[ARM32] Eliminate red zone usage in runtime stubs by cshung · Pull Request #129398 · dotnet/runtime

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung · 2026-06-14T22:53:29Z

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment.

This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop):

NativeAOT interop thunks (ThunksMapping.cpp) — use ldr pc dispatch directly from r12, no stack intermediate. This also shrinks THUNK_SIZE from 20 to 12 bytes.
NativeAOT UniversalTransition — caller pushes args onto stack before branching; prolog reads them from known stack offsets after saving argument registers.
NativeAOT interface dispatch (DispatchResolve.S, StubDispatch.S) — PROLOG_PUSH/EPILOG_POP instead of red zone stores.
CoreCLR VTableCallStub — pre-indexed str / post-indexed ldr (actual push/pop).

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. Replace red zone reads/writes with explicit stack adjustments (push/pop) in: - NativeAOT interop thunks (ldr pc dispatch, no stack intermediate) - NativeAOT UniversalTransition (caller pushes args onto stack) - NativeAOT interface dispatch stubs (PROLOG_STACK_ALLOC instead of sub-SP stores) - CoreCLR VTableCallStub (pre-indexed str/post-indexed ldr) Guarded by FEATURE_AVOID_RED_ZONE, enabled for ARM32 non-Windows targets. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-06-14T22:54:52Z

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

MichalPetryka · 2026-06-15T05:29:23Z

Windows ARM32 has a well-defined red zone guarantee

Windows ARM32 is no longer supported.

Windows ARM32 is no longer supported, so every ARM32 target is Linux. The red zone avoidance is always needed — remove the preprocessor guard and delete the old red zone code paths entirely. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ldr pc dispatch needs only 12 bytes (mov r12 + ldr pc), no padding required. This increases thunks per page from 204 to 341 (67% more). Also shorten verbose comments per review feedback. Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- StubDispatch: use PROLOG_PUSH/EPILOG_POP {r1,r2} instead of manual STACK_ALLOC + str/ldr - UniversalTransition: replace interleaved ldr/push dance with a single PROLOG_PUSH {r0-r3} then load caller args from known stack offsets - Clean up stale red zone comments Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas · 2026-06-15T20:10:00Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-15T20:10:10Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas

Thanks!

jkotas · 2026-06-16T01:10:43Z

Segfaults in many linux arm32 NAOT tests

pushd .
chmod +rwx Microsoft.Extensions.Configuration.FileExtensions.Tests ^&^& ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml popd
===========================================================================================================
/root/helix/work/workitem/e /root/helix/work/workitem/e
DOTNET_DbgEnableMiniDump is set and the createdump binary does not exist: ./createdump
./RunTests.sh: line 173: 18 Segmentation fault (core dumped) ./Microsoft.Extensions.Configuration.FileExtensions.Tests -notrait category=IgnoreForCI -notrait category=OuterLoop -notrait category=failing -xml testResults.xml $RSP_FILE
/root/helix/work/workitem/e
----- end Mon Jun 15 09:53:04 PM UTC 2026 ----- exit code 139 ----------------------------------------------------------

Could you please take a look?

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas · 2026-06-17T06:16:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T06:16:51Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-17T17:04:41Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-17T17:04:52Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T00:17:41Z

All Arm32 test failures are known

Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

…USH/EPILOG_POP - DispatchResolve.S: use PROLOG_PUSH/EPILOG_POP for {r3,r4,r5,r6,r8}, add .save {r1,r2} at Hashtable entry, drop lr from push list - UniversalTransition.S: rewrite prolog to preserve original frame layout (push r0-r1, capture caller args, store r2-r3 into caller slots) - StackFrameIterator.cpp: revert to original UniversalTransitionStackFrame layout (no m_callerPushedArgs) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jkotas · 2026-06-18T18:00:05Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T18:00:18Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-06-18T21:07:21Z

/azp run runtime-nativeaot-outerloop

azure-pipelines · 2026-06-18T21:07:32Z

Azure Pipelines successfully started running 1 pipeline(s).

cshung · 2026-06-19T15:09:32Z

Thanks @jkotas and @MichalStrehovsky for the thorough review and guidance! The prolog trick to preserve the original frame layout was particularly elegant — avoiding the TransitionBlock.cs changes made this much cleaner. Appreciated the push toward idiomatic ARM patterns (PROLOG_PUSH/EPILOG_POP, r8 as scratch) and the thunk size reduction as a bonus. Learned a lot from this one.

jkotas · 2026-06-19T15:20:40Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

cshung · 2026-06-19T16:06:15Z

@cshung Thank you for fixing this! It is likely source of some of the intermittent arm32 crashes. How did you find the problem?

