Meeting 31 March 2016

George mentions adding the mbuf SDTs and asks if anyone has used them. The answer is that they have not been used as yet but will be useful eventually. George will add example scripts showing how to use these to the main repo.

There is now a bit of stack unwinding code for ARM64. Mark has reviewed this code and seems happy with t.

Brendan has given both his blessing and his scripts to the OpenDTrace effort. Mark asks if these are mostly fbt based or not. The idea is that they are but we'll have to check.

Robert points out that moving away from fbt probes to probes that are more stable will help with OpenDTrace.

Mark has a Python provider at Isilon.

Mark has been testing the hot patching and finds that it only has 1% overhead in his testing. There is an LLVM intrinsic called a tracepoint that might be helpful in dealing with the overhead. Mark has also been working on PPC and ARM with his patching code and will update the mailing list about that. Mark will also reach out to Bjoern about figuring out overheads in his patch.

Robert asks about MIPS but no one knows if it works at all. It's possible Robert will look at the MIPS side of DTrace.

The rate of changes in Illumos is very low so we're only getting bug fixes from them.

The SmartOS folks have a new libctf which integrates support for ctfmerge etc. Mark is going to look at the BSD licensed libctf from Daniel Lovasko as well.

Mark has boot time dtrace working as of last night. This works by running dtrace to create a DOF file before reboot. On boot the loader loads the DOF file and runs the script. Robert asks if this happens before or after kernel modules are loaded. Mark says it happens afterwards.

George will put up his fds[] path support in a Phabricator patch.

George will also be updating the Handbook section on DTrace. Mark has additions as well. We also need manual pages for the SDTs.

Mark will put up his lock provider patch and George will work on a patch for the rm_locks. Robert points out that rm locks may not be cheap and that we ought to look at the issues as well.

George will put up the machine readable output code as a Phabricator patch as well.

DTrace/31March2016 (last edited 2016-03-31T16:57:10+0000 by GeorgeNevilleNeil)