Re: [dpdk-dev] [PATCH] examples/kni: fix crash during MTU set

[Thomas Monjalon <thomas@monjalon.net>]

21/05/2020 17:46, Ferruh Yigit:
> On 5/21/2020 4:37 PM, Thomas Monjalon wrote:
> > 21/05/2020 17:10, Ferruh Yigit:
> >> During MTU set (kni_change_mtu) sample application setup queues, which
> >> can free and re-allocate queues.
> >> Meanwhile sample application keeps continues in Rx/Tx burst calls in
> >> different threads, which may cause crash during queue setup.
> >>
> >> Pausing application Rx/Tx calls before MTU set and starts it back
> >> afterwards.
> >>
> >> Bugzilla ID: 482
> >> Fixes: a26b116749a3 ("examples/kni: fix MTU change to setup Tx queue")
> >> Cc: stable@dpdk.org
> >>
> >> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
> >> ---
> >> -/* Callback for request of changing MTU */
> >>  static int
> >> -kni_change_mtu(uint16_t port_id, unsigned int new_mtu)
> >> +kni_change_mtu_(uint16_t port_id, unsigned int new_mtu)
> > [...]
> >> +/* Callback for request of changing MTU */
> >> +static int
> >> +kni_change_mtu(uint16_t port_id, unsigned int new_mtu)
> >> +{
> >> +	int ret;
> >> +
> >> +	rte_atomic32_inc(&kni_pause);
> >> +	ret =  kni_change_mtu_(port_id, new_mtu);
> >> +	rte_atomic32_dec(&kni_pause);
> >> +
> >> +	return ret;
> >> +}
> > 
> > Why creating a new function which is called only once?
> > 
> 
> Just wrapping the existing one with stop/start the forwarding.
> 
> These can be added into the existing function but there are many exit points in
> the function, so it will create more clutter and it is more error prone. I think
> this way more simple and clear.

OK to go on the safe side at this stage of the release :-)