similar to: [LLVMdev] Weird msan problem

I assume there are transitions between JITted code and native helper functions. How are you handling them? Are native functions MSan-instrumented? MSan is passing shadow across function calls in TLS slots. Does your TLS implementation guarantee that accesses to __msan_param_tls from JITted and from native code map to the same memory? On Mon, Jan 27, 2014 at 11:36 PM, Evgeniy Stepanov

I have verified that both TLS implementations indeed find the same area of memory. Anything else I could look for? On Tue, Jan 28, 2014 at 4:28 PM, Keno Fischer <kfischer at college.harvard.edu>wrote: > Yes, both JIT code and the native runtime are instrumented. I am under the > impressions that the the C library should guarantee that from the way the > relocations are

How is ccall() implemented? If it manually sets up a stack frame, then it also needs to store argument shadow values in paramtls. I don't think there is an overflow, unless you have a _lot_ of arguments in a function call. On Sun, Feb 2, 2014 at 9:26 AM, Keno Fischer <kfischer at college.harvard.edu> wrote: > Also, I was looking at the instrumented LLVM code and I noticed that the

The code for ccall looks right. Sounds like you have a very small range of instructions where an uninitialized value appear. You could try debugging at asm level. Shadow for b should be passed at offset 0 in __msan_param_tls. MSan could propagate shadow through arithmetic and even some logic operations (like select). It could be that b is clean on function entry, but then something uninitialized

Looks like when you materialize the stores, you should check the size of the the store and emit an appropriate amount of stores to the origin shadow (or just a memset intrinsic?). On Wed, Feb 5, 2014 at 2:13 PM, Keno Fischer <kfischer at college.harvard.edu>wrote: > The @entry stuff is just a gdb artifact. I've been tracking this back a > little further, and it seems there's

Yes, it would be great to get that fixed. On Wed, Feb 5, 2014 at 4:09 PM, Evgeniy Stepanov <eugeni.stepanov at gmail.com>wrote: > On Thu, Feb 6, 2014 at 12:21 AM, Keno Fischer > <kfischer at college.harvard.edu> wrote: > > Looks like when you materialize the stores, you should check the size of > the > > the store and emit an appropriate amount of stores to the

Indeed, removing that flag works fine, the only question is why this is added in tools/llvm-shlib/Makefile in the first place then and what to do about it: ifeq ($(HOST_OS), $(filter $(HOST_OS), Linux GNU GNU/kFreeBSD)) # Don't allow unresolved symbols. LLVMLibsOptions += -Wl,--no-undefined endif On Thu, Nov 21, 2013 at 11:53 AM, Alexander Potapenko <glider at

Hello, I'm trying to fix a long standing issue we are having in Julia where when the file information switched, we weren't recording that correctly, so line information showed up in the wrong file. Basically we would always create a scope with the DISubprogram and the given line number. What I tried was to change the scope to be a DIFile instead so we'd get the correct file

Actually it only link the shlib fine. Linking any executable against it fails: llvm[2]: Linking Release+Asserts executable llvm-lto (without symbols) /home/kfischer/julia/deps/llvm-svn/build_Release+Asserts/Release+Asserts/bin/clang++ -fsanitize=address -O3 -Wl,-R -Wl,'$ORIGIN/../lib' -L/home/kfischer/julia/deps/llvm-svn/build_Release+Asserts+Sanitize/Release+Asserts/lib

Hello everybody, after moving from OS X to Linux build llvm with asan enabled (I also updated to trunk, but not sure if that's related). However, it's totally possible that I missed a step that I took back when I set this up for me, so I might be doing something very stupid. Anyway, I'm configuring LLVM with ../configure --prefix=/home/kfischer/julia/usr --build=x86_64-pc-linux-gnu

Sorry, I didn't have a small IR example and I was sure I was just doing something stupid. Thanks for the help, I'll try it out and report back. Maybe it would be good to add an assertion or something that tells people what's wrong in this case, since the generated DWARF seems to be invalid? On Wed, Aug 13, 2014 at 5:53 PM, David Blaikie <dblaikie at gmail.com> wrote: > Use

Oh, I see. Sorry I misunderstood. I'll try to come up with some minimal IR. The assertion stems from the fact that getCompileUnitDIE() returns null and then crashes at DWARFUnit.cpp:301. I admit I don't know if this problem is on the parsing or the generation side. While I come up with the IR, basically what I was doing was using a DebugLoc with scope being a DIFile rather than a

What I meant to say was that it worked for me on OS X on a slightly older version of LLVM. Anyway, here's the ld line: "/usr/bin/ld" -export-dynamic -z relro --hash-style=gnu --build-id --eh-frame-hdr -m elf_x86_64 -shared -o /home/kfischer/julia/deps/llvm-svn/build_Release+Asserts+Sanitize/Release+Asserts/lib/

On Thu, Nov 21, 2013 at 6:18 AM, Keno Fischer <kfischer at csail.mit.edu> wrote: > Hello everybody, > > after moving from OS X to Linux build llvm with asan enabled Sorry, I failed to parse this. You're compiling Clang on Linux, and OSX is unrelated, right? > (I also updated > to trunk, but not sure if that's related). However, it's totally possible > that I

Yeah, it's not linking the runtime because you're building a shared library. This is totally ok. In this case there are undefined __asan_* symbols that are resolved when an instrumented binary loads your library. What's wrong here is that some flag passed to ld prevents it from creating undefined symbols (--no-undefined?). This flag must be removed when building with ASan. This

We don't currently apply any relocations (that I know of) for debug info in LLDB. > On Jun 2, 2014, at 12:35 PM, Keno Fischer <kfischer at college.harvard.edu> wrote: > > I think I'm getting closer. The debug_info section is being relocated correctly (I think): > > 0x00000000: Compile Unit: length = 0x00000045 version = 0x0003 abbr_offset = 0x00000000 addr_size =

I didn't get to work on this more last week, but I'll look at incorporating that suggestion. The other question of course is how to do this in LLDB. Right, now what I'm doing is going through and adjusting the load address of every leaf in the section tree. That basically works and gets me backtraces with the correct function names and the ability to set breakpoints at functions in

> On 2014-Dec-18, at 01:43, Keno Fischer <kfischer at college.harvard.edu> wrote: > > Hi Duncan, > > I'm in the following situation for which this change caused an assertion failure: > > Three modules, let's say A B C > Two function, F in A, G in B > > Now I CloneFunction F into B (call the new function F') and inline F' into G. > Now,

Hi Keno, Could you give a short high-level overview of the way Julia works now with MCJIT instead the JIT: What I gather so far... * Compiled IR functions are emitted to a shadow module. * Any used function is cloned into its own new module and the module is added to MCJIT. * Called functions or globalvars are only declared in that module. * Modules are never removed meaning "old"

On Wed, May 7, 2014 at 11:33 AM, Eric Christopher <echristo at gmail.com>wrote: > Have you checked out llvm-symbolize? It's what the asan folk > (including Alexey) have created for backtrace symbolication. > Yeah, we potentially can add some kind of option: "llvm-symbolizer -print-short-function-names", I don't yet see why this would be valuable. What's wrong