diff --git a/src/cmd/compile/internal/base/debug.go b/src/cmd/compile/internal/base/debug.go
index 5a40b1bb2a9414e5d6aafb50a60ed59f5812d56a..4a92364332951f484b1f5573d543e1fd1adfce23 100644
--- a/src/cmd/compile/internal/base/debug.go
+++ b/src/cmd/compile/internal/base/debug.go
@@ -73,7 +73,7 @@ type DebugFlags struct {
 	PGODevirtualize       int    `help:"enable profile-guided devirtualization; 0 to disable, 1 to enable interface devirtualization, 2 to enable function devirtualization" concurrent:"ok"`
 	CFGODebug             int    `help:"debug continuous feature guided optimizations"`
 	CFGOHash              string `help:"hash value for debugging continuous feature guided optimizations" concurrent:"ok"`
-	CFGOInline            int    `help:"enable continuous feature guided inlining" concurrent:"ok"`
+	CFGOInline            int    `help:"enable continuous feature guided inlining (0 => off, 1 => default, 2 => multi-level mode)" concurrent:"ok"`
 	CFGOInlineCDFThreshold string `help:"cumulative threshold percentage for determining call sites as hot candidates for inlining" concurrent:"ok"`
 	CFGOInlineBudget      int    `help:"inline budget for hot functions" concurrent:"ok"`
 	CFGODevirtualize      int    `help:"enable continuous feature guided devirtualization; 0 to disable, 1 to enable interface devirtualization, 2 to enable function devirtualization" concurrent:"ok"`
diff --git a/src/cmd/compile/internal/inline/inl.go b/src/cmd/compile/internal/inline/inl.go
index 2c518243e8b203346763a7de2cc3d86b922e4b07..96156da5fec2f428b4c8b390d82efe64c9a34379 100644
--- a/src/cmd/compile/internal/inline/inl.go
+++ b/src/cmd/compile/internal/inline/inl.go
@@ -945,7 +945,20 @@ func inlineCostOK(n *ir.CallExpr, caller, callee *ir.Func, bigCaller, closureCal
 	}
 
 	lineOffset := pgoir.NodeLineOffset(n, caller)
-	csi := pgoir.CallSiteInfo{LineOffset: lineOffset, Caller: caller}
+	parentCaller := caller
+	baseOffset := lineOffset
+	if base.ENABLE_CFGO && base.Debug.CFGOInline == 2 {
+		parentCallerName := ir.LinkFuncName(caller)
+		parentInlIdx := base.Ctxt.PosTable.Pos(n.Pos()).Base().InliningIndex()
+		if parentInlIdx >= 0 {
+			parentCallerName = base.Ctxt.InlTree.InlinedFunction(parentInlIdx).Name
+		}
+		if f, err := pgoir.LookupFunc(parentCallerName); err == nil {
+			parentCaller = f
+			lineOffset = pgoir.NodeLineOffset(n, parentCaller)
+		}
+	}
+	csi := pgoir.CallSiteInfo{LineOffset: lineOffset, Caller: parentCaller}
 	_, hot := candHotEdgeMap[csi]
 
 	if metric <= maxCost {
@@ -961,6 +974,12 @@ func inlineCostOK(n *ir.CallExpr, caller, callee *ir.Func, bigCaller, closureCal
 		return false, maxCost, metric, false
 	}
 
+	if base.ENABLE_CFGO && base.Debug.CFGODebug > 0 && base.Debug.CFGOInline == 2 && baseOffset != lineOffset {
+		fmt.Printf("changed offset from %v to %v; callee = %v, parent caller = %v (start line: %v), root caller = %v\n",
+		baseOffset, lineOffset, ir.LinkFuncName(callee), ir.LinkFuncName(parentCaller),
+		int(base.Ctxt.InnermostPos(parentCaller.Pos()).RelLine()), ir.LinkFuncName(caller))
+	}
+
 	// Hot
 
 	if bigCaller {
@@ -970,6 +989,16 @@ func inlineCostOK(n *ir.CallExpr, caller, callee *ir.Func, bigCaller, closureCal
 		return false, maxCost, metric, false
 	}
 
+	if base.ENABLE_CFGO && base.Debug.CFGOInline == 2 {
+		if caller.Pragma&ir.Nosplit != 0 && parentCaller != caller {
+			if base.Debug.CFGODebug > 0 {
+				fmt.Printf("%v: prevent inlining of hot call %s (cost %d) in function %s: root function is nosplit. Parent caller = %v\n", ir.Line(n),
+				ir.PkgFuncName(callee), callee.Inl.Cost, ir.PkgFuncName(caller), ir.LinkFuncName(parentCaller))
+			}
+			return false, maxCost, metric, false
+		}
+	}
+
 	if metric > inlineHotMaxBudget {
 		return false, inlineHotMaxBudget, metric, false
 	}