[openib-general] IBAT kernel oops with latest 3432 svn drop....

[Arlin Davis]

Hal Rosenstock wrote:

>On Wed, 2005-09-14 at 19:16, Arlin Davis wrote:
>  
>
>>>>>My SM was down (ports in INIT state)
>>>>>     
>>>>>
>>>>>          
>>>>>
>>>Had an SM been up previously ? Just wondering how to recreate this.
>>> 
>>>
>>>      
>>>
>>SM was actually running, but for some reason it was not sweeping and 
>>configuring my ports.
>>    
>>
>
>OpenSM 1.0.0 or 1.1.0 or some other SM ?
>  
>

1.0.0

>  
>
>>>>Question: What can I expect as a result of ib_at_cancel()?
>>>>   
>>>>
>>>>        
>>>>
>>>Return 0 if canceled, -1 if cancel failed (e.g. bad ID)
>>>
>>> 
>>>
>>>      
>>>
>>>>Will I always 
>>>>get an event with -EINTR for the cancelled request id? 
>>>>   
>>>>
>>>>        
>>>>
>>>Is that what you are seeing ? 
>>> 
>>>
>>>      
>>>
>>sometimes I see an event with -EINTR and sometimes I don't see any event.
>>    
>>
>
>Yes, it looks like in kernel AT it depends on whether the request is
>still pending or not as to what occurs (at least right now).
>  
>
We need some deterministic results. Maybe a guarantee that all events 
for the cancelled req_id
will be processed before ib_at_cancel returns? Is that possible?

-arlin

>-- Hal
>
>  
>