Re: [RFC PATCH 09/11] IB/Verbs: Use management helper has_ipoib() and, cap_ipoib() for ipoib-check

[Michael Wang <yun.wang-EIkl63zCoXaH+58JC4qpiA@public.gmane.org>]

On 03/27/2015 05:06 PM, Jason Gunthorpe wrote:
> On Fri, Mar 27, 2015 at 04:48:22PM +0100, Michael Wang wrote:
>
>> diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
>> index 3341754..fcd7558 100644
>> +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
>> @@ -1655,7 +1655,7 @@ static void ipoib_add_one(struct ib_device *device)
>>      struct ipoib_dev_priv *priv;
>>      int s, e, p;
>>  
>> -    if (!rdma_transport_is_ib(device))
>> +    if (!has_ipoib(device))
>>          return;
> This is a good example of a test that doesn't really make sense, IPoIB
> is certainly a port specific attribute.
According to my understanding, seems like the logical is:

    if 'device have no port need ipoib initialization'
            return
>
>>      dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
>> @@ -1673,7 +1673,7 @@ static void ipoib_add_one(struct ib_device *device)
>>      }
>>  
>>      for (p = s; p <= e; ++p) {
>> -        if (!rdma_port_ll_is_ib(device, p))
>> +        if (!cap_ipoib(device, p))
>>              continue;
> And down here we test every port.
>
> The routine should just test every port and do nothing if no ports
> need IPoIB.
And here since there are some initial work to do, it iterate all the
port to find out which one need to be initialized.

>
>>          dev = ipoib_add_port("ib%d", device, p);
>>          if (!IS_ERR(dev)) {
>> @@ -1690,7 +1690,7 @@ static void ipoib_remove_one(struct ib_device *device)
>>      struct ipoib_dev_priv *priv, *tmp;
>>      struct list_head *dev_list;
>>  
>> -    if (!rdma_transport_is_ib(device))
>> +    if (!has_ipoib(device))
>>          return;
> This test should be made redundant by having ib_get_client_data return
> null if ipoib_add_one didn't do anything. Maybe that already happens?
Here if we have done nothing when add a device, then nothing
need to be cleanup when remove it.

I'm not sure if it's a good idea, but seems like the idea is use twice
check on different level to save some cycles?

Regards,
Michael Wang
>
> Jason

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html