[Interpreter] Fix localloc to guarantee 16-byte stack alignment by Copilot · Pull Request #125720 · dotnet/runtime

Copilot · 2026-03-18T09:39:54Z

FrameDataAllocator::Alloc() was aligning localloc buffers to sizeof(void*) (4–8 bytes) instead of INTERP_STACK_ALIGNMENT (16 bytes), causing the ShowLocallocAlignment regression test to fail on any platform using the CoreCLR interpreter (e.g., WASM).

Description

Fragment backing memory: Replace malloc with VMToOSInterface::AlignedAllocate(INTERP_STACK_ALIGNMENT, size) so every fragment's base address is 16-byte aligned. Pair free → VMToOSInterface::AlignedFree accordingly.
Bump allocation alignment: Change ALIGN_UP(size, sizeof(void*)) → ALIGN_UP(size, INTERP_STACK_ALIGNMENT) so pFramePos advances by a multiple of 16, keeping it aligned across all subsequent allocations and after PopInfo restores.

VMToOSInterface is already available in all cee_wks_core VM files via common.h, which is force-included as a precompiled header (see src/coreclr/vm/wks/CMakeLists.txt). No additional include is needed.

Customer Impact

Programs using stackalloc (localloc) under the CoreCLR interpreter receive misaligned pointers. On WASM and other platforms requiring 16-byte alignment (e.g., for SIMD), this causes incorrect behavior or crashes.

Regression

Not a regression in a released build; this is a correctness gap in the interpreter that has existed since its introduction.

Testing

Existing JIT/Regression/Dev11/External/dev11_239804/ShowLocallocAlignment test covers this scenario. Full clr+libs build verified clean on Linux x64.

Risk

Low. The change is isolated to interpframeallocator.cpp — a single file only compiled under FEATURE_INTERPRETER. Both allocation paths (AlignedAllocate/AlignedFree) are already used elsewhere in the interpreter for the main stack (interpexec.cpp). No behavioral change outside the interpreter.

Original prompt

This section details on the original issue you should resolve

<issue_title>[Interpreter] localloc does not guarantee platform stack alignment</issue_title>
<issue_description>FrameDataAllocator::Alloc() in src/coreclr/vm/interpframeallocator.cpp aligns localloc buffer sizes to sizeof(void*) (4 bytes on WASM, 8 bytes on 64-bit platforms):
size = ALIGN_UP(size, sizeof(void*));
The JIT guarantees localloc returns pointers aligned to STACK_ALIGN (16 bytes on most platforms including WASM — see src/coreclr/jit/targetwasm.h), but the interpreter does not honor this. The fragment backing memory comes from raw malloc() with no explicit alignment, and subsequent bump allocations within a fragment are only pointer-size aligned.

This causes the ShowLocallocAlignment regression test to fail on WASM (and would fail on any platform using the interpreter) because stackalloc returns 8-byte aligned addresses instead of the expected 16-byte alignment.

Repro

Run JIT/Regression/Dev11/External/dev11_239804/ShowLocallocAlignment on a platform using the CoreCLR interpreter (e.g. WASM).

Expected

Localloc returns a 16-byte aligned pointer (matching STACK_ALIGN).

Actual

Localloc returns a pointer aligned only to sizeof(void*).

Suggested fix

Change the alignment in FrameDataAllocator::Alloc() to use INTERP_STACK_ALIGNMENT (16) instead of sizeof(void*), and ensure the returned pointer (not just the size) is aligned.</issue_description>

Comments on the Issue (you are @copilot in this section)
@radekdoulik @janvorli @BrzVlad should we align FrameDataAllocator::Alloc() to INTERP_STACK_ALIGNMENT? @BrzVlad Sounds good. We could use AlignedAllocate here as well.

Fixes [Interpreter] localloc does not guarantee platform stack alignment #125657

📱 Kick off Copilot coding agent tasks wherever you are with GitHub Mobile, available on iOS and Android.

dotnet-policy-service · 2026-03-18T10:01:38Z

Tagging subscribers to this area: @BrzVlad, @janvorli, @kg
See info in area-owners.md if you want to be subscribed.

…NMENT Co-authored-by: radekdoulik <4147484+radekdoulik@users.noreply.github.com>

src/coreclr/vm/interpframeallocator.cpp

Co-authored-by: BrzVlad <4720621+BrzVlad@users.noreply.github.com>

Copilot

Pull request overview

Fixes CoreCLR interpreter localloc alignment to guarantee 16-byte alignment (matching INTERP_STACK_ALIGNMENT), addressing failures in ShowLocallocAlignment and preventing misaligned stackalloc buffers on interpreter platforms (e.g., WASM).

