Re: [PATCH 5/7] ftrace: Store direct called addresses in their ops

[Mark Rutland <mark.rutland@arm.com>]

On Mon, Mar 20, 2023 at 05:31:55PM -0400, Steven Rostedt wrote:
> On Mon, 20 Mar 2023 18:45:08 +0100
> Florent Revest <revest@chromium.org> wrote:
> 
> > On Sun, Mar 19, 2023 at 7:55 PM Jiri Olsa <olsajiri@gmail.com> wrote:
> > >
> > > On Sun, Mar 19, 2023 at 01:54:43PM -0400, Steven Rostedt wrote:  
> > > > On Sun, 19 Mar 2023 16:29:22 +0100
> > > > Jiri Olsa <olsajiri@gmail.com> wrote:
> > > >  
> > > > > > +++ b/kernel/trace/ftrace.c
> > > > > > @@ -2582,9 +2582,8 @@ ftrace_add_rec_direct(unsigned long ip, unsigned long addr,
> > > > > >  static void call_direct_funcs(unsigned long ip, unsigned long pip,
> > > > > >                         struct ftrace_ops *ops, struct ftrace_regs *fregs)
> > > > > >  {
> > > > > > - unsigned long addr;
> > > > > > + unsigned long addr = ops->direct_call;  
> > > > >
> > > > > nice, should it be read with READ_ONCE ?  
> > > >
> > > > Is there a "read tearing" too?  
> > >
> > > don't know, saw the comment in __modify_ftrace_direct and got curious
> > > why it's not in here.. feel free to ignore, I'll look it up
> > >
> > > jirka  
> > 
> > Mhh, that's a good question. Based on my current understanding, it
> > seems that it should have a READ_ONCE, indeed. However, I'd like Mark
> > to confirm/deny this. :)
> > 
> > If this should be a READ_ONCE, I can send a v2 series with this fixed.
> 
> After re-reading: https://lwn.net/Articles/793253/
> 
> I think we should add the READ_ONCE() (also with a comment).

I think so, too.

AFAICT there's nothing that prevents __modify_ftrace_direct() and
call_direct_funcs() from concurrently accessing ftrace_ops::direct_call, so we
need READ_ONCE() in call_direct_funcs() to prevent load tearing and other
issues mentioned in the article linked above.

The existing code has a similar pattern where __modify_ftrace_direct() and
ftrace_find_rec_direct() access ftrace_func_entry::direct concurrently. Do we
want a preparatory patch fixing that for stable?

Thanks,
Mark.