[PATCH] D86528: [4/5] [MC] [Win64EH] Fill in FuncletOrFuncEnd if missing

[Martin Storsjö via Phabricator via llvm-commits]

mstorsjo added inline comments.


================
Comment at: llvm/lib/MC/MCStreamer.cpp:696
+  if (!CurFrame->FuncletOrFuncEnd)
+    CurFrame->FuncletOrFuncEnd = CurFrame->End;
 }
----------------
efriedma wrote:
> Instead of setting FuncletOrFuncEnd here, should we be calling EmitWinCFIFuncletOrFuncEnd somewhere?
I don't think so. EmitWinCFIEndProc is essentially called from two places; either from lib/CodeGen/AsmPrinter/WinException.cpp, which also does call EmitWinCFIFuncletOrFuncEnd properly, so there it shouldn't be possible to reach EndProc without FuncletOrFuncEnd being set. The only other case is MC/MCParser/COFFAsmParser.cpp, which has a 1:1 mapping between directives and calling these methods.


================
Comment at: llvm/lib/MC/MCWin64EH.cpp:531
+    // This can happen if unwind info is forced to be generated (with a
+    // .seh_handlerdata directive) directly after the start of the function.
     RawFuncLength = 0;
----------------
efriedma wrote:
> With the other changes in this patch, is it still possible for FuncletOrFuncEnd to be null?
Ah, right, yes, it shouldn't be possible. Even if we'd trigger `.seh_handlerdata` immediately, we would have a FuncletOrFuncEnd symbol (with a zero distance from the start), and we'd skip writing this due to it being empty anyway. So we should be able to make this a `report_fatal_error`, as it shouldn't be triggerable even with specially crafted input.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D86528/new/

https://reviews.llvm.org/D86528