gh-145615: Fix mimalloc page leak in the free-threaded build#145626
gh-145615: Fix mimalloc page leak in the free-threaded build#145626colesbury merged 4 commits intopython:mainfrom
Conversation
Fix three issues that caused mimalloc pages to be leaked until the owning thread exited: 1. In _PyMem_mi_page_maybe_free(), move pages out of the full queue when relying on QSBR to defer freeing the page. Pages in the full queue are never searched by mi_page_queue_find_free_ex(), so a page left there is unusable for allocations. 2. Move _PyMem_mi_page_clear_qsbr() from _mi_page_free_collect() to _mi_page_thread_free_collect() where it only fires when all blocks on the page are free (used == 0). The previous placement was too broad: it cleared QSBR state whenever local_free was non-NULL, but _mi_page_free_collect() is called from non-allocation paths (e.g., page visiting in mi_heap_visit_blocks) where the page is not being reused. 3. In _PyMem_mi_page_maybe_free(), use the page's heap tld to find the correct thread state for QSBR list insertion instead of PyThreadState_GET(). During stop-the-world pauses, the function may process pages belonging to other threads, so the current thread state is not necessarily the owner of the page.
This comment was marked as outdated.
This comment was marked as outdated.
|
🤖 New build scheduled with the buildbot fleet by @colesbury for commit 2516a1a 🤖 Results will be shown at: https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F145626%2Fmerge If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again. |
|
The changes look good to me, though I am not intimately familiar with mimalloc or our modifications to it. The buildbot failure seems like it might be real though. It looks like |
|
I don't think the buildbot failure is related. The PR only touches mimalloc and |
|
Thanks @colesbury for the PR 🌮🎉.. I'm working now to backport this PR to: 3.14. |
…ythongh-145626) Fix three issues that caused mimalloc pages to be leaked until the owning thread exited: 1. In _PyMem_mi_page_maybe_free(), move pages out of the full queue when relying on QSBR to defer freeing the page. Pages in the full queue are never searched by mi_page_queue_find_free_ex(), so a page left there is unusable for allocations. 2. Move _PyMem_mi_page_clear_qsbr() from _mi_page_free_collect() to _mi_page_thread_free_collect() where it only fires when all blocks on the page are free (used == 0). The previous placement was too broad: it cleared QSBR state whenever local_free was non-NULL, but _mi_page_free_collect() is called from non-allocation paths (e.g., page visiting in mi_heap_visit_blocks) where the page is not being reused. 3. In _PyMem_mi_page_maybe_free(), use the page's heap tld to find the correct thread state for QSBR list insertion instead of PyThreadState_GET(). During stop-the-world pauses, the function may process pages belonging to other threads, so the current thread state is not necessarily the owner of the page. (cherry picked from commit d76df75f51e662fd15ebe00e107058841de94860) Co-authored-by: Sam Gross <colesbury@gmail.com>
|
GH-145691 is a backport of this pull request to the 3.14 branch. |
…h-145626) (#145691) Fix three issues that caused mimalloc pages to be leaked until the owning thread exited: 1. In _PyMem_mi_page_maybe_free(), move pages out of the full queue when relying on QSBR to defer freeing the page. Pages in the full queue are never searched by mi_page_queue_find_free_ex(), so a page left there is unusable for allocations. 2. Move _PyMem_mi_page_clear_qsbr() from _mi_page_free_collect() to _mi_page_thread_free_collect() where it only fires when all blocks on the page are free (used == 0). The previous placement was too broad: it cleared QSBR state whenever local_free was non-NULL, but _mi_page_free_collect() is called from non-allocation paths (e.g., page visiting in mi_heap_visit_blocks) where the page is not being reused. 3. In _PyMem_mi_page_maybe_free(), use the page's heap tld to find the correct thread state for QSBR list insertion instead of PyThreadState_GET(). During stop-the-world pauses, the function may process pages belonging to other threads, so the current thread state is not necessarily the owner of the page. (cherry picked from commit d76df75) Co-authored-by: Sam Gross <colesbury@gmail.com>
Fix three issues that caused mimalloc pages to be leaked until the owning thread exited:
In
_PyMem_mi_page_maybe_free(), move pages out of the full queue when relying on QSBR to defer freeing the page. Pages in the full queue are never searched bymi_page_queue_find_free_ex(), so a page left there is unusable for allocations.Move
_PyMem_mi_page_clear_qsbr()from_mi_page_free_collect()to_mi_page_thread_free_collect()where it only fires when all blocks on the page are free (used == 0). The previous placement was too broad: it cleared QSBR state whenever local_free was non-NULL, but_mi_page_free_collect()is called from non-allocation paths (e.g., page visiting inmi_heap_visit_blocks) where the page is not being reused.In
_PyMem_mi_page_maybe_free(), use the page's heap tld to find the correct thread state for QSBR list insertion instead ofPyThreadState_GET(). During stop-the-world pauses, the function may process pages belonging to other threads, so the current thread state is not necessarily the owner of the page.