NVIDIA roadmap for 22.11

NVIDIA roadmap for 22.11

[Lior Margalit]

[-- Attachment #1: Type: text/plain, Size: 4930 bytes --]

Please find below NVIDIA roadmap for 22.11 release:


rte_flow new APIs:

===============

[1] Add async flow connection tracking configuration

https://patches.dpdk.org/project/dpdk/patch/20220809132534.24441-1-suanmingm@nvidia.com/<https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatches.dpdk.org%2Fproject%2Fdpdk%2Fpatch%2F20220809132534.24441-1-suanmingm%40nvidia.com%2F&data=05%7C01%7Clmargalit%40nvidia.com%7C461f54315657470a9a9208da90c3e33d%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637981468244215709%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=mRF7%2FiEZL0t9ToOr0izTqnZmnDE3c2cFnP33c%2BvfnCU%3D&reserved=0>

Configuration of number of Connection Tracking actions to optimize flows insertion.



[2] Add indirect action async query

https://patches.dpdk.org/project/dpdk/patch/20220809132824.25890-1-suanmingm@nvidia.com/<https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatches.dpdk.org%2Fproject%2Fdpdk%2Fpatch%2F20220809132824.25890-1-suanmingm%40nvidia.com%2F&data=05%7C01%7Clmargalit%40nvidia.com%7C461f54315657470a9a9208da90c3e33d%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637981468244371938%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=Me9fzhcX%2Bag3USorvEtK9uHFfMwSCBcXpQiYuKdYwKA%3D&reserved=0>

Support async query call, similar to the other operations (create, update and destroy). The call to the query API returns immediately to free the CPU for other tasks. The Application can get the query results from rte_flow_pull() asynchronously.



[3] Add direction info when creating the transfer table.

https://patches.dpdk.org/project/dpdk/patch/20220811071315.851032-1-rongweil@nvidia.com/<https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatches.dpdk.org%2Fproject%2Fdpdk%2Fpatch%2F20220811071315.851032-1-rongweil%40nvidia.com%2F&data=05%7C01%7Clmargalit%40nvidia.com%7C461f54315657470a9a9208da90c3e33d%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637981468244371938%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=q%2FOXEVEZ%2BdPGclRxNgqpXAWura0FMMvgYRvsaJwuL4Y%3D&reserved=0>

Add option to define the transfer_mode (traffic originates from uplink/vport/both) when creating the HW table to reduce memory footprint.



[4] Add send to kernel action

https://patches.dpdk.org/project/dpdk/patch/20220811113544.1718643-1-michaelsav@nvidia.com/<https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatches.dpdk.org%2Fproject%2Fdpdk%2Fpatch%2F20220811113544.1718643-1-michaelsav%40nvidia.com%2F&data=05%7C01%7Clmargalit%40nvidia.com%7C461f54315657470a9a9208da90c3e33d%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637981468244371938%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=Ito5mGfiY%2BjPL2wsKL37bJt0rN0uK8Kyx19Oz7%2B3%2FT4%3D&reserved=0>

Introduce new rte_flow action to route the packet to the kernel.



[5] Introduce hairpin memory capabilities

https://patches.dpdk.org/project/dpdk/patch/20220811120530.191683-1-dsosnowski@nvidia.com/<https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatches.dpdk.org%2Fproject%2Fdpdk%2Fpatch%2F20220811120530.191683-1-dsosnowski%40nvidia.com%2F&data=05%7C01%7Clmargalit%40nvidia.com%7C461f54315657470a9a9208da90c3e33d%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637981468244371938%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=urbKk4eBdVoPvzsIhHcjoK2%2BhnAzPlTubjfbq2R6Y5M%3D&reserved=0>

Introduce new configuration options specific to hairpin SQs/RQs which result in reduction of ICMC misses on hairpin queues reducing latency and increasing ZPL throughput.​

locked_device_memory - this flag corresponds to locking HW memory available for RQs.​

rte_memory - this flag corresponds to usage of host memory buffer as WQ buffer for SQs.



[6] Support GTP PSC QFI in modify_field flow action

https://patches.dpdk.org/project/dpdk/patch/20220921101839.1711058-2-dsosnowski@nvidia.com/

Add RTE_FLOW_FIELD_GTP_PSC_QFI to enum rte_flow_field_id.



[7] Extend indirect action update for AGE action

Add a new structure for indirect AGE action update to support updating the timeout value and start, stop or restart the aging check.



 

Net/mlx5 PMD updates:

====================

[1] Support for flow template API and asynchronous queue-based management.



[2] Support port_representor match item​

Match traffic originated from specific port representor. Creating a rule with this match item will ensure that other representors are not affected.

 

Testpmd updates:

===============

[1] Support the changes in rte_flow listed above.




[-- Attachment #2: Type: text/html, Size: 23634 bytes --]