>On Fri, 29 May 2015 09:15:09 -0400 >Liang-Min Larry Wang <liang-min.wang at intel.com> wrote: > >> + drvinfo->n_stats = sizeof(struct rte_eth_stats) / sizeof(uint64_t); > >+ drvinfo->testinfo_len = 0; > > >Providing a hook to access more functionality is good and compatiablity with >old API's is helpful. Too bad ethtool is kind of a clunky old inflexible, and >easily broken with changes. But >that is not your fault. > >Duplicating eth_stats in ethtool is not helpful. >This is actively discouraged for Linux device drivers.
(I have a typo of David's email, and I corrected it in this email) This new library is designed to support ethtool alike interface, some of the implementation will be just a placeholder. The rte_ethtool_net_get_stats64 is designed as a place holder to support net_device_ops::ndo_get_stats64. For the legacy support, it is required to provide # of statistic parameters are supported, so caller can allocate right size of buffer to get statistic from device driver.