[Wasm] Write barriers by kg · Pull Request #128225 · dotnet/runtime

kg · 2026-05-14T20:11:15Z

Sufficient for the R2R'd version of this to run without crashing:

    static string? TestField;

    [MethodImpl(MethodImplOptions.NoOptimization | MethodImplOptions.NoInlining)]
    public static void ConcatTwoStrings (string a, string b) {
        TestField = a + b;
    }

    [MethodImpl(MethodImplOptions.NoOptimization | MethodImplOptions.NoInlining)]
    public static int Main () {
        Console.WriteLine("Hello world!");
        ConcatTwoStrings("A", "B");
        Console.WriteLine(TestField);
        return 42;
    }

Currently the write barrier in ConcatTwoStrings fails due to the calling convention being wrong and then once the calling convention is fixed, it fails because we didn't implement the write barriers yet.

This PR:

Updates the calling convention of the write barriers to match that expected by the JIT (bare calling convention without sp/pep args) by adding new FCDECL2_RAW / FCIMPL2_RAW macros
Fixes RhpByRefAssignRef's signature being incorrect (missing *)
Implements C++ versions of the byref assign helpers for TARGET_WASM
Enables our C++ write barriers for TARGET_WASM (we were defining FEATURE_USE_ASM_GC_WRITE_BARRIERS before)

…ing convention on Wasm

dotnet-policy-service · 2026-05-14T20:12:37Z

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

Copilot

Pull request overview

Initial scaffolding to make the WASM portable write barriers (RhpAssignRef, RhpCheckedAssignRef, RhpByRefAssignRef) use a "raw" calling convention that matches what RyuJIT expects for write barriers, rather than the trampolined WASM FCALL convention that threads callersStackPointer / portableEntryPointContext through. The actual barrier implementations still assert and will be filled in later.

Changes:

Add new FCDECL2_RAW / FCIMPL2_RAW macros that bypass the WASM stack-pointer/entrypoint-context plumbing and define the function with the plain (a1, a2) signature; provide non-WASM fallbacks that alias to FCDECL2 / FCIMPL2.
Switch the three Rhp*AssignRef declarations in jitinterface.h and definitions in WriteBarriers.cpp to the new _RAW variants.
Trivial whitespace cleanup in jitinterface.h.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
`src/coreclr/vm/fcall.h`	Adds WASM-specific `FCDECL2_RAW` / `FCIMPL2_RAW` macros and non-WASM fallbacks aliasing to the standard `FCDECL2` / `FCIMPL2`.
`src/coreclr/vm/jitinterface.h`	Switches `Rhp{Checked,ByRef,}AssignRef` declarations to `FCDECL2_RAW`; trims trailing whitespace.
`src/coreclr/runtime/portable/WriteBarriers.cpp`	Updates the three portable write-barrier stubs to use `FCDECL2_RAW` / `FCIMPL2_RAW`.

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

 #define FCDECL1(rettype, funcname, a1) rettype F_CALL_CONV funcname(uintptr_t callersStackPointer, a1, PCODE portableEntryPointContext); rettype F_CALL_CONV funcname ## _IMPL(uintptr_t callersStackPointer, a1, PCODE portableEntryPointContext)
 #define FCDECL1_V(rettype, funcname, a1) rettype F_CALL_CONV funcname(uintptr_t callersStackPointer, a1, PCODE portableEntryPointContext); rettype F_CALL_CONV funcname ## _IMPL(uintptr_t callersStackPointer, a1, PCODE portableEntryPointContext)
 #define FCDECL2(rettype, funcname, a1, a2) rettype F_CALL_CONV funcname(uintptr_t callersStackPointer, a1, a2, PCODE portableEntryPointContext); rettype F_CALL_CONV funcname ## _IMPL(uintptr_t callersStackPointer, a1, a2, PCODE portableEntryPointContext)
+#define FCDECL2_RAW(rettype, funcname, a1, a2) rettype F_CALL_CONV funcname(a1, a2); rettype F_CALL_CONV funcname(a1, a2)


+#ifdef TARGET_WASM
+    *dst = ref;
+    JIT_WriteBarrier(dst, ref);
+#else
    PORTABILITY_ASSERT("RhpAssignRef is not yet implemented");
+#endif // TARGET_WASM
 }
 FCIMPLEND

-EXTERN_C FCDECL2(VOID, RhpCheckedAssignRef, Object **dst, Object *ref);
-FCIMPL2(VOID, RhpCheckedAssignRef, Object **dst, Object *ref)
+EXTERN_C FCDECL2_RAW(VOID, RhpCheckedAssignRef, Object **dst, Object *ref);
+FCIMPL2_RAW(VOID, RhpCheckedAssignRef, Object **dst, Object *ref)
 {
+#ifdef TARGET_WASM
+    *dst = ref;
+    JIT_CheckedWriteBarrier(dst, ref);
+#else
    PORTABILITY_ASSERT("RhpCheckedAssignRef is not yet implemented");
+#endif // TARGET_WASM
 }
 FCIMPLEND

-EXTERN_C FCDECL2(VOID, RhpByRefAssignRef, Object **dst, Object *ref);
-FCIMPL2(VOID, RhpByRefAssignRef, Object **dst, Object *ref)
+EXTERN_C FCDECL2_RAW(VOID, RhpByRefAssignRef, Object **dst, Object **ref);
+FCIMPL2_RAW(VOID, RhpByRefAssignRef, Object **dst, Object **ref)
 {
+#ifdef TARGET_WASM
+    Object *tmp = *ref;
+    *dst = tmp;
+    JIT_CheckedWriteBarrier(dst, tmp);


jkotas · 2026-05-14T21:02:39Z

 {
+#ifdef TARGET_WASM
+    *dst = ref;
+    JIT_WriteBarrier(dst, ref);


