Re: [PATCH rdma-next 6/6] RDMA/core: Unify style of IOCTL commands

On Tue, Aug 16, 2016 at 05:09:27PM +0000, Dalessandro, Dennis wrote:

On Tue, 2016-08-16 at 19:50 +0300, Leon Romanovsky wrote:

quoted

On Tue, Aug 16, 2016 at 02:31:32PM +0000, Dalessandro, Dennis wrote:

quoted

On Tue, 2016-08-16 at 16:45 +0300, Leon Romanovsky wrote:

quoted

From: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

MAD and HFI1 have different naming convention, this patch
simplifies and unifies their defines and names.

I don't know that I agree that it simplifies things. It changes a
lot
of code for not much real value in my opinion. 

Was this something that was discussed in the verbs call? I have not
been able to attend that the last few weeks.

It was discussed over mails, for example Jason's opinion [1],
Christopher Lameter's opinion [2], Christoph Hellwig's opinion [3].

I'm not opposed to trying to unify things, however this seems to be
more than plop down the hfi1 stuff from here and put it over there. It
is certainly not simplifying anything.

It is "dark side" of UAPI - inability to change legacy declarations.
This is why I didn't remove anything except _NUM() macro.

The simplification comes from definition of one place for
declaration of IOCTLs numbers and exporting it to users. It gives
visibility for user space authors too.

quoted

quoted

quoted

As part of cleanup, the HFI1 _NUM() macro was removed and MAD
indexes were renamed. It has a potential to break application
which use these defines directly.

Why do you want to remove the _NUM() macro that Doug just put in?

It is not used after refactoring and IMHO this macro doesn't
belong to UAPI, since it wouldn't in use by any users, but I'll be
glad to
get an examples of its usage in real user space applications
(libfabric???), if any.

That's a fair point. I can see getting rid of it now.

-Denny