Re: [PATCH v5 02/14] perf tools: Prevent calling machine__delete() on non-allocated machine

[Arnaldo Carvalho de Melo]

Em Wed, Dec 16, 2015 at 09:37:18AM +0800, Wangnan (F) escreveu:
> 
> 
> On 2015/12/15 20:36, Jiri Olsa wrote:
> >On Mon, Dec 14, 2015 at 10:39:11AM +0000, Wang Nan wrote:
> >
> >SNIP
> >
> >>@@ -137,12 +138,15 @@ void machine__exit(struct machine *machine)
> >>  void machine__delete(struct machine *machine)
> >>  {
> >>  	machine__exit(machine);
> >>-	free(machine);
> >>+	if (machine->allocated)
> >>+		free(machine);
> >>+	else
> >>+		pr_warning("WARNING: delete a non-allocated machine. Skip.\n");
> >we used WARN_ONCE several times already in similar cases
> >
> >jirka
> 
> Will switch to:
> 
> @@ -136,13 +138,13 @@ void machine__exit(struct machine *machine)
> 
>  void machine__delete(struct machine *machine)
>  {
> -       machine__exit(machine);

Better keep the above.

And I wonder if we would go on sprinkling these kinds of checks for all
classes we have :-\

I think this is a job for some static analisys tool, that or we figure
out a way to find out if an address is for a stack or heap and use that
instead, and in a bpf based tool, perhaps, one that would hook into all
*__delete() tools and check if the object it is using should or not be
in fact free()ed.

I could think about hooking __new*() calls, hashing the return value,
then at __delete() time check it, for instance.

- Arnaldo

> -       free(machine);
> +       WARN_ONCE((machine->allocated ? free(machine), 0 : -1),
> +                 "WARNING: deleting a non-allocated machine. Skip.\n");
>  }
> 
> Thank you.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/