[Codegen][X86] Fix /HOTPATCH with clang-cl and inline asm (llvm#87639)

This fixes an edge case where functions starting with inline assembly
would assert while trying to lower that inline asm instruction.

After this PR, for now we always add a no-op (xchgw in this case) without
considering the size of the next inline asm instruction. We might want
to revisit this in the future.

This fixes Unreal Engine 5.3.2 compilation with clang-cl and /HOTPATCH.

Should close llvm#56234

(cherry picked from commit ec1af63)

llvm/lib/Target/X86/X86MCInstLower.cpp

@@ -959,8 +959,10 @@ void X86AsmPrinter::LowerPATCHABLE_OP(const MachineInstr &MI,
   SmallString<256> Code;
   unsigned MinSize = MI.getOperand(0).getImm();
 
-  if (NextMI != MI.getParent()->end()) {
+  if (NextMI != MI.getParent()->end() && !NextMI->isInlineAsm()) {
     // Lower the next MachineInstr to find its byte size.
+    // If the next instruction is inline assembly, we skip lowering it for now,
+    // and assume we should always generate NOPs.
     MCInst MCI;
     MCIL.Lower(&*NextMI, MCI);
 

llvm/test/CodeGen/X86/patchable-prologue.ll

@@ -193,3 +193,20 @@ do.body:                                          ; preds = %do.body, %entry
 do.end:                                           ; preds = %do.body
   ret void
 }
+
+
+; Test that inline asm is properly hotpatched. We currently don't examine the
+; asm instruction when printing it, thus we always emit patching NOPs.
+
+; 64: inline_asm:
+; 64-NEXT: # %bb.0:
+; 64-NEXT: xchgw   %ax, %ax                        # encoding: [0x66,0x90]
+; 64-NEXT: #APP
+; 64-NEXT: int3                                    # encoding: [0xcc]
+; 64-NEXT: #NO_APP
+
+define dso_local void @inline_asm() "patchable-function"="prologue-short-redirect" {
+entry:
+  call void asm sideeffect "int3", "~{dirflag},~{fpsr},~{flags}"()
+  ret void
+}