[openib-general] Re: rtr & post_recv

[yipee]

Roland Dreier <rolandd <at> cisco.com> writes:

> What do you mean by "some HCAs"?  Can you give details about which
> HCAs it works with and which it doesn't?

Ok, I have a bit more details about the HCAs.

****
First node info:
yip1 ] lspci | grep Mell
6:00.0 InfiniBand: Mellanox Technologies MT25208 InfiniHost III Ex (Tavor 
compatibility mode) (rev a0)
yip1 ] cat /sys/class/infiniband/mthca0/fw_ver
4.7.0
yip1 ] cat /sys/class/infiniband/mthca0/node_guid
0002:c902:0000:56ac

****
Second node info:
yip2 ] lspci | grep Mell
06:00.0 InfiniBand: Mellanox Technologies MT25208 InfiniHost III Ex (Tavor 
compatibility mode) (rev a0)
yip2 ] cat /sys/class/infiniband/mthca0/fw_ver
4.6.2
yip2 ] cat /sys/class/infiniband/mthca0/node_guid
0002:c902:0020:3ed8

****
Working nodes info:
yip3 ] lspci | grep Mell
06:00.0 InfiniBand: Mellanox Technology MT25208 InfiniHost III Ex (rev 20)
yip3 ] cat /sys/class/infiniband/mthca0/node_guid
0002:c902:0040:2764
yip3 ] cat /sys/class/infiniband/mthca0/fw_ver
5.1.0

yip4 ] lspci | grep Mell
06:00.0 InfiniBand: Mellanox Technology MT25208 InfiniHost III Ex (rev 20)
yip4 ] cat /sys/class/infiniband/mthca0/node_guid
0002:c902:0040:2768
yip4 ] cat /sys/class/infiniband/mthca0/fw_ver
5.1.0

****

Is there any more info that can help you in resolving this issue?
I tried reproducing this problem using the user-space cmpost.c. In the function 
rep_handler() I moved the post_recvs() call before the modify_to_rtr() and the 
problem reproduced itself.
Any idea?

thanks,
y