[PATCH 2/3] net/nfp: fix free resource problem

[Ferruh Yigit]

On 12/18/2023 1:50 AM, Chaoyong He wrote:
>> On 12/14/2023 10:24 AM, Chaoyong He wrote:
>>> From: Long Wu <long.wu at corigine.com>
>>>
>>> Set the representor array to NULL to avoid that close interface does
>>> not free some resource.
>>>
>>> Fixes: a135bc1644d6 ("net/nfp: fix resource leak for flower firmware")
>>> Cc: chaoyong.he at corigine.com
>>> Cc: stable at dpdk.org
>>>
>>> Signed-off-by: Long Wu <long.wu at corigine.com>
>>> Reviewed-by: Chaoyong He <chaoyong.he at corigine.com>
>>> Reviewed-by: Peng Zhang <peng.zhang at corigine.com>
>>> ---
>>>  drivers/net/nfp/flower/nfp_flower_representor.c | 15 ++++++++++++++-
>>>  1 file changed, 14 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/net/nfp/flower/nfp_flower_representor.c
>>> b/drivers/net/nfp/flower/nfp_flower_representor.c
>>> index 27ea3891bd..5f7c1fa737 100644
>>> --- a/drivers/net/nfp/flower/nfp_flower_representor.c
>>> +++ b/drivers/net/nfp/flower/nfp_flower_representor.c
>>> @@ -294,17 +294,30 @@ nfp_flower_repr_tx_burst(void *tx_queue,  static
>>> int  nfp_flower_repr_uninit(struct rte_eth_dev *eth_dev)  {
>>> +	uint16_t index;
>>>  	struct nfp_flower_representor *repr;
>>>
>>>  	repr = eth_dev->data->dev_private;
>>>  	rte_ring_free(repr->ring);
>>>
>>> +	if (repr->repr_type == NFP_REPR_TYPE_PHYS_PORT) {
>>> +		index = NFP_FLOWER_CMSG_PORT_PHYS_PORT_NUM(repr-
>>> port_id);
>>> +		repr->app_fw_flower->phy_reprs[index] = NULL;
>>> +	} else {
>>> +		index = repr->vf_id;
>>> +		repr->app_fw_flower->vf_reprs[index] = NULL;
>>> +	}
>>> +
>>>  	return 0;
>>>  }
>>>
>>>  static int
>>> -nfp_flower_pf_repr_uninit(__rte_unused struct rte_eth_dev *eth_dev)
>>> +nfp_flower_pf_repr_uninit(struct rte_eth_dev *eth_dev)
>>>  {
>>> +	struct nfp_flower_representor *repr = eth_dev->data->dev_private;
>>> +
>>> +	repr->app_fw_flower->pf_repr = NULL;
>>>
>>
>> Here it is assigned to NULL but is it freed? If freed, why not set to NULL where
>> it is freed?
>>
>> Same for above phy_reprs & vf_reprs.
> 
> The whole invoke view:
> rte_eth_dev_close()
>     --> nfp_flower_repr_dev_close()
>         --> nfp_flower_repr_free()
>             --> nfp_flower_pf_repr_uninit()
>             --> nfp_flower_repr_uninit()
>            // In these two functions, we just assigned to NULL but not freed yet.
>            // It is still refer by the `eth_dev->data->dev_private`.
>     --> rte_eth_dev_release_port()
>         --> rte_free(eth_dev->data->dev_private);
>         // And here it is really freed (by the rte framework).
> 

'rte_eth_dev_release_port()' frees the device private data, but not all
pointers, like 'repr->app_fw_flower->pf_repr', in the struct are freed,
it is dev_close() or unint() functions responsibility.

Can you please double check if
'eth_dev->data->dev_private->app_fw_flower->pf_repr' freed or not?