Re: [rfc/patch] wake_up_info() draft ...

[Manfred Spraul <manfred@colorfullife.com>]

Hi Davide,

I think the patch adds unnecessary bloat, and mandates one particular 
use of the wait queue info interface.
For example, why does remove_wait_queue_info copy the wakeup info 
around? That's now how I would use it for fasync: I would send the 
necessary signals directly from the wakeup handler, and 
remove_wait_queue_info is called during sys_close handling, info discarded.

I'm thinking about a simpler approach: add a wake_up_info() function, 
and forward the info parameter to the wait_queue_func_t. This means 
changing the prototype of this function - there shouldn't be that many 
instances. NULL is passed if the normal wake_up functions are used. No 
additional fields in the wait queue entry are required. Then I would 
convert kill_fasync to that interface, with the band value from 
kill_fasync as the info parameter. A custom wait queue func does the 
signal sending. fasync_helper would be kmalloc+add_wait_queue.

--
    Manfred