Re: [PATCH 1/2] ice: restore DDP state during PFR recovery

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

On 9/3/26 9:47 AM, Aaron Ma wrote:
> The firmware package and switch recipes are shared by all PFs of an
> adapter. However, each PF rebuilds independently after system resume. The
> resulting package downloads and recipe updates can interleave, causing
> firmware timeouts and leaving interfaces unusable.
> 
> PFR also clears package-derived firmware state on affected devices. The
> existing PFR path only rebuilds the driver's block tables, so VLAN recipe
> programming can fail after the reset.
> 
> The failure is reported as:
> 
>    ice 0000:04:00.0: Update pkg failed: err -5
>    ice 0000:04:00.0: package load failed, -12
>    ice 0000:04:00.0: Rebuild failed, unload and reload driver
> 
> This was observed on an Intel E810-XXV-2 adapter (PCI ID 8086:159b,
> revision 02) with NVM package 1.0.0.18 and ICE OS Default DDP package
> 1.3.43.0.
> 
> Serialize rebuilds across PFs of the same adapter, while allowing each PFR
> to complete independently. Reload the DDP package from its cached copy and
> restore the default DVM recipes before rebuilding the remaining PF state.
> Abort recovery if either operation fails.
> 
> Fixes: 462acf6aca85 ("ice: Enable DDP package download")
> Signed-off-by: Aaron Ma <aaron.ma@canonical.com>


we have FW-level lock for global (per-card) changes,
we have also per-PF SW lock to protect global-update from given PF
against unrelated AQ stuff

You have added a protection for the case PF1 starts its SW preparation,
then other PF starts the same and its first to grab the FW global lock,
so PF1 could be confused.

I fully support this approach, thank you!

Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>