|
From: Jeremy F. <je...@go...> - 2005-01-12 23:13:48
|
OK, I went through and cleaned things up, and I think its ready for checkin. Under NPTL/glibc2.3.3/FC2, the only regcheck failures are the helgrind tests and massif (it's getting an assertion failure). Helgrind needs pthread knowledge before it can work again. Massif looks like it should be relatively simple to fix. I poked at it a bit, but I didn't quite work out what's going on. I'll poke at it a bit in the next few days, unless Nick gets to it first. With LinuxThreads (LD_ASSUME_KERNEL=2.4.0), there are a few more failures, but they're generally just things like different library function names &c. There seems to be a libpthread bug which prevents susphello from running properly under LinuxThreads (I get the same results with and without Valgrind). I've done some preliminary testing on my RH7.3 system, and it seems to work. I'm still waiting for the regtest to finish, but so far it looks the same as the LinuxThreads regtest run. The changes don't add any XXX comments (or at least, it certainly removes more than it adds). I added a couple of suppressions to deal with messages coming out of libpthread, but most of the messages were quietened by fixing sysctl's syscall wrapper. I added a few new features as well. All the PRE_MEM_READ/WRITE functions check to see if all the addresses are mapped before touching their contents, so its a bit more robust against bad syscall args (scalar completes without crashing, which it never did before for me). Patches at http://www.goop.org/~jeremy/valgrind/ I think the next step is to scatter some VALGRIND_* macros around the source and give it a good introspection. J |
|
From: Jeremy F. <je...@go...> - 2005-01-15 01:59:44
|
On Wed, 2005-01-12 at 15:09 -0800, Jeremy Fitzhardinge wrote: > OK, I went through and cleaned things up, and I think its ready for > checkin. Well, I'll take the absence of any protest that everyone thinks this is OK. I'll start checking them in time for today's regression tests (assuming they get picked up in time). J |