[ofw] RE: [PATCH] UAL: Fix NULL dereference when polling a CEP

[Fab Tillier]

>> Is issuing IOCTLs by non-driver components something that is/should be
>> supported (let alone done?)
>
> I don't consider the ND provider library a driver component. :)

It gets installed with IPoIB so currently it's a driver component.

> My short answer is yes.  Users should be free to write their own
> libraries (for whatever reason) to interface to the IOCTL interfaces.
> Our goal should be to provide simple, well-defined IOCTL interfaces
> that are supportable long term.

Isn't that what WinVerbs is striving to do?  Aren't the IBAL IOCTLs slated for elimination long term?

The correct fix would be to remove the field from the IOCTL altogether.  We can do that, but then we have to decide whether that change should require another bump in the IOCTL version.  Revision 2079 changed the IOCTL version and there hasn't been a fully functional build since, so we could eliminate the field from the IOCTL under the same ABI change (though a different check-in.)

Thoughts?

-Fab