On Tue, May 31, 2016 at 08:01:31PM +0000, Hefty, Sean wrote:
> > If I understand your proposal correctly, the proposed IOCTL framework will be
> > able to handle all types of objects without any limitations in order and locking,
> > everything is going to be processed by the driver, which will decide if
> > such object/method is supported or not.
> 
> For the core ioctl framework - yes.
> 
> An ioctl can be dispatched to some other common handling, which can validate object ordering and type.

Agree and IMHO it is where we should be focused.

> 
> > What will stop the GPU with DRMs to use it?
> 
> It would have to hook into the same file.  But if it's generic enough, nothing would prevent other subsystems (e.g. Xeon Phi or NVM) from re-using the code.  I don't see this as a negative.

It is not negative, just out of the scope of RDMA subsystem.

> 
> - Sean