[dpdk-dev] [PATCH v8] net/iavf: support flex desc metadata extraction

[Ferruh Yigit]

On 10/13/2020 9:17 AM, Jeff Guo wrote:
> Enable metadata extraction for flexible descriptors in AVF, that would
> allow network function directly get metadata without additional parsing
> which would reduce the CPU cost for VFs. The enabling metadata
> extractions involve the metadata of VLAN/IPv4/IPv6/IPv6-FLOW/TCP/MPLS
> flexible descriptors, and the VF could negotiate the capability of
> the flexible descriptor with PF and correspondingly configure the
> specific offload at receiving queues.
> 
> Signed-off-by: Jeff Guo <jia.guo at intel.com>
> Acked-by: Haiyue Wang <haiyue.wang at intel.com>
> ---
> v8:
> rebase patch for apply issue
> 
> v7:
> clean some useless and add doc
> 
> v6:
> rebase patch
> 
> v5:
> remove ovs configure since ovs is not protocol extraction
> 
> v4:
> add flex desc type in rx queue for handling vector path
> handle ovs flex type
> 
> v3:
> export these global symbols into .map
> 
> v2:
> remove makefile change and modify the rxdid handling
> ---
>   config/rte_config.h                       |   3 +
>   doc/guides/nics/intel_vf.rst              |  16 +
>   doc/guides/rel_notes/release_20_11.rst    |   6 +
>   drivers/net/iavf/iavf.h                   |  24 +-
>   drivers/net/iavf/iavf_ethdev.c            | 394 ++++++++++++++++++++++
>   drivers/net/iavf/iavf_rxtx.c              | 252 ++++++++++++--
>   drivers/net/iavf/iavf_rxtx.h              | 168 +++++----
>   drivers/net/iavf/iavf_rxtx_vec_common.h   |   3 +
>   drivers/net/iavf/iavf_vchnl.c             |  22 +-
>   drivers/net/iavf/meson.build              |   2 +
>   drivers/net/iavf/rte_pmd_iavf.h           | 250 ++++++++++++++
>   drivers/net/iavf/rte_pmd_iavf_version.map |  13 +
>   12 files changed, 1039 insertions(+), 114 deletions(-)
>   create mode 100644 drivers/net/iavf/rte_pmd_iavf.h
> 
> diff --git a/config/rte_config.h b/config/rte_config.h
> index 03d90d78bc..2c53072c3d 100644
> --- a/config/rte_config.h
> +++ b/config/rte_config.h
> @@ -127,6 +127,9 @@
>   #define RTE_LIBRTE_I40E_QUEUE_NUM_PER_VF 4
>   #define RTE_LIBRTE_I40E_QUEUE_NUM_PER_VM 4
>   
> +/* iavf defines */
> +#undef RTE_LIBRTE_IAVF_16BYTE_RX_DESC
> +

Hi Jeff,

The 'RTE_LIBRTE_IAVF_16BYTE_RX_DESC' was already there, not introduced with this 
patch, so I think better to add this change as different patch.

Also not sure if we want to add more config options to the 'rte_config.h', 
indeed otherway around and we are trying to get rid of as much as compile time 
optios.
cc'ed Bruce too.

>   /* Ring net PMD settings */
>   #define RTE_PMD_RING_MAX_RX_RINGS 16
>   #define RTE_PMD_RING_MAX_TX_RINGS 16
> diff --git a/doc/guides/nics/intel_vf.rst b/doc/guides/nics/intel_vf.rst
> index ade5152595..207f456143 100644
> --- a/doc/guides/nics/intel_vf.rst
> +++ b/doc/guides/nics/intel_vf.rst
> @@ -615,3 +615,19 @@ which belongs to the destination VF on the VM.
>   .. figure:: img/inter_vm_comms.*
>   
>      Inter-VM Communication
> +
> +
> +Pre-Installation Configuration
> +------------------------------
> +
> +Config File Options
> +~~~~~~~~~~~~~~~~~~~
> +
> +The following options can be modified in the ``config`` file.
> +Please note that enabling debugging options may affect system performance.
> +
> +- ``CONFIG_RTE_LIBRTE_IAVF_16BYTE_RX_DESC`` (default ``n``)

There is no 'CONFIG_RTE_LIBRTE_IAVF_16BYTE_RX_DESC' anymore, this is from make 
days naming.

Instead, what do you think not adding the 'RTE_LIBRTE_IAVF_16BYTE_RX_DESC' to 
the 'rte_config.h', but document how this flag can be provided by meson during 
build:
meson -Dc_args="-DRTE_LIBRTE_IAVF_16BYTE_RX_DESC"

And we should plan for long term to convert this compile time flag to runtime 
devargs.

What do you think?

> +
> +  Toggle to use a 16-byte RX descriptor, by default the RX descriptor is 32 byte.
> +  Configure to 16-byte Rx descriptor may cause a negotiation failure during VF driver initialization
> +  if the PF driver doesn't support.
> diff --git a/doc/guides/rel_notes/release_20_11.rst b/doc/guides/rel_notes/release_20_11.rst
> index e7691ee732..93d3ccc60a 100644
> --- a/doc/guides/rel_notes/release_20_11.rst
> +++ b/doc/guides/rel_notes/release_20_11.rst
> @@ -160,6 +160,12 @@ New Features
>     packets with specified ratio, and apply with own set of actions with a fate
>     action. When the ratio is set to 1 then the packets will be 100% mirrored.
>   
> +* **Updated Intel iavf driver.**
> +
> +  Updated iavf PMD with new features and improvements, including:
> +
> +  * Added support for flexible descriptor metadata extraction.
> +

Can you please move the update to the net drivers block, instead of very bottom.
There is an order in the release notes (as commented in section header) like:
- core libs
- ethdev lib related changes
- ethdev PMDS change
- ...

<...>

> +
> +EXPERIMENTAL {
> +        global:
> +
> +        # added in 20.11
> +        rte_net_iavf_dynfield_proto_xtr_metadata_offs;
> +        rte_net_iavf_dynflag_proto_xtr_vlan_mask;
> +        rte_net_iavf_dynflag_proto_xtr_ipv4_mask;
> +        rte_net_iavf_dynflag_proto_xtr_ipv6_mask;
> +        rte_net_iavf_dynflag_proto_xtr_ipv6_flow_mask;
> +        rte_net_iavf_dynflag_proto_xtr_tcp_mask;
> +        rte_net_iavf_dynflag_proto_xtr_ip_offset_mask;

As a namespace previously "rte_pmd_xxx" was used for PMD specific APIs, can you 
please switch to that?
'rte_net_' is used by the 'librte_net' library.

Above list is the dynfield values, what is the correct usage for dynfields,
1- Put dynfileds names in to the header, and application does a lookup 
('rte_mbuf_dynfield_lookup()') to get the dynfield values.
or
2- Expose dynfield values to be accessed directly from application, as done above.

@Oliver, can you please support.

I can see (1) has advantage of portability if more than one PMD supports same 
dynfield names, but that sees not a case for above ones.