cuda.core: reject register() on a non-IPC memory resource - #2569
Open
fedonman wants to merge 5 commits into
Open
cuda.core: reject register() on a non-IPC memory resource#2569fedonman wants to merge 5 commits into
fedonman wants to merge 5 commits into
Conversation
MP_register read self._ipc_data._alloc_handle._uuid unconditionally. _ipc_data is None whenever IPC is not enabled, and Cython compiles the chained access without a none check, so the call terminated the process with a segmentation fault instead of raising. No exception reached the caller, so the crash could not be guarded with try. Check is_ipc_enabled first and raise RuntimeError, matching the wording the allocation_handle property already uses. The check runs before the registry insertion, so a rejected registration no longer leaves an entry behind. Closes NVIDIA#2568 Signed-off-by: Vyron Vasileiadis <hi@fedonman.com>
Contributor
lijinf2
approved these changes
Aug 10, 2026
lijinf2
left a comment
lijinf2
left a comment
Contributor
There was a problem hiding this comment.
Thanks for the contribution — this turns a hard segfault into a clear RuntimeError that callers can actually handle.
Contributor
|
/ok to test a7da9e2 |
Contributor
|
@fedonman You may select cuda core next 1.2.0 under "Milestone" which should resolve the failing CI checks. |
|
Author
Contributor
Contributor
|
/ok to test ecb4ecf |
Andy-Jost
approved these changes
Aug 17, 2026
Contributor
|
The new test needs to be gated on mempool availability: |
Andy-Jost
requested changes
Aug 17, 2026
Andy-Jost
left a comment
Andy-Jost
left a comment
Contributor
There was a problem hiding this comment.
The new test needs to be gated on mempool availability.
Author
|
@Andy-Jost Thanks. Done. |
test_register_rejects_non_ipc_memory_resource constructed a DeviceMemoryResource from a bare Device(), which raises CUDA_ERROR_NOT_SUPPORTED on devices without memory pool support. Every win-64 TCC job in CI failed there before reaching the assertion under test. Take the mempool_device fixture instead, matching how the other mempool-dependent tests are gated. The fixture skips when memory_pools_supported is false and already makes the device current, so the manual Device()/set_current() pair is dropped.
fedonman
force-pushed
the
fix-ipc-register-non-ipc-guard
branch
from
August 18, 2026 08:03
f76c70f to
ff819a6
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Closes #2568.
DeviceMemoryResource.register()andPinnedMemoryResource.register()terminated the process with a segmentation fault when the memory resource did not have IPC enabled, including the default resource fromDevice().memory_resource. Both route throughMP_register, which readself._ipc_dataunconditionally although it isNoneunless IPC is enabled.Changes
_ipc.pyx: checkis_ipc_enabledinMP_registerand raiseRuntimeError("Memory resource is not IPC-enabled"), the wording theallocation_handleproperty already uses. The check runs before the registry insertion, so a rejected registration no longer leaves an entry behind.tests/memory_ipc/test_errors.py: regression test covering the raise and the registry.docs/source/release/1.2.0-notes.rst: release note. The defect is present in released 1.0.1, 1.1.0 and 1.1.1, so this is user-visible rather than confined to the current cycle.Verification
Without the production change the new test does not merely fail, it takes the interpreter down (
exit=139, SIGSEGV). With the change:No regressions in the surrounding suites: