On Thu, Jul 9, 2020 at 1:56 PM Jakub Kicinski <k...@kernel.org> wrote:
>
> On Wed, 8 Jul 2020 22:27:13 -0700 Michael Chan wrote:
> > On Wed, Jul 8, 2020 at 6:18 PM Jakub Kicinski <k...@kernel.org> wrote:
> > > @@ -1752,10 +1752,8 @@ struct bnxt {
> > > #define BNXT_FW_MAJ(bp) ((bp)->fw_ver_code >> 48)
> > >
> > > __be16 vxlan_port;
> >
> > We can also do away with vxlan_port and nge_port, now that we no
> > longer need to pass the port from NDO to workqueue. We just need to
> > initialize the 2 firmware tunnel IDs to INVALID_HW_RING_ID before use
> > and after free. But it is ok the way you have it too.
>
> Seems like I need to send a v3 anyway - the only reason I kept them is
> to know if bnxt_hwrm_free_tunnel_ports() has to issue its commands or
> not.
>
> Are you suggesting I just add a flag that'd say "tunnel in use" instead
> of holding onto the ports? Or free unconditionally?
Here's what I would do:
1. Change vxlan_fw_dst_port_id and nge_fw_dst_port_id to u16 and
initialize them to INVALID_HW_RING_ID. This will indicate that the
ports are not offloaded. The assignments of these fields to the fw
message will now need cpu_to_le16().
2. bnxt_hwrm_free_tunnel_ports() will check if
bp->vxlan_fw_dst_port_id == INVALID_HW_RING_ID before freeing. After
freeing, it will set bp->vxlan_fw_dst_port_id to INVALID_HW_RING_ID.
Same for the Geneve port.
