[openib-general] Re: [PATCH][RFC] kDAPL: use linux native timers instead of own

[James Lentini]

On Tue, 24 May 2005, Sean Hefty wrote:

> James Lentini wrote:
> > - Could we use cm timeouts?
>> 
>>   The timers are only used to implement 2 DAT API calls:   dat_ep_connect 
>> and dat_ep_dup_connect. Both use a timer to   implement the same feature 
>> (cancel connection attempts   after some period of time) and both 
>> eventually call   ib_send_cm_req().  ib_send_cm_req() allows the consumer 
>> to specify   timeouts. What happens when these timeouts expire? Could DAPL 
>> use   these instead of kernel timers? 
>>   I don't see an ib_cm_event_type that corresponds to a timeout. This 
>> implies that the CM user's ib_cm_handler is not invoked when one of   these 
>> timeout values expires. If so kDAPL wouldn't be able to make
>>   use of them. Does anyone know how these work?
>
> The CM will invoke the user's ib_cm_handler with an "ERROR" status if a 
> timeout occurs.
>
> For example, a user calls ib_send_cm_req().  If a REP is not received within 
> the timeout/retry window, the REQ will fail with IB_CM_REQ_ERROR.  You 
> shouldn't need any timers to cancel the connection requests.

Excellent. I guess the original DAPL timer wasn't necessary for gen2.

Sean, what unit of time are the timeouts expected in?

Tom, are you willing to rework the patch to use the CM timeout values?