>I think I agree with Michael that having ib_sa_register_client()
>allocate the structure is going to far in the name of encapsulation.
>How about making the structure public, but providing a macro
>DEFINE_IB_SA_CLIENT() to wrap up the initialization? (Similar to
>DEFINE_MUTEX, DEFINE_IDR, etc)
Currently what I have is:
struct ib_sa_client {
atomic_t users;
struct completion comp;
};
I tried not to change it too much from what Michael had, but I haven't added
canceling of requests yet either. Canceling requests isn't strictly needed; we
just end up waiting longer for deregister to complete. But doing so should only
add a list_head to the structure. So, yes, I can move the definition back to
being public.
>If we want to allow for dynamic allocation of clients, then we can
>have a ib_sa_init_client(struct ib_sa_client *client) function, but I
>don't see a need for that right now.
I do use dynamic allocation for userspace support.
- Sean
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
