Mellanox ConnectX-4 Lx Ethernet Adapter Firmware

Fixes:
- Fixed an issue that caused the card not to be detected during power cycle.
- Set the stateless offloads cap to be permanently '1'.
- Fixed an issue that prevented the system from counting multicast/broadcast traffic on the ETH unicast vport counter when the driver did not specify the MAC address in the FTE match criteria of the Flow Table Entry in the eswitch FDB table.
- Fixed a standby deadlock that was caused when the PCIe reset arrived during traffic.
- Fixed an issue that caused the mlxconfig tool to present all possible expansion ROM images, instead of presenting only the existing images.
- Fixed an issue that prevented the SX error handling mechanism from clearing the counters when the max WQE re-read attempts were completed successfully.
- Fixed an issue related to the VF EQ error that caused the health buffer to report to the host (PF) when the VF released the EQ buffer memory before the EQ was destroyed.
- Modifying VMQoS rate limiter parameters during traffic might cause transmission failure.
- Aligned the default tuning type in PHY TEST MODE to the device protocol.
- When a device is operating in Safe Mode state, and the user issues the mlxfwreset command, the device might fail to come-up correctly after the reset. Note: Do not run mlxfwreset when operating in a Safe Mode state.
- Fixed an issue that caused the max_qp_retry_freq_exceeded counter (including a CQE with error syndrome 0x97, and the QP moving to error state) to be activated only after exceeding the NIC Vport context max_qp_retry_limit, and not when reaching it.
- Fixed an issue related to PCIe 'Surprise link down' event reporting capability.
- Fixed an issue that resulted in the link partner experiencing false active linkup when plugging in a base-T cable to a closed port.
- Modifying VMQoS rate limiter parameters during traffic might cause transmission failure.

Enhancements:
- Added support for send opcode operations targeting a SRQ/RMP with the receive WQEs using ODP memory. In case the receive WQE receives an ODP, the device will generate ODP notifications (EQE) and PFAULT will abort CQEs.
Note: It is recommended to prefetch the memory used by the receive WQEs to reduce ODP occurrence as these have significant latencies and will cause a performance degradation.
- Added support for an additional layer (Layer 3) of packet processing at the hypervisor level that enables adding and removing protocol headers (e.g., the MAC address is removed during encapsulation, and added during decapsulation) for the encapsulated traffic.
- Added support for the transmission histogram counter set as part of the Ethernet extended group counters.
- Enabled the hardware to generate an event upon counter incrementation, in order to reduce an overhead from the software from reading rarely updated counters such as error counters.
- Enabled NOIDNIC connectivity to the network through the e-switch and not directly to the physical port.
- Enabled setting the QP and the Mkey values by the software upon these resources creation.
- Enabled advanced PCIe atomic operations. The HCA will perform PCIe atomic operations as a requestor towards the host memory when receiving compatible atomic messages from the network, and according to the configuration of NV_SW_OFFLOAD_CONFIG pci_atomic_mode field and the PCI AtomicOpRequester Enable bit in the Device Control 2 register.
- Enabled a single TIR destination from the FDB.
- Changed the WRED default mode to OFF for Multi-Host adapter cards.
- Added support for TX steering rule on flow_table_metadata in WQE Ethernet segment.
- Added L3 encapsulation/decapsulation support in the reformat context allocation.
- L3 encapsulation removes L2 headers and adds generic L3 tunnel encapsulation.
- L3 decapsulation removes the generic L3 tunnel decapsulation and L2 header.
- Added support for flow steering header modification (header rewrite) for IPv4 TTL header for loopback traffic (VF-VF/VF-PF).
Note: TTL modification for tra