[ovs-dev] [PATCH] tc: Use skip_hw flag when probing tc features

[Ilya Maximets]

On 8/3/20 12:58 PM, Roi Dayan wrote:
> There is no need to pass tc rules to hw when just probing
> for tc features. this will avoid redundant errors from hw drivers
> that may happen.

That makes sense.  Thanks!
Few comments inline.

> 
> Signed-off-by: Roi Dayan <roid at mellanox.com>
> ---
>  lib/netdev-offload-tc.c |  2 ++
>  lib/tc.c                | 13 ++++++-------
>  lib/tc.h                |  8 ++++++++
>  3 files changed, 16 insertions(+), 7 deletions(-)
> 
> diff --git a/lib/netdev-offload-tc.c b/lib/netdev-offload-tc.c
> index 2c9c6f4cae8b..18ff380f9861 100644
> --- a/lib/netdev-offload-tc.c
> +++ b/lib/netdev-offload-tc.c
> @@ -1918,6 +1918,7 @@ probe_multi_mask_per_prio(int ifindex)
>  
>      memset(&flower, 0, sizeof flower);
>  
> +    flower.tc_policy = TC_POLICY_SKIP_HW;
>      flower.key.eth_type = htons(ETH_P_IP);
>      flower.mask.eth_type = OVS_BE16_MAX;
>      memset(&flower.key.dst_mac, 0x11, sizeof flower.key.dst_mac);
> @@ -1965,6 +1966,7 @@ probe_tc_block_support(int ifindex)
>  
>      memset(&flower, 0, sizeof flower);
>  
> +    flower.tc_policy = TC_POLICY_SKIP_HW;
>      flower.key.eth_type = htons(ETH_P_IP);
>      flower.mask.eth_type = OVS_BE16_MAX;
>      memset(&flower.key.dst_mac, 0x11, sizeof flower.key.dst_mac);
> diff --git a/lib/tc.c b/lib/tc.c
> index c96d095381d7..8761304c92bb 100644
> --- a/lib/tc.c
> +++ b/lib/tc.c
> @@ -65,12 +65,6 @@ VLOG_DEFINE_THIS_MODULE(tc);
>  
>  static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
>  
> -enum tc_offload_policy {
> -    TC_POLICY_NONE,
> -    TC_POLICY_SKIP_SW,
> -    TC_POLICY_SKIP_HW
> -};
> -
>  static enum tc_offload_policy tc_policy = TC_POLICY_NONE;
>  
>  struct tc_pedit_key_ex {
> @@ -2757,6 +2751,7 @@ nl_msg_put_flower_options(struct ofpbuf *request, struct tc_flower *flower)
>      bool is_vlan = eth_type_vlan(flower->key.eth_type);
>      bool is_qinq = is_vlan && eth_type_vlan(flower->key.encap_eth_type[0]);
>      bool is_mpls = eth_type_mpls(flower->key.eth_type);
> +    enum tc_offload_policy policy = flower->tc_policy;
>      int err;
>  
>      /* need to parse acts first as some acts require changing the matching
> @@ -2882,7 +2877,11 @@ nl_msg_put_flower_options(struct ofpbuf *request, struct tc_flower *flower)
>          }
>      }
>  
> -    nl_msg_put_u32(request, TCA_FLOWER_FLAGS, tc_get_tc_cls_policy(tc_policy));
> +    if (policy == TC_POLICY_NONE) {
> +        policy = tc_policy;
> +    }
> +
> +    nl_msg_put_u32(request, TCA_FLOWER_FLAGS, tc_get_tc_cls_policy(policy));
>  
>      if (flower->tunnel) {
>          nl_msg_put_flower_tunnel(request, flower);
> diff --git a/lib/tc.h b/lib/tc.h
> index 028eed5d0658..78f433ceef77 100644
> --- a/lib/tc.h
> +++ b/lib/tc.h
> @@ -312,6 +312,12 @@ is_tcf_id_eq(struct tcf_id *id1, struct tcf_id *id2)
>             && id1->chain == id2->chain;
>  }
>  
> +enum tc_offload_policy {
> +    TC_POLICY_NONE,

Since now the code actually depends on this value to be zero, it might be
good to explicitly set it, or have a build assertion.  This will protect us
from possible issues if someone will try to modify this enum later.
i.e.
    TC_POLICY_NONE = 0,

> +    TC_POLICY_SKIP_SW,
> +    TC_POLICY_SKIP_HW
> +};
> +
>  struct tc_flower {
>      struct tc_flower_key key;
>      struct tc_flower_key mask;
> @@ -337,6 +343,8 @@ struct tc_flower {
>      bool needs_full_ip_proto_mask;
>  
>      enum tc_offloaded_state offloaded_state;
> +    /* used to force skip_hw when probing tc features */

I see that there is a comment below in this file that doesn't follow OVS coding
style, but, please, use the OVS-style comments for a new code, i.e. start with
a capital letter and end with a period.

> +    enum tc_offload_policy tc_policy;
>  };
>  
>  /* assert that if we overflow with a masked write of uint32_t to the last byte
>