On Fri, 6 Dec 2024 09:11:32 +0000 Russell King (Oracle) wrote: > Maybe: > > * The input structure is pre-initialised with ETHTOOL_STAT_NOT_SET and > * the implementation must only change implemented statistics. Yup, that's better! FWIW I think my brain goes to talking about zero-init because for per-queue or per-cpu stats some drivers do: for each q: struct->stat += q->stat; without first setting to 0. And it _seems_ fine since NOT_SET is -1, and the off-by-one is hard to spot. But for PHY stats this sort of iteration is very unlikely.