Messages in this thread | | | From | (Eric W. Biederman) | Date | Tue, 08 Sep 2009 19:02:37 -0700 | Subject | Re: [PATCH 2/2] tracing/events: Add kexec tracepoints |
| |
Steven Rostedt <rostedt@goodmis.org> writes:
> On Tue, 2009-09-08 at 18:27 -0700, Eric W. Biederman wrote: >> Li Zefan <lizf@cn.fujitsu.com> writes: >> >> > When panic, save in ring buffer the time when crash happened, >> > the panic message, etc. And the data can be retrieved from >> > core dump file using flight-recorder, which is going to be a >> > module of the crash utility. >> > >> > Also if we implement flight-recorder in perf tool, those trace >> > outputs can be used by perf tool. >> >> Please let's keep dynamic code out of crash_kexec if we can. > > What dynamic code? It's a tracepoint not something like the function > tracer. It is activated via a bit test and a jmp. Not very dynamic.
The philosophy of the crash dump code is that the kernel is broken trust as little as possible.
Within a few seconds we should be able to get the time from the kernel that saves the crash dump so I don't see how adding a trace point adds anything into that code path except an unnecessary dependency that might have broken.
Or in short you have a full user space to add features to. Please don't features to the critical kernel magic that gets you to userspace.
Eric
| |