[openib-general] Couple of questions on struct ib_device

[Roland Dreier]

    Krishna> Hi, I was looking at ib_verbs.h and I am confused about
    Krishna> one thing : why are entry points like open_hca,
    Krishna> modify_hca and close_hca missing ?

Not sure which version of ib_verbs.h you were looking at.  If you were
looking in my branch then the structure is incomplete.  In any case,
modify_hca will be split into modify_port and modify_device functions.

I'm not sure we really need open_hca and close_hca (at least as
defined in the spec), and in any case I think this will be handled in
the access layer (the low-level driver doesn't need to handle the
reference counting or anything else).  I expect to get a better idea
about this as I implement device model/sysfs stuff.

    Krishna> Also, is there any value in encapsulating the various
    Krishna> functions of ib_device into a different structure, like
    Krishna> ib_device_ops or something rather than having a visually
    Krishna> huge structure that we have today ?

Maybe, but the kernel currently has far bigger structures such as
struct net_device or struct task_struct.  We can just keep the
function pointers separate in the declaration of struct ib_device.

 - R.