Svadev mailing list

[John Criswell <criswell AT illinois.edu>]

On 12/6/11 1:33 AM, Santosh Nagarakatte wrote:

[snip]


Let me know if there is a way I can inline the SafeCode checks?

First, please do an svn up.  I believe I fixed a bug the other day in which we weren't running mem2reg before the SAFECode instrumentation passes.

What I think we really need to do is to compile the run-time libraries into LLVM bitcode using the libLTO infrastructure.  That may be as simple as telling it to generate archives using -O4 level optimization or higher.

Come to think of it, I think this approach would work for people who are using libLTO and people who are not.  If libLTO isn't installed before SAFECode is built, then -O4 should just generate machine code.  If libLTO is installed, then -O4 will generate bitcode.

Another thing one could do is to remove some of the _ui() calls (such as funccheck_ui() and poolcheck_ui()).  Some of them do nothing at all (because there's no reliable check to do if the compiler doesn't know about all the memory objects to which the pointer can point); others can do some checks (such as poolcheck_ui(), which can check that a memory dereference only accesses bytes within a valid memory object is a memory object can be found), but those checks may not be worth the overhead when in debug mode.

Finally, PR#10774 needs to be fixed.

-- John T.

Santosh


--
Santosh G Nagarakatte,
PhD Student,
Computer and Information Science Department
University of Pennsylvania,
Philadelphia-19104
http://www.cis.upenn.edu/~santoshn

_______________________________________________
svadev mailing list
svadev AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/svadev