Changes:

Allocate frame data fragments with VMToOSInterface::AlignedAllocate(INTERP_STACK_ALIGNMENT, ...) and free with AlignedFree.
Align bump allocations in FrameDataAllocator::Alloc() to INTERP_STACK_ALIGNMENT instead of sizeof(void*).

Comments suppressed due to low confidence (1)

src/coreclr/vm/interpframeallocator.cpp:94

If the first fragment allocation fails (or a fragment’s backing allocation fails), pFirst can be left non-null while pCurrent is still null, and the partially-constructed fragment isn’t deleted. This can corrupt the allocator state: after the nullptr return triggers OutOfMemoryException, subsequent Alloc calls will skip the pFirst == nullptr path and then dereference pCurrent.

Consider allocating into a temporary FrameDataFragment* and only assigning to pFirst/pCurrent after verifying pFrameStart != nullptr; otherwise delete the fragment before returning failure (same pattern should be applied for later fragment allocations too).

    if (pFirst == nullptr)
    {
        pFirst = new (nothrow) FrameDataFragment(size);
        if (pFirst == nullptr || pFirst->pFrameStart == nullptr)
        {
            return nullptr;
        }

radekdoulik

LGTM

@radekdoulik

…et#125720) `FrameDataAllocator::Alloc()` was aligning localloc buffers to `sizeof(void*)` (4–8 bytes) instead of `INTERP_STACK_ALIGNMENT` (16 bytes), causing the `ShowLocallocAlignment` regression test to fail on any platform using the CoreCLR interpreter (e.g., WASM). ## Description - **Fragment backing memory**: Replace `malloc` with `VMToOSInterface::AlignedAllocate(INTERP_STACK_ALIGNMENT, size)` so every fragment's base address is 16-byte aligned. Pair `free` → `VMToOSInterface::AlignedFree` accordingly. - **Bump allocation alignment**: Change `ALIGN_UP(size, sizeof(void*))` → `ALIGN_UP(size, INTERP_STACK_ALIGNMENT)` so `pFramePos` advances by a multiple of 16, keeping it aligned across all subsequent allocations and after `PopInfo` restores. `VMToOSInterface` is already available in all `cee_wks_core` VM files via `common.h`, which is force-included as a precompiled header (see `src/coreclr/vm/wks/CMakeLists.txt`). No additional include is needed. ## Customer Impact Programs using `stackalloc` (localloc) under the CoreCLR interpreter receive misaligned pointers. On WASM and other platforms requiring 16-byte alignment (e.g., for SIMD), this causes incorrect behavior or crashes. ## Regression Not a regression in a released build; this is a correctness gap in the interpreter that has existed since its introduction. ## Testing Existing `JIT/Regression/Dev11/External/dev11_239804/ShowLocallocAlignment` test covers this scenario. Full `clr+libs` build verified clean on Linux x64. ## Risk Low. The change is isolated to `interpframeallocator.cpp` — a single file only compiled under `FEATURE_INTERPRETER`. Both allocation paths (`AlignedAllocate`/`AlignedFree`) are already used elsewhere in the interpreter for the main stack (`interpexec.cpp`). No behavioral change outside the interpreter.  <details> <summary>Original prompt</summary> > > ---- > > *This section details on the original issue you should resolve* > > <issue_title>[Interpreter] localloc does not guarantee platform stack alignment</issue_title> > <issue_description>`FrameDataAllocator::Alloc()` in `src/coreclr/vm/interpframeallocator.cpp` aligns localloc buffer sizes to `sizeof(void*)` (4 bytes on WASM, 8 bytes on 64-bit platforms): > > ```cpp > size = ALIGN_UP(size, sizeof(void*)); > ``` > > The JIT guarantees localloc returns pointers aligned to `STACK_ALIGN` (16 bytes on most platforms including WASM — see `src/coreclr/jit/targetwasm.h`), but the interpreter does not honor this. The fragment backing memory comes from raw `malloc()` with no explicit alignment, and subsequent bump allocations within a fragment are only pointer-size aligned. > > This causes the `ShowLocallocAlignment` regression test to fail on WASM (and would fail on any platform using the interpreter) because `stackalloc` returns 8-byte aligned addresses instead of the expected 16-byte alignment. > > ### Repro > > Run `JIT/Regression/Dev11/External/dev11_239804/ShowLocallocAlignment` on a platform using the CoreCLR interpreter (e.g. WASM). > > ### Expected > > Localloc returns a 16-byte aligned pointer (matching `STACK_ALIGN`). > > ### Actual > > Localloc returns a pointer aligned only to `sizeof(void*)`. > > ### Suggested fix > > Change the alignment in `FrameDataAllocator::Alloc()` to use `INTERP_STACK_ALIGNMENT` (16) instead of `sizeof(void*)`, and ensure the returned pointer (not just the size) is aligned.</issue_description> > > ## Comments on the Issue (you are @copilot in this section) > > <comments> > <comment_new><author>@radekdoulik</author><body> > @janvorli @BrzVlad should we align FrameDataAllocator::Alloc() to INTERP_STACK_ALIGNMENT?</body></comment_new> > <comment_new><author>@BrzVlad</author><body> > Sounds good. We could use AlignedAllocate here as well. </body></comment_new> > </comments> > </details>  - Fixes dotnet#125657  --- 📱 Kick off Copilot coding agent tasks wherever you are with [GitHub Mobile](https://gh.io/cca-mobile-docs), available on iOS and Android. --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: radekdoulik <4147484+radekdoulik@users.noreply.github.com> Co-authored-by: BrzVlad <4720621+BrzVlad@users.noreply.github.com>

