[wasm][R2R] Fix calls through MethodImpl-redirected slots on portable entrypoints by lewing · Pull Request #136050 · dotnet/runtime

lewing · 2026-10-19T01:52:36Z

With a same-type .override, a declaration's vtable slot holds the implementation's entrypoint, so on non-portable targets callers that start from the declaration still reach the implementation through the slot. With portable entrypoints (Wasm), three paths ran or prepared the declaration instead (self_override5):

ExternalMethodFixupWorker prestubbed the declaration before reading the slot, publishing the declaration's own R2R body and asserting in Checked. It now reads the slot, prepares the method that owns that entry, and patches the cell with it, like the non-portable branch.
ExecuteInterpretedMethodWithArgs_PortableEntryPoint_Complex (R2R → interpreter) prepared and ran the declaration, for example for a delegate bound by name to it. It now maps decl→impl first, as PrepareInterpreterCode already does.
The same entry's early path asserted in Checked when the implementation's interpreter code had been stored on the declaration (pre-existing). It now runs as the method that owns that code.

Removes the #134949 quarantine on self_override5.

Contributes to #134949

Scope and validation

The other #134949 failure (overflow0[1-4]_div) occurs only when R2R code runs with an interpreted CoreLib, which review settled is unsupported, so those quarantines stay. CallDescrWorkerInternal's portable path also prestubs without mapping; not reproduced, possible follow-up.

Validated on WASI (Checked runtime, wasmtime) with both an R2R CoreLib and the CI configuration: the MethodImpl tests, 20 CovariantReturns tests and jit64_5 pass under R2R and the interpreter. Not validated: browser and Release runtime builds.

Note

This pull request description was drafted with the help of GitHub Copilot.

… targets Crossgen2 emitted R2R-to-interpreter thunks only for call-site signatures. When a JIT helper is backed by a managed CoreLib method that is not in the image, the method runs in the interpreter, and the helper's import cell can only reach it through a thunk of the helper's signature. Small composites that lacked a call site with a matching signature left the helper's portable entrypoint without native code, so R2R code trapped with "uninitialized element". Register the thunk at each helper use. On portable entrypoints, ExternalMethodFixupWorker prestubbed the declared method even when a MethodImpl in the same type redirected its slot to another method, tripping the !ShouldCallPrestub() assert. Prepare and call the MethodImpl target instead. TryGetMultiCallableAddrOfCode likewise handed out the declared method's entrypoint, so delegates and ldftn ran the substituted body; return the MethodImpl target's entrypoint, as FuncPtrStubs does on other platforms. Remove the WASI R2R quarantines for overflow0[1-4]_div and self_override5. Fixes #134949 Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: fc3c5462-a1aa-4727-b4d2-d43b9bbb16d9

azure-pipelines · 2026-10-19T01:53:26Z

Azure Pipelines: Successfully started running 3 pipeline(s). 13 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run.

dotnet-policy-service · 2026-10-19T01:53:49Z

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

dotnet-policy-service · 2026-10-19T01:53:51Z

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

jkotas · 2026-10-19T02:32:47Z

When that method isn't in the image, it runs in the interpreter

Can we declare this configuration to be invalid? I do not think it is interesting to chase bugs in configurations where Corelib is interpreted, but the app is R2R.

R2R code outside CoreLib can be used only when CoreLib is R2R too
CoreLib R2R image must contain code for all JIT helpers

lewing · 2026-10-19T16:08:57Z

When that method isn't in the image, it runs in the interpreter

Can we declare this configuration to be invalid? I do not think it is interesting to chase bugs in configurations where Corelib is interpreted, but the app is R2R.

R2R code outside CoreLib can be used only when CoreLib is R2R too

CoreLib R2R image must contain code for all JIT helpers

Yeah that is probably the better choice, this is mostly a synthetic test problem due to the runtime tests being difficult to produce in r2r composite mode. Right now apple mobile runtime tests only r2r CoreLib, and wasi only r2rs the test assembly. I'm double checking and will update shortly.

Managed JIT helpers only lack an R2R-to-interpreter thunk when CoreLib is interpreted while other code is R2R. Treat that configuration as unsupported (R2R code outside CoreLib requires an R2R CoreLib that contains all JIT helpers) instead of emitting helper thunks into every image. Restore the WASI R2R quarantines for overflow0[1-4]_div, which hit this in the current runtime-test lane where CoreLib is interpreted. The MethodImpl fixes are independent of that configuration and remain. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: fc3c5462-a1aa-4727-b4d2-d43b9bbb16d9

lewing · 2026-10-19T16:26:17Z

Agreed, and I've updated the PR accordingly. The crossgen2 helper-thunk change is gone, and the overflow*_div quarantines are back. What's left are the two MethodImpl fixes in ExternalMethodFixupWorker and TryGetMultiCallableAddrOfCode. Those don't depend on the CoreLib configuration: self_override5 hits the same !ShouldCallPrestub() assert with an R2R CoreLib, and passes with these changes.

Follow-ups:

WASI test lane. It currently compiles only the test assemblies, because WASI has no runtime linking and the composer can merge just one R2R image into corerun. I have a working prototype that merges the prebuilt R2R System.Private.CoreLib.wasm alongside the test composite. That puts tests on R2R CoreLib without recompiling CoreLib per test.
Browser publishing. It can currently produce this configuration: PublishReadyToRunExclude of System.Private.CoreLib.dll (non-composite is the default on browser) publishes an IL CoreLib next to R2R app code. With only the app R2R, that hits the same null-funcref trap. I'll add an SDK error for excluding CoreLib from ReadyToRun on wasm, so the unsupported configuration can't be produced.

