On 28/10/2016 09:12, Stephen Hemminger wrote: > On Fri, 28 Oct 2016 09:04:30 +0800 > Remy Horton <remy.horton at intel.com> wrote: > >> + >> +struct rte_stats_bitrate_s { >> + uint64_t last_ibytes; >> + uint64_t last_obytes; >> + uint64_t peak_ibits; >> + uint64_t peak_obits; >> + uint64_t ewma_ibits; >> + uint64_t ewma_obits; >> +}; >> + > > Reader/write access of 64 bit values is not safe on 32 bit platforms. > I think you need to add a generation counter (see Linux kernel syncp) > to handle 32 bit architecture. If done correctly, it would be a nop > on 64 bit platforms.
I don't see a problem since this is private persistent data that is only read/written from rte_stats_bitrate_calc(), and once calculated it pushes them into the metrics library using rte_metrics_update_metrics(). The idea is that downstream consumers get the values using rte_metrics_get_values() rather than reading rte_stats_bitrate_s directly. Having said that, what you mention quite likley affects the metrics library itself.. :)