[openib-general] Re: ib_uat kernel panic without async callback

[Arlin Davis]

Hal Rosenstock wrote:

>Hi Arlin,
>
>On Thu, 2005-07-14 at 19:23, Arlin Davis wrote:
>  
>
>>>OK. I looked into this and it appears to me that the route is not copied
>>>back to userspace until the callback is made.
>>>
>>>-- Hal
>>> 
>>>
>>>      
>>>
>>ok, I will always get the callback for now and ignore a return of 1. 
>>Is this the expected behavior because it does not match the documentation.
>>    
>>
>
>I can look at changing the implementation but I'm curious about what
>specifically in the documentation you are referring to.
>  
>

at.h

* The following asynchronous resolution function behavior is as follows:
 *    If the resolve operation can be fulfilled immediately, then the output
 *    structures are set and the number of filled structures is returned.
 *
 *    If the resolve operation cannot by fulfilled immediately and
 *    an ib_at_completion structure is not provided,
 *    then the function immediately returns -EWOULDBLOCK.
 *
 *     If ib_at_completion structure is provided and an asynchronous
 *    operation is started, the function immediately returns zero,
 *    and the request ID field (req_id) is set if the pointer is
 *    non NULL. This request ID may be used to cancel the operation,
 *    or to poll its status.

Your implementation is fine, as long as the correct return values are used
and the sync version works. I currently get returned a 1 but the ib_route
data is not filled in and the req_id == 0. I would expect a 0 returned 
and a
valid req_id if the ib_route cannot be processed.

for ret = ib_at_route_by_ip( async_cb provided ), I read this as: 

  if ret > 0  then ib_route output data is written, resolve complete
  if ret == 0 then req_id is written with valid id to cancel or poll
  if ret < 0 error

Am I reading this right?
 
-arlin

>-- Hal
>
>  
>