Re: [PATCH iwl-next v3 3/5] i40e: Add helpers to find VSI and VEB by SEID and use them

[Ivan Vecera <ivecera@redhat.com>]

On 20. 11. 23 12:42, Simon Horman wrote:
> On Thu, Nov 16, 2023 at 04:21:12PM +0100, Ivan Vecera wrote:
>> Add two helpers i40e_(veb|vsi)_get_by_seid() to find corresponding
>> VEB or VSI by their SEID value and use these helpers to replace
>> existing open-coded loops.
>>
>> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
>> Signed-off-by: Ivan Vecera <ivecera@redhat.com>
> 
> Hi Ivan,
> 
> some minor feedback from my side.
> 
> ...
> 
>> diff --git a/drivers/net/ethernet/intel/i40e/i40e.h b/drivers/net/ethernet/intel/i40e/i40e.h
>> index 1e9266de270b..ca8997d29c02 100644
>> --- a/drivers/net/ethernet/intel/i40e/i40e.h
>> +++ b/drivers/net/ethernet/intel/i40e/i40e.h
>> @@ -1360,4 +1360,38 @@ static inline struct i40e_pf *i40e_hw_to_pf(struct i40e_hw *hw)
>>   
>>   struct device *i40e_hw_to_dev(struct i40e_hw *hw);
>>   
>> +/**
>> + * i40e_pf_get_vsi_by_seid - find VSI by SEID
>> + * @pf: pointer to a PF
> 
> nit: @seid is missing here
> 
>> + **/
>> +static inline struct i40e_vsi *
>> +i40e_pf_get_vsi_by_seid(struct i40e_pf *pf, u16 seid)
>> +{
>> +	struct i40e_vsi *vsi;
>> +	int i;
>> +
>> +	i40e_pf_for_each_vsi(pf, i, vsi)
>> +		if (vsi->seid == seid)
>> +			return vsi;
>> +
>> +	return NULL;
>> +}
>> +
>> +/**
>> + * i40e_pf_get_veb_by_seid - find VEB by SEID
>> + * @pf: pointer to a PF
> 
> Ditto
> 
> ...
> 
>> diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c
> 
> ...
> 
>> @@ -14848,23 +14831,16 @@ struct i40e_veb *i40e_veb_setup(struct i40e_pf *pf, u16 flags,
>>   	}
>>   
>>   	/* make sure there is such a vsi and uplink */
>> -	i40e_pf_for_each_vsi(pf, vsi_idx, vsi)
>> -		if (vsi->seid == vsi_seid)
>> -			break;
>> -
>> -	if (vsi_idx == pf->num_alloc_vsi && vsi_seid != 0) {
>> -		dev_info(&pf->pdev->dev, "vsi seid %d not found\n",
>> -			 vsi_seid);
>> -		return NULL;
>> +	if (vsi_seid) {
>> +		vsi = i40e_pf_get_vsi_by_seid(pf, vsi_seid);
>> +		if (!vsi) {
>> +			dev_err(&pf->pdev->dev, "vsi seid %d not found\n",
>> +				vsi_seid);
>> +			return NULL;
>> +		}
>>   	}
>> -
>>   	if (uplink_seid && uplink_seid != pf->mac_seid) {
>> -		i40e_pf_for_each_veb(pf, veb_idx, veb) {
>> -			if (veb->seid == uplink_seid) {
>> -				uplink_veb = veb;
>> -				break;
>> -			}
>> -		}
>> +		uplink_veb = i40e_pf_get_veb_by_seid(pf, uplink_seid);
>>   		if (!uplink_veb) {
>>   			dev_info(&pf->pdev->dev,
>>   				 "uplink seid %d not found\n", uplink_seid);
> 
> The next part of this function looks like this:
> 
> 		if (!uplink_veb) {
> 			dev_info(&pf->pdev->dev,
> 				 "uplink seid %d not found\n", uplink_seid);
> 			return NULL;
> 		}
> 	}
> 	/* get veb sw struct */
> 	veb_idx = i40e_veb_mem_alloc(pf);
> 	if (veb_idx < 0)
> 		goto err_alloc;
> 	veb = pf->veb[veb_idx];
> 	veb->flags = flags;
> 	veb->uplink_seid = uplink_seid;
> 	veb->veb_idx = (uplink_veb ? uplink_veb->idx : I40E_NO_VEB);
> 	veb->enabled_tc = (enabled_tc ? enabled_tc : 0x1);
> 
> 	/* create the VEB in the switch */
> 	ret = i40e_add_veb(veb, vsi);
> 
> Smatch complains that vsi may be used uninitialised here.
> Which does seem possible to me if vsi_seid is 0.

Yes, the support for floating VEBs is and WAS broken prior this patch 
and it is fixed by the following patch.

Prior this patch... Let's vsi_seid == 0:

	/* make sure there is such a vsi and uplink */
	i40e_pf_for_each_vsi(pf, vsi_idx, vsi)
		if (vsi->seid == vsi_seid)
			break;
-> here vsi_idx == pf->num_alloc_vsi because there cannot be VSI with 
SEID == 0... and VSI points after the pf->vsi[] array.

	if (vsi_idx == pf->num_alloc_vsi && vsi_seid != 0) {
		dev_info(&pf->pdev->dev, "vsi seid %d not found\n",
			 vsi_seid);
		return NULL;
	}

This condition is not met, although vsi_idx == pf->num_alloc_vsi but 
vsi_seid == 0 so the rest of code ended by:

	ret = i40e_add_veb(veb, vsi);

and vsi pointer points to invalid area (item after the last one from 
pf->vsi).

As I mentioned the broken floating VEB functionality (where vsi_seid == 
0 and uplink_seid == 0) is fixed by the following patch.

Thanks,
Ivan