diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -3395,6 +3395,10 @@
     if (CallerPAL.hasParamAttr(i, Attribute::SwiftError))
       return false;
 
+    if (!CallerPAL.hasParamAttr(i, Attribute::ByVal) &&
+        Callee->getAttributes().hasParamAttr(i, Attribute::ByVal))
+      return false; // Cannot transform to byval.
+
     // If the parameter is passed as a byval argument, then we have to have a
     // sized type and the sized type has to have the same size as the old type.
     if (ParamTy != ActTy && CallerPAL.hasParamAttr(i, Attribute::ByVal)) {
diff --git a/llvm/test/LTO/X86/cast-to-byval.ll b/llvm/test/LTO/X86/cast-to-byval.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/LTO/X86/cast-to-byval.ll
@@ -0,0 +1,33 @@
+; Check that cross-module function calls involving conversion to byval aren't
+; inlined.
+
+; RUN: split-file %s %t
+; RUN: llvm-as %t/t1.ll -o %t1
+; RUN: llvm-as %t/t2.ll -o %t2
+; RUN: llvm-lto %t1 %t2 -o %t0 -filetype=asm -save-merged-module -exported-symbol=testFoo
+; RUN: llvm-dis -o - %t0.merged.bc | FileCheck %s
+
+
+;--- t1.ll
+target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
+target triple = "i686-unknown-linux-gnu"
+
+define dso_local noundef i32 @testFoo(ptr nocapture noundef nonnull readonly align 4 dereferenceable(4) %0) local_unnamed_addr {
+  %2 = load i32, ptr %0, align 4
+  %3 = tail call i32 @getFooValue(i32 noundef %2)
+; CHECK: %3 = tail call i32 @getFooValue(i32 noundef %2)
+  ret i32 %3
+}
+
+declare i32 @getFooValue(i32 noundef) local_unnamed_addr
+
+;--- t2.ll
+target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
+target triple = "i686-unknown-linux-gnu"
+
+%0 = type { i32 }
+
+define i32 @getFooValue(ptr noalias nocapture noundef readonly byval(%0) dereferenceable(4) %0) unnamed_addr {
+  %2 = load i32, ptr %0, align 4
+  ret i32 %2
+}