[openib-general] [PATCH] kDAPL: cleanup dat/ a bit more

[Caitlin Bestler]

On 6/3/05, Christoph Hellwig <hch at lst.de> wrote:

> 
> > The next question is to define what if any handshake is desired.
> > My guess that the consumer would acknowledge this by closing
> > the RNIC, and that there would be some sort of deadline for doing
> > so (much like a shutdown, you have so long to clean up, after that
> > -- well you were warned).
> 
> There should be no handshake involved.  After ->remove is called the
> midlayer must fail all operations.  This is total standard behaviour
> you get when using the driver model.  There's no reason the rdma
> code should be any different from other subsystems.
> 
> 
Since you haven't bothered to learn much about RDMA, let me 
explain that unlike typical devices an RDMA device can be
pefrforming IO even when the OS is not aware of it. That is
why the Quadrics extensions for zero placement are not
applicable to an RDMA device. That extension, for example,
only remaps memory when the IO device is not actively
doing a transfer. Something that the kernel does not
know with an RDMA device.

In any event, "after ->remove is called" *is* a handshake.
The question is whether the tolerances on how long the
call can take are acceptable to this type of device.