Tue, Oct 06, 2015 at 07:16:29AM CEST, sfel...@gmail.com wrote: >On Mon, Oct 5, 2015 at 10:43 AM, Jiri Pirko <j...@resnulli.us> wrote: >> From: Jiri Pirko <j...@mellanox.com> >> >> This is another step on the way to per-world clean cut. Introduce world >> ops hooks which each world can implement in world-specific way. >> Also introduce world infrastructure including function for port worlds >> change. >> >> Signed-off-by: Jiri Pirko <j...@mellanox.com> >> --- >> v1->v2: >> - removed functions to change worlds based on name, as rtnl mode >> set patch is removed from patchset. >> --- >> drivers/net/ethernet/rocker/rocker.h | 56 ++++ >> drivers/net/ethernet/rocker/rocker_main.c | 464 >> +++++++++++++++++++++++++++++- >> 2 files changed, 519 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/net/ethernet/rocker/rocker.h >> b/drivers/net/ethernet/rocker/rocker.h >> index 650caa0..d49bc5d 100644 >> --- a/drivers/net/ethernet/rocker/rocker.h >> +++ b/drivers/net/ethernet/rocker/rocker.h >> @@ -12,7 +12,11 @@ >> #ifndef _ROCKER_H >> #define _ROCKER_H >> >> +#include <linux/kernel.h> >> #include <linux/types.h> >> +#include <linux/netdevice.h> >> +#include <net/neighbour.h> >> +#include <net/switchdev.h> >> >> #include "rocker_hw.h" >> >> @@ -24,4 +28,56 @@ struct rocker_desc_info { >> dma_addr_t mapaddr; >> }; >> >> +struct rocker; >> +struct rocker_port; >> + >> +struct rocker_world_ops { >> + const char *kind; >> + size_t priv_size; >> + size_t port_priv_size; >> + u8 mode; >> + int (*init)(struct rocker *rocker, void *priv); >> + void (*fini)(void *priv); >> + int (*port_init)(struct rocker_port *rocker_port, void *priv, >> + void *port_priv); >> + void (*port_fini)(void *port_priv); >> + int (*port_open)(void *port_priv); >> + void (*port_stop)(void *port_priv); >> + int (*port_attr_stp_state_set)(void *port_priv, u8 state, >> + struct switchdev_trans *trans); >> + int (*port_attr_bridge_flags_set)(void *port_priv, >> + unsigned long brport_flags, >> + struct switchdev_trans *trans); >> + int (*port_attr_bridge_flags_get)(void *port_priv, >> + unsigned long *p_brport_flags); >> + int (*port_obj_vlan_add)(void *port_priv, >> + const struct switchdev_obj_port_vlan *vlan, >> + struct switchdev_trans *trans); >> + int (*port_obj_vlan_del)(void *port_priv, >> + const struct switchdev_obj_port_vlan *vlan); >> + int (*port_obj_vlan_dump)(void *port_priv, >> + struct switchdev_obj_port_vlan *vlan, >> + switchdev_obj_dump_cb_t *cb); >> + int (*port_obj_fib4_add)(void *port_priv, >> + const struct switchdev_obj_ipv4_fib *fib4, >> + struct switchdev_trans *trans); >> + int (*port_obj_fib4_del)(void *port_priv, >> + const struct switchdev_obj_ipv4_fib *fib4); >> + int (*port_obj_fdb_add)(void *port_priv, >> + const struct switchdev_obj_port_fdb *fdb, >> + struct switchdev_trans *trans); >> + int (*port_obj_fdb_del)(void *port_priv, >> + const struct switchdev_obj_port_fdb *fdb); >> + int (*port_obj_fdb_dump)(void *port_priv, >> + struct switchdev_obj_port_fdb *fdb, >> + switchdev_obj_dump_cb_t *cb); >> + int (*port_master_linked)(void *port_priv, struct net_device >> *master); >> + int (*port_master_unlinked)(void *port_priv, struct net_device >> *master); >> + int (*port_neigh_update)(void *port_priv, struct neighbour *n); >> + int (*port_neigh_destroy)(void *port_priv, struct neighbour *n); >> + int (*port_ev_mac_vlan_seen)(void *port_priv, >> + const unsigned char *addr, >> + __be16 vlan_id); >> +}; > >Yuck, void *. Can we do better?
I see nothing wrong with this priv usage. It's done like this on many places. I think it is completely legit, since the call points are well defined and wrapped. > >How about using base struct rocker_port and then sub-classing >world-specific rocker ports. And make rocker_world_ops pass struct >rocker_port * (for the port-centric ops). > >struct rocker_port { > // common stuff for all worlds >}; > >struct my_world_port { > struct rocker_port rocker_port; > // world-specific stuff for port >}; > >#define MY_WORLD_PORT(rocker_port) \ > container_of(rocker_port, struct my_world_port, rocker_port) > >Same for world-centric ops: > >struct rocker_world { > // common stuff for all worlds >}; > >struct my_world { > struct rocker_world rocker_world; > // stuff for this world >}; This would be unnecessary complex. + this cannot be done because you would have to know the world in advance of creating the structs and you would not be able to switch it. No, not a good idea. > > >> +static int rocker_world_port_ev_mac_vlan_seen(struct rocker_port >> *rocker_port, >> + const unsigned char *addr, >> + __be16 vlan_id); >> >> static int rocker_event_mac_vlan_seen(const struct rocker *rocker, >> const struct rocker_tlv *info) >> @@ -1246,6 +1252,7 @@ static int rocker_event_mac_vlan_seen(const struct >> rocker *rocker, >> const unsigned char *addr; >> int flags = ROCKER_OP_FLAG_NOWAIT | ROCKER_OP_FLAG_LEARNED; >> __be16 vlan_id; >> + int err; >> >> rocker_tlv_parse_nested(attrs, ROCKER_TLV_EVENT_MAC_VLAN_MAX, info); >> if (!attrs[ROCKER_TLV_EVENT_MAC_VLAN_PPORT] || >> @@ -1262,6 +1269,10 @@ static int rocker_event_mac_vlan_seen(const struct >> rocker *rocker, >> >> rocker_port = rocker->ports[port_number]; >> >> + err = rocker_world_port_ev_mac_vlan_seen(rocker_port, addr, vlan_id); >> + if (err) >> + return err; >> + > >Why is this in this patch? > >> +static int rocker_world_port_attr_stp_state_set(struct rocker_port >> *rocker_port, >> + u8 state, >> + struct switchdev_trans >> *trans) >> +{ >> + struct rocker_world *world = rocker_port->world; >> + >> + if (!world->ops->port_attr_stp_state_set) >> + return 0; > >Should return -EOPNOTSUPP. Returning 0 means the operation >succeeded. I would like to avoid silent failures. If the user does >something to a port, and the operation is not supported on the port, >return -EOPNOTSUPP so user gets some feedback that operation failed. >Returning 0 gives the user feedback that the operation succeeded, when >in fact nothing happened at all. At this stage of the patchset, this has to return 0 in order not to break the flows. -EOPNOTSUPP is treated like 0 upper in callers anyway. But I agree that we can return -OPNOTSUPP in these cases. Will add a patch that changes this to the end of the patchset. > >> + return >> world->ops->port_attr_stp_state_set(rocker_port->world_port_priv, >> + state, trans); >> +} >> + >> +static int >> +rocker_world_port_attr_bridge_flags_set(struct rocker_port *rocker_port, >> + unsigned long brport_flags, >> + struct switchdev_trans *trans) >> +{ >> + struct rocker_world *world = rocker_port->world; >> + >> + if (!world->ops->port_attr_bridge_flags_set) >> + return 0; > >Same comment, pretty much down the list... -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html