Re: [RFC PATCH 06/10] ftrace: Release the function hlist if we don't need it anymore

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Jan 25, 2010 at 02:41:59PM +0800, Lai Jiangshan wrote:
> Frederic Weisbecker wrote:
> > After we disable the function profiler, the function hashlist
> > stays in memory. This is wasteful as nobody needs it anymore,
> > until the next use if any.
> > 
> > Release it when we disable the function profiler instead of
> > resetting it in the next use.
> > 
> > Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> > Cc: Steven Rostedt <rostedt@goodmis.org>
> > Cc: Li Zefan <lizf@cn.fujitsu.com>
> > Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
> > ---
> >  kernel/trace/ftrace.c          |    1 +
> >  kernel/trace/functions_hlist.c |   61 +++++++++++++++++-----------------------
> >  kernel/trace/functions_hlist.h |    1 +
> >  3 files changed, 28 insertions(+), 35 deletions(-)
> > 
> > diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> > index dfd8f7c..0ded01c 100644
> > --- a/kernel/trace/ftrace.c
> > +++ b/kernel/trace/ftrace.c
> > @@ -509,6 +509,7 @@ ftrace_profile_write(struct file *filp, const char __user *ubuf,
> 
> >			/*
> >			 * unregister_ftrace_profiler calls stop_machine
> >			 * so this acts like an synchronize_sched.
> >			 */
> >  			unregister_ftrace_profiler();
> 
> unluckily, when !CONFIG_DYNAMIC_FTRACE, it does not call stop_machine()
> nor synchronize_sched(), bug here?



Yeah. I guess the synchronize_sched() is here to ensure
that if a user reenables the profiler, there is no
pending tracing callback that touches the hlist
that can become reset.

We indeed may need an explicit synchronize_sched() for
!CONFIG_DYNAMIC_FTRACE.