Re: [RFC PATCH] use snprintf instead of sprintf in rcu_torture_printk

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Thu, Jun 19, 2014 at 08:13:22PM -0400, Pranith Kumar wrote:
> On 06/19/2014 07:49 PM, Paul E. McKenney wrote:
> > On Thu, Jun 19, 2014 at 07:24:48PM -0400, Pranith Kumar wrote:
> >> (dropping some CCs)
> >>
> >> On 06/19/2014 05:00 PM, Paul E. McKenney wrote:
> >>> On Wed, Jun 18, 2014 at 12:49:42PM -0700, Joe Perches wrote:
> >>>>
> >>>> I believe the function doesn't work well.
> >>>>
> >>>> static void
> >>>> rcu_torture_stats_print(void)
> >>>> {
> >>>> 	int size = nr_cpu_ids * 200 + 8192;
> >>>> 	char *buf;
> >>>>
> >>>> 	buf = kmalloc(size, GFP_KERNEL);
> >>>> 	if (!buf) {
> >>>> 		pr_err("rcu-torture: Out of memory, need: %d\n", size);
> >>>> 		return;
> >>>> 	}
> >>>> 	rcu_torture_printk(buf);
> >>>> 	pr_alert("%s", buf);
> >>>> 	kfree(buf);
> >>>> }
> >>>>
> >>>> rcu_torture_printk simply fills buf
> >>>>
> >>>> btw: I believe the arguments should pass size and
> >>>>      rcu_torture_printk should use snprintf/size
> >>>>
> >>>> but all printks are limited to a maximum of 1024
> >>>> bytes so the large allocation is senseless and
> >>>> would even if it worked, would likely need to be
> >>>> vmalloc/vfree
> >>>
> >>> Fair point!
> >>>
> >>> Pranith, Romanov, if you would like part of RCU that is less touchy
> >>> about random hacking, this would be a good place to start.  Scripts in
> >>> tools/testing/selftests/rcutorture/bin do care about some of the format,
> >>> but the variable-length portion generated by cur_ops->stats() is as far
> >>> as I know only parsed by human eyes.
> >>>
> >>
> >> Here is a first run of the change. Please let me know if I am totally off. RFC. :)
> > 
> > Thank you for taking this on!
> 
> You are most welcome :)
> 
> > 
> >> Three things on Todo list:
> >>
> >> * We need to check that we are using less than the allocated size of the buffer (used > size). (we are allocating a big buffer, so not sure if necessary)
> >> * Need to check with the scripts if they are working.
> >> * I used a loop for pr_alert(). I am not sure this is right, there should be a better way for printing large buffers
> >>
> >> If the overall structure is ok I will go ahead and check how the scripts are handling these changes. 
> > 
> > One other thing...  Convince this function (and a few others that it
> > calls) that the system really has 4096 CPUs, run this code, and see what
> > actually happens both before and after.  Just to get a bit of practice
> > mixed in with the theory.  ;-)
> > 
> 
> OK. I think to do this I need to use 4096 instead of nr_cpu_ids. I will try this and see how it goes :)

That's the spirit!  You will probably have to adjust a few other things
as well.

							Thanx, Paul