Changes and New Features

• ConnectX-8 Counters: Added PCIe latency histogram performance reporting.

• Argus:

  • Now available in NGC

  • Added support for 5-level paging

• DPF:

  • Support DOCA Argus and SNAP Virtio-fs

  • Support for zero-trust deployments (BMaaS) [Beta]

  • DPF VPC service

  • Bluefield auto-discovery via OOB

  • Non-accelerated K8s CNIs support (Calico/Cilium)

  • Scaling enhancements - up to 1K BlueFields in a cluster

  • SNAP Virtio-fs Data volumes provisioning

• DMS:

  • Firmware installation on SuperNICs with autodetection.

  • Support for adaptive retransmission timeout (RTO) configuration.

• HBN:

  • Support up to 64 next-hops in ECMP groups

  • Improved DPF integration

• Target Offload Service: General Availability (GA) with high QP scale support.

• OVS-DOCA:

  • OVS metrics Integration with DTS

  • Increased scale – up to 256K megaflow rules

  • Hitless slowpath metering configuration

  • Real-world scenario benchmarking across multiple frameworks

• DTS: New examples for Real time distributed analysis

• Firefly: GA support for the Security Extension (Linux PTP 4.3).

• DOCA Flow: Added support for modifying TCP/ACK sequence numbers.

• FWCtrl: GA-level support for firmware control subsystem driver.

• SOS Report: Added SOS Report functionality to the NVDebug tool.

• GPUNetIO: Provided header files with new CUDA function implementations.

• DOCA Core: DOCA RDMA Verbs are now publicly available.

• BlueField BMC:

  • Updated DPU BMC Linux Kernel to version 6.12.

  • Unified logging system for BlueField-3 BMC ensuring time-synchronized and consistent event logs across all management interfaces.

• BlueField Firmware Reset: Replaced power cycle mechanism in NIC mode with firmware reset.

• BFB-FW-Bundle Upgrade Tool: Enhanced to read firmware component versions from the host.

• DOCA Sample Code: Migrated all DOCA sample code snippets to GitHub for public access: https://github.com/NVIDIA-DOCA/doca-samples.

• DOCA Chatbot: Integrated a chatbot to DOCA documentation for ease of use and access (see bottom right corner of this page)

• Hardware Steering (HWS): Enable HWS as a secondary option for the device steering mode. If Software Steering (SWS) is not supported, HWS will be used by default, assuming it is available. Firmware Steering (FW) will now serve as the default only when both SWS and HWS are unsupported.

• Hardware-Managed Flow Steering (HMFS): HMFS is a new approach to managing steering rules, where STEs are written to the ICM by hardware instead of software (as in software-managed steering). This enables a higher rate of rule insertion.

• Queue Affinity Bonding: Added support for 4-port bonding to allow users to add both p0 and p1 to a bond without requiring SSH access to the host or bringing down the driver beforehand.

• Multiple Doorbells per PF: Increased the number of doorbells from 1 to 8, distributing them in a round-robin manner across channel send and receive queues. This enhancement improves scalability on multi-core systems with high packet rates by reducing MMIO contention on doorbell addresses. In high packet-rate scenarios, Sub-NUMA Clustering is also required to maintain consistent channel packet rates.

• HQoS for Virtualization Solution: Added support for hierarchical group structures to enable multi-tier resource control, where parent groups enforce overall limits and child groups dynamically share available bandwidth. This design is essential for effective QoS enforcement, burst handling, and fair bandwidth allocation, ensuring workloads receive appropriate resources.

• Number of lanes Configuration when Auto-Negotiation is Disabled: Added support for specifying the number of lanes (protocol width) when auto-negotiation is disabled. Since each combination of speed and lane count corresponds to a unique link mode, this allows the user to select a specific link mode when auto-negotiation is off. This change aligns with a recent firmware update that requires the driver to provide a single link mode, rather than a bitmap based on the user-defined speed.

• Extend ODP Statistics with Operation Count: Added the following new page-granularity counters:

  • num_page_faults: counts the number of page fault events handled

  • num_invalidations: counts the number of page invalidation events handled

• RSS with Crypto Offload: Added support for RSS with crypto offload enabling the NIC to parallelize packet processing across CPU cores while performing encryption/decryption in hardware. Additionally, introduced a new l4_type_ext parameter with values: 0 (None), 1 (TCP), 2 (UDP), 3 (ICMP).

• Safely Identify DPUs/SmartNICs is a Machine and PCIe Slot: A new access register is introduced that accepts a type, length, and R/W command.

  • Write operation: Allocates a new ICMC buffer of the specified size (aligned to 64B) and stores the provided data. If a buffer for the given type already exists, the data in the ICMC is overwritten, and the locked area is adjusted accordingly

  • Read operation: If a buffer exists, its data is copied out. If not, the access register returns a size of 0 or an explicit error

  The length can be stored within the data in the ICMC, and the type is mapped to 256B chunks (due to access register limitations), so the VA of the buffer is calculated as (base + (type << 8)). The first 4 bytes store a validity flag and the length. If length storage is unnecessary (e.g., null-terminated data), a hardware read can use a cache-line hit as a validity bit. This feature is designed for limited use cases and does not address multi-host scenarios or broader ICMC utilization implications.

• Power Measurements: Enabled Arm power measurements such as voltage, current and power in MVCR.

• RDMA QP: When an RDMA QP encounters a memory access an issue caused by address translation, it can recover without transitioning to an error state. The QP will send an error CQE to notify the software while continuing to serve other VMs and functions.

• RSS with Crypto Offload: Added support for RSS with crypto offload enabling the NIC to parallelize packet processing across CPU cores while performing encryption/decryption in hardware. Additionally, introduced a new l4_type_ext parameter with values: 0 (None), 1 (TCP), 2 (UDP), 3 (ICMP).

The following table details the API changes that were introduced in this version of DOCA: