This series of patches implements the mlx4-pmd with Tx data path that directly
access the device queues for transmitting packets, bypassing the ibverbs Tx
data path altogether.
Using this scheme allows the PMD to work with upstream rdma-core package
instead of the Mellanox OFED one without sacrificing Tx functionality.

These patches should be applied in the order listed below as each depends on
its predecessor to work.

This implementation allows rapid deployment of new features without the need to
update the underlying OFED.

This work depends on
        http://dpdk.org/ml/archives/dev/2017-August/072281.html
        [dpdk-dev] [PATCH v1 00/48] net/mlx4: trim and refactor entire PMD
by Adrien Mazarguil

It had been built and tested using rdma-core-15-1 from
 https://github.com/linux-rdma/rdma-core
and kernel-ml-4.12.0-1.el7.elrepo.x86_64

It had been built and tested using rdma-core-15-1 from
 https://github.com/linux-rdma/rdma-core
and kernel-ml-4.12.0-1.el7.elrepo.x86_64

Moti Haimovsky (5):
  net/mlx4: add simple Tx bypassing ibverbs
  net/mlx4: support multi-segments Tx
  net/mlx4: refine setting Tx completion flag
  net/mlx4: add Tx checksum offloads
  net/mlx4: add loopback Tx from VF

 drivers/net/mlx4/mlx4.c        |   7 +
 drivers/net/mlx4/mlx4.h        |   2 +
 drivers/net/mlx4/mlx4_ethdev.c |   6 +
 drivers/net/mlx4/mlx4_prm.h    | 249 ++++++++++++++++++++++
 drivers/net/mlx4/mlx4_rxtx.c   | 456 +++++++++++++++++++++++++++++++++--------
 drivers/net/mlx4/mlx4_rxtx.h   |  39 +++-
 drivers/net/mlx4/mlx4_txq.c    |  66 +++++-
 mk/rte.app.mk                  |   2 +-
 8 files changed, 734 insertions(+), 93 deletions(-)
 create mode 100644 drivers/net/mlx4/mlx4_prm.h

-- 
1.8.3.1

Reply via email to