyObjectSteal(stop),
+                    Py_None);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
                 int err;
                 if (slice == NULL) {
diff --git a/Objects/sliceobject.c b/Objects/sliceobject.c
index 95f10815687757e..96ff3118dc4405d 100644
--- a/Objects/sliceobject.c
+++ b/Objects/sliceobject.c
@@ -117,8 +117,8 @@ PyObject _Py_EllipsisObject = _PyObject_HEAD_INIT(&PyEllipsis_Type);
    index is present.
 */
 
-static PySliceObject *
-_PyBuildSlice_Consume2(PyObject *start, PyObject *stop, PyObject *step)
+PyObject *
+_PyBuildSlice_ConsumeRefs(PyObject *start, PyObject *stop, PyObject *step)
 {
     assert(start != NULL && stop != NULL && step != NULL);
     PySliceObject *obj = _Py_FREELIST_POP(PySliceObject, slices);
@@ -131,13 +131,14 @@ _PyBuildSlice_Consume2(PyObject *start, PyObject *stop, PyObject *step)
 
     obj->start = start;
     obj->stop = stop;
-    obj->step = Py_NewRef(step);
+    obj->step = step;
 
     _PyObject_GC_TRACK(obj);
-    return obj;
+    return (PyObject *)obj;
 error:
     Py_DECREF(start);
     Py_DECREF(stop);
+    Py_DECREF(step);
     return NULL;
 }
 
@@ -153,15 +154,8 @@ PySlice_New(PyObject *start, PyObject *stop, PyObject *step)
     if (stop == NULL) {
         stop = Py_None;
     }
-    return (PyObject *)_PyBuildSlice_Consume2(Py_NewRef(start),
-                                              Py_NewRef(stop), step);
-}
-
-PyObject *
-_PyBuildSlice_ConsumeRefs(PyObject *start, PyObject *stop)
-{
-    assert(start != NULL && stop != NULL);
-    return (PyObject *)_PyBuildSlice_Consume2(start, stop, Py_None);
+    return _PyBuildSlice_ConsumeRefs(Py_NewRef(start),
+                                              Py_NewRef(stop), Py_NewRef(step));
 }
 
 PyObject *
@@ -177,9 +171,7 @@ _PySlice_FromIndices(Py_ssize_t istart, Py_ssize_t istop)
         return NULL;
     }
 
-    slice = PySlice_New(start, end, NULL);
-    Py_DECREF(start);
-    Py_DECREF(end);
+    slice = _PyBuildSlice_ConsumeRefs(start, end, Py_None);
     return slice;
 }
 
diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index 7cca137f74be58f..fc679ef747e8567 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -6490,9 +6490,25 @@ set_flags_recursive(PyTypeObject *self, unsigned long mask, unsigned long flags)
 void
 _PyType_SetFlagsRecursive(PyTypeObject *self, unsigned long mask, unsigned long flags)
 {
+    BEGIN_TYPE_LOCK();
+    /* Ideally, changing flags and invalidating the old version tag would
+       happen in one step. But type_modified_unlocked() is re-entrant and
+       cannot run with the world stopped, so we must invalidate first.
+       Immutable/static-builtin types are skipped because
+       set_flags_recursive() does not modify them. */
+    if (!PyType_HasFeature(self, Py_TPFLAGS_IMMUTABLETYPE) &&
+        (self->tp_flags & mask) != flags)
+    {
+        type_modified_unlocked(self);
+    }
+    /* Keep TYPE_LOCK held while waiting for stop-the-world so no thread
+       can reassign a version tag before the flag update. */
+    type_lock_prevent_release();
     types_stop_world();
     set_flags_recursive(self, mask, flags);
     types_start_world();
+    type_lock_allow_release();
+    END_TYPE_LOCK();
 }
 
