[dpdk-dev] [PATCH v2 1/2] test/mempool_perf: Free mempool on exit

[Olivier MATZ]

Hi Santosh,

On Mon, 10 Apr 2017 01:13:43 +0530
santosh <santosh.shukla at caviumnetworks.com> wrote:

> Hi Olivier,
> 
> On Monday 10 April 2017 12:47 AM, Shukla, Santosh wrote:
> 
> >
> >
> > --------------------------------------------------------------------------------
> > *From:* Olivier Matz <olivier.matz at 6wind.com>
> > *Sent:* Friday, April 7, 2017 9:21 PM
> > *To:* Shukla, Santosh
> > *Cc:* dev at dpdk.org; hemant.agrawal at nxp.com; shreyansh.jain at nxp.com; stable at dpdk.org
> > *Subject:* Re: [PATCH v2 1/2] test/mempool_perf: Free mempool on exit
> > Hi Santosh,
> >
> > On Thu,  6 Apr 2017 12:15:48 +0530, Santosh Shukla 
> > <santosh.shukla at caviumnetworks.com> wrote:  
> > > Mempool_perf test not freeing pool memory.
> > >
> > > Cc: stable at dpdk.org
> > > Signed-off-by: Santosh Shukla <santosh.shukla at caviumnetworks.com>
> > > Acked-by: Shreyansh Jain <shreyansh.jain at nxp.com>
> > > ---
> > > v1 --> v2:
> > >  * Fixed patch context
> > >
> > >  test/test/test_mempool_perf.c | 31 +++++++++++++++++++------------
> > >  1 file changed, 19 insertions(+), 12 deletions(-)
> > >
> > > diff --git a/test/test/test_mempool_perf.c b/test/test/test_mempool_perf.c
> > > index ebf1721ac..3c45971ab 100644
> > > --- a/test/test/test_mempool_perf.c
> > > +++ b/test/test/test_mempool_perf.c
> > > @@ -312,6 +312,8 @@ do_one_mempool_test(unsigned cores)
> > >  static int
> > >  test_mempool_perf(void)
> > >  {
> > > +     int ret = -1;
> > > +
> > >        rte_atomic32_init(&synchro);
> > >
> > >        /* create a mempool (without cache) */
> > > @@ -322,7 +324,7 @@ test_mempool_perf(void)
> > >                                                my_obj_init, NULL,
> > >                                                SOCKET_ID_ANY, 0);
> > >        if (mp_nocache == NULL)
> > > -             return -1;
> > > +             goto err;
> > >
> > >        /* create a mempool (with cache) */
> > >        if (mp_cache == NULL)  
> >
> > [...]
> >  
> > >
> > > -     return 0;
> > > +     ret = 0;
> > > +
> > > +err:
> > > +     rte_mempool_free(mp_cache);
> > > +     rte_mempool_free(mp_nocache);
> > > +     return ret;  
> >
> >
> > Since mp_cache and mp_nocache are global variables, this won't
> > work properly due to the way mempool are created:
> >
> >          /* create a mempool (without cache) */
> >          if (mp_nocache == NULL)
> >                  mp_nocache = rte_mempool_create("perf_test_nocache", MEMPOOL_SIZE,
> >                                                  MEMPOOL_ELT_SIZE, 0, 0,
> >                                                  NULL, NULL,
> >                                                  my_obj_init, NULL,
> >                                                  SOCKET_ID_ANY, 0);
> >
> > The if() should be removed, else we'll have a use after free the next
> > time.  
> 
> I understand your point.
> But I think problem is rte_mempool_free() not referencing mp = null
> after freeing resources. Result of that is mp_nocache still has valid 
> address, Although internal resources (mz/_ops_handle) were actually freed by
> rte_mempool_free(), right?
> 
> So rather removing above if(), why not
> - Application explicit set mp_nocache = NULL after mempool_free().
> ie.. 
> 
> err:
> 	rte_mempool_free(xxx);
> 	xxx = NULL;
> 
> 
> Or
> - Let rte_mempool_free() { - do mp = null; }
> 
> And yes remove that if condition anyway. As its a dead-code
>  for either of above 2 options.
> 
> Does that make sense to you? If so then which one you prefer?

Yes, it makes sense.
My first preference would be removing the global vars (as suggested
below). Else your proposition is ok too.

> 
> > If you want to do more clean-up, you can try to remove the global variables,
> > but it's maybe harder.  
> 
> Removing global var won't be harder imo, May be you know more but
> here is my point of view, after going through code:
> 
> - All test_func like
> do_one_mempool_test -> launch_cores --> per_lcore_mempool_test -> using 'mp'
> 
> where 'mp' is global. 
> 
> how about,
> - As you said Yes - remove global var ie.. mp_cache/nocache, default_pool, mp
> - Add 'rte_mempool *mp' as argument in do_one_mempool_test() func and other func too.
> 
> Thus get rid of globals from app.
> 
> Does that make sense to you? 

Yes, looks good. It would be clearer without global vars.

Historically, it was not possible to free a mempool, that's why it was
done like this.


Thanks!
Olivier