On 22 Jan 2026, at 23:24, Ilya Maximets wrote:
> On 1/22/26 12:55 PM, Eelco Chaudron wrote:
>> The original series allowed the provider priority to be configured
>> on the port using the other-config:hw-offload-priority option.
>>
>> However, this is not the correct level. The configuration should be
>> applied at the interface level, allowing, for example, specific bond
>> members to have different configurations.
>>
>> This patch moves the configuration from the port to the interface.
>>
>> Fixes: 5c9b96ed5abd ("dpif-offload: Allow per-port offload provider priority
>> config.")
>> Signed-off-by: Eelco Chaudron <[email protected]>
>> ---
>> NEWS | 7 +++---
>> lib/dpif-offload.c | 23 +++++++++++-------
>> tests/ofproto-dpif.at | 8 +++----
>> vswitchd/vswitch.xml | 56 ++++++++++++++++++++++++-------------------
>> 4 files changed, 54 insertions(+), 40 deletions(-)
>>
>> diff --git a/NEWS b/NEWS
>> index 3ff5b3633b..c3470b84ec 100644
>> --- a/NEWS
>> +++ b/NEWS
>> @@ -27,9 +27,10 @@ v3.7.0 - xx xxx xxxx
>> * New global configuration option 'other-config:hw-offload-priority'
>> that allows to set the order of the hardware offload providers
>> to try when multiple exist for a given datapath implementation.
>> - * New per-port configuration option 'other-config:hw-offload-priority'
>> - that allows to set order of the hardware offload providers or disable
>> - flow offloading on a particular port.
>> + * New per-interface configuration option
>> + 'other-config:hw-offload-priority' that allows to set order of the
>> + hardware offload providers or disable flow offloading on a particular
>> + interface.
>> - OVSDB-IDL:
>> * New ovsdb_idl_txn_assert_read_only() interface to mark transactions
>> as read-only and trigger assertion failure when application attempts
>> diff --git a/lib/dpif-offload.c b/lib/dpif-offload.c
>> index f6ab65e980..0ee2faf08f 100644
>> --- a/lib/dpif-offload.c
>> +++ b/lib/dpif-offload.c
>> @@ -59,7 +59,7 @@ static const struct dpif_offload_class
>> *base_dpif_offload_classes[] = {
>> static char *dpif_offload_provider_priority_list = NULL;
>> static atomic_bool offload_global_enabled = false;
>> static atomic_bool offload_rebalance_policy = false;
>> -static struct smap port_order_cfg = SMAP_INITIALIZER(&port_order_cfg);
>> +static struct smap iface_order_cfg = SMAP_INITIALIZER(&iface_order_cfg);
>>
>> static int
>> dpif_offload_register_provider__(const struct dpif_offload_class *class)
>> @@ -527,7 +527,7 @@ dpif_offload_port_add(struct dpif *dpif, struct netdev
>> *netdev,
>> odp_port_t port_no)
>> {
>> struct dpif_offload_provider_collection *collection;
>> - const char *port_priority = smap_get(&port_order_cfg,
>> + const char *port_priority = smap_get(&iface_order_cfg,
>> netdev_get_name(netdev));
>> struct dpif_offload *offload;
>>
>> @@ -773,21 +773,26 @@ dpif_offload_set_global_cfg(const struct
>> ovsrec_open_vswitch *cfg)
>> }
>> }
>>
>> - /* Filter out the 'hw-offload-priority' per port setting we need it
>> before
>> - * ports are added, so we can assign the correct offload-provider.
>> + /* Filter out the 'hw-offload-priority' per interface setting we need it
>> + * before ports are added, so we can assign the correct
>> offload-provider.
>
> nit: Should there be a period before 'we' ?
>
> Otherwise,
> Acked-by: Ilya Maximets <[email protected]>
With the nit fixes, applied to main and branch-3.7.
Thanks for the review!
//Eelco
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
