Added software parser for packet type.
Improved offloads support in mbuf.
Added vhost-user dequeue zero copy support.
The copy in the dequeue path is avoided in order to improve the performance. In the VM2VM case, the boost is quite impressive. The bigger the packet size, the bigger performance boost you may get. However, for the VM2NIC case, there are some limitations, so the boost is not as impressive as the VM2VM case. It may even drop quite a bit for small packets.
For that reason, this feature is disabled by default. It can be enabled when the RTE_VHOST_USER_DEQUEUE_ZERO_COPY flag is set. Check the VHost section of the Programming Guide for more information.
Added vhost-user indirect descriptors support.
If the indirect descriptor feature is enabled, each packet sent by the guest will take exactly one slot in the enqueue virtqueue. Without this feature, as in the current version, even 64 bytes packets take two slots with Virtio PMD on guest side.
The main impact is better performance for 0% packet loss use-cases, as it behaves as if the virtqueue size was enlarged, so more packets can be buffered in the case of system perturbations. On the downside, small performance degradations were measured when running micro-benchmarks.
Added vhost PMD xstats.
Added extended statistics to vhost PMD from a per port perspective.
Supported offloads with virtio.
Added support for the following offloads in virtio:
Added virtio NEON support for ARM.
Added NEON support for ARM based virtio.
Updated the ixgbe base driver.
Updated the ixgbe base driver, including the following changes:
Added APIs for VF management to the ixgbe PMD.
Eight new APIs have been added to the ixgbe PMD for VF management from the PF. The declarations for the API’s can be found in rte_pmd_ixgbe.h.
Updated the enic driver.
Updated the mlx5 driver.
Updated the QAT PMD.
The QAT PMD was updated with additional support for:
Added openssl PMD.
A new crypto PMD has been added, which provides several ciphering and hashing algorithms. All cryptography operations use the Openssl library crypto API.
Updated the IPsec example.
Updated the IPsec example with the following support:
Added support for new gcc -march option.
The GCC 4.9 -march option supports the Intel processor code names. The config option RTE_MACHINE can be used to pass code names to the compiler via the -march flag.
L3fwd-power app does not work properly when Rx vector is enabled.
The L3fwd-power app doesn’t work properly with some drivers in vector mode since the queue monitoring works differently between scalar and vector modes leading to incorrect frequency scaling. In addition, L3fwd-power application requires the mbuf to have correct packet type set but in some drivers the vector mode must be disabled for this.
Therefore, in order to use L3fwd-power, vector mode should be disabled via the config file.
Digest address must be supplied for crypto auth operation on QAT PMD.
The cryptodev API specifies that if the rte_crypto_sym_op.digest.data field, and by inference the digest.phys_addr field which points to the same location, is not set for an auth operation the driver is to understand that the digest result is located immediately following the region over which the digest is computed. The QAT PMD doesn’t correctly handle this case and reads and writes to an incorrect location.
Callers can workaround this by always supplying the digest virtual and physical address fields in the rte_crypto_sym_op for an auth operation.
The driver naming convention has been changed to make them more consistent. It especially impacts --vdev arguments. For example eth_pcap becomes net_pcap and cryptodev_aesni_mb_pmd becomes crypto_aesni_mb.
For backward compatibility an alias feature has been enabled to support the original names.
The log history has been removed.
The rte_ivshmem feature (including library and EAL code) has been removed in 16.11 because it had some design issues which were not planned to be fixed.
The file_name data type of struct rte_port_source_params and struct rte_port_sink_params is changed from char * to const char *.
Improved device/driver hierarchy and generalized hotplugging.
The device and driver relationship has been restructured by introducing generic classes. This paves the way for having PCI, VDEV and other device types as instantiated objects rather than classes in themselves. Hotplugging has also been generalized into EAL so that Ethernet or crypto devices can use the common infrastructure.