I am trying to get it to run on low-end devices without virtual memory support. On those platforms, using red zone will fail pretty easily.

On ARM32 Linux, the area below SP is not guaranteed to be preserved across signal delivery. The runtime previously used the red zone (writing below SP without adjusting it) in several stubs, which can cause silent corruption or crashes when a signal is delivered at the wrong moment. This PR eliminates all red zone usage in ARM32 runtime stubs by replacing sub-SP reads/writes with explicit stack adjustments (push/pop): --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

cshung requested a review from MichalStrehovsky as a code owner June 14, 2026 22:53

github-actionsBot added the area-NativeAOT-coreclr label Jun 14, 2026

dotnet-policy-serviceBot added the community-contribution Indicates that the PR has been added by a community member label Jun 14, 2026

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/vm/CMakeLists.txt Outdated

jkotas added the arch-arm32 label Jun 15, 2026

build-analysisBot mentioned this pull request Jun 15, 2026
The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008
Open
3 tasks

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/nativeaot/Runtime/ThunksMapping.cpp Outdated

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/nativeaot/Runtime/ThunksMapping.cpp Outdated

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/vm/arm/virtualcallstubcpu.hpp Outdated
Comment threadsrc/coreclr/vm/arm/virtualcallstubcpu.hpp Outdated

cshung force-pushed the feature-avoid-red-zone branch 2 times, most recently from 7cc9b73 to 59bc77cCompare June 15, 2026 17:31

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/nativeaot/Runtime/arm/UniversalTransition.S Outdated

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/runtime/arm/StubDispatch.S Outdated

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/runtime/arm/StubDispatch.S Outdated

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/runtime/arm/StubDispatch.S Outdated

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/runtime/arm/StubDispatch.S Outdated

jkotas reviewed Jun 15, 2026
View reviewed changes

Comment threadsrc/coreclr/nativeaot/Runtime/arm/UniversalTransition.S Outdated

cshungand others added 2 commits June 15, 2026 18:16

cshung force-pushed the feature-avoid-red-zone branch from 59bc77c to 87288dfCompare June 15, 2026 18:27

jkotas reviewed Jun 15, 2026
View reviewed changes

Apply suggestions from code review
a3f998e
Co-authored-by: Jan Kotas <jkotas@microsoft.com>

jkotas approved these changes Jun 15, 2026
View reviewed changes

Potential fix for pull request finding
5ffee39
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

jkotas reviewed Jun 17, 2026
View reviewed changes

Comment threadsrc/coreclr/nativeaot/Runtime/arm/DispatchResolve.S

Apply suggestion from@jkotas
bf61b67

build-analysisBot mentioned this pull request Jun 18, 2026
TestNativeDigits fails for ur-IN on Apple platforms (xunit v3 exposed previously non-running test) #125933
Closed

build-analysisBot mentioned this pull request Jun 18, 2026
slow macOS - "##[error]The job running on agent Azure Pipelines 9 ran longer than the maximum time of 60 minutes." dotnet/dnceng#1883
Open
3 tasks

MichalStrehovsky reviewed Jun 18, 2026
View reviewed changes

Comment threadsrc/coreclr/nativeaot/Runtime/arm/DispatchResolve.S
Comment threadsrc/coreclr/nativeaot/Runtime/EHHelpers.cpp
Comment threadsrc/coreclr/nativeaot/Runtime/StackFrameIterator.cpp Outdated

jkotasand others added 2 commits June 17, 2026 21:23

Apply suggestion from@MichalStrehovsky
bcae72e
Co-authored-by: Michal Strehovský <MichalStrehovsky@users.noreply.github.com>

jkotas approved these changes Jun 18, 2026
View reviewed changes

jkotas requested a review from MichalStrehovsky June 18, 2026 17:59

Merge branch 'main' into feature-avoid-red-zone
dc3a328

jkotas merged commit 083ad8b into dotnet:mainJun 19, 2026
137 checks passed

cshung deleted the feature-avoid-red-zone branch June 19, 2026 03:40

dotnet-maestroBot mentioned this pull request Jun 20, 2026
[main] Source code updates from dotnet/runtime dotnet/dotnet#7303
Merged

dotnet-milestone-botBot added this to the 11.0-preview6 milestone Jun 22, 2026

github-actionsBot locked and limited conversation to collaborators Jul 23, 2026