Messages in this thread | | | Date | Thu, 27 Aug 2009 13:32:53 -0400 | From | Masami Hiramatsu <> | Subject | Re: [GIT PULL] tracing/kprobes: Add dynamic tracepoints + instruction decoder |
| |
Frederic Weisbecker wrote: > On Thu, Aug 27, 2009 at 11:00:38AM -0400, Masami Hiramatsu wrote: >> Frederic Weisbecker wrote: >>> Ingo, >>> >>> This is the kprobes tracing pile of patches. I've tested it >>> successfully by setting some kprobes through debugfs by hand. >>> >>> It brings no known regressions. >>> >>> However, the stress test provided by Masami have revealed some >>> unstable points. Some symbols are unsafe to probe and raise >>> probing recursion. >>> >>> I've added a tiny patch in the series that helps identifying the >>> kprobe that has raised such situation. >>> For example it has learned me today that it's unsafe to trace >>> ret_from_exception() (obviously: it's on the int 3 handler path). >> >> Thank you, Frederic. >> >> I'm currently fixing those problems on x86-64 (on kvm). >> I'll also check it on x86-32, and fix it. >> >> Thanks > > > > Cool. > Don't hesitate to interate through small patches that fix these > sites. > > So that I can continue to test it over time and report you what I find, > may be I can try to fix some of them too. > > I haven't fixed ret_from_exception because I wasn't sure about the > right thing to do: putting it in the kprobes section or blacklist its > name. It's a bit more delicate to put assembly code into sections so... :)
It depends on implementation, but I think blacklist is the last resort. :) AFAIK, some assembly codes are already pushed into kprobes section.
Thank you,
-- Masami Hiramatsu
Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division
e-mail: mhiramat@redhat.com
| |