 /* This is similar to PyObject_GenericGetAttr(),
diff --git a/Python/asm_trampoline_aarch64.h b/Python/asm_trampoline_aarch64.S
similarity index 76%
rename from Python/asm_trampoline_aarch64.h
rename to Python/asm_trampoline_aarch64.S
index bc83aa460b6860d..b3aeb728de200c6 100644
--- a/Python/asm_trampoline_aarch64.h
+++ b/Python/asm_trampoline_aarch64.S
@@ -1,6 +1,3 @@
-#ifndef ASM_TRAMPOLINE_AARCH_64_H_
-#define ASM_TRAMPOLINE_AARCH_64_H_
-
 /*
  * References:
  *  - https://developer.arm.com/documentation/101028/0012/5--Feature-test-macros
@@ -38,6 +35,31 @@
   #define GNU_PROPERTY_AARCH64_GCS 0
 #endif
 
+    .text
+#if defined(__aarch64__) && defined(__AARCH64EL__) && !defined(__ILP32__)
+#if defined(__APPLE__)
+    .globl	__Py_trampoline_func_start
+__Py_trampoline_func_start:
+#else
+    .globl	_Py_trampoline_func_start
+_Py_trampoline_func_start:
+#endif
+    SIGN_LR
+    stp     x29, x30, [sp, -16]!
+    mov     x29, sp
+    blr     x3
+    ldp     x29, x30, [sp], 16
+    VERIFY_LR
+    ret
+#if defined(__APPLE__)
+    .globl	__Py_trampoline_func_end
+__Py_trampoline_func_end:
+#else
+    .globl	_Py_trampoline_func_end
+_Py_trampoline_func_end:
+    .section        .note.GNU-stack,"",@progbits
+#endif
+
 /* Add the BTI, PAC and GCS support to GNU Notes section */
 #if GNU_PROPERTY_AARCH64_BTI != 0 || GNU_PROPERTY_AARCH64_POINTER_AUTH != 0 || GNU_PROPERTY_AARCH64_GCS != 0
     .pushsection .note.gnu.property, "a"; /* Start a new allocatable section */
@@ -52,5 +74,4 @@
     .long 0; /* padding for 8 byte alignment */
     .popsection; /* end the section */
 #endif
-
-#endif
+#endif // __aarch64__ && __AARCH64EL__ && !__ILP32__
diff --git a/Python/asm_trampoline_riscv64.S b/Python/asm_trampoline_riscv64.S
new file mode 100644
index 000000000000000..6125ba95373f954
--- /dev/null
+++ b/Python/asm_trampoline_riscv64.S
@@ -0,0 +1,12 @@
+    .text
+    .globl	_Py_trampoline_func_start
+_Py_trampoline_func_start:
+    addi    sp,sp,-16
+    sd      ra,8(sp)
+    jalr    a3
+    ld      ra,8(sp)
+    addi    sp,sp,16
+    jr      ra
+    .globl	_Py_trampoline_func_end
+_Py_trampoline_func_end:
+    .section        .note.GNU-stack,"",@progbits
diff --git a/Python/asm_trampoline.S b/Python/asm_trampoline_x86_64.S
similarity index 51%
rename from Python/asm_trampoline.S
rename to Python/asm_trampoline_x86_64.S
index 9f3ca909ab7d852..0e6b11589eafc82 100644
--- a/Python/asm_trampoline.S
+++ b/Python/asm_trampoline_x86_64.S
@@ -1,24 +1,12 @@
-#include "asm_trampoline_aarch64.h"
-
     .text
+#ifdef __x86_64__
 #if defined(__APPLE__)
     .globl	__Py_trampoline_func_start
-#else
-    .globl	_Py_trampoline_func_start
-#endif
-# The following assembly is equivalent to:
-# PyObject *
-# trampoline(PyThreadState *ts, _PyInterpreterFrame *f,
-#            int throwflag, py_evaluator evaluator)
-# {
-#     return evaluator(ts, f, throwflag);
-# }
-#if defined(__APPLE__)
 __Py_trampoline_func_start:
 #else
+    .globl	_Py_trampoline_func_start
 _Py_trampoline_func_start:
 #endif
-#ifdef __x86_64__
 #if defined(__CET__) && (__CET__ & 1)
     endbr64
 #endif
@@ -27,26 +15,6 @@ _Py_trampoline_func_start:
     call   *%rcx
     pop    %rbp
     ret
-#endif // __x86_64__
-#if defined(__aarch64__) && defined(__AARCH64EL__) && !defined(__ILP32__)
-    // ARM64 little endian, 64bit ABI
-    // generate with aarch64-linux-gnu-gcc 12.1
-    SIGN_LR
-    stp     x29, x30, [sp, -16]!
-    mov     x29, sp
-    blr     x3
-    ldp     x29, x30, [sp], 16
-    VERIFY_LR
-    ret
-#endif
-#ifdef __riscv
-    addi    sp,sp,-16
-    sd      ra,8(sp)
-    jalr    a3
-    ld      ra,8(sp)
-    addi    sp,sp,16
-    jr      ra
-#endif
 #if defined(__APPLE__)
     .globl	__Py_trampoline_func_end
 __Py_trampoline_func_end:
@@ -56,7 +24,7 @@ _Py_trampoline_func_end:
     .section        .note.GNU-stack,"",@progbits
 #endif
 # Note for indicating the assembly code supports CET
-#if defined(__x86_64__) && defined(__CET__) && (__CET__ & 1)
+#if defined(__CET__) && (__CET__ & 1)
     .section    .note.gnu.property,"a"
     .align 8
     .long    1f - 0f
@@ -73,4 +41,5 @@ _Py_trampoline_func_end:
 3:
     .align 8
 4:
+#endif
 #endif // __x86_64__
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index f7487c7136962f1..300b7da753c2baf 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -1116,7 +1116,8 @@ dummy_func(
 
