On 6 Oct 2026, at 16:29, Eli Britstein wrote:
> On 29/09/2026 11:34, David Marchand wrote: >> External email: Use caution opening links or attachments >> >> >> Sending this as a RFC because it is still not polished (especially on >> the structure layout side, in the last patch). >> >> This series aims at splitting the netdev-dpdk.c file and the netdev_dpdk >> structure that intermix all the various concerns for "normal" >> (I call them ethernet) DPDK ports and vhost DPDK ports. >> >> Until the last patch, there should be no performance impact, >> but I did not go far enough to back that claim. >> >> My hope is to have this series in a good shape to ease integration of >> netdev-doca.c. >> >> Comments welcome. > Hi David, > Thanks for this series. > I reviewed it. It achieves the eth/vhost separation, but keeps the eth part > static in netdev-dpdk.c. > This doesn't quite get us towards easy netdev-doca integration: doca needs > the whole eth part > exposed for reuse and some helpers factored out. > > That exposure and factoring is already done in the netdev-doca patches, not > by this series. > > I did rebase netdev-doca on top of your series as a check. I re-did the > exposing and helper > factoring and adapted netdev-doca on top. Can you share the dev branch with this change? I’ll be going over the RFC, and it would be useful to easily see where it overlaps with the existing work, where it adds value, and where it doesn’t. > It works, but it means redoing a fair amount of what the netdev-doca patches > already do (and that Eelco already reviewed), just in a different place. > The netdev-doca patches were already acked by Eelco (with a few nits > remaining), > so I think it makes more sense to get those reviewed and merged first, and > have the eth/vhost I guess that if we do apply David’s work, it will be on main, which would then break the already-reviewed series on dpdk-latest, and would warrant another review of netdev-doca anyway. However, let’s first look at the benefits of the RFC compared to your changes and see where there is overlap. > separation come later as an independent cleanup. It's a reasonable cleanup on > its own. > > Thanks, Eli > >> >> -- >> David Marchand >> >> David Marchand (11): >> netdev-dpdk: Hide some rte_flow error helper. >> netdev-dpdk: Cleanup netdev ops. >> netdev-dpdk: Separate ethernet and vhost netdev ops. >> netdev-dpdk: Insulate mempool creation from netdev_dpdk. >> netdev-dpdk: Separate ethernet and vhost specific fields. >> netdev-dpdk: Move common fields and associated helpers. >> netdev-dpdk: Move mempool code. >> netdev-dpdk: Move QoS code. >> netdev-dpdk: Move transmission code. >> netdev-dpdk: Store vhost ports in a separate list. >> netdev-dpdk: Split ethernet and vhost netdevs. >> >> lib/automake.mk | 4 +- >> lib/netdev-dpdk-common.c | 1789 +++++++++++++++++ >> lib/netdev-dpdk-common.h | 517 +++++ >> lib/netdev-dpdk.c | 4018 ++++++++++++-------------------------- >> lib/netdev-dpdk.h | 11 - >> tests/system-dpdk.at | 1 + >> 6 files changed, 3563 insertions(+), 2777 deletions(-) >> create mode 100644 lib/netdev-dpdk-common.c >> create mode 100644 lib/netdev-dpdk-common.h >> >> -- >> 2.54.0 _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
