Re: [PATCH V2 1/4] net/bonding: fix non-active slaves aren't stopped

["Min Hu (Connor)" <humin29@huawei.com>]

Hi, Ferruh,

在 2022/5/4 3:04, Ferruh Yigit 写道:
> On 5/3/2022 7:54 AM, Min Hu (Connor) wrote:
>> Hi, Ferruh,
>>
>> 在 2022/4/29 21:31, Ferruh Yigit 写道:
>>> On 4/29/2022 7:45 AM, Min Hu (Connor) wrote:
>>>> Hi, Ferruh,
>>>>
>>>> 在 2022/4/27 2:19, Ferruh Yigit 写道:
>>>>> On 3/24/2022 3:00 AM, Min Hu (Connor) wrote:
>>>>>> From: Huisong Li <lihuisong@huawei.com>
>>>>>>
>>>>>> When stopping a bonded port, all slaves should be deactivated. But 
>>>>>> only
>>>>>
>>>>> s/deactivated/stopped/ ?
>>>> not agreed. deactivated and stopped are different state for slave.
>>>>
>>>
>>> Just to clarify the sentences, otherwise I see the 'stopped' and 
>>> 'deactivated' states are different.
>>> Next sentences complains that not all ports are stopped: "But only 
>>> active slaves are stopped.", so I thought intention in this sentences 
>>> to claim that all slaves should be stopped (but it mentions all 
>>> slaves should be 'deactivated').
>>> As long as you address the disconnection between two sentences, I 
>>> don't mind the wording.
>> Actually, there is something wrong with the wording.
>> Yes, I should take your advice.
>>
>>>
>>>>>
>>>>>> active slaves are stopped. So fix it and do "deactivae_slave()" 
>>>>>> for active
>>>>>
>>>>> s/deactivae_slave()/deactivate_slave()/
>>>>>
>>>> agreed.
>>>>
>>>>>> slaves.
>>>>>
>>>>> Hi Connor,
>>>>>
>>>>> When a bonding port is closed, is it clear if all slave ports or 
>>>>> active slave ports should be stopped?
>>>> Yes, I think all the slave ports should be stopped(or try to be 
>>>> stopped).
>>>>>
>>>>>>
>>>>>> Fixes: 0911d4ec0183 ("net/bonding: fix crash when stopping mode 4 
>>>>>> port")
>>>>>> Cc: stable@dpdk.org
>>>>>>
>>>>>> Signed-off-by: Huisong Li <lihuisong@huawei.com>
>>>>>> Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
>>>>>> ---
>>>>>>   drivers/net/bonding/rte_eth_bond_pmd.c | 20 +++++++++++---------
>>>>>>   1 file changed, 11 insertions(+), 9 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/net/bonding/rte_eth_bond_pmd.c 
>>>>>> b/drivers/net/bonding/rte_eth_bond_pmd.c
>>>>>> index b305b6a35b..469dc71170 100644
>>>>>> --- a/drivers/net/bonding/rte_eth_bond_pmd.c
>>>>>> +++ b/drivers/net/bonding/rte_eth_bond_pmd.c
>>>>>> @@ -2118,18 +2118,20 @@ bond_ethdev_stop(struct rte_eth_dev *eth_dev)
>>>>>>       internals->link_status_polling_enabled = 0;
>>>>>>       for (i = 0; i < internals->slave_count; i++) {
>>>>>>           uint16_t slave_id = internals->slaves[i].port_id;
>>>>>> +
>>>>>> +        internals->slaves[i].last_link_status = 0;
>>>>>> +        ret = rte_eth_dev_stop(slave_id);
>>>>>> +        if (ret != 0) {
>>>>>> +            RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
>>>>>> +                     slave_id);
>>>>>> +            return ret;
>>>>>
>>>>> Should it return here or try to stop all ports?
>>>>> What about to record the return status, but keep continue to stop 
>>>>> all ports. And return error if any of the stop failed?
>> Well, I am glad you have found something unreasaonable about 'stop'.
>> Let us see API 'rte_eth_dev_stop'
>>
>> rte_eth_dev_stop(dev)
>> {
>>      ....
>>      dev->data->dev_started = 0;
>>      ret = (*dev->dev_ops->dev_stop)(dev)
>>      retur ret;
>> }
>> This is unreasaonable. No matter 'dev_ops->dev_stop' succeed or fail,
>> the state 'dev_started ' will always set to be '0'.
>>
>> But this does not only influence bonding device but other devices like
>> eth dev or vdev.
>> This is the bug in rte ethdev level. I will send another patch to fix
>> it.
>>
> 
> Hi Connor,
> 
> I agree this is an issue in the API, cc'ed Andrew and Thomas.
> 
> I vaguely remember that "dev_started = 0" was required for some dev_ops, 
> but not quite sure, let me check this.
> At worst we can do as following to be sure:
> 
>    dev->data->dev_started = 0;
>    ret = (*dev->dev_ops->dev_stop)(dev)
>    if (ret)
>      dev->data->dev_started = 1;
> 
> Also we need to clarify in the API documentation (.h file), what is the 
> status of the device if 'rte_eth_dev_stop()' returned error.
> 
> 
> Btw, would you be OK to separate this ethdev patch from your bonding 
> patch, to not stuck your series because of ethdev one.
Yes, this patch can be abandoned from this set.

> 
> 
>>
>>>> I think no need to do this. APP only see the bonded device. If bonded
>>>> device stop failed, it means it works failed. And the number of 
>>>> "stopped" successfully slave does not make any sense.
>>>>
>>>
>>> OK if trying to stop as much as possible 'slave' devices doesn't make 
>>> sense, we can keep as it is.
>>>
>>> Btw, when functions fails at this point, bonding device itself 
>>> already marked as stopped, right? And some of the slave devices may 
>>> be stopped already before failure.
>>> I don't know how confusing this is for the user, that stop() function 
>>> is failed but bonding device state is 'stopped'. I don't know if 
>>> function should recover at least bonding device status (back to 
>>> started) on failure, what do you think?
>>>
>>>>>
>>>>>> +        }
>>>>>> +
>>>>>> +        /* active slaves need to deactivate. */
>>>>>
>>>>> " active slaves need to be deactivated. " ?
>>>> agreed.
>>>>>
>>>>>>           if (find_slave_by_id(internals->active_slaves,
>>>>>>                   internals->active_slave_count, slave_id) !=
>>>>>> -                        internals->active_slave_count) {
>>>>>> -            internals->slaves[i].last_link_status = 0;
>>>>>> -            ret = rte_eth_dev_stop(slave_id);
>>>>>> -            if (ret != 0) {
>>>>>> -                RTE_BOND_LOG(ERR, "Failed to stop device on port 
>>>>>> %u",
>>>>>> -                         slave_id);
>>>>>> -                return ret;
>>>>>> -            }
>>>>>> +                internals->active_slave_count)
>>>>>
>>>>> I think original indentation for this line is better.
>>>>>
>>>> agreed.
>>>>>>               deactivate_slave(eth_dev, slave_id);
>>>>>> -        }
>>>>>>       }
>>>>>>       return 0;
>>>>>
>>>>> .
>>>
>>> .
> 
> .