Re: [dpdk-dev] [PATCH 4/5] net/iavf: fix protocol size for virtchnl copy

[Jack Min <jackmin@nvidia.com>]

> -----Original Message-----
> From: Ferruh Yigit <ferruh.yigit@intel.com>
> Sent: Tuesday, November 17, 2020 00:23
> To: Xiaoyu Min <jackmin@mellanox.com>; Jingjing Wu <jingjing.wu@intel.com>;
> Beilei Xing <beilei.xing@intel.com>
> Cc: dev@dpdk.org; Jack Min <jackmin@nvidia.com>; NBU-Contact-Thomas
> Monjalon <thomas@monjalon.net>; Andrew Rybchenko
> <arybchenko@solarflare.com>; Ori Kam <orika@nvidia.com>; Dekel Peled
> <dekelp@nvidia.com>
> Subject: Re: [dpdk-dev] [PATCH 4/5] net/iavf: fix protocol size for virtchnl copy
> 
> On 11/16/2020 7:55 AM, Xiaoyu Min wrote:
> > From: Xiaoyu Min <jackmin@nvidia.com>
> >
> > The rte_flow_item_vlan items are refined.
> > The structs do not exactly represent the packet bits captured on the
> > wire anymore so should only copy real header instead of the whole struct.
> >
> > Replace the rte_flow_item_* with the existing corresponding rte_*_hdr.
> >
> > Fixes: 09315fc83861 ("ethdev: add VLAN attributes to ethernet and VLAN
> items")
> >
> > Signed-off-by: Xiaoyu Min <jackmin@nvidia.com>
> > ---
> >   drivers/net/iavf/iavf_fdir.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/net/iavf/iavf_fdir.c b/drivers/net/iavf/iavf_fdir.c
> > index d683a468c1..7054bde0b9 100644
> > --- a/drivers/net/iavf/iavf_fdir.c
> > +++ b/drivers/net/iavf/iavf_fdir.c
> > @@ -541,7 +541,7 @@ iavf_fdir_parse_pattern(__rte_unused struct
> iavf_adapter *ad,
> >   				VIRTCHNL_ADD_PROTO_HDR_FIELD_BIT(hdr,
> ETH, ETHERTYPE);
> >
> >   				rte_memcpy(hdr->buffer,
> > -					eth_spec, sizeof(*eth_spec));
> > +					eth_spec, sizeof(struct rte_ether_hdr));
> 
> This requires 'struct rte_flow_item_eth' should have 'struct rte_ether_hdr' as
> first element, and I suspect this usage exists in a few more locations, but I
> wonder if this assumption is real and documented in somewhere?
> I am not just talking about 'struct rte_flow_item_eth', but for all
> 'rte_flow_item_*'...
> 
I think this is not documented and this assumption is not real.
I've created one ticket on Bugzilla (https://bugs.dpdk.org/show_bug.cgi?id=581) to track.

> 
> 
> btw, while checking for the 'struct rte_flow_item_eth', pahole shows it is using
> 20 bytes, and I suspect this is not the intention with the reserved field:
> 
> struct rte_flow_item_eth {
> 	struct rte_ether_addr      dst;                  /*     0     6 */
> 	struct rte_ether_addr      src;                  /*     6     6 */
> 	uint16_t                   type;                 /*    12     2 */
> 
> 	/* Bitfield combined with previous fields */
> 
> 	uint32_t                   has_vlan:1;           /*    12:15  4 */
> 
> 	/* XXX 31 bits hole, try to pack */
> 
> 	uint32_t                   reserved:31;          /*    16: 1  4 */
> 
> 	/* size: 20, cachelines: 1, members: 5 */
> 	/* bit holes: 1, sum bit holes: 31 bits */
> 	/* bit_padding: 1 bits */
> 	/* last cacheline: 20 bytes */
> };
> 
> 'has_vlan' seems combined with previous field to make together 32 bits. So the
> 'reserved' field is occupying a new 32 bits all by itself.
> 
> What about changing the struct as following, while we can change the ABI:
> struct rte_flow_item_eth {
> 	struct rte_ether_addr      dst;                  /*     0     6 */
> 	struct rte_ether_addr      src;                  /*     6     6 */
> 	uint16_t                   type;                 /*    12     2 */
> 	uint16_t                   has_vlan:1;           /*    14:15  2 */
> 	uint16_t                   reserved:15;          /*    14: 0  2 */
> 
> 	/* size: 16, cachelines: 1, members: 5 */
> 	/* last cacheline: 16 bytes */
> };
> 

Well we probably need to discuss this in next release. 
It's too late to change this API at this moment.

-Jack