brcmfmac & DEL_INTERFACE

brcmfmac & DEL_INTERFACE

[Denis Kenzior]

Hi Arend,

We noticed that brcmfmac doesn't support .del_virtual_intf for 
non-p2p/ap interface types.  Any chance this can be added?

We currently remove all wifi interfaces and re-create the needed ones 
with SOCKET_OWNER set, and it would be nice if we didn't need to treat 
brcmfmac specially.

Regards,
-Denis

Re: brcmfmac & DEL_INTERFACE

[Arend Van Spriel]

On 5/28/2019 8:16 PM, Denis Kenzior wrote:
> Hi Arend,
> 
> We noticed that brcmfmac doesn't support .del_virtual_intf for 
> non-p2p/ap interface types.  Any chance this can be added?
> 
> We currently remove all wifi interfaces and re-create the needed ones 
> with SOCKET_OWNER set, and it would be nice if we didn't need to treat 
> brcmfmac specially.

This came up recently. During probe the driver creates a network 
interface that we refer to as primary interface. We consider this 
non-virtual and ownership is with the driver. My guess is that this 
concept comes from the WEXT era, where we did not have the ieee80211 phy 
objects to interact with the driver from user-space. I suppose you don't 
mind the creation of this interface and just want to allow removing it, 
right?

Regards,
Arend

Re: brcmfmac & DEL_INTERFACE

[Denis Kenzior]

Hi Arend,

On 05/28/2019 03:27 PM, Arend Van Spriel wrote:
> On 5/28/2019 8:16 PM, Denis Kenzior wrote:
>> Hi Arend,
>>
>> We noticed that brcmfmac doesn't support .del_virtual_intf for 
>> non-p2p/ap interface types.  Any chance this can be added?
>>
>> We currently remove all wifi interfaces and re-create the needed ones 
>> with SOCKET_OWNER set, and it would be nice if we didn't need to treat 
>> brcmfmac specially.
> 
> This came up recently. During probe the driver creates a network 
> interface that we refer to as primary interface. We consider this 
> non-virtual and ownership is with the driver. My guess is that this 
> concept comes from the WEXT era, where we did not have the ieee80211 phy 
> objects to interact with the driver from user-space. I suppose you don't 
> mind the creation of this interface and just want to allow removing it, 
> right?

Correct.  If we can at least get the DEL_INTERFACE supported, that would 
solve our immediate use case.

I do think that the drivers should not be creating a netdev by default 
and should wait until userspace asks for it.  But that is a separate 
topic, with backwards compatibility concerns, so I'll leave it for the 
future :)

Regards,
-Denis