On May 20, 2014, at 4:03 PM, Jarno Rajahalme <jrajaha...@nicira.com> wrote:
>
> On May 8, 2014, at 4:50 PM, Ben Pfaff <b...@nicira.com> wrote:
>
>> Signed-off-by: Ben Pfaff <b...@nicira.com>
>> ---
>> lib/dpif-netdev.c | 163
>> +++++++++++++++++++++++++++--------------------------
>> 1 file changed, 82 insertions(+), 81 deletions(-)
>>
>> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
>> index 55712dd..b682876 100644
>> --- a/lib/dpif-netdev.c
>> +++ b/lib/dpif-netdev.c
>> @@ -32,6 +32,7 @@
>>
>
> (snip)
>
>> static void
>> -port_unref(struct dp_netdev_port *port)
>> +port_destroy__(struct dp_netdev_port *port)
>> {
>> - if (port && ovs_refcount_unref(&port->ref_cnt) == 1) {
>> - int i;
>> + int i;
>>
>> - netdev_close(port->netdev);
>> - netdev_restore_flags(port->sf);
>> + netdev_close(port->netdev);
>> + netdev_restore_flags(port->sf);
>>
>> - for (i = 0; i < netdev_n_rxq(port->netdev); i++) {
>> - netdev_rxq_close(port->rxq[i]);
>> - }
>> - free(port->type);
>> - free(port);
>> + for (i = 0; i < netdev_n_rxq(port->netdev); i++) {
>> + netdev_rxq_close(port->rxq[i]);
>
> Running with the v1, I saw a core here. It appears that the port->netdev is
> freed before the postponed port_destroy__ is executed. Maybe the netdev
> deletion needs also be ovsrcu_postponed()?
>
Or maybe the closing stuff should really be done when unreeling, and only the
freeing of memory should be postponed?
> Jarno
>
>> + }
>> + free(port->type);
>> + free(port);
>> +}
>> +
>> +static void
>> +port_unref(struct dp_netdev_port *port)
>> +{
>> + if (port && ovs_refcount_unref(&port->ref_cnt) == 1) {
>> + ovsrcu_postpone(port_destroy__, port);
>> }
>> }
>
>
> Jarno
>
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
