[dpdk-dev] [PATCH] ethdev: fix memory leak on removing eth rxtx callback

Ananyev, Konstantin konstantin.ananyev at intel.com
Fri Nov 23 10:13:30 CET 2018



> -----Original Message-----
> From: Yigit, Ferruh
> Sent: Wednesday, November 21, 2018 2:49 PM
> To: wanlebing <wanlebing at gmail.com>; dev at dpdk.org; Ananyev, Konstantin <konstantin.ananyev at intel.com>
> Cc: wanlebing <wanlebing at didichuxing.com>
> Subject: Re: [dpdk-dev] [PATCH] ethdev: fix memory leak on removing eth rxtx callback
> 
> On 11/21/2018 1:11 PM, wanlebing wrote:
> > eth rxtx callback is dynamically allocated using rte_zmalloc()
> > but not released. Fix it by calling rte_free() to free callback
> > when removing rxtx callback.
> 
> I guess they are not freed intentionally because there is no safe way to do it.

That's right - we can't free it here without extra synchronization.

> I remember Konstantin sent an RFC in the past to rework the callbacks, cc'ed him
> for more details.

Following RFC was submitted:
http://patches.dpdk.org/patch/31867/

Though there was a concern that it would introduce some performance
degradation on low-end arm machines, so it is deferred right now.
Konstantin 

> 
> >
> > Signed-off-by: wanlebing <wanlebing at didichuxing.com>
> > Signed-off-by: wanlebing <wanlebing at gmail.com>
> > ---
> >  lib/librte_ethdev/rte_ethdev.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/lib/librte_ethdev/rte_ethdev.c b/lib/librte_ethdev/rte_ethdev.c
> > index 5f858174b..f00311047 100644
> > --- a/lib/librte_ethdev/rte_ethdev.c
> > +++ b/lib/librte_ethdev/rte_ethdev.c
> > @@ -3926,6 +3926,7 @@ rte_eth_remove_rx_callback(uint16_t port_id, uint16_t queue_id,
> >  		if (cb == user_cb) {
> >  			/* Remove the user cb from the callback list. */
> >  			*prev_cb = cb->next;
> > +			rte_free(cb);
> >  			ret = 0;
> >  			break;
> >  		}
> > @@ -3960,6 +3961,7 @@ rte_eth_remove_tx_callback(uint16_t port_id, uint16_t queue_id,
> >  		if (cb == user_cb) {
> >  			/* Remove the user cb from the callback list. */
> >  			*prev_cb = cb->next;
> > +			rte_free(cb);
> >  			ret = 0;
> >  			break;
> >  		}
> >



More information about the dev mailing list