         op(_STORE_SLICE, (v, container, start, stop -- )) {
             PyObject *slice = _PyBuildSlice_ConsumeRefs(PyStackRef_AsPyObjectSteal(start),
-                                                        PyStackRef_AsPyObjectSteal(stop));
+                                                        PyStackRef_AsPyObjectSteal(stop),
+                                                        Py_None);
             int err;
             if (slice == NULL) {
                 err = 1;
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index efa61d7de74e88c..952860f01b8a682 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -6811,7 +6811,8 @@
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             _PyFrame_SetStackPointer(frame, stack_pointer);
             PyObject *slice = _PyBuildSlice_ConsumeRefs(PyStackRef_AsPyObjectSteal(start),
-                PyStackRef_AsPyObjectSteal(stop));
+                PyStackRef_AsPyObjectSteal(stop),
+                Py_None);
             stack_pointer = _PyFrame_GetStackPointer(frame);
             int err;
             if (slice == NULL) {
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 53e09a8f4523c7c..83051cf41cc043b 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -12004,7 +12004,8 @@
                 v = stack_pointer[-4];
                 _PyFrame_SetStackPointer(frame, stack_pointer);
                 PyObject *slice = _PyBuildSlice_ConsumeRefs(PyStackRef_AsPyObjectSteal(start),
-                    PyStackRef_AsPyObjectSteal(stop));
+                    PyStackRef_AsPyObjectSteal(stop),
+                    Py_None);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
                 int err;
                 if (slice == NULL) {
diff --git a/Python/hamt.c b/Python/hamt.c
index e4719e71a5259a5..95998ae5062ac7e 100644
--- a/Python/hamt.c
+++ b/Python/hamt.c
@@ -702,6 +702,7 @@ hamt_node_bitmap_assoc(PyHamtNode_Bitmap *self,
 
             PyHamtNode_Bitmap *ret = hamt_node_bitmap_clone(self);
             if (ret == NULL) {
+                Py_DECREF(sub_node);
                 return NULL;
             }
             Py_SETREF(ret->b_array[val_idx], (PyObject*)sub_node);
@@ -994,6 +995,7 @@ hamt_node_bitmap_without(PyHamtNode_Bitmap *self,
 
