Re: git: 91fcacc35597 - main - if_bnxt: Add support for VLAN on Thor

On Tue, Aug 22, 2023 at 2:07 PM Kristof Provost <kp@freebsd.org> wrote:

> On 27 Jun 2023, at 11:17, Kristof Provost wrote:
>
> On 4 Nov 2022, at 23:55, Warner Losh wrote:
>
> The branch main has been updated by imp:
>
> URL:
> https://cgit.FreeBSD.org/src/commit/?id=91fcacc355971f74aa26fc7861020dc3a2a2d717
>
> commit 91fcacc355971f74aa26fc7861020dc3a2a2d717
> Author: Sumit Saxena <sumit.saxena@broadcom.com>
> AuthorDate: 2022-11-04 22:24:32 +0000
> Commit: Warner Losh <imp@FreeBSD.org>
> CommitDate: 2022-11-04 22:24:32 +0000
>
> if_bnxt: Add support for VLAN on Thor
>
> Reviewed by: imp
> Differential Revision: https://reviews.freebsd.org/D36443
> ---
> sys/dev/bnxt/bnxt.h | 3 +-
> sys/dev/bnxt/bnxt_hwrm.c | 110
> ++++++++++++++++++++++++++++-------------------
> sys/dev/bnxt/bnxt_hwrm.h | 7 ++-
> sys/dev/bnxt/if_bnxt.c | 7 ++-
> 4 files changed, 74 insertions(+), 53 deletions(-)
>
> This commit appears to have broken vlan on these interfaces.
> My hardware is a Broadcom BCM57416 NetXtreme-E 10GBase-T Ethernet, (in a
> Dell T640, if that helps).
>
> A simple vlan creation on top of one:
> ifconfig vlan create
> ifconfig vlan0 vlan 201 vlandev bnxt0
>
> Results in a loss of connectivity. The kernel logs this, which I assume is
> related:
>
> `bnxt0: HWRM_CFA_L2_FILTER_ALLOC command returned INVALID_PARAMS error.`
>
> After reverting this (and 72e9dbb58cad5262190cf2eae47f764021072128) I can
> create vlan interfaces on top of bnxt0 without losing connectivity.
>
> Should we revert this before 14.0?
>
> There’s also a report of what I think is the same issue in
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=269133
>
> There’s a proposed patch there, but it did not work for me.
>

In markj’s analysis he mentions the issue of the reinit.

It sounds like there are multiple bugs in this driver but one quick band
aid may be to mark the driver as not needing re-init for VLAN
modifications.. that sounds unlikely and was probably an accident when the
if reset functionality was added.  You can check e1000 for that.

Note I don’t have this hw and haven’t audited the code so this could be
noise but it’s an easy thing to check.

Best regards,
> Kristof
>