This patchset introduce a new structure that will contain all the
infiniband device operations, the structure will be used by the
providers to initialize their supported operations. This patchset also
includes the required changes in the core and ulps to start using it.
I'm also not sure if this is a real improvement, it was done in
userspace because we needed to inject actual dummy handlers that do
ENOSUPP and had a complication where the ops were exposed as part of
the ABI, but we needed to revise the driver ops. None of these reasons
really apply to the kernel..
It is a lot of churn and will hurt backporting.. But it is a little
tidier in the drivers.
I guess if nobody is opposed it could go forward
I think its OK'ish... I agree with Parav that it would be much more
appealing if we could actually sort fast-path vs. slow-path
references (which would also move stuff like device name and various
lists and locks to live behind it). But that could be done as an
incremental patch..
