Re: [PATCH for-next v3 2/2] RDMA/rxe: Call rxe_set_mtu after rxe_register_device

[Zhu Yanjun <yanjun.zhu@linux.dev>]

在 2023/9/19 11:25, Zhijian Li (Fujitsu) 写道:
>
> On 19/09/2023 09:11, Zhu Yanjun wrote:
>> On Tue, Sep 19, 2023 at 8:57 AM Zhijian Li (Fujitsu)
>> <lizhijian@fujitsu.com> wrote:
>>>
>>>
>>> On 18/09/2023 20:37, Leon Romanovsky wrote:
>>>> On Mon, Sep 18, 2023 at 10:05:43AM +0800, Li Zhijian wrote:
>>>>> rxe_set_mtu() will call rxe_info_dev() to print message, and
>>>>> rxe_info_dev() expects dev_name(rxe->ib_dev->dev) has been assigned.
>>>>>
>>>>> Previously since dev_name() is not set, when a new rxe link is being
>>>>> added, 'null' will be used as the dev_name like:
>>>>>
>>>>> "(null): rxe_set_mtu: Set mtu to 1024"
>>>>>
>>>>> Move rxe_register_device() earlier to assign the correct dev_name
>>>>> so that it can be read by rxe_set_mtu() later.
>>>> I would expect removal of that print line instead of moving
>>>> rxe_register_device().
>>>
>>> I also struggled with this point. The last option is keep it as it is.
>>> Once rxe is registered, this print will work fine.
>> I delved into the source code. About moving rxe_register_device, I
>> could not find any harm to the driver.
> The point i'm struggling was that, it's strange/opaque to move rxe_register_device().
> There is no doubt that the original order was more clear.


Please check the source code. It is very clear.


> In terms of the message content, is it valuable to print(pr_info) this message, i noticed
> that there is a duplicate pr_dbg() in rxe_notify().


rxe_notify is for netdevice event, not for rdma event.


>
> rxe's mtu is always same with the NIC, isn't it ?

Please check the following source code.

void rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu)

{
         struct rxe_port *port = &rxe->port;
         enum ib_mtu mtu;

         mtu = eth_mtu_int_to_enum(ndev_mtu);

         /* Make sure that new MTU in range */
         mtu = mtu ? min_t(enum ib_mtu, mtu, IB_MTU_4096) : IB_MTU_256;

         port->attr.active_mtu = mtu;
         port->mtu_cap = ib_mtu_enum_to_int(mtu);

         rxe_info_dev(rxe, "Set mtu to %d", port->mtu_cap);
}

Zhu Yanjun


>
> Thanks
> Zhijian
>
>
>
>> So I think this is also a solution to this problem.
>>
>> Zhu Yanjun
>>
>>> Thanks
>>> Zhijian
>>>
>>>
>>>> Thanks
>>>>
>>>>> And it's safe to do such change since mtu will not be used during the
>>>>> rxe_register_device()
>>>>>
>>>>> After this change, the message becomes:
>>>>> "rxe_eth0: rxe_set_mtu: Set mtu to 4096"
>>>>>
>>>>> Fixes: 9ac01f434a1e ("RDMA/rxe: Extend dbg log messages to err and info")
>>>>> Reviewed-by: Daisuke Matsuda <matsuda-daisuke@fujitsu.com>
>>>>> Reviewed-by: Zhu Yanjun <yanjun.zhu@linux.dev>
>>>>> Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
>>>>> ---
>>>>>     drivers/infiniband/sw/rxe/rxe.c | 5 ++++-
>>>>>     1 file changed, 4 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/infiniband/sw/rxe/rxe.c b/drivers/infiniband/sw/rxe/rxe.c
>>>>> index a086d588e159..8a43c0c4f8d8 100644
>>>>> --- a/drivers/infiniband/sw/rxe/rxe.c
>>>>> +++ b/drivers/infiniband/sw/rxe/rxe.c
>>>>> @@ -169,10 +169,13 @@ void rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu)
>>>>>      */
>>>>>     int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name)
>>>>>     {
>>>>> +    int ret;
>>>>> +
>>>>>        rxe_init(rxe);
>>>>> +    ret = rxe_register_device(rxe, ibdev_name);
>>>>>        rxe_set_mtu(rxe, mtu);
>>>>>
>>>>> -    return rxe_register_device(rxe, ibdev_name);
>>>>> +    return ret;
>>>>>     }
>>>>>
>>>>>     static int rxe_newlink(const char *ibdev_name, struct net_device *ndev)
>>>>> --
>>>>> 2.29.2
>>>> >