Re: [PATCH net-next v5 00/15] devlink, mlx5, iavf, ice: XLVF for iavf

[Przemek Kitszel <przemyslaw.kitszel@intel.com>]

On 11/10/26 15:07, Przemek Kitszel wrote:
> Also available here
> https://github.com/pkitszel/linux/tree/xlvf-v5
> 
> The purpose of this series is to allow iavf to use more than 16 queue
> pairs, in two modes: up to 64 and up to 256 queue pairs.

Sashiko review [1], summary of the problems:

1. Division by zero in iavf_fill_rss_lut() (10/15): the only real bug,
    but triggered only in weird conditions: a failed VF reset
    (iavf_disable_vf()), and then the PF reporting a different RSS LUT
    size when the VF comes back. One additional if () resolves it, best
    if it could be done as a follow-up.

2. Bisection: a non-default setting (sriov_vf_msix_count above 17, as
    the default MSI-X tiers give a VF at most 16 queues) is needed to
    cause minor problems during bisection (VF queues above 16 not
    working at a few commits in the middle of the series). Fine at the
    end of the series, and still much better to have the patches split.

3. Minor wording problems in commit messages and code comments: AI is
    way too picky here, e.g. it reads the casual
    "valid when lut_type == GLOBAL_LUT" as a wrong enum used. A wrong
    struct name in a comment, or a detail that is in the changelog but
    not in the commit message, is not worth a respin of 3k lines.

4. Multiple pre-existing bugs. Of those, the lock ordering ones, which
    I have hit during development, are fixed in a separate series
    (including both lock inversions listed in the cover letter):
https://lore.kernel.org/netdev/20261102105346.30739-1przemyslaw.kitszel@intel.com

5. The rest is either in the cover letter already (VF has to trust its
    PF), intended (ethtool -L in the LARGE mode is async, as the PF
    grants a request by resetting the VF), or an error path corner case
    with a benign outcome (e.g. a VF left disabled after its VSI rebuild
    fails, as ice_reset_vf() already does).


1st one is also reported by Petr, with a stress-test reproducer.
I would like to merge as-is with a followup.

All other are rather "cosmetics/nitpicks" or pre-existing.

[1] 
https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261110141534.95614-1-przemyslaw.kitszel%40intel.com

>   45 files changed, 3099 insertions(+), 351 deletions(-)
>   create mode 100644 drivers/net/ethernet/intel/ice/devlink/resource.h
>   create mode 100644 drivers/net/ethernet/intel/ice/devlink/resource.c
>