On 11/27/2023 10:19 AM, Ferruh Yigit wrote: > On 11/24/2023 9:18 AM, Edwin Brossette wrote: >> Hello, >> >> I noticed a small inconsistency in the virtio pmd's xstats. >> The stat "rx_q0_errors" appears twice. >> I also think the stats "rx_q0_packets", "rx_q0_bytes", "tx_q0_packets" >> and "tx_q0_bytes" are duplicates of "rx_q0_good_packets", >> "rx_q0_good_bytes", "tx_q0_good_packets" and "tx_q0_good_bytes" >> > > Ack, I didn't reproduce but from code I can see this inconsistency can > occur. > > >> I believe this issue probably appeared after this commit: >> >> f30e69b41f94: ethdev: add device flag to bypass auto-filled queue xstats >> http://scm.6wind.com/vendor/dpdk.org/dpdk/commit/?id=f30e69b41f949cd4a9afb6ff39de196e661708e2 >> >> <http://scm.6wind.com/vendor/dpdk.org/dpdk/commit/?id=f30e69b41f949cd4a9afb6ff39de196e661708e2> >> >> From what I understand, the rxq0_error stat was originally reported by >> the librte. However, changes were made so it is reported by the pmd instead. >> The flag RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS was temporarily set to keep >> the old behaviour so that every pmd could have time to adapt the change. >> But it seems the flag was forgotten in the virtio pmd and as a result, >> some stats are fetched at two different times when displaying xstats. >> > > Motivation is move queue stats from "basic stats" to "xstats". > > 'RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS' is a temporary flag, as you said, > and when driver sets it, ethdev layer copies queue stats from basic > stats to xstats. > In long term expectation is driver itself implement queue stats in > xstats and unset the flag, when all drivers implement this, flag will be > removed. > > > The issue with virtio is it already has queue stats in the xstats, with > the flag some queue stats duplicated. > It seems it is not forgotten to be removed, but it was mistake to add > flag at first place. So OK to remove the flag from virtio PMD. >
Fixed Maxime's email.