Note

This comment was drafted with the help of GitHub Copilot.

Address review feedback on the portable-entrypoint MethodImpl handling: - ExternalMethodFixupWorker: like the non-portable branch, call through pMD's slot (GetMethodEntryPoint) and prepare the method that owns the slot's entry, instead of mapping pMD with MapMethodDeclToMethodImpl. - Drop the TryGetMultiCallableAddrOfCode change. Non-portable targets hand out the declared method's own entrypoint there too; the redirection happens when ThePreStub jumps to DoBackpatch's result, the slot target. The portable equivalent is the R2R-to-interpreter entry, which now maps decl to impl (as the interpreter's PrepareInterpreterCode does) before preparing and running the method, instead of ignoring the redirection. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: fc3c5462-a1aa-4727-b4d2-d43b9bbb16d9

When the interpreter reaches a declaration whose slot a same-type MethodImpl redirects, PrepareInterpreterCode stores the implementation's interpreter code on the declaration. A later call from R2R code through the declaration's portable entrypoint took ExecuteInterpretedMethodWithArgs_ PortableEntryPoint_Complex's early path with that code, and in Checked builds asserted that the declaration's entrypoint had interpreter data. Run as the method that owns the interpreter code instead. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: fc3c5462-a1aa-4727-b4d2-d43b9bbb16d9

Apply review suggestions: when the method already has interpreter code, run as the method that code was compiled for; otherwise prestub. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: fc3c5462-a1aa-4727-b4d2-d43b9bbb16d9

lewing · 2026-10-20T02:28:46Z

/ba-g All failures unrelated: JIT_r is #135966; TlsSessionTests is #136096; System.Runtime.Tests timeouts on wasm are the ControlledExecutionTests regression fixed by #136116.

lewing requested a review from MichalStrehovsky as a code owner October 19, 2026 01:52

github-actions Bot added the area-ReadyToRun label Oct 19, 2026

lewing added the arch-wasm WebAssembly architecture label Oct 19, 2026

dotnet-policy-service Bot assigned lewing Oct 19, 2026

lewing mentioned this pull request Oct 19, 2026

[wasm][R2R] Interpreted managed JIT helpers depend on per-image R2R-to-interpreter thunks #136051

Closed

lewing requested a review from davidwrighton October 19, 2026 02:31

lewing changed the title ~~[wasi][R2R] Fix traps calling interpreted managed helpers and MethodImpl targets~~ [wasm][R2R] Fix calls through MethodImpl-redirected slots on portable entrypoints Oct 19, 2026

davidwrighton reviewed Oct 19, 2026

View reviewed changes

Comment thread src/coreclr/vm/method.cpp Outdated

Comment thread src/coreclr/vm/prestub.cpp Outdated

lewing and others added 2 commits October 19, 2026 13:09

davidwrighton approved these changes Oct 19, 2026

View reviewed changes

lewing enabled auto-merge (squash) October 19, 2026 19:11

build-analysis Bot mentioned this pull request Oct 19, 2026

AcceptWithDefaultValidation_CertificateContextTrust_HonorsChainPolicy test fails with TimeoutException #136096

Closed

jkotas reviewed Oct 19, 2026

View reviewed changes

Comment thread src/coreclr/vm/prestub.cpp Outdated

jkotas reviewed Oct 19, 2026

View reviewed changes

Comment thread src/coreclr/vm/prestub.cpp Outdated

jkotas approved these changes Oct 19, 2026

View reviewed changes

lewing merged commit 046f365 into main Oct 20, 2026
107 of 109 checks passed

lewing deleted the lewing-wasi-r2r-uninitialized-element-fix branch October 20, 2026 02:29

lewing mentioned this pull request Oct 20, 2026

[wasi][R2R] Runtime test failures: prestub assert in self_override5, uninitialized element in overflow01_div #134949

Open

dotnet-maestro Bot mentioned this pull request Oct 21, 2026

[main] Source code updates from dotnet/runtime dotnet/dotnet#10348

Merged

dotnet-milestone-bot Bot added this to the 12.0-preview1 milestone Oct 21, 2026

Navigation Menu

Repository navigation

[wasm][R2R] Fix calls through MethodImpl-redirected slots on portable entrypoints - #136050

[wasm][R2R] Fix calls through MethodImpl-redirected slots on portable entrypoints#136050
lewing merged 5 commits into
mainfrom
lewing-wasi-r2r-uninitialized-element-fix

lewing commented Oct 19, 2026 •

edited

Loading

Uh oh!

azure-pipelines Bot commented Oct 19, 2026

Uh oh!

dotnet-policy-service Bot commented Oct 19, 2026

Uh oh!

dotnet-policy-service Bot commented Oct 19, 2026

Uh oh!

jkotas commented Oct 19, 2026

Uh oh!

lewing commented Oct 19, 2026

Uh oh!

lewing commented Oct 19, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

lewing commented Oct 20, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

lewing commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

azure-pipelines Bot commented Oct 19, 2026

Uh oh!

dotnet-policy-service Bot commented Oct 19, 2026

Uh oh!

dotnet-policy-service Bot commented Oct 19, 2026

Uh oh!

jkotas commented Oct 19, 2026

Uh oh!

lewing commented Oct 19, 2026

Uh oh!

lewing commented Oct 19, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

lewing commented Oct 20, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

lewing commented Oct 19, 2026 •

edited

Loading