Re: [GIT PULL v2] tracing: Updates for v6.14

[Steven Rostedt]

On Tue, 21 Jan 2025 14:11:27 -0500
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Tue, 21 Jan 2025 14:09:02 -0500
> Sasha Levin <sashal@xxxxxxxxxx> wrote:
> 
> > >Let's see if your config makes that different.    
> > 
> > Hmm... LKFT reproduces it fairly easily even on -next. From the log, it
> > looks like it reproduces under LTP's read_all_sys test.  
> 
> I'm able to reproduce it with random writes into set_event and then running
> "trace-cmd stat" which does a lot of reads of tracefs.
> 
> I re-enabled KASAN and kmemleak to see if it gives a hint to what is going wrong.
> 

Bah! I found it.

 static void *s_start(struct seq_file *m, loff_t *pos)
 {
-	struct trace_event_file *file;
 	struct trace_array *tr = m->private;
+	struct set_event_iter *iter;
 	loff_t l;
 
+	iter = kzalloc(sizeof(iter), GFP_KERNEL);
                              ^^^^^

Should be (*iter) :-p

+	if (!iter)
+		return NULL;
+

I caught this before I pushed this to linux-next and I thought I fixed it.
No idea how the older version made it out to linux-next and my pull request.

I tested the fixed version, but sent out the broken one! I was also
updating my test suite at the time, so I may have screwed something up in
my workflow.

Since this is already out, I'll send a separate fix to it. Thanks for
reporting.

-- Steve