Initial plan

96f1543

Copilot AI assigned Copilot and radekdoulik Mar 18, 2026

Copilot started work on behalf of radekdoulik March 18, 2026 09:39 View session

github-actions bot added the area-CodeGen-Interpreter-coreclr label Mar 18, 2026

Fix localloc alignment in FrameDataAllocator to use INTERP_STACK_ALIG…

bd89e7c

…NMENT Co-authored-by: radekdoulik <4147484+radekdoulik@users.noreply.github.com>

Copilot AI changed the title ~~[WIP] [Interpreter] Fix localloc alignment issue for stack~~ [Interpreter] Fix localloc to guarantee 16-byte stack alignment Mar 18, 2026

Copilot AI requested a review from radekdoulik March 18, 2026 10:26

Copilot finished work on behalf of radekdoulik March 18, 2026 10:26

BrzVlad reviewed Mar 18, 2026

View reviewed changes

src/coreclr/vm/interpframeallocator.cpp Outdated Show resolved Hide resolved

Copilot started work on behalf of BrzVlad March 18, 2026 10:45 View session

Remove spurious #include <minipal.h> from interpframeallocator.cpp

648462e

Co-authored-by: BrzVlad <4720621+BrzVlad@users.noreply.github.com>

Copilot AI requested a review from BrzVlad March 18, 2026 10:56

Copilot finished work on behalf of BrzVlad March 18, 2026 10:56

This was referenced Mar 18, 2026

System.IO.Tests.Directory_Create_Tests.FileSystemWatcher_Directory_Create_InNestedDirectory fails with "Created event did not occur as expected" #103630

Closed

MsQuic fails with QUIC_STATUS_OUT_OF_MEMORY on AzureLinux #123216

Open

radekdoulik marked this pull request as ready for review March 18, 2026 16:19

Copilot AI review requested due to automatic review settings March 18, 2026 16:19

radekdoulik added this to the Future milestone Mar 18, 2026

Copilot started reviewing on behalf of radekdoulik March 18, 2026 16:24 View session

Copilot AI reviewed Mar 18, 2026

View reviewed changes

build-analysis bot mentioned this pull request Mar 19, 2026

XHarness package install failure on iOS due to devicectl NSPOSIXErrorDomain error 49 #123796

Open

radekdoulik approved these changes Mar 30, 2026

View reviewed changes

radekdoulik enabled auto-merge (squash) March 30, 2026 21:48

radekdoulik mentioned this pull request Mar 30, 2026

[wasm][coreclr] fix and run more runtime tests, run them on CI #120708

Open

31 tasks

BrzVlad approved these changes Mar 31, 2026

View reviewed changes

radekdoulik merged commit 2f94e9d into main Mar 31, 2026
107 of 110 checks passed

radekdoulik deleted the copilot/fix-localloc-stack-alignment branch March 31, 2026 09:21

dotnet-maestro bot mentioned this pull request Apr 1, 2026

[main] Source code updates from dotnet/runtime dotnet/dotnet#5823

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Interpreter] Fix localloc to guarantee 16-byte stack alignment#125720

[Interpreter] Fix localloc to guarantee 16-byte stack alignment#125720
radekdoulik merged 3 commits intomainfrom
copilot/fix-localloc-stack-alignment

Copilot AI commented Mar 18, 2026 •

edited

Loading

Uh oh!

dotnet-policy-service bot commented Mar 18, 2026

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

radekdoulik left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

Copilot AI commented Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Customer Impact

Regression

Testing

Risk

Repro

Expected

Actual

Suggested fix

Comments on the Issue (you are @copilot in this section)

Uh oh!

dotnet-policy-service bot commented Mar 18, 2026

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

radekdoulik left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Copilot AI commented Mar 18, 2026 •

edited

Loading