We should just move the implementation to here, and delete it where it is currently. We do not want to pay for the extra call.

jkotas · 2026-05-14T21:03:02Z

+EXTERN_C FCDECL2_RAW(VOID, RhpAssignRef, Object **dst, Object *ref);
+FCIMPL2_RAW(VOID, RhpAssignRef, Object **dst, Object *ref)
 {
+#ifdef TARGET_WASM


These TARGET_WASM ifdefs should not be needed.

jkotas · 2026-05-14T21:06:00Z

 #define HCIMPL1_RAW(rettype, funcname, a1) rettype F_CALL_CONV funcname(uintptr_t callersStackPointer, a1, PCODE portableEntryPointContext) {
 #define HCIMPL1_V(rettype, funcname, a1) rettype F_CALL_CONV funcname(uintptr_t callersStackPointer, a1, PCODE portableEntryPointContext) { HCIMPL_PROLOG(funcname)
 #define HCIMPL2(rettype, funcname, a1, a2) rettype F_CALL_CONV funcname(uintptr_t callersStackPointer, a1, a2, PCODE portableEntryPointContext) { HCIMPL_PROLOG(funcname)
-#define HCIMPL2_RAW(rettype, funcname, a1, a2) rettype F_CALL_CONV funcname(uintptr_t callersStackPointer, a1, a2, PCODE portableEntryPointContext) {


I do not think you can drop callersStackPointer. It would be assuming that the C/C++ compiler does not spill anything to the stack. It is very fragile assumption.

If we wanted to drop callersStackPointer arg, we would have to implement these in wasm assembly to make sure that nothing is spilled.

If we wanted to drop callersStackPointer arg, we would have to implement these in wasm assembly to make sure that nothing is spilled.

That is how I would expect these to be implemented. Inserting the SP argument is somewhat complex for the Jit in case of barriers, and it's worse CQ of course.

jkotas · 2026-05-14T21:07:11Z

+    Object *tmp = *ref;
+    *dst = tmp;
+    JIT_CheckedWriteBarrier(dst, tmp);


Suggested change

Object *tmp = *ref;

*dst = tmp;

JIT_CheckedWriteBarrier(dst, tmp);

JIT_CheckedWriteBarrier(dst, *ref);

JIT_CheckedWriteBarrier does the ref->dst assignment.

Interesting, thank you. I guess that's a difference between our WB and the nativeaot one.

jkotas · 2026-05-14T21:09:31Z

+FCIMPL2_RAW(VOID, RhpAssignRef, Object **dst, Object *ref)
 {
+#ifdef TARGET_WASM
+    *dst = ref;


Suggested change

*dst = ref;

jkotas · 2026-05-14T21:09:38Z

+FCIMPL2_RAW(VOID, RhpCheckedAssignRef, Object **dst, Object *ref)
 {
+#ifdef TARGET_WASM
+    *dst = ref;


Suggested change

*dst = ref;

Add FCDECL2_RAW and FCIMPL2_RAW so write barriers can use native call…

26af0e2

…ing convention on Wasm

Copilot AI review requested due to automatic review settings May 14, 2026 20:11

kg added the arch-wasm WebAssembly architecture label May 14, 2026

Copilot started reviewing on behalf of kg May 14, 2026 20:11 View session

github-actions Bot added the area-VM-coreclr label May 14, 2026

dotnet-policy-service Bot assigned kg May 14, 2026

Copilot AI reviewed May 14, 2026

View reviewed changes

kg added 3 commits May 14, 2026 13:48

Fix us not building the GC write barriers at all on Wasm

de3c6db

Make the write barriers FCDECL2_RAW

259144d

Implement C++ versions of RhpAssignRef et al for TARGET_WASM

a35761f

kg requested a review from Copilot May 14, 2026 20:56

Copilot started reviewing on behalf of kg May 14, 2026 20:56 View session

Copilot AI reviewed May 14, 2026

View reviewed changes

jkotas reviewed May 14, 2026

View reviewed changes

EgorBo mentioned this pull request May 14, 2026

Remove CORINFO_HELP_ASSIGN_BYREF ? #128227

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Wasm] Write barriers#128225

[Wasm] Write barriers#128225
kg wants to merge 4 commits into
dotnet:mainfrom
kg:wasm-writebarriers-callconv

kg commented May 14, 2026 •

edited

Loading

Uh oh!

dotnet-policy-service Bot commented May 14, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

jkotas May 14, 2026

Uh oh!

jkotas May 14, 2026

Uh oh!

jkotas May 14, 2026 •

edited

Loading

Uh oh!

SingleAccretion May 15, 2026

Uh oh!

jkotas May 14, 2026

Uh oh!

kg May 14, 2026

Uh oh!

jkotas May 14, 2026

Uh oh!

jkotas May 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

kg commented May 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dotnet-policy-service Bot commented May 14, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

jkotas May 14, 2026

Choose a reason for hiding this comment

Uh oh!

jkotas May 14, 2026

Choose a reason for hiding this comment

Uh oh!

jkotas May 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SingleAccretion May 15, 2026

Choose a reason for hiding this comment

Uh oh!

jkotas May 14, 2026

Choose a reason for hiding this comment

Uh oh!

kg May 14, 2026

Choose a reason for hiding this comment

Uh oh!

jkotas May 14, 2026

Choose a reason for hiding this comment

Uh oh!

jkotas May 14, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

kg commented May 14, 2026 •

edited

Loading

jkotas May 14, 2026 •

edited

Loading