kprobes slide from 4/20 meeting

[Jim Keniston]

Kprobes Status

Multi-probes (multiple probes at same address):
Ananth has posted 4 "takes."
Prasanna posted 1, but on LKML, and it was pulled into the -mm tree.
Prasanna, Suparna oppose any approach that significantly changes the
kprobes base.
[Update from Ananth at the 4/20 meeting: Ananth has posted a new patch
to LKML, based on his takes 2 and 3.  Ananth is confident that this
patch will replace Prasanna's in the -mm tree, and that Prasanna and
Suparna will go along.]

Return Probes:
Has evolved some since LKML posting:
- retprobes registered/unregistered independently of kprobes/jprobes
- recycling of outstanding instance objects when stack is torn down by
do_exit, do_exec
- will be based on multi-probes; waiting for this to settle out
Non-coexistence of multi-probes with jprobes affects retprobes.
Prasanna hasn't bought in.

x86_64 RIP-relative addressing bug
Roland's fix is in 2.6.12-rc2.

x86_64 wart removal
Multi-probes and return probes keep tripping over this wart.
Coded, but not tested.  (Drops about 60 lines in x86_64 kprobes.c)
Need to verify that this isn't a performance hit.