Sean Hefty wrote:
I reviewed the file mad_rmpp.c and it seems that there is a leak of
the Address Handle.
The AH that is being created in the function "alloc_response_msg" is
never being destroyed.
The AH is destroyed in ib_rmpp_send_handler().
I checked this issue again and I added the following prints:
the AH handler which is being created in alloc_response_msg()
the AH handler which is being destroyed in ib_rmpp_send_handler()
It seems that the AHs which are being created in alloc_response_msg()
(which is being called from
ack_ds_ack()) are not being destroyed because the rmpp_type of this
packet is
IB_MGMT_RMPP_TYPE_ACK, so the destroy AH is not being executed.
We saw this issue in our daily regression during the osmtest, here are
the reproduction instructions:
Start the openSM (during the following commands, the openSM needs to be
online)
execute: # osmtest -f c
execute: # osmtest -f a
during the test, the AHs which were mentioned above will be created.
thanks
Dotan
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
