[llvm-bugs] [Bug 35142] New: Mishandling SEH filters for available_externally functions

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=35142

            Bug ID: 35142
           Summary: Mishandling SEH filters for available_externally
                    functions
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Core LLVM classes
          Assignee: unassignedbugs at nondot.org
          Reporter: peter at pcc.me.uk
                CC: llvm-bugs at lists.llvm.org

$ cat 1.cc
static void try_body(int numerator, int denominator, int *myres) {
  *myres = numerator / denominator;
}

inline int safe_div(int numerator, int denominator, int *res) {
  int myres = 0;
  int success = 1;
  __try {
    try_body(numerator, denominator, &myres);
  } __except (1) {
    success = -42;
  }
  *res = myres;
  return success;
}

void *f() { return (void *)&safe_div; }
$ cat 2.cc
static void try_body(int numerator, int denominator, int *myres) {
  *myres = numerator / denominator;
}

inline int safe_div(int numerator, int denominator, int *res) {
  int myres = 0;
  int success = 1;
  __try {
    try_body(numerator, denominator, &myres);
  } __except (1) {
    success = -42;
  }
  *res = myres;
  return success;
}

void *g() { return (void *)&safe_div; }
$ clang-cl /c 1.cc 2.cc -flto=thin -m32
$ lld-link 1.obj 2.obj /nodefaultlib /entry:f /opt:lldlto=0 /lldsavetemps
LLVM ERROR: assembler label 'L?safe_div@@YAHHHPAH at Z$parent_frame_offset' can
not be undefined

What's going on here is that the linker is choosing 1.obj's definition of
safe_div as prevailing, and requesting that 2.obj's definition be dropped. It
does this by setting linkage to available_externally. This reveals two bugs.
The first is that the localrecover check in Verifier.cpp is insufficient; under
the current semantics it would need to account for available_externally linkage
by calling isDeclarationForLinker instead of isDeclaration.

Even if we did that, though, we'd still hit a verifier check for this test
case, and the fix for that bug is unclear. At the object file level, 2.obj
contains an SEH filter with information about the non-prevailing definition of
safe_seh. The fix I have in mind is that because it is impossible for there to
be an active stack frame for said non-prevailing definition, the behaviour of
the SEH filter is undefined, and we can simply treat the intrinsic call as
equivalent to unreachable if it refers to an undefined function.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20171031/2cf818a7/attachment.html>