Krishna Kumar wrote:
Hi,
I am not sure if this is a good idea, but since I am new to this area, here it goes :-)
I think that the idea is valid. :)
Section 11.2.6.3, C11-16 states that resize of qp must be permitted. In the patch I am submitting, I don't understand why so many parameters are expected by driver/verbs. I thought the qp_handle and ib_qp_attr is enough, atleast according to the spec.
I didn't follow what you were trying to reference here. Are you referring to the QP or CQ?
Along with this, I am going to submit another patch to catch "catastrophic" errors in return value of the resize operation. This is due to the need to check for 2 special cases : "CQ overrun" and "CQ inaccessible". For these two errors, I think the queues should be deallocated and error returned. This is in the second patch. I am not sure of the error numbers, I guessed it from mthca_eq.c and could be wrong here.
I'm adding in code to handle QP errors and overrun. If we are unable to resize the CQ, we can prevent CQ overrun by limited the number of work requests posted to the corresponding QPs, rather than completely disabling the port. I'll have a better idea of what we can do in this case when I get more of the code in place.
_______________________________________________
openib-general mailing list
[EMAIL PROTECTED]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
