From 76595b904513c56bf919805e6fc666455ce4b31c Mon Sep 17 00:00:00 2001
From: David Wrighton <davidwr@microsoft.com>
Date: Thu, 24 Sep 2026 18:36:05 +0000
Subject: [PATCH] Avoid frame identity for leaf Wasm funclets

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---
 src/coreclr/jit/codegenwasm.cpp | 58 ++++++++++++++++++++++++++++++---
 src/coreclr/jit/compiler.h      |  1 +
 src/coreclr/jit/emitwasm.cpp    |  4 +++
 3 files changed, 58 insertions(+), 5 deletions(-)

diff --git a/src/coreclr/jit/codegenwasm.cpp b/src/coreclr/jit/codegenwasm.cpp
index 96d1443f5f75b1..aa39dbd52f2139 100644
--- a/src/coreclr/jit/codegenwasm.cpp
+++ b/src/coreclr/jit/codegenwasm.cpp
@@ -91,6 +91,49 @@ void CodeGen::genMarkLabelsForCodegen()
     //
     for (FuncInfoDsc* const func : m_compiler->Funcs())
     {
+        // Finalize call presence from the lowered IR before generating any code.
+        // BBF_HAS_CALL is a quick path, but funclet creation does not preserve it
+        // in every case, so inspect LIR for GT_CALL nodes too. Account separately
+        // for flow and codegen constructs that emit calls without a GT_CALL node.
+        //
+        for (BasicBlock* const block : func->Blocks(m_compiler))
+        {
+            if (block->HasFlag(BBF_HAS_CALL) || block->KindIs(BBJ_CALLFINALLY))
+            {
+                func->hasCalls = true;
+                break;
+            }
+
+            for (GenTree* const node : LIR::AsRange(block))
+            {
+                const bool hasCall = node->OperIs(GT_CALL);
+                const bool hasInlineThrowHelper =
+                    !m_compiler->fgUseThrowHelperBlocks() && ((node->gtFlags & GTF_EXCEPT) != 0);
+                const bool hasWriteBarrier =
+                    node->OperIs(GT_STOREIND) &&
+                    (gcInfo.gcIsWriteBarrierCandidate(node->AsStoreInd()) != GCInfo::WBF_NoBarrier);
+                const bool hasCpObjHelper = node->OperIs(GT_STORE_BLK) &&
+                                            (node->AsBlk()->gtBlkOpKind != GenTreeBlk::BlkOpKindNativeOpcode) &&
+                                            (node->AsBlk()->gtBlkOpKind != GenTreeBlk::BlkOpKindLoop);
+
+                if (hasCall || hasInlineThrowHelper || hasWriteBarrier || hasCpObjHelper)
+                {
+                    func->hasCalls = true;
+                    break;
+                }
+            }
+
+            if (func->hasCalls)
+            {
+                break;
+            }
+        }
+
+        if (func->hasCalls)
+        {
+            func->ensureUnwindableFrame(m_compiler);
+        }
+
         BasicBlock* const firstBlock = func->GetStartBlock(m_compiler);
         firstBlock->SetFlags(BBF_HAS_LABEL);
 
@@ -521,8 +564,7 @@ void CodeGen::genFuncletProlog(BasicBlock* block)
     // All the funclet params are used from their home registers, so nothing
     // needs homing here.
     //
-    // If the funclet needs to be unwindable (contains any calls), set up
-    // what we need.
+    // If the funclet needs to be unwindable, set up what we need.
     //
     if (func->needsUnwindableFrame)
     {
@@ -540,9 +582,15 @@ void CodeGen::genFuncletProlog(BasicBlock* block)
         GetEmitter()->emitIns(INS_I_sub);
         GetEmitter()->emitIns_I(INS_local_set, EA_PTRSIZE, GetStackPointerRegIndex());
 
-        GetEmitter()->emitIns_I(INS_local_get, EA_PTRSIZE, GetStackPointerRegIndex());
-        GetEmitter()->emitFuncletAddressConstant((cnsval_ssize_t)funcletIndex);
-        GetEmitter()->emitIns_I(ins_Store(TYP_I_IMPL), EA_PTRSIZE, 0);
+        // A leaf funclet can require a frame for EH virtual IP tracking, but no
+        // suspended caller needs its function identity for stack walking.
+        //
+        if (func->hasCalls)
+        {
+            GetEmitter()->emitIns_I(INS_local_get, EA_PTRSIZE, GetStackPointerRegIndex());
+            GetEmitter()->emitFuncletAddressConstant((cnsval_ssize_t)funcletIndex);
+            GetEmitter()->emitIns_I(ins_Store(TYP_I_IMPL), EA_PTRSIZE, 0);
+        }
     }
 }
 
diff --git a/src/coreclr/jit/compiler.h b/src/coreclr/jit/compiler.h
index 8392a0a609a4af..4b12d6d37cec3a 100644
--- a/src/coreclr/jit/compiler.h
+++ b/src/coreclr/jit/compiler.h
@@ -1841,6 +1841,7 @@ struct FuncInfoDsc
     jitstd::vector<WasmLocalsDecl>* funWasmLocalDecls;
     unsigned funWasmFrameSize;
     unsigned funWasmExnRefLocalIndex = UINT_MAX;
+    bool hasCalls;
     bool needsUnwindableFrame;
     emitLocation* startLoc;
     emitLocation* endLoc;
diff --git a/src/coreclr/jit/emitwasm.cpp b/src/coreclr/jit/emitwasm.cpp
index e8eb2e047730a7..63718c274a64b0 100644
--- a/src/coreclr/jit/emitwasm.cpp
+++ b/src/coreclr/jit/emitwasm.cpp
@@ -287,6 +287,10 @@ void emitter::emitIns_Call(const EmitCallParams& params)
 
     assert(params.wasmSignature != nullptr);
 
+    FuncInfoDsc* const func = m_compiler->funCurrentFunc();
+    assert(func->hasCalls);
+    func->hasCalls = true;
+
     /*
         We need to allocate the appropriate instruction descriptor based
         on whether this is a direct/indirect call, and whether we need to