Re: [dpdk-dev] [RFC PATCH 0/1] Dataplane Workload Accelerator library

[Jerin Jacob <jerinjacobk@gmail.com>]

On Mon, Nov 1, 2021 at 3:25 AM Thomas Monjalon <thomas@monjalon.net> wrote:
>
> 31/10/2021 22:13, Jerin Jacob:
> > On Mon, Nov 1, 2021 at 1:04 AM Thomas Monjalon <thomas@monjalon.net> wrote:
> > >
> > > 31/10/2021 15:01, Jerin Jacob:
> > > > Since rte_flow already has the TLV concept it may not be new to DPDK.
> > >
> > > Where is there TLV in rte_flow?
> >
> > struct rte_flow_item {
> >         enum rte_flow_item_type type; /**< Item type. */
> >         const void *spec; /**< Pointer to item specification structure. */
> >
> > Type is the tag here and the spec is the value here. Length is the
> > size of the specification structure.
> > rte_flows spec does not support/need zero length variable at the end
> > of spec structure,
> > that reason for not embedding explicit length value as it is can be
> > derived from sizeof(specification structure).
>
> Ah OK I see what you mean.
> But rte_flow_item is quite limited,
> it is not the kind of TLV with multiple levels of nesting.
> Do you need nesting of objects in DWA?

No. Currently, ethernet-based on host port has the following
prototype[1] and it has array
of TLV(not in continuous memory). For simplicity, we could remove
legth value from rte_dwa_tlv and just
keep like rte_flow and let the payload contain the length of the
message if the message has
a variable length. See rte_dwa_profile_l3fwd_d2h_exception_pkts::nb_pkts below.


[1]
+/**
+ * Receive a burst of TLVs of type `TYPE_USER_PLANE` from the Rx queue
+ * designated by its *queue_id* of DWA object *obj*.
+ *
+ * @param obj
+ *   DWA object.
+ * @param queue_id
+ *   The identifier of Rx queue id. The queue id should in the range of
+ *   [0 to rte_dwa_port_host_ethernet_config::nb_rx_queues].
+ * @param[out] tlvs
+ *   Points to an array of *nb_tlvs* tlvs of type *rte_dwa_tlv* structure
+ *   to be received.
+ * @param nb_tlvs
+ *   The maximum number of TLVs to received.
+ *
+ * @return
+ * The number of TLVs actually received on the Rx queue. The return
+ * value can be less than the value of the *nb_tlvs* parameter when the
+ * Rx queue is not full.
+ */
+uint16_t rte_dwa_port_host_ethernet_rx(rte_dwa_obj_t obj, uint16_t queue_id,
+ struct rte_dwa_tlv **tlvs, uint16_t nb_tlvs);


[2]
example TLV for TYPE_USER_PLANE traffic.


+ /**
+ * Attribute |  Value
+ * ----------|--------
+ * Tag       | RTE_DWA_TAG_PROFILE_L3FWD
+ * Stag      | RTE_DWA_STAG_PROFILE_L3FWD_D2H_EXCEPTION_PACKETS
+ * Direction | D2H
+ * Type      | TYPE_USER_PLANE
+ * Payload   | struct rte_dwa_profile_l3fwd_d2h_exception_pkts
+ * Pair TLV  | NA
+ *
+ * Response from DWA of exception packets.
+ */

+/**
+ * Payload of RTE_DWA_STAG_PROFILE_L3FWD_D2H_EXCEPTION_PACKETS message.
+ */
+struct rte_dwa_profile_l3fwd_d2h_exception_pkts {
+ uint16_t nb_pkts;
+ /**< Number of packets in the variable size array.*/
+ uint16_t rsvd16;
+ /**< Reserved field to make pkts[0] to be 64bit aligned.*/
+ uint32_t rsvd32;
+ /**< Reserved field to make pkts[0] to be 64bit aligned.*/
+ struct rte_mbuf *pkts[0];
+ /**< Array of rte_mbufs of size nb_pkts. */
+} __rte_packed;


>
> > > > I really liked rte_flow enablement of ABI combability and its ease of adding
> > > > new stuff. Try to follow similar stuff which is proven in DPDK.
> > > > Ie. New profile creation will very easy, it will be a matter of identifying
> > > > the TLVs and their type and payload, rather than everyone comes with
> > > > new APIs in every profile.
> > > >
> > > > > Why not use protobuf and its IDL to specify the interface?
> > >
> > > Yes I think it is important to discuss alternatives,
> > > and at least get justifications of why TLV is chosen among others.
> >
> > Yes. Current list is
> >
> > 1) Very easy to enable ABI compatibility.
> > 2) If it needs to be transported over network etc it needs to be
> > packed so that way it is easy for implementation to do that
> > with TLV also gives better performance in such
> > cases by avoiding reformatting or possibly avoiding memcpy etc.
> > 3) It is easy to plugin with another high-level programing language as
> > just one API.
> > 4) Easy to decouple DWA core library functionalities from profile.
> > 5) Easy to enable asynchronous scheme using request and response TLVs.
> > 6) Most importantly, We could introduce type notion with TLV
> > (connected with the type of message  See TYPE_ATTACHED, TYPE_STOPPED,
> > TYPE_USER_PLANE etc ),
> > That way, we can have a uniform outlook of profiles instead of each profile
> > coming with a setup of its own APIs and __rules__ on the state machine.
> > I think, for a framework to leverage communication mechanisms and other
> > aspects between profiles, it's important to have some synergy between profiles.
> > 7) No Additional library dependencies like gRPC, protobuf
> > 8) Provide driver to implement the optimized means of supporting different
> > transport such as Ethernet, Shared memory, PCIe DMA style HW etc.
> > 9) Avoid creating endless APIs and their associated driver function
> > calls for each
> > profile APIs.
>
>
>