Re: [PATCH 0/7] IB/hfi1: Remove write() and use ioctl() for user access

[Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

[-- Attachment #1.1: Type: text/plain, Size: 1647 bytes --]

On 4/27/2016 1:21 AM, Weiny, Ira wrote:
>> 
>> On 4/20/2016 4:46 PM, Doug Ledford wrote:

>> We already have precedent for requiring an IPoIB device, and it's
>> associate netdevice, in order to manage some non-IP, non-Ethernet,
>> IB specific items (the recent SRIOV patches being a perfect
>> example).
> 
> I'm not sure this is always going to be the case.  I have heard a
> number of reports of IPoIBs inability to scale and a desire to not
> run it.

Then don't.  If you set the configuration file for the IPoIB device to
not start on bootup, the module will get loaded, the device will get
allocated, and then it will not transmit on the wire or join any
multicast groups.  It will then be a perfectly fine vessel for accessing
configuration, eeproms, etc. and it won't impact the fabric in any way.

If, at a later point in time, you wish to actually use it, then we can
look at making it better for scale.

> Isn't there a way we can create a netdevice that does not have the
> associated overhead of IPoIB?

Yes, load IPoIB with the device turned off.  Zero overhead.  At the
point at which you have a new IP emulation device that can be hooked
into the netstack, you can use it as the ndo entry point device instead.
 It makes no difference in the long run, the user space tools are the
same and the training for admins is the same, only the layer used to
create the user visible net device changes.

I snipped the rest of your email, it all centers around the incorrect
assumption that IPoIB devices must be up and running in order for the
configuration entry points to be used by user space tools.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]