lkml.org 
[lkml]   [2016]   [Dec]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [for-next][PATCH 7/8] fgraph: Handle a case where a tracer ignores set_graph_notrace
On Mon, Dec 12, 2016 at 11:49:20AM -0500, Steven Rostedt wrote:
> On Tue, 13 Dec 2016 01:30:01 +0900
> Namhyung Kim <namhyung@kernel.org> wrote:
>
>
> > Sorry to miss updating those tracers. I guess it's no more necessary once
> > the patch 8 is applied so that functions in the notrace filter will not be
> > recorded.
> >
> > Or maybe we need to change the prepare_ftrace_return() so that the
> > graph_entry callback should be called after ftrace_push_return_trace() as
> > some archs do.
>
> I plan on updating fgraph in general so this should all be handled then.

ok

>
> >
> > >
> > > Have the print functions handle cases where a tracer still records functions
> > > even when they are in set_graph_notrace.
> >
> > I think it'd be better (or consistent, at least) not printing negative index
> > records rather than showing entry only.
>
> I thought about this too, but I'm more concerned about it not crashing
> the kernel than to show a proper trace. The fix will just make sure it
> doesn't crash.

ok

>
> >
> > >
> > > Also add warnings if the depth is below zero before accessing the array.
> > >
> > > Note, the function graph logic will still prevent the return of these
> > > functions from being recorded, which means that they will be left hanging
> > > without a return. For example:
> > >
> > > # echo '*spin*' > set_graph_notrace
> > > # echo 1 > options/display-graph
> > > # echo wakeup > current_tracer
> > > # cat trace
> > > [...]
> > > _raw_spin_lock() {
> > > preempt_count_add() {
> > > do_raw_spin_lock() {
> > > update_rq_clock();
> > >
> > > Where it should look like:
> > >
> > > _raw_spin_lock() {
> > > preempt_count_add();
> > > do_raw_spin_lock();
> > > }
> > > update_rq_clock();
> >
> > If set_graph_notrace works correctly, it should be just:
> >
> > update_rq_clock();
>
> Which is what it should look like after patch 8. But I didn't mark 8 as
> stable as that's more of a feature. As wakeup and irqsoff doesn't use
> notrace yet. Yeah, notrace may break it a bit, but since this is the
> first someone noticed it, I don't think it's used much.
>
> I wanted the simplest fix for stable.

I think a simpler fix is just to return when it sees a negative record..


diff --git a/kernel/trace/trace_functions_graph.c b/kernel/trace/trace_functions_graph.c
index 52fb1e21b86b..2fb73c2e35b5 100644
--- a/kernel/trace/trace_functions_graph.c
+++ b/kernel/trace/trace_functions_graph.c
@@ -844,7 +844,7 @@ print_graph_entry_leaf(struct trace_iterator *iter,

/* If a graph tracer ignored set_graph_notrace */
if (call->depth < -1)
- call->depth += FTRACE_NOTRACE_DEPTH;
+ return TRACE_TYPE_HANDLED;

/*
* Comments display at + 1 to depth. Since
@@ -887,7 +887,7 @@ print_graph_entry_nested(struct trace_iterator *iter,

/* If a graph tracer ignored set_graph_notrace */
if (call->depth < -1)
- call->depth += FTRACE_NOTRACE_DEPTH;
+ return TRACE_TYPE_HANDLED;

cpu_data = per_cpu_ptr(data->cpu_data, cpu);
cpu_data->depth = call->depth;

Thanks,
Namhyung

\
 
 \ /
  Last update: 2016-12-12 18:10    [W:0.060 / U:22.188 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site