We were not properly accounting for interpreter memory leaks at
shutdown and had two sources of leaks:
* Objects that use deferred reference counting and were reachable via
static types outlive the final GC. We now disable deferred reference
counting on all objects if we are calling the GC due to interpreter
shutdown.
* `_PyMem_FreeDelayed` did not properly check for interpreter shutdown
so we had some memory blocks that were enqueued to be freed, but
never actually freed.
* `_PyType_FinalizeIdPool` wasn't called at interpreter shutdown.
This implements the delayed reuse of mimalloc pages that contain Python
objects in the free-threaded build.
Allocations of the same size class are grouped in data structures called
pages. These are different from operating system pages. For thread-safety, we
want to ensure that memory used to store PyObjects remains valid as long as
there may be concurrent lock-free readers; we want to delay using it for
other size classes, in other heaps, or returning it to the operating system.
When a mimalloc page becomes empty, instead of immediately freeing it, we tag
it with a QSBR goal and insert it into a per-thread state linked list of
pages to be freed. When mimalloc needs a fresh page, we process the queue and
free any still empty pages that are now deemed safe to be freed. Pages
waiting to be freed are still available for allocations of the same size
class and allocating from a page prevent it from being freed. There is
additional logic to handle abandoned pages when threads exit.
The previous code had two bugs. First, the debug offset in the mimalloc
heap includes the two pymalloc debug words, but the pointer passed to
fill_mem_debug does not include them. Second, the current object heap is
correct source for allocations, but not deallocations.
This adds `_PyMem_FreeDelayed()` and supporting functions. The
`_PyMem_FreeDelayed()` function frees memory with the same allocator as
`PyMem_Free()`, but after some delay to ensure that concurrent lock-free
readers have finished.
Fixes a few issues related to refleak tracking in the free-threaded build:
- Count blocks in abandoned segments
- Call `_mi_page_free_collect` earlier during heap traversal in order to get an accurate count of blocks in use.
- Add missing refcount tracking in `_Py_DecRefSharedDebug` and `_Py_ExplicitMergeRefcount`.
- Pause threads in `get_num_global_allocated_blocks` to ensure that traversing the mimalloc heaps is safe.
For interpreters that share state with the main interpreter, this points
to the same static memory structure. For interpreters with their own
obmalloc state, it is heap allocated. Add free_obmalloc_arenas() which
will free the obmalloc arenas and radix tree structures for interpreters
with their own obmalloc state.
Co-authored-by: Eric Snow <ericsnowcurrently@gmail.com>
* gh-112529: Track if debug allocator is used as underlying allocator
The GC implementation for free-threaded builds will need to accurately
detect if the debug allocator is used because it affects the offset of
the Python object from the beginning of the memory allocation. The
current implementation of `_PyMem_DebugEnabled` only considers if the
debug allocator is the outer-most allocator; it doesn't handle the case
of "hooks" like tracemalloc being used on top of the debug allocator.
This change enables more accurate detection of the debug allocator by
tracking when debug hooks are enabled.
* Simplify _PyMem_DebugEnabled
This fixes `_PyInterpreterState_GetAllocatedBlocks()` and
`_Py_GetGlobalAllocatedBlocks()` in the free-threaded builds. The
gh-113263 change that introduced multiple mimalloc heaps per-thread
broke the logic for counting the number of allocated blocks. For subtle
reasons, this led to reported reference count leaks in the refleaks
buildbots.
* gh-112532: Use separate mimalloc heaps for GC objects
In `--disable-gil` builds, we now use four separate heaps in
anticipation of using mimalloc to find GC objects when the GIL is
disabled. To support this, we also make a few changes to mimalloc:
* `mi_heap_t` and `mi_tld_t` initialization is split from allocation.
This allows us to have a `mi_tld_t` per-`PyThreadState`, which is
important to keep interpreter isolation, since the same OS thread may
run in multiple interpreters (using different PyThreadStates.)
* Heap abandoning (mi_heap_collect_ex) can now be called from a
different thread than the one that created the heap. This is necessary
because we may clear and delete the containing PyThreadStates from a
different thread during finalization and after fork().
* Use enum instead of defines and guard mimalloc includes.
* The enum typedef will be convenient for future PRs that use the type.
* Guarding the mimalloc includes allows us to unconditionally include
pycore_mimalloc.h from other header files that rely on things like
`struct _mimalloc_thread_state`.
* Only define _mimalloc_thread_state in Py_GIL_DISABLED builds
This replaces some usages of PyThread_type_lock with PyMutex, which does not require memory allocation to initialize.
This simplifies some of the runtime initialization and is also one step towards avoiding changing the default raw memory allocator during initialize/finalization, which can be non-thread-safe in some circumstances.
Don't declare _PyMem_MimallocEnabled() if WITH_PYMALLOC macro is not
defined (./configure --without-pymalloc).
Fix also a typo in _PyInterpreterState_FinalizeAllocatedBlocks().
* Add mimalloc v2.12
Modified src/alloc.c to remove include of alloc-override.c and not
compile new handler.
Did not include the following files:
- include/mimalloc-new-delete.h
- include/mimalloc-override.h
- src/alloc-override-osx.c
- src/alloc-override.c
- src/static.c
- src/region.c
mimalloc is thread safe and shares a single heap across all runtimes,
therefore finalization and getting global allocated blocks across all
runtimes is different.
* mimalloc: minimal changes for use in Python:
- remove debug spam for freeing large allocations
- use same bytes (0xDD) for freed allocations in CPython and mimalloc
This is important for the test_capi debug memory tests
* Don't export mimalloc symbol in libpython.
* Enable mimalloc as Python allocator option.
* Add mimalloc MIT license.
* Log mimalloc in Lib/test/pythoninfo.py.
