Doc/c-api/perfmaps.rst

.. highlight:: c

.. _perfmaps:

Support for Perf Maps
----------------------

On supported platforms (Linux and macOS), the runtime can take
advantage of *perf map files* to make Python functions visible to an external
profiling tool (such as `perf <https://perf.wiki.kernel.org/index.php/Main_Page>`_ or
`samply <https://github.com/mstange/samply/>`_). A running process may create a
file in the ``/tmp`` directory, which contains entries that can map a section
of executable code to a name. This interface is described in the
`documentation of the Linux Perf tool <https://git.kernel.org/pub/scm/linux/
kernel/git/torvalds/linux.git/tree/tools/perf/Documentation/jit-interface.txt>`_.

In Python, these helper APIs can be used by libraries and features that rely
on generating machine code on the fly.

Note that holding an :term:`attached thread state` is not required for these APIs.

.. c:function:: int PyUnstable_PerfMapState_Init(void)

   Open the ``/tmp/perf-$pid.map`` file, unless it's already opened, and create
   a lock to ensure thread-safe writes to the file (provided the writes are
   done through :c:func:`PyUnstable_WritePerfMapEntry`). Normally, there's no need
   to call this explicitly; just use :c:func:`PyUnstable_WritePerfMapEntry`
   and it will initialize the state on first call.

   Returns ``0`` on success, ``-1`` on failure to create/open the perf map file,
   or ``-2`` on failure to create a lock. Check ``errno`` for more information
   about the cause of a failure.

.. c:function:: int PyUnstable_WritePerfMapEntry(const void *code_addr, size_t code_size, const char *entry_name)

   Write one single entry to the ``/tmp/perf-$pid.map`` file. This function is
   thread safe. Here is what an example entry looks like::

      # address      size  name
      7f3529fcf759 b     py::bar:/run/t.py

   Will call :c:func:`PyUnstable_PerfMapState_Init` before writing the entry, if
   the perf map file is not already opened. Returns ``0`` on success, or the
   same error codes as :c:func:`PyUnstable_PerfMapState_Init` on failure.

.. c:function:: void PyUnstable_PerfMapState_Fini(void)

   Close the perf map file opened by :c:func:`PyUnstable_PerfMapState_Init`.
   This is called by the runtime itself during interpreter shut-down. In
   general, there shouldn't be a reason to explicitly call this, except to
   handle specific scenarios such as forking.

.. c:function:: int PyUnstable_CopyPerfMapFile(const char *parent_filename)

   Open the ``/tmp/perf-$pid.map`` file and append the content of *parent_filename*
   to it.

   This function is available on all platforms but only generates output on platforms
   that support perf maps (currently only Linux). On other platforms, it does nothing.

   .. versionadded:: 3.13

.. c:function:: int PyUnstable_PerfTrampoline_CompileCode(PyCodeObject *code)

   Compile the given code object using the current perf trampoline.

   The "current" trampoline is the one set by the runtime or the most recent
   :c:func:`PyUnstable_PerfTrampoline_SetPersistAfterFork` call.

   If no trampoline is set, falls back to normal compilation (no perf map entry).

   :param code: The code object to compile.
   :return: 0 on success, -1 on failure.

   .. versionadded:: 3.13

.. c:function:: int PyUnstable_PerfTrampoline_SetPersistAfterFork(int enable)

   Set whether the perf trampoline should persist after a fork.

   * If ``enable`` is true (non-zero): perf map file remains open/valid post-fork.
     Child process inherits all existing perf map entries.
   * If ``enable`` is false (zero): perf map closes post-fork.
     Child process gets empty perf map.

   Default: false (clears on fork).

   :param enable: 1 to enable, 0 to disable.
   :return: 0 on success, -1 on failure.

   .. versionadded:: 3.13