DTrace Hangout Notes 02 March 2017
- gnn@ is tracking down how to fix the gethrtime() issue on ARM32 and ARM64 platforms
- turns out that fork() does not return the 0 to the child. It should return twice, once to the parent and once to the child.
- gnn@ is starting the process of abstracting the OS dependent bits of code from the machine independent code in libdtrace. The ultimate goal is to have a smaller number of ifdefs but to get as much of the code to be independent of the OS.
- domagoj@ discussed with folks at Cambridge doing optimizations along the lines of those in EBPF. This might be handled by a new D subroutine.
- samuel@ is chasing some race condition bugs.
- markj@ wants to attach attributes to probes which is a new idea.
- markj@ also wants to work on instrumenting individual instructions in the kernel which would give us a feature that we have in user space
markj@ will point gnn@ at a branch with changes to bring in libctf from Joyent. Whoever imports this must review the commit log to make sure that we get all the fixes. We will create a smartos alongside opensolaris for now.
- markj@ will finish his hot patching work sometime in the near future.
- gnn@ will put up reviews of the libxo machine readable output code
- domagoj@ has been trying to create a virtio driver that operates with DTrace between a guest and a host on bhyve and might need to create some new ioctls for use with this