pythongh-118518: Allow perf to work without frame pointers (python#112254)

Author: pablogsal

Doc/c-api/init_config.rst

@@ -1251,7 +1251,10 @@ PyConfig
       for more information.
 
       Set by :option:`-X perf <-X>` command line option and by the
-      :envvar:`PYTHONPERFSUPPORT` environment variable.
+      :envvar:`PYTHONPERFSUPPORT` environment variable for perf support
+      with stack pointers and :option:`-X perfjit <-X>` command line option
+      and by the :envvar:`PYTHONPERFJITSUPPORT` environment variable for perf
+      support with DWARF JIT information.
 
       Default: ``-1``.
 

Doc/howto/perf_profiling.rst

@@ -205,3 +205,36 @@ You can check if your system has been compiled with this flag by running::
 If you don't see any output it means that your interpreter has not been compiled with
 frame pointers and therefore it may not be able to show Python functions in the output
 of ``perf``.
+
+
+How to work without frame pointers
+----------------------------------
+
+If you are working with a Python interpreter that has been compiled without frame pointers
+you can still use the ``perf`` profiler but the overhead will be a bit higher because Python
+needs to generate unwinding information for every Python function call on the fly. Additionally,
+``perf`` will take more time to process the data because it will need to use the DWARF debugging
+information to unwind the stack and this is a slow process.
+
+To enable this mode, you can use the environment variable :envvar:`PYTHONPERFJITSUPPORT` or the
+:option:`-X perfjit <-X>` option, which will enable the JIT mode for the ``perf`` profiler.
+
+When using the perf JIT mode, you need an extra step before you can run ``perf report``. You need to
+call the ``perf inject`` command to inject the JIT information into the ``perf.data`` file.
+
+    $ perf record -F 9999 -g --call-graph dwarf -o perf.data python -Xperfjit my_script.py
+    $ perf inject -i perf.data --jit
+    $ perf report -g -i perf.data
+
+or using the environment variable::
+
+    $ PYTHONPERFJITSUPPORT=1 perf record -F 9999 -g --call-graph dwarf -o perf.data python my_script.py
+    $ perf inject -i perf.data --jit
+    $ perf report -g -i perf.data
+
+Notice that when using ``--call-graph dwarf`` the ``perf`` tool will take snapshots of the stack of
+the process being profiled and save the information in the ``perf.data`` file. By default the size of
+the stack dump is 8192 bytes but the user can change the size by passing the size after comma like
+``--call-graph dwarf,4096``. The size of the stack dump is important because if the size is too small
+``perf`` will not be able to unwind the stack and the output will be incomplete.
+

Doc/using/cmdline.rst

@@ -586,6 +586,15 @@ Miscellaneous options
 
      .. versionadded:: 3.12
 
+   * ``-X perfjit`` enables support for the Linux ``perf`` profiler with DWARF
+     support. When this option is provided, the ``perf`` profiler will be able
+     to report Python calls using DWARF ifnormation. This option is only available on
+     some platforms and will do nothing if is not supported on the current
+     system. The default value is "off". See also :envvar:`PYTHONPERFJITSUPPORT`
+     and :ref:`perf_profiling`.
+
+     .. versionadded:: 3.13
+
    * :samp:`-X cpu_count={n}` overrides :func:`os.cpu_count`,
      :func:`os.process_cpu_count`, and :func:`multiprocessing.cpu_count`.
      *n* must be greater than or equal to 1.
@@ -1127,6 +1136,21 @@ conflict.
 
    .. versionadded:: 3.12
 
+.. envvar:: PYTHONPERFJITSUPPORT
+
+   If this variable is set to a nonzero value, it enables support for
+   the Linux ``perf`` profiler so Python calls can be detected by it
+   using DWARF information.
+
+   If set to ``0``, disable Linux ``perf`` profiler support.
+
+   See also the :option:`-X perfjit <-X>` command-line option
+   and :ref:`perf_profiling`.
+
+   .. versionadded:: 3.13
+
+
+
 .. envvar:: PYTHON_CPU_COUNT
 
    If this variable is set to a positive integer, it overrides the return

Doc/whatsnew/3.13.rst

@@ -231,6 +231,11 @@ Other Language Changes
   equivalent of the :option:`-X frozen_modules <-X>` command-line option.
   (Contributed by Yilei Yang in :gh:`111374`.)
 
