Re: [PATCH for-next] RDMA/core: Assign the name of device when allocating ib_device

[liweihang <liweihang@xxxxxxxxxx>]

On 2020/4/27 20:03, Leon Romanovsky wrote:
>>>>  /**
>>>>   * _ib_alloc_device - allocate an IB device struct
>>>>   * @size:size of structure to allocate
>>>> + * @name: unique string device name. This may include a '%' which will
>>> It looks like all drivers are setting "%" in their name and "name" can
>>> be changed to be "prefix".
>> Does hfi? I thought the name was forced there for some port swapped
>> reason?
> This patch doesn't touch HFI, nothing prohibits from us to make this
> conversion work for all drivers except HFI and for the HFI add some
> different callback. There is no need to make API harder just because
> one driver needs it.
> 
> Thanks
> 
>> Jason

Hi Jason and Leon,

I missed some codes related to assign_name() in this series including
hfi/qib as Shiraz pointed. And I found a "name" without a "%" in following
funtions in core/nldev.c, and ibdev_name will be used for rxe/siw later.

	static int nldev_newlink(struct sk_buff *skb, struct nlmsghdr *nlh,
				  struct netlink_ext_ack *extack)
	{
		...
	
		nla_strlcpy(ibdev_name, tb[RDMA_NLDEV_ATTR_DEV_NAME],
			    sizeof(ibdev_name));
		if (strchr(ibdev_name, '%') || strlen(ibdev_name) == 0)
			return -EINVAL;
	
		...
	}

I'm not familiar with these codes, but I think the judgment in assign_name()
is for the situaion like above.

	if (strchr(name, '%'))
		ret = alloc_name(device, name);
	else
		ret = dev_set_name(&device->dev, name);

So is it a better idea to keep using "name" instead of "prefix"?

Thanks
Weihang