[LLVMbugs] [Bug 17303] New: name of parameter Alloca is lost in LLVM_ENABLE_ASSERTIONS=OFF build of clang

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=17303

            Bug ID: 17303
           Summary: name of parameter Alloca is lost in
                    LLVM_ENABLE_ASSERTIONS=OFF build of clang
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: kcc at google.com
                CC: chandlerc at gmail.com, llvmbugs at cs.uiuc.edu,
                    samsonov at google.com
    Classification: Unclassified

AddressSanitizer relies on the AllocaInst name to produce informative results 
for stack-buffer-overflow and stack-use-after-return bugs. 

For local variables the AllocaInstr always has a name, whether the build 
has LLVM_ENABLE_ASSERTIONS=OFF or ON:

void foo(int *a);
void bar() {
 int ABCDE;
 foo(&ABCDE);
}

% clang++ -O -S -o - -emit-llvm stack2.cc | grep alloca
 %ABCDE = alloca i32, align 4
% noassert_clang++ -O -S -o - -emit-llvm stack2.cc | grep alloca
 %ABCDE = alloca i32, align 4

But for parameters the AllocaInst has a name only with assertion
LLVM_ENABLE_ASSERTIONS=ON, 
so with the release clang build AddressSanitizer produces a less informative
report.

void bar(int *a);
void foo(int ABC) {
 bar(&ABC);
}

% clang++ -O -S -o - -emit-llvm param.cc | grep alloca
 %ABC.addr = alloca i32, align 4
% noassert_clang++ -O -S -o - -emit-llvm param.cc | grep alloca
 %1 = alloca i32, align 4



IMHO, the names in AllocaInst should be present in all builds, and a very
simple patch is sufficient:

--- lib/CodeGen/CGExpr.cpp      (revision 191075)
+++ lib/CodeGen/CGExpr.cpp      (working copy)
@@ -52,8 +52,6 @@
 /// block.
 llvm::AllocaInst *CodeGenFunction::CreateTempAlloca(llvm::Type *Ty,
                                                     const Twine &Name) {
-  if (!Builder.isNamePreserving())
-    return new llvm::AllocaInst(Ty, 0, "", AllocaInsertPt);
   return new llvm::AllocaInst(Ty, 0, Name, AllocaInsertPt);
 }

Chandler however thinks that we need to extract the names from debug info 
and never rely on the SSA variable names.
This is potentially doable, but will require a special subset of -g to 
be always on with asan/msan, which is quite a bit of maintenance work 
compared to a very simple patch above.

Thoughts?

-- 
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/20130920/6b98f849/attachment.html>