Re: [dpdk-dev] [RFC PATCH DRAFT 1/2] ethdev: add buffered single pkt TX function to API

["Richardson, Bruce" <bruce.richardson@intel.com>]

> -----Original Message-----
> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> Sent: Tuesday, June 24, 2014 4:07 PM
> To: Richardson, Bruce
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [RFC PATCH DRAFT 1/2] ethdev: add buffered single pkt
> TX function to API
> 
> On Tue, 24 Jun 2014 23:32:15 +0100
> Bruce Richardson <bruce.richardson@intel.com> wrote:
> 
> > 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 inside the core ethdev API.
> > The new APIs include three functions:
> > * 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
> 
> This must be per-queue to be MP safe.
> 
> Internally, we do something similar but it proved to be fastest to
> user per-cpu data in some cases.

The implementation is per-queue. If you look at the changes to the tx queue setup function, we set up a buffer space for each TX queue inside the eth_dev_data structure that is then used for buffering that queue. The fact that it's handled per-queue inside the eth_dev means that it's as thread-safe as the existing tx_burst function, and also is completely driver agnostic.