                 PyHamtNode_Bitmap *clone = hamt_node_bitmap_clone(self);
                 if (clone == NULL) {
+                    Py_DECREF(sub_node);
                     return W_ERROR;
                 }
 
diff --git a/configure b/configure
index a1b635ffd15c4a3..00e3ca25c86e8f9 100755
--- a/configure
+++ b/configure
@@ -14411,17 +14411,35 @@ printf "%s\n" "$SHLIBS" >&6; }
 
 { printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking perf trampoline" >&5
 printf %s "checking perf trampoline... " >&6; }
+PERF_TRAMPOLINE_OBJ=""
 case $PLATFORM_TRIPLET in #(
   x86_64-linux-gnu) :
-    perf_trampoline=yes ;; #(
+    perf_trampoline=yes
+                        PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_x86_64.o ;; #(
   aarch64-linux-gnu) :
-    perf_trampoline=yes ;; #(
+    perf_trampoline=yes
+                         PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_aarch64.o ;; #(
   darwin) :
     case $MACOSX_DEPLOYMENT_TARGET in #(
   10.[0-9]|10.1[0-1]) :
     perf_trampoline=no ;; #(
   *) :
     perf_trampoline=yes
+                 if test "${enable_universalsdk}" && test "$UNIVERSAL_ARCHS" = "universal2"; then
+                   PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_universal2.o
+                 else
+                   case "$host_cpu" in
+                     x86_64)
+                       PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_x86_64.o
+                       ;;
+                     aarch64|arm64)
+                       PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_aarch64.o
+                       ;;
+                     *)
+                       perf_trampoline=no
+                       ;;
+                   esac
+                 fi
              ;;
 esac ;; #(
   *) :
@@ -14437,7 +14455,6 @@ then :
 
 printf "%s\n" "#define PY_HAVE_PERF_TRAMPOLINE 1" >>confdefs.h
 
-  PERF_TRAMPOLINE_OBJ=Python/asm_trampoline.o
 
 fi
 
diff --git a/configure.ac b/configure.ac
index 082c6c2d756cdca..bb2a7eee9a116e3 100644
--- a/configure.ac
+++ b/configure.ac
@@ -3866,12 +3866,30 @@ AC_MSG_RESULT([$SHLIBS])
 dnl perf trampoline is Linux and macOS specific and requires an arch-specific
 dnl trampoline in assembly.
 AC_MSG_CHECKING([perf trampoline])
+PERF_TRAMPOLINE_OBJ=""
 AS_CASE([$PLATFORM_TRIPLET],
-  [x86_64-linux-gnu], [perf_trampoline=yes],
-  [aarch64-linux-gnu], [perf_trampoline=yes],
+  [x86_64-linux-gnu], [perf_trampoline=yes
+                        PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_x86_64.o],
+  [aarch64-linux-gnu], [perf_trampoline=yes
+                         PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_aarch64.o],
   [darwin], [AS_CASE([$MACOSX_DEPLOYMENT_TARGET],
                 [[10.[0-9]|10.1[0-1]]], [perf_trampoline=no],
-                [perf_trampoline=yes]
+                [perf_trampoline=yes
+                 if test "${enable_universalsdk}" && test "$UNIVERSAL_ARCHS" = "universal2"; then
+                   PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_universal2.o
+                 else
+                   case "$host_cpu" in
+                     x86_64)
+                       PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_x86_64.o
+                       ;;
+                     aarch64|arm64)
+                       PERF_TRAMPOLINE_OBJ=Python/asm_trampoline_aarch64.o
+                       ;;
+                     *)
+                       perf_trampoline=no
+                       ;;
+                   esac
+                 fi]
             )],
   [perf_trampoline=no]
 )
@@ -3879,7 +3897,6 @@ AC_MSG_RESULT([$perf_trampoline])
 
 AS_VAR_IF([perf_trampoline], [yes], [
   AC_DEFINE([PY_HAVE_PERF_TRAMPOLINE], [1], [Define to 1 if you have the perf trampoline.])
-  PERF_TRAMPOLINE_OBJ=Python/asm_trampoline.o
 ])
 AC_SUBST([PERF_TRAMPOLINE_OBJ])