Messages in this thread | | | Date | Tue, 21 Apr 2009 09:13:05 -0400 (EDT) | From | Steven Rostedt <> | Subject | Re: [PATCH 1/5] ftrace: use module notifier for function tracer |
| |
On Tue, 21 Apr 2009, Rusty Russell wrote:
> On Mon, 20 Apr 2009 11:27:35 pm Steven Rostedt wrote: > > > > On Sun, 19 Apr 2009, Rusty Russell wrote: > > > > > On Thu, 16 Apr 2009 11:48:31 am Steven Rostedt wrote: > > > > From: Steven Rostedt <srostedt@redhat.com> > > > > > > > > Impact: fix and clean up > > > > > > > > The hooks in the module code for the function tracer must be called > > > > before any of that module code runs. The function tracer hooks > > > > modify the module (replacing calls to mcount to nops). If the code > > > > is executed while the change occurs, then the CPU can take a GPF. > > > > > > > > To handle the above with a bit of paranoia, I originally implemented > > > > the hooks as calls directly from the module code. > > > > > > > > After examining the notifier calls, it looks as though the start up > > > > notify is called before any of the module's code is executed. This makes > > > > the use of the notify safe with ftrace. > > > > > > Hi Steven, > > > > > > Unfortunately not: we do parse_args, which can call into the module code. > > > (Though it shouldn't do anything "significant", as it won't get a chance to > > > clean up if module load fails later). > > > > > > I think you need to do something else in general. Share the module_mutex for > > > the ftrace code? The ksplice guys have a similar issue, so maybe we should > > > generalize this into a "kernel_text" mutex? > > > > Hi Rusty, > > > > Thanks, for the update. I think we may still be OK. > > Agreed, just wanted to make sure you were aware. > > > Can those parse_args kick off threads? Hmm, probably. Sounds nasty to > > me. > > Not without a bug. Imagine you have a "create_threads" module_param, someone > loads the module with two args "create_threads crap". We call the > create_threads parse function via parse_args, then hit the crap parameter > and free the module. Oops. > > > The other thing is, if the parse_args code is only in "__init" then they > > also will not be touched. > > It can be non-init for sysfs access. > > FWIW: > Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Thanks Rusty!
Ingo,
This change also fixes a possible deadlock in mainline between the ftrace_lock and the module_mutex. Perhaps we should push this to Linus?
The possible deadlock is if a user unloads/loads modules at the same time starts the function graph tracer. Highly unlikely to happen, but it does spit out a lockdep warning.
-- Steve
| |