* Document new mimalloc support.
* Use macro defs for exports as done in:
https://github.com/python/cpython/pull/31164/
Co-authored-by: Sam Gross <colesbury@gmail.com>
Co-authored-by: Christian Heimes <christian@python.org>
Co-authored-by: Victor Stinner <vstinner@python.org>
Move private debug _PyObject functions to the internal C API
(pycore_object.h):
* _PyDebugAllocatorStats()
* _PyObject_CheckConsistency()
* _PyObject_DebugTypeStats()
* _PyObject_IsFreed()
No longer export most of these functions, except of
_PyObject_IsFreed().
Move test functions using _PyObject_IsFreed() from _testcapi to
_testinternalcapi. check_pyobject_is_freed() test no longer catch
_testcapi.error: the tested function cannot raise _testcapi.error.
The risk of a race with this state is relatively low, but we play it safe anyway. We do avoid using the lock in performance-sensitive cases where the risk of a race is very, very low.
This is strictly about moving the "obmalloc" runtime state from
`_PyRuntimeState` to `PyInterpreterState`. Doing so improves isolation
between interpreters, specifically most of the memory (incl. objects)
allocated for each interpreter's use. This is important for a
per-interpreter GIL, but such isolation is valuable even without it.
FWIW, a per-interpreter obmalloc is the proverbial
canary-in-the-coalmine when it comes to the isolation of objects between
interpreters. Any object that leaks (unintentionally) to another
interpreter is highly likely to cause a crash (on debug builds at
least). That's a useful thing to know, relative to interpreter
isolation.
Add `MS_WINDOWS_DESKTOP`, `MS_WINDOWS_APPS`, `MS_WINDOWS_SYSTEM` and `MS_WINDOWS_GAMES` preprocessor definitions to allow switching off functionality missing from particular API partitions ("partitions" are used in Windows to identify overlapping subsets of APIs).
CPython only officially supports `MS_WINDOWS_DESKTOP` and `MS_WINDOWS_SYSTEM` (APPS is included by normal desktop builds, but APPS without DESKTOP is not covered). Other configurations are a convenience for people building their own runtimes.
`MS_WINDOWS_GAMES` is for the Xbox subset of the Windows API, which is also available on client OS, but is restricted compared to `MS_WINDOWS_DESKTOP`. These restrictions may change over time, as they relate to the build headers rather than the OS support, and so we assume that Xbox builds will use the latest available version of the GDK.
The global allocators were stored in 3 static global variables: _PyMem_Raw, _PyMem, and _PyObject. State for the "small block" allocator was stored in another 13. That makes a total of 16 global variables. We are moving all 16 to the _PyRuntimeState struct as part of the work for gh-81057. (If PEP 684 is accepted then we will follow up by moving them all to PyInterpreterState.)
https://github.com/python/cpython/issues/81057
Move _Py_GetAllocatedBlocks() and _PyObject_DebugMallocStats()
declarations to pycore_pymem.h. These functions are related to memory
allocators, not to the PyObject structure.
MAP_BOT_LENGTH was incorrectly used to compute MAP_TOP_MASK instead of
MAP_TOP_LENGTH. On 64-bit machines, the error causes the tree to hold
46-bits of virtual addresses, rather than the intended 48-bits.
* Remove 'zombie' frames. We won't need them once we are allocating fixed-size frames.
* Add co_nlocalplus field to code object to avoid recomputing size of locals + frees + cells.
* Move locals, cells and freevars out of frame object into separate memory buffer.
* Use per-threadstate allocated memory chunks for local variables.
* Move globals and builtins from frame object to per-thread stack.
* Move (slow) locals frame object to per-thread stack.
* Move internal frame functions to internal header.
When printing stats, move radix tree info to its own section.
Restore that the breakdown of bytes in arenas exactly accounts for the total of arena bytes allocated.
Add an assert so that invariant doesn't break again.
The radix tree approach is a relatively simple and memory sanitary
alternative to the old (slightly) unsanitary address_in_range().
To disable the radix tree map, set a preprocessor flag as follows:
-DWITH_PYMALLOC_RADIX_TREE=0.
Co-authored-by: Tim Peters <tim.peters@gmail.com>
The PEP 353, written in 2005, introduced PY_FORMAT_SIZE_T. Python no
longer supports macOS 10.4 and Visual Studio 2010, but requires more
recent macOS and Visual Studio versions. In 2020 with Python 3.10, it
is now safe to use directly "%zu" to format size_t and "%zi" to
format Py_ssize_t.
The Py_FatalError() function is replaced with a macro which logs
automatically the name of the current function, unless the
Py_LIMITED_API macro is defined.
Changes:
* Add _Py_FatalErrorFunc() function.
* Remove the function name from the message of Py_FatalError() calls
which included the function name.
* Update tests.
bpo-3605, bpo-38733: Optimize _PyErr_Occurred(): remove "tstate ==
NULL" test.
Py_FatalError() no longer calls PyErr_Occurred() if called without
holding the GIL. So PyErr_Occurred() no longer has to support
tstate==NULL case.
_Py_CheckFunctionResult(): use directly _PyErr_Occurred() to avoid
explicit "!= NULL" test.
pymalloc_alloc() now returns directly the pointer, return NULL on
memory allocation error.
allocate_from_new_pool() already uses NULL as marker for "allocation
failed".
Sam Gross
Stefano Rivera
Xie Yanbo
Dino Viehland
Neil Schemenauer
Dino Viehland
Victor Stinner
Eric Snow
Brandt Bucher
Max Bachmann
Benjamin Peterson
neonene
oda-gitso
Mark Shannon
Christian Clauss
Tim Peters
Batuhan Taşkaya
Victor Stinner