On Thu, 2008-11-13 at 00:17 -0500, Chris Moller wrote: > > Jim Keniston wrote: ... > > Non-requirements: > > ----------------- > > ... > > > > Just off hand, I'm fairly sure froggy will support all the > "non-requirements." It already keeps per-client (I'm not sure what you > mean by client--in froggyworld, a client is a userspace thing like gdb)
Ubp provides an in-kernel API, so a ubp client would be a kernel subsystem or kernel module -- e.g., froggy or uprobes. > and per-attached-process information in the froggy module and thus could > keep the various objects and "SSOL slots" (whatever they are--I'll read > that pdf, I promise) and already does the utrace attach/quiesce/whatever > stuff. And since all froggy clients use the same froggy module, > coordinating multiple ubp operations should be possible. > > Any chance there's some actual working ubp code I could tinker with? There is no implementation of ubp yet. We could probably hack together an x86 implementation from x86 uprobes in a few days. But we'd need a client to test it. One approach would be to gut that part of (a copy of) uprobes and make uprobes call ubp. But if you want to give it a try with froggy, I suspect Ananth can be persuaded to assign somebody to work with you. > > Chris > Jim
