On 12/02/15 19:57, John McNamara wrote: > This patchset is for a small addition to the ethdev library, to > add in support for callbacks at the RX and TX stages. This allows > packet processing to be done on packets before they get returned > to applications using rte_eth_rx_burst call. > > See the RFC cover letter for the use cases: > > http://dpdk.org/ml/archives/dev/2014-December/010491.html > > For this version we spent some time investigating Stephen Hemminger's > suggestion of using the userspace RCU (read-copy-update) library for > SMP safety: > > http://urcu.so/ > > The default liburcu (which defaulted to liburcu-mb) requires the least > interaction from the end user but showed a 25% drop in packet throughput > in the callback sample app. > > The liburcu-qsbr (quiescent state) variant showed a 1% drop in packet > throughput in the callback sample app. However it requires registered > RCU threads in the program to periodically announce quiescent states. > This makes it more difficult to implement for end user applications. > > For this release we will document that callbacks should be added/removed > on stopped ports. > > Version 1 changes: > * Added callback removal functions. > * Minor fixes. > > > Richardson, Bruce (3): > ethdev: rename callbacks field to intr_cbs > ethdev: Add in data rxtx callback support > examples: example showing use of callbacks. > > app/test/virtual_pmd.c | 2 +- > examples/rxtx_callbacks/Makefile | 57 ++++++++ > examples/rxtx_callbacks/basicfwd.c | 222 > ++++++++++++++++++++++++++++++++ > examples/rxtx_callbacks/basicfwd.h | 46 +++++++ > lib/librte_ether/rte_ethdev.c | 177 ++++++++++++++++++++++++-- > lib/librte_ether/rte_ethdev.h | 175 +++++++++++++++++++++++++- > lib/librte_pmd_bond/rte_eth_bond_api.c | 2 +- > 7 files changed, 667 insertions(+), 14 deletions(-) > create mode 100644 examples/rxtx_callbacks/Makefile > create mode 100644 examples/rxtx_callbacks/basicfwd.c > create mode 100644 examples/rxtx_callbacks/basicfwd.h >
Looks good to me, I'll ack the next version which has addressed the ABI issues Neil raised. Also, it should probably be noted in the doxygen comments for the add/remove rxtx callbacks that as currently implemented the addition/removal of callbacks isn't thread safe Declan