+* Add :ref:`support for the perf profiler <perf_profiling>` working without
+  frame pointers through the new environment variable
+  :envvar:`PYTHONPERFJITSUPPORT` and command-line option :option:`-X perfjit
+  <-X>` (Contributed by Pablo Galindo in :gh:`118518`.)
+
 * The new :envvar:`PYTHON_HISTORY` environment variable can be used to change
   the location of a ``.python_history`` file.
   (Contributed by Levi Sabah, Zackery Spytz and Hugo van Kemenade in

Include/internal/pycore_ceval.h

@@ -108,6 +108,7 @@ extern int _PyIsPerfTrampolineActive(void);
 extern PyStatus _PyPerfTrampoline_AfterFork_Child(void);
 #ifdef PY_HAVE_PERF_TRAMPOLINE
 extern _PyPerf_Callbacks _Py_perfmap_callbacks;
+extern _PyPerf_Callbacks _Py_perfmap_jit_callbacks;
 #endif
 
 static inline PyObject*

Include/internal/pycore_ceval_state.h

@@ -75,6 +75,7 @@ struct trampoline_api_st {
                         unsigned int code_size, PyCodeObject* code);
     int (*free_state)(void* state);
     void *state;
+    Py_ssize_t code_padding;
 };
 #endif
 
@@ -83,6 +84,7 @@ struct _ceval_runtime_state {
     struct {
 #ifdef PY_HAVE_PERF_TRAMPOLINE
         perf_status_t status;
+        int perf_trampoline_type;
         Py_ssize_t extra_code_index;
         struct code_arena_st *code_arena;
         struct trampoline_api_st trampoline_api;

Lib/test/test_perf_profiler.py

@@ -5,6 +5,7 @@
 import sysconfig
 import os
 import pathlib
+import shutil
 from test import support
 from test.support.script_helper import (
     make_script,
@@ -76,14 +77,27 @@ def baz():
         perf_file = pathlib.Path(f"/tmp/perf-{process.pid}.map")
         self.assertTrue(perf_file.exists())
         perf_file_contents = perf_file.read_text()
-        perf_lines = perf_file_contents.splitlines();
-        expected_symbols = [f"py::foo:{script}", f"py::bar:{script}", f"py::baz:{script}"]
+        perf_lines = perf_file_contents.splitlines()
+        expected_symbols = [
+            f"py::foo:{script}",
+            f"py::bar:{script}",
+            f"py::baz:{script}",
+        ]
         for expected_symbol in expected_symbols:
-            perf_line = next((line for line in perf_lines if expected_symbol in line), None)
-            self.assertIsNotNone(perf_line, f"Could not find {expected_symbol} in perf file")
+            perf_line = next(
+                (line for line in perf_lines if expected_symbol in line), None
+            )
+            self.assertIsNotNone(
+                perf_line, f"Could not find {expected_symbol} in perf file"
+            )
             perf_addr = perf_line.split(" ")[0]
-            self.assertFalse(perf_addr.startswith("0x"), "Address should not be prefixed with 0x")
-            self.assertTrue(set(perf_addr).issubset(string.hexdigits), "Address should contain only hex characters")
+            self.assertFalse(
+                perf_addr.startswith("0x"), "Address should not be prefixed with 0x"
+            )
+            self.assertTrue(
+                set(perf_addr).issubset(string.hexdigits),
+                "Address should contain only hex characters",
+            )
 
     def test_trampoline_works_with_forks(self):
         code = """if 1:
@@ -212,7 +226,7 @@ def test_sys_api_get_status(self):
         assert_python_ok("-c", code)
 
 
-def is_unwinding_reliable():
+def is_unwinding_reliable_with_frame_pointers():
     cflags = sysconfig.get_config_var("PY_CORE_CFLAGS")
     if not cflags:
         return False
@@ -259,24 +273,49 @@ def perf_command_works():
     return True
 
