Re: [ofa-general] Re: linux-next: Tree for October 23

[Randy Dunlap]

On Sun, 26 Oct 2008 09:19:59 +0200 Or Gerlitz wrote:

> Randy Dunlap wrote:
> > Building with CONFIG_INFINIBAND=m, kconfig allows CONFIG_NET_9P_RDMA=m,
> > so one module wants symbols from the other (net/9p wants symbols from rmda_*).
> >
> > ERROR: "rdma_destroy_id" [net/9p/9pnet_rdma.ko] undefined!
> > ERROR: "rdma_connect" [net/9p/9pnet_rdma.ko] undefined!
> > ...
> > Is this supposed to be allowed/possible?  Otherwise NET_9P_RDMA might have to depend on INFINBAND=y...
> No, there's no need to config INFINIBAND at built it.  What's the value 
> of CONFIG_INFINIBAND_ADDR_TRANS ?

That should teach me to include the .config file.
However, it's not difficult to recreate.
CONFIG_INFINIBAND_ADDR_TRANS=n because it depends on INFINIBAND && INET,
(INET being TCