> -----Original Message-----
> From: Yan, Zhirun <zhirun....@intel.com>
> Sent: Wednesday, November 18, 2020 11:15 AM
> To: Cao, Yahui <yahui....@intel.com>
> Cc: dev@dpdk.org; Zhang, Qi Z <qi.z.zh...@intel.com>; Wang, Xiao W
> <xiao.w.w...@intel.com>; Su, Simei <simei...@intel.com>
> Subject: [PATCH v1] net/ice: refactor flow pattern parser
>
> * Cao, Yahui <yahui....@intel.com> [2020-11-17 23:31:40 +0800]:
>
> > Hi Zhirun,
> >
> > I think it is great to refactor ice fdir to differentiate inner
> > and outer field vector in a more clear way. Thanks for your commit.
> > It seems there still needs some effort to complete this patch.
> >
> > > -----Original Message-----
> > > From: Yan, Zhirun <zhirun....@intel.com>
> > > Sent: Tuesday, November 17, 2020 4:45 PM
> > > To: dev@dpdk.org; Zhang, Qi Z <qi.z.zh...@intel.com>; Cao, Yahui
> > > <yahui....@intel.com>; Wang, Xiao W
> <xiao.w.w...@intel.com>;
> > > Su, Simei <simei...@intel.com>
> > > Cc: Yan, Zhirun <zhirun....@intel.com>
> > > Subject: [PATCH v1] net/ice: refactor flow pattern parser
> > >
> > > Distinguish inner/outer fields. And avoid too many nested conditionals
> > > in each type's parser.
> > [Cao, Yahui]
> > Since this is quite a huge refactor, could you give a more detailed
> > description in the commit message? Thanks.
> > It would be better if you can make them a patchset.
> >
> Thanks Yahui, will add more details. All changes in one function.
>
> > >
> > > Signed-off-by: Zhirun Yan <zhirun....@intel.com>
> > > ---
> > > drivers/net/ice/ice_fdir_filter.c | 504 ++++++++++++++++--------------
> > > 1 file changed, 269 insertions(+), 235 deletions(-)
> > >
> > > diff --git a/drivers/net/ice/ice_fdir_filter.c
> > > b/drivers/net/ice/ice_fdir_filter.c
> > > index 175abcdd5c..b53ed30b1c 100644
> > > --- a/drivers/net/ice/ice_fdir_filter.c
> > > +++ b/drivers/net/ice/ice_fdir_filter.c
> > > @@ -1646,7 +1646,9 @@ ice_fdir_parse_pattern(__rte_unused struct
> > > ice_adapter *ad,
> > > const struct rte_flow_item_vxlan *vxlan_spec, *vxlan_mask;
> > > const struct rte_flow_item_gtp *gtp_spec, *gtp_mask;
> > > const struct rte_flow_item_gtp_psc *gtp_psc_spec, *gtp_psc_mask;
> > > - uint64_t input_set = ICE_INSET_NONE;
> > > + uint64_t inner_input_set = ICE_INSET_NONE;
> > > + uint64_t outer_input_set = ICE_INSET_NONE;
> > > + uint64_t *input_set;
> > > uint8_t flow_type = ICE_FLTR_PTYPE_NONF_NONE;
> > > uint8_t ipv6_addr_mask[16] = {
> > > 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
> > > @@ -1655,289 +1657,315 @@ ice_fdir_parse_pattern(__rte_unused struct
> > > ice_adapter *ad,
> > > uint32_t vtc_flow_cpu;
> > > uint16_t ether_type;
> > > enum rte_flow_item_type next_type;
> > > + bool is_outer = true;
> > > + struct ice_fdir_extra *p_ext_data;
> > > + struct ice_fdir_v4 *p_v4;
> > > + struct ice_fdir_v6 *p_v6;
> > >
> > > + for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
> > > + if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
> > > + tunnel_type = ICE_FDIR_TUNNEL_TYPE_VXLAN;
> > > + break;
> > > + }
> > [Cao, Yahui]
> > You should take both of VXLAN and GTP cases into consideration
>
> Yes, will add GTP in v2.
>
> > > + }
> > > +
> > > + /* This loop parse flow pattern and distinguish Non-tunnel and tunnel
> > > + * flow. For tunnel flow, reuse non-tunnel structure to track inner
> > > + * part.
> > > + *
> > > + * is_outer tunnel_type p_input_set input_set_bit data_struct
> > > + * Non-Tun Y N inner outer origin
> > > + * Tun-out Y Y outer outer outer
> > > + * Tun-in N Y inner inner origin
> > > + */
> > > for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
> > > if (item->last) {
> > > rte_flow_error_set(error, EINVAL,
> > > - RTE_FLOW_ERROR_TYPE_ITEM,
> > > - item,
> > > - "Not support range");
> > > + RTE_FLOW_ERROR_TYPE_ITEM,
> > > + item,
> > > + "Not support range");
> > > return -rte_errno;
> > > }
> > > item_type = item->type;
> > >
> > > + input_set = (tunnel_type && is_outer) ?