 
-def run_perf(cwd, *args, **env_vars):
+def run_perf(cwd, *args, use_jit=False, **env_vars):
     if env_vars:
         env = os.environ.copy()
         env.update(env_vars)
     else:
         env = None
     output_file = cwd + "/perf_output.perf"
-    base_cmd = ("perf", "record", "-g", "--call-graph=fp", "-o", output_file, "--")
+    if not use_jit:
+        base_cmd = ("perf", "record", "-g", "--call-graph=fp", "-o", output_file, "--")
+    else:
+        base_cmd = (
+            "perf",
+            "record",
+            "-g",
+            "--call-graph=dwarf,65528",
+            "-F99",
+            "-k1",
+            "-o",
+            output_file,
+            "--",
+        )
     proc = subprocess.run(
         base_cmd + args,
         stdout=subprocess.PIPE,
         stderr=subprocess.PIPE,
         env=env,
     )
     if proc.returncode:
-        print(proc.stderr)
+        print(proc.stderr, file=sys.stderr)
         raise ValueError(f"Perf failed with return code {proc.returncode}")
 
+    if use_jit:
+        jit_output_file = cwd + "/jit_output.dump"
+        command = ("perf", "inject", "-j", "-i", output_file, "-o", jit_output_file)
+        proc = subprocess.run(
+            command, stderr=subprocess.PIPE, stdout=subprocess.PIPE, env=env
+        )
+        if proc.returncode:
+            print(proc.stderr)
+            raise ValueError(f"Perf failed with return code {proc.returncode}")
+        # Copy the jit_output_file to the output_file
+        os.rename(jit_output_file, output_file)
+
     base_cmd = ("perf", "script")
     proc = subprocess.run(
         ("perf", "script", "-i", output_file),
@@ -290,20 +329,9 @@ def run_perf(cwd, *args, **env_vars):
     )
 
 
-@unittest.skipUnless(perf_command_works(), "perf command doesn't work")
-@unittest.skipUnless(is_unwinding_reliable(), "Unwinding is unreliable")
-class TestPerfProfiler(unittest.TestCase):
-    def setUp(self):
-        super().setUp()
-        self.perf_files = set(pathlib.Path("/tmp/").glob("perf-*.map"))
-
-    def tearDown(self) -> None:
-        super().tearDown()
-        files_to_delete = (
-            set(pathlib.Path("/tmp/").glob("perf-*.map")) - self.perf_files
-        )
-        for file in files_to_delete:
-            file.unlink()
+class TestPerfProfilerMixin:
+    def run_perf(self, script_dir, perf_mode, script):
+        raise NotImplementedError()
 
     def test_python_calls_appear_in_the_stack_if_perf_activated(self):
         with temp_dir() as script_dir:
@@ -322,14 +350,14 @@ def baz(n):
                 baz(10000000)
                 """
             script = make_script(script_dir, "perftest", code)
-            stdout, stderr = run_perf(script_dir, sys.executable, "-Xperf", script)
+            stdout, stderr = self.run_perf(script_dir, script)
             self.assertEqual(stderr, "")
 
             self.assertIn(f"py::foo:{script}", stdout)
             self.assertIn(f"py::bar:{script}", stdout)
             self.assertIn(f"py::baz:{script}", stdout)
 
-    def test_python_calls_do_not_appear_in_the_stack_if_perf_activated(self):
+    def test_python_calls_do_not_appear_in_the_stack_if_perf_deactivated(self):
         with temp_dir() as script_dir:
             code = """if 1:
                 def foo(n):
@@ -346,13 +374,38 @@ def baz(n):
                 baz(10000000)
                 """
             script = make_script(script_dir, "perftest", code)
-            stdout, stderr = run_perf(script_dir, sys.executable, script)
+            stdout, stderr = self.run_perf(
+                script_dir, script, activate_trampoline=False
+            )
             self.assertEqual(stderr, "")
 
             self.assertNotIn(f"py::foo:{script}", stdout)
             self.assertNotIn(f"py::bar:{script}", stdout)
             self.assertNotIn(f"py::baz:{script}", stdout)
 
+@unittest.skipUnless(perf_command_works(), "perf command doesn't work")
+@unittest.skipUnless(
+    is_unwinding_reliable_with_frame_pointers(),
+    "Unwinding is unreliable with frame pointers",
+)
+class TestPerfProfiler(unittest.TestCase, TestPerfProfilerMixin):
+    def run_perf(self, script_dir, script, activate_trampoline=True):
+        if activate_trampoline:
+            return run_perf(script_dir, sys.executable, "-Xperf", script)
+        return run_perf(script_dir, sys.executable, script)
+
+    def setUp(self):
+        super().setUp()
+        self.perf_files = set(pathlib.Path("/tmp/").glob("perf-*.map"))
+
+    def tearDown(self) -> None:
+        super().tearDown()
+        files_to_delete = (
+            set(pathlib.Path("/tmp/").glob("perf-*.map")) - self.perf_files
+        )
+        for file in files_to_delete:
+            file.unlink()
+
     def test_pre_fork_compile(self):
         code = """if 1:
                 import sys
@@ -370,7 +423,7 @@ def bar_fork():
                     foo_fork()
 
                 def foo():
-                    pass
+                    import time; time.sleep(1)
 
                 def bar():
                     foo()
@@ -423,12 +476,41 @@ def compile_trampolines_for_all_functions():
         # identical in both the parent and child perf-map files.
         perf_file_lines = perf_file_contents.split("\n")
         for line in perf_file_lines:
-            if (
-                f"py::foo_fork:{script}" in line
-                or f"py::bar_fork:{script}" in line
-            ):
+            if f"py::foo_fork:{script}" in line or f"py::bar_fork:{script}" in line:
                 self.assertIn(line, child_perf_file_contents)
 
+def _is_kernel_version_at_least(major, minor):
+    try:
+        with open("/proc/version") as f:
+            version = f.readline().split()[2]
+    except FileNotFoundError:
+        return False
+    version = version.split(".")
+    return int(version[0]) > major or (int(version[0]) == major and int(version[1]) >= minor)
+
+@unittest.skipUnless(perf_command_works(), "perf command doesn't work")
+@unittest.skipUnless(_is_kernel_version_at_least(6, 6), "perf command may not work due to a perf bug")
+class TestPerfProfilerWithDwarf(unittest.TestCase, TestPerfProfilerMixin):
+    def run_perf(self, script_dir, script, activate_trampoline=True):
+        if activate_trampoline:
+            return run_perf(
+                script_dir, sys.executable, "-Xperfjit", script, use_jit=True
+            )
+        return run_perf(script_dir, sys.executable, script, use_jit=True)
+
+    def setUp(self):
+        super().setUp()
+        self.perf_files = set(pathlib.Path("/tmp/").glob("jit*.dump"))
+        self.perf_files |= set(pathlib.Path("/tmp/").glob("jitted-*.so"))
+
+    def tearDown(self) -> None:
+        super().tearDown()
+        files_to_delete = set(pathlib.Path("/tmp/").glob("jit*.dump"))
+        files_to_delete |= set(pathlib.Path("/tmp/").glob("jitted-*.so"))
+        files_to_delete = files_to_delete - self.perf_files
+        for file in files_to_delete:
+            file.unlink()
+
 
 if __name__ == "__main__":
     unittest.main()

Makefile.pre.in

@@ -488,6 +488,7 @@ PYTHON_OBJS=	\
 		Python/fileutils.o \
 		Python/suggestions.o \
 		Python/perf_trampoline.o \
+		Python/perf_jit_trampoline.o \
 		Python/$(DYNLOADFILE) \
 		$(LIBOBJS) \
 		$(MACHDEP_OBJS) \

Misc/NEWS.d/next/Core and Builtins/2024-05-02-20-32-42.gh-issue-118518.m-JbTi.rst

@@ -0,0 +1,4 @@
+Allow the Linux perf support to work without frame pointers using perf's
+advanced JIT support. The feature is activated when using the
+``PYTHONPERFJITSUPPORT`` environment variable or when running Python with
+``-Xperfjit``. Patch by Pablo Galindo

PCbuild/_freeze_module.vcxproj

@@ -240,6 +240,7 @@
     <ClCompile Include="..\Python\parking_lot.c" />
     <ClCompile Include="..\Python\pathconfig.c" />
     <ClCompile Include="..\Python\perf_trampoline.c" />
+    <ClCompile Include="..\Python\perf_jit_trampoline.c" />
     <ClCompile Include="..\Python\preconfig.c" />
     <ClCompile Include="..\Python\pyarena.c" />
     <ClCompile Include="..\Python\pyctype.c" />