Re: [PATCH net-next 0/4] net: stmmac: new features

[Sebastian Basierski <sebastian.basierski@intel.com>]

On 8/30/2025 4:46 AM, Joseph Steel wrote:
> On Fri, Aug 29, 2025 at 02:23:24PM -0700, Jacob Keller wrote:
>>
>> On 8/28/2025 7:45 AM, Konrad Leszczynski wrote:
>>> This series adds four new patches which introduce features such as ARP
>>> Offload support, VLAN protocol detection and TC flower filter support.
>>>
>>> Patchset has been created as a result of discussion at [1].
>>>
>>> [1] 
>>> https://lore.kernel.org/netdev/20250826113247.3481273-1-konrad.leszczynski@intel.com/ 
>>>
>>>
>>> v1 -> v2:
>>> - add missing SoB lines
>>> - place ifa_list under RCU protection
>>>
>>> Karol Jurczenia (3):
>>>    net: stmmac: enable ARP Offload on mac_link_up()
>>>    net: stmmac: set TE/RE bits for ARP Offload when interface down
>>>    net: stmmac: add TC flower filter support for IP EtherType
>>>
>>> Piotr Warpechowski (1):
>>>    net: stmmac: enhance VLAN protocol detection for GRO
>>>
>>>   drivers/net/ethernet/stmicro/stmmac/stmmac.h  |  1 +
>>>   .../net/ethernet/stmicro/stmmac/stmmac_main.c | 35 
>>> ++++++++++++++++---
>>>   .../net/ethernet/stmicro/stmmac/stmmac_tc.c   | 19 +++++++++-
>>>   include/linux/stmmac.h                        |  1 +
>>>   4 files changed, 50 insertions(+), 6 deletions(-)
>>>
>> The series looks good to me.
>>
>> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
> Not a single comment? Really? Three Rb and three Sb tags from Intel
> staff and nobody found even a tiny problem? Sigh...
Hi Joseph,
Thank you for your time and valuable review
>
> Let's start with an easiest one. What about introducing an unused
> platform flag for ARP-offload?
Right, this patch should not be here. Will be removed in next revision.
> Next is more serious one. What about considering a case that
> IP-address can be changed or removed while MAC link is being up?
>
> Why does Intel want to have ARP requests being silently handled even
> when a link is completely set down by the host, when PHY-link is
> stopped and PHY is disconnected, after net_device::ndo_stop() is
> called?

While trying to enable ARP offload,
we found out that when interface was set down and up,
MAC_ARP_Address and ARP offload enable bit were reset to default values,
the address was set to 0xFFFFFFFF and ARP offload was disabled.
There was two possible solutions out of this:
a) caching address and ARP offload bit state
b) enabling ARP while interface is down.
We choose to go with second solution.
But given that fact this code depends on unused STMMAC_ARP_OFFLOAD_EN flag,
i guess whether it is fine or not, should not be placed in patchset.

> Finally did anyone test out the functionality of the patches 1 and
> 2? What does arping show for instance for just three ARP requests?
> Nothing strange?
Yes, we have a validation team that verified proposed solution.
> So to speak at this stage I'd give NAK at least for the patches 1 and
> 2.
>
> BTW I've been working with the driver for quite some time and AFAICS
> Intel contributed if not half but at least quarter of it' mess.
>
> Joseph

BR,
Sebastian