> > > + &outer_input_set :
> > > + &inner_input_set;
> > > +
> > > + if (l3 == RTE_FLOW_ITEM_TYPE_IPV4)
> > > + p_v4 = (tunnel_type && is_outer) ?
> > > + &filter->input.ip_outer.v4 :
> > > + &filter->input.ip.v4;
> > > + if (l3 == RTE_FLOW_ITEM_TYPE_IPV6)
> > > + p_v6 = (tunnel_type && is_outer) ?
> > > + &filter->input.ip_outer.v6 :
> > > + &filter->input.ip.v6;
> > > +
> > [Cao, Yahui]
> > Why do you put p_v4 value assignment out of switch case
> > RTE_FLOW_ITEM_TYPE_IPV4?
> > Why do you put p_v6 value assignment out of switch case
> > RTE_FLOW_ITEM_TYPE_IPV6?
> >
>
> Yes, for RTE_FLOW_ITEM_TYPE_IPV4/6, p_v4/6 assigned in each case as you
> say.
>
> But this is only for L4 layer parser. For L4 layer, the p_v4/6 is the shared
> pointer
> will be used after L3 loop.
>
>
> > > switch (item_type) {
> > > case RTE_FLOW_ITEM_TYPE_ETH:
> > > + flow_type = ICE_FLTR_PTYPE_NON_IP_L2;
> > > eth_spec = item->spec;
> > > eth_mask = item->mask;
> > > - next_type = (item + 1)->type;
> > >
> > > - if (eth_spec && eth_mask) {
> > > - if (!rte_is_zero_ether_addr(ð_mask->dst)) {
> > > - input_set |= ICE_INSET_DMAC;
> > > -
> > > rte_memcpy(&filter->input.ext_data.dst_mac,
> > > - ð_spec->dst,
> > > - RTE_ETHER_ADDR_LEN);
> > > - }
> > > + if (!(eth_spec && eth_mask))
> > > + break;
> > >
> > > - if (!rte_is_zero_ether_addr(ð_mask->src)) {
> > > - input_set |= ICE_INSET_SMAC;
> > > -
> > > rte_memcpy(&filter->input.ext_data.src_mac,
> > > - ð_spec->src,
> > > - RTE_ETHER_ADDR_LEN);
> > > - }
> > > + *input_set |= is_outer ? ICE_PROT_MAC_OUTER :
> > > ICE_PROT_MAC_INNER;
> > [Cao, Yahui]
> > ICE_PROT_XXX is internally used. You should use ICE_INSET_XXX version.
> > The same comment applies for similar case elsewhere.
>
> How to distinguish internally used? ICE_PROT_XXX and ICE_INSET_XXX are
> defined in same file ice_generic_flow.h. And this kind of macro tries
> to introduce inner/outer bit to distinguish inner/outer, but actually
> it will mix field with its location. For FDIR, it seems not a good way.
>
[Cao, Yahui]
The distinguish rule is that only ICE_INSET_XXX is allowed to be used for
input_set value.
Otherwise, it may cause malfunction.
> There is much more cases for inner/outer and src/dst. So I try to
> use this line to set outer/inner bit first, and src/dst bit can be reused.
>
[Cao, Yahui]
It is a historical reason that input_set have both inner and outer field
defined.
To not break the original design and also compatible with current design,
I suggest we can just deprecate the tunnel version field ICE_INSET_TUN_XXX and
instead use the non-tunnel version like ICE_INSET_XXX.
For example, if you specify a vxlan filter rule with outter src ip and inner
dst udp,
The value is like:
outer_input_set = ICE_INSET_IPV4_SRC
inner_input_set = ICE_INSET_UDP_DST_PORT
> > > + if (!rte_is_zero_ether_addr(ð_mask->dst))
> > > + *input_set |= ICE_DMAC;
> > [Cao, Yahui]
> > You should not use ICE_DMAC here. You should use ICE_INSET_XXX version.
> > The same comment applies for similar case elsewhere.
>
> I guess this part should re-design if we want to use. So I prefer not to use
> it.
>
> > > + if (!rte_is_zero_ether_addr(ð_mask->src))
> > > + *input_set |= ICE_SMAC;
> > >
> > > - /* Ignore this field except for
> > > ICE_FLTR_PTYPE_NON_IP_L2 */
> > > - if (eth_mask->type == RTE_BE16(0xffff) &&
> > > - next_type == RTE_FLOW_ITEM_TYPE_END) {
> > > - input_set |= ICE_INSET_ETHERTYPE;
> > > - ether_type =
> > > rte_be_to_cpu_16(eth_spec->type);
> > > -
> > > - if (ether_type == RTE_ETHER_TYPE_IPV4 ||
> > > - ether_type == RTE_ETHER_TYPE_IPV6) {
> > > - rte_flow_error_set(error,
> > > EINVAL,
> > > -
> > > RTE_FLOW_ERROR_TYPE_ITEM,
> > > - item,
> > > - "Unsupported
> > > ether_type.");
> > > - return -rte_errno;
> > > - }
> > > -
> > > -
> > > rte_memcpy(&filter->input.ext_data.ether_type,
> > ....
> > > }
> > > --
> > > 2.25.1
> >
>
> --
> Best regards,
> Zhirun Yan