RE: [PATCH net v6 3/3] igb: do not split AF_XDP zero-copy frames over several Rx buffers

From: Tjerk Kusters

Date: Fri Nov 20 2026 - 06:08:00 EST


> - [High] The new XSK queue check sits in igb_sriov_reinit(), and that
> function is shared by the SR-IOV enable and disable paths.

Right. In v7 the check is at the start of igb_enable_sriov(), before any
VF state is set up, and does not run when VFs are disabled.

> - [Medium] The new -EBUSY return from igb_sriov_reinit() gives userspace
> a deterministic way into igb_enable_sriov()'s err_out path while the…

Right. In v7 the -EBUSY return comes before vfs_allocated_count and
vf_data are set, so it does not go through err_out. The MAC filter
leak and the restore of the DMA coalescing flag are gone with it.

> - [Low] This patch raises the XSK ring hardware buffer in
> igb_setup_srrctl() to 2 KiB or 3 KiB, while the XSK frame can be as…

Patch 1 is marked for stable as well, and
Documentation/process/stable-kernel-rules.rst says that patches in the
same series do not need to be listed as prerequisites then. Patch 1's
Fixes: commit is older than patch 3's, so every tree that takes patch 3
can also take patch 1. The commit message names the dependency.

> - [High] This problem wasn't introduced by this patch, but it sits in
> igb_xsk_pool(), which the patch edits.

Right, and older than this series: the ethtool offline self-test sets
up its test rings with queue index 0, so igb_xsk_pool() gives them the
pool bound to queue 0. It needs its own fix, separate from this series,
so I listed it in the v7 cover letter as an older issue.

> - [Medium] This problem wasn't introduced by this patch.

That is the failed pci_enable_sriov() path, which keeps the VF queue
layout. It is older than this series and unchanged; I listed it in the
v7 cover letter. The new -EBUSY return in v7 does not go through it.