[ofa-general] timeout question

[Dotan Barak]

Rui Machado wrote:
>>
>> Can you replace it with a write (from the other side)?
>> READ has "higher price" than a WRITE.
>>
>>     
>
> Can you please, shortly explain why this higher price?
>   
Basically, when a Read request is being sent the HCA need to preserve 
resources in order to accept the response.

>   
>> Anyway, you should get the mentioned behavior anyway..
>>
>> When the sender get the error, what is the status of the receiver QP?
>> (did you try to execute ibv_query_qp and get its status?)
>>
>>     
>
> I tried to get the qp state right after the error and it is 6 (which I
> believe is IBV_QPS_ERR).
> Why do you ask?
>   
If the QP in the receiver side is in ERROR state, it means that there 
was an error and this is the root
cause that caused to the retry exceeded from the first place ...

maybe there was a remote address/key violation that caused the receiver 
QP to be transitioned to the
ERROR state.

Dotan