Re: [PATCH v2] IGC: Remove I225_I_PHY_ID checking

[Kevin Traynor <ktraynor@redhat.com>]

On 19/10/2022 09:34, Kevin Traynor wrote:
> On 18/10/2022 23:45, Mah, Yock Gen wrote:
>>
>>
>> -----Original Message-----
>> From: Kevin Traynor <ktraynor@redhat.com>
>> Sent: Tuesday, 18 October, 2022 8:54 PM
>> To: Mah, Yock Gen <yock.gen.mah@intel.com>; IOTG DPDK Ref App <iotg.dpdk.ref.app@intel.com>; Zhang, Qi Z <qi.z.zhang@intel.com>
>> Cc: dev@dpdk.org
>> Subject: Re: [PATCH v2] IGC: Remove I225_I_PHY_ID checking
>>
>> On 12/10/2022 09:39, Kevin Traynor wrote:
>>> On 12/10/2022 08:45, Mah, Yock Gen wrote:
>>>>
>>>>
>>>> -----Original Message-----
>>>> From: Kevin Traynor <ktraynor@redhat.com>
>>>> Sent: Tuesday, 11 October, 2022 7:05 PM
>>>> To: IOTG DPDK Ref App <iotg.dpdk.ref.app@intel.com>; Zhang, Qi Z
>>>> <qi.z.zhang@intel.com>
>>>> Cc: dev@dpdk.org
>>>> Subject: Re: [PATCH v2] IGC: Remove I225_I_PHY_ID checking
>>>>
>>>> On 31/08/2022 23:51, iotg.dpdk.ref.app@intel.com wrote:
>>>>> From: NSWE SWS DPDK Dev <iotg.dpdk.ref.app@intel.com>
>>>>>
>>>>> i225 devices have only one PHY vendor. There is unnecessary to check
>>>>> _I_PHY_ID during the link establishment and auto-negotiation
>>>>> process, the checking also caused devices like i225-IT failed. This
>>>>> patch is to remove the mentioned unnecessary checking.
>>>>>
>>>>> Cc: stable@dpdk.org
>>>>> Signed-off-by: NSWE SWS DPDK Dev <iotg.dpdk.ref.app@intel.com>
>>>>> ---
>>>>>       drivers/net/igc/base/igc_i225.c | 15 ++-------------
>>>>>       drivers/net/igc/base/igc_phy.c  |  6 ++----
>>>>>       2 files changed, 4 insertions(+), 17 deletions(-)
>>>>>
>>>>> diff --git a/drivers/net/igc/base/igc_i225.c
>>>>> b/drivers/net/igc/base/igc_i225.c index 5f3d535490..af26602afb
>>>>> 100644
>>>>> --- a/drivers/net/igc/base/igc_i225.c
>>>>> +++ b/drivers/net/igc/base/igc_i225.c
>>>>> @@ -173,19 +173,8 @@ static s32 igc_init_phy_params_i225(struct igc_hw *hw)
>>>>>       	phy->ops.write_reg = igc_write_phy_reg_gpy;
>>>>>       
>>>>>       	ret_val = igc_get_phy_id(hw);
>>>>> -	/* Verify phy id and set remaining function pointers */
>>>>> -	switch (phy->id) {
>>>>> -	case I225_I_PHY_ID:
>>>>> -	case I226_LM_PHY_ID:
>>>>> -		phy->type		= igc_phy_i225;
>>>>> -		phy->ops.set_d0_lplu_state = igc_set_d0_lplu_state_i225;
>>>>> -		phy->ops.set_d3_lplu_state = igc_set_d3_lplu_state_i225;
>>>>
>>>>> - The commit log says it is removing a check on the ID, but it does not say why these function pointers are being removed.
>>>>
>>>>> - Why are they removed, were they not needed?
>>>>
>>>>
>>>> i225 devices have only one PHY vendor. There is no point checking _I_PHY_ID during the link establishment and auto-negotiation process.
>>>>
>>>
>>> Right, that's clear about the vendor ID check. But it's not clear to
>>> me why the the resulting code like this:
>>>
>>> phy->type = igc_phy_i225;
>>>
>>>      and not like this:
>>>
>>> phy->type = igc_phy_i225;
>>> phy->ops.set_d0_lplu_state = igc_set_d0_lplu_state_i225;
>>> phy->ops.set_d3_lplu_state = igc_set_d3_lplu_state_i225;
>>>
>>> So it is using dummy null functions instead:
>>> https://git.dpdk.org/dpdk/tree/drivers/net/igc/base/igc_phy.c#n61
>>>
>>> Do the device registers not need to be set anymore?
>>>
>>
>>> For main branch, it would be nice to have an answer to above.
>>
>>> It only adds a small readability benefit by removing some code branches, but does change functionality which adds risk, so I don't think it's a good candidate for stable branches.
>>
>> This is not only added readability, but to fix real world issue, we were experiencing i225-IT not runnable issue without patching another case checking as below:
>>
>> +++ b/drivers/net/igc/base/igc_phy.c
>> @@ -1881,6 +1881,7 @@ s32 igc_phy_force_speed_duplex_m88(struct igc_hw *hw)
>>                                   case I210_I_PHY_ID:
>>                                   /* fall-through */
>>                                   case I225_I_PHY_ID:
>> +                              case I225_IT_PHY_ID:
>>
>> However, cleaner solution is to remove those unnecessary checking completely as it does in kernel also https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=7c496de538eebd8212dc2a3c9a468386b264d0d4
>>
> 
> ok, you are correct that it is more than readability. I can apply but
> first it would be good if you can answer the original question and
> explain why the functionality is changed for I225_I_PHY_ID by removing
> the function pointers that set the registers.
> 

Hi, I didn't apply this to DPDK 21.11.3 because there was no explanation 
of why the registers were changing after asking multiple times.

If it is required for a later 21.11 LTS release, please send a backport 
to stable mailing list with an updated commit message explaining the 
changes.

thanks,
Kevin.

>>>>> -		/* TODO - complete with GPY PHY information */
>>>>> -		break;
>>>>> -	default:
>>>>> -		ret_val = -IGC_ERR_PHY;
>>>>> -		goto out;
>>>>> -	}
>>>>> +        phy->type = igc_phy_i225;
>>>>> +
>>>>>       
>>>>>       out:
>>>>>       	return ret_val;
>>>>> diff --git a/drivers/net/igc/base/igc_phy.c
>>>>> b/drivers/net/igc/base/igc_phy.c index 43bbe69bca..2906bae21a 100644
>>>>> --- a/drivers/net/igc/base/igc_phy.c
>>>>> +++ b/drivers/net/igc/base/igc_phy.c
>>>>> @@ -1474,8 +1474,7 @@ s32 igc_phy_setup_autoneg(struct igc_hw *hw)
>>>>>       			return ret_val;
>>>>>       	}
>>>>>       
>>>>> -	if ((phy->autoneg_mask & ADVERTISE_2500_FULL) &&
>>>>> -	    hw->phy.id == I225_I_PHY_ID) {
>>>>> +	if (phy->autoneg_mask & ADVERTISE_2500_FULL) {
>>>>>       	/* Read the MULTI GBT AN Control Register - reg 7.32 */
>>>>>       		ret_val = phy->ops.read_reg(hw, (STANDARD_AN_REG_MASK <<
>>>>>       					    MMD_DEVADDR_SHIFT) |
>>>>> @@ -1615,8 +1614,7 @@ s32 igc_phy_setup_autoneg(struct igc_hw *hw)
>>>>>       		ret_val = phy->ops.write_reg(hw, PHY_1000T_CTRL,
>>>>>       					     mii_1000t_ctrl_reg);
>>>>>       
>>>>> -	if ((phy->autoneg_mask & ADVERTISE_2500_FULL) &&
>>>>> -	    hw->phy.id == I225_I_PHY_ID)
>>>>> +	if (phy->autoneg_mask & ADVERTISE_2500_FULL)
>>>>>       		ret_val = phy->ops.write_reg(hw,
>>>>>       					     (STANDARD_AN_REG_MASK <<
>>>>>       					     MMD_DEVADDR_SHIFT) |
>>>>
>>>
>>
>