Re: [PATCH v6 net-next 6/7] net: ethernet: ti: am65-cpsw-qos: Add Frame Preemption MAC Merge support

[Roger Quadros]

On 30/11/2023 15:22, Vladimir Oltean wrote:
> On Thu, Nov 30, 2023 at 01:49:03PM +0200, Roger Quadros wrote:
>> Thanks for the debug instructions. Indeed lldpad tries to enable MM TX and the
>> network drivers set_mm() hook gets called and returns success but still
>> lldpad sees some error.
>>
>> I've also confirmed that ethnl_set_mm() runs successfully and returns 1.
>> I suppose something is going wrong in user-space with libnl?
>>
>> Nov 21 11:50:02 am62xx lldpad[708]: eth0: Link partner preemption capability supported
>> Nov 21 11:50:02 am62xx lldpad[708]: eth0: Link partner preemption capability not enabled
>> Nov 21 11:50:02 am62xx lldpad[708]: eth0: Link partner preemption capability not active
>> Nov 21 11:50:02 am62xx lldpad[708]: eth0: Link partner minimum fragment size: 124 octets
>> Nov 21 11:50:02 am62xx lldpad[708]: eth0: initiating MM verification with a retry interval of 134 ms...
>> Nov 21 11:50:02 am62xx lldpad[708]: ethtool: kernel reports: integer out of range
>>
>>
>> full debug log is below.
> 
> Ah, you got confused. Openlldp issues multiple ETHTOOL_MSG_MM_SET
> netlink messages. What you observe is that one of them succeeds, and
> then another one returns -ERANGE before even calling the driver's
> set_mm() method.
> 
> And that comes from here in net/ethtool/mm.c:
> 
> 149 const struct nla_policy ethnl_mm_set_policy[ETHTOOL_A_MM_MAX + 1] = {
> 150 »       [ETHTOOL_A_MM_HEADER]»  »       = NLA_POLICY_NESTED(ethnl_header_policy),
> 151 »       [ETHTOOL_A_MM_VERIFY_ENABLED]»  = NLA_POLICY_MAX(NLA_U8, 1),
> 152 »       [ETHTOOL_A_MM_VERIFY_TIME]»     = NLA_POLICY_RANGE(NLA_U32, 1, 128), // <---- here
> 153 »       [ETHTOOL_A_MM_TX_ENABLED]»      = NLA_POLICY_MAX(NLA_U8, 1),
> 154 »       [ETHTOOL_A_MM_PMAC_ENABLED]»    = NLA_POLICY_MAX(NLA_U8, 1),
> 155 »       [ETHTOOL_A_MM_TX_MIN_FRAG_SIZE]»= NLA_POLICY_RANGE(NLA_U32, 60, 252),
> 156 };
> 
> You are reporting in .get_mm() a maximum verify time which is larger
> than the core ethtool is willing to accept in a further .set_mm() call.
> And openlldp will try to max out on the verify time. Hence the -ERANGE.

You are spot on on this. Thanks. :)

> 
> The range I chose for the policy comes from 802.3-2018 clause 30.14.1.6,
> which says that the aMACMergeVerifyTime variable has a range between 1
> and 128 ms inclusive.

I forced driver state->max_verify_time = 128; and now that -ERANGE
error is gone and the lldp test case passes.

I also applied your patch to ethtool_mm.sh and don't see the error with 
'addFragSize 0' anymore

Should I include your patch in the next revision of this series?

-- 
cheers,
-roger