[dpdk-dev] [PATCH v2 0/2] add support for buffered tx to ethdev

[Tomasz Kulasek]

Many sample apps include internal buffering for single-packet-at-a-time
operation. Since this is such a common paradigm, this functionality is
better suited to being implemented in the ethdev API.

The new APIs in the ethdev library are:
* rte_eth_tx_buffer_init - initialize buffer
* rte_eth_tx_buffer - buffer up a single packet for future transmission
* rte_eth_tx_buffer_flush - flush any unsent buffered packets
* rte_eth_tx_buffer_set_err_callback - set up a callback to be called in
  case transmitting a buffered burst fails. By default, we just free the
  unsent packets.

As well as these, an additional reference callback is provided, which
frees the packets (as the default callback does), as well as updating a
user-provided counter, so that the number of dropped packets can be
tracked.

Due to the feedback from mailing list, that buffer management facilities
in the user application are more preferable than API simplicity, we decided
to move internal buffer table, as well as callback functions and user data,
from rte_eth_dev/rte_eth_dev_data to the application space.
It prevents ABI breakage and gives some more flexibility in the buffer's
management such as allocation, dynamical size change, reuse buffers on many
ports or after fail, and so on.


The following steps illustrate how tx buffers can be used in application:

1) Initialization

a) Allocate memory for a buffer

   struct rte_eth_dev_tx_buffer *buffer = rte_zmalloc_socket("tx_buffer",
           RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0, socket_id);

   RTE_ETH_TX_BUFFER_SIZE(size) macro computes memory required to store
   "size" packets in buffer.

b) Initialize allocated memory and set up default values. Threshold level
   must be lower than or equal to the MAX_PKT_BURST from 1a)

   rte_eth_tx_buffer_init(buffer, threshold);


c) Set error callback (optional)

   rte_eth_tx_buffer_set_err_callback(buffer, callback_fn, userdata);


2) Store packet "pkt" in buffer and send them all to the queue_id on
   port_id when number of packets reaches threshold level set up in 1b)

   rte_eth_tx_buffer(port_id, queue_id, buffer, pkt);


3) Send all stored packets to the queue_id on port_id

   rte_eth_tx_buffer_flush(port_id, queue_id, buffer);


4) Flush buffer and free memory

   rte_eth_tx_buffer_flush(port_id, queue_id, buffer);
   ...
   rte_free(buffer);


v2 changes:
 - reworked to use new buffer model
 - buffer data and callbacks are removed from rte_eth_dev/rte_eth_dev_data,
   so this patch doesn't brake an ABI anymore
 - introduced RTE_ETH_TX_BUFFER macro and rte_eth_tx_buffer_init
 - buffers are not attached to the port-queue
 - buffers can be allocated dynamically during application work
 - size of buffer can be changed without port restart


Tomasz Kulasek (2):
  ethdev: add buffered tx api
  examples: rework to use buffered tx

 examples/l2fwd-jobstats/main.c                     |  104 +++++------
 examples/l2fwd-keepalive/main.c                    |  100 ++++-------
 examples/l2fwd/main.c                              |  104 +++++------
 examples/l3fwd-acl/main.c                          |   92 ++++------
 examples/l3fwd-power/main.c                        |   89 ++++------
 examples/link_status_interrupt/main.c              |  107 +++++-------
 .../client_server_mp/mp_client/client.c            |  101 ++++++-----
 examples/multi_process/l2fwd_fork/main.c           |   97 +++++------
 examples/packet_ordering/main.c                    |  122 +++++++++----
 examples/qos_meter/main.c                          |   61 ++-----
 lib/librte_ether/rte_ethdev.c                      |   36 ++++
 lib/librte_ether/rte_ethdev.h                      |  182 +++++++++++++++++++-
 lib/librte_ether/rte_ether_version.map             |    9 +
 13 files changed, 662 insertions(+), 542 deletions(-)

-- 
1.7.9.5