Messages in this thread | | | Date | Tue, 18 Jan 2011 17:41:48 +0100 | From | Jiri Olsa <> | Subject | Re: [PATCH] tracing: remove superfluous sub instructions |
| |
On Tue, Jan 18, 2011 at 11:22:42AM -0500, Steven Rostedt wrote: > On Tue, 2011-01-18 at 16:52 +0100, Frederic Weisbecker wrote: > > On Tue, Jan 18, 2011 at 04:28:18PM +0100, Jiri Olsa wrote: > > > hi, > > > > > > I think there's no need for substracting MCOUNT_INSN_SIZE from the > > > IP parameter before calling the function trace (/graph) handler. > > > > > > Maybe I overlooked something, but all the IP usage I saw ended > > > up in the kallsyms_lookup function, which does the lookup using the > > > functions' start/end boundaries to find the correct symbol for pointer. > > > > > > Thus it seems to me there's no point in substracting the > > > MCOUNT_INSN_SIZE value from the IP parameter. > > > > > > I tested for x86_64 and got proper results, I believe it's > > > the same case for x86_32. > > > > > > wbr, > > > jirka > > > > > > > > > Signed-off-by: Jiri Olsa <jolsa@redhat.com> > > > > Well, that sounds right after all. If we are only interested in the > > symbol, the instruction that follows "call mcount" is still relevant > > as it must belong to the same function. > > > > Steve? > > NAK, it will break function triggers/probes. The "func:traceon" and > "func:traceoff". They compare the ip to the call location of mcount. > > -- Steve > >
ops, missed this one..
would it make sense to update the IP inside the function_trace_probe_call function to save one instruction in the entry code used by all? or it's not worth it..
jirka
| |