13/07/2021 15:42, Matan Azrad: > Hi Thomas > > From: Thomas Monjalon > > When registering a new event callback, if allocation fails, there is no > > need for > > unregistering the callback, because it is not registered. > > > > Fixes: 9ec0b3869d8d ("ethdev: allow event registration for all ports") > > Cc: sta...@dpdk.org > > > > Signed-off-by: Thomas Monjalon <tho...@monjalon.net> > > --- > > lib/ethdev/rte_ethdev.c | 2 -- > > 1 file changed, 2 deletions(-) > > > > diff --git a/lib/ethdev/rte_ethdev.c b/lib/ethdev/rte_ethdev.c index > > 9d95cd11e1..1731854628 100644 > > --- a/lib/ethdev/rte_ethdev.c > > +++ b/lib/ethdev/rte_ethdev.c > > @@ -4649,8 +4649,6 @@ rte_eth_dev_callback_register(uint16_t port_id, > > user_cb, next); > > } else { > > rte_spinlock_unlock(ð_dev_cb_lock); > > - rte_eth_dev_callback_unregister(port_id, > > event, > > - cb_fn, > > cb_arg); > > Please pay attention to the case of port_id=RTE_ETH_ALL where the user wants > to register the event for all the ports. > > In this case, when a failure happens for one of the ports, this unregister > call cleans the callback from all the ports.
Yes I missed it. Now I better understand the intent, thanks. Next question: do we really want to rollback already registered ports? Anyway, if we are out of memory, I think it is better not doing more operations. There can be various opinions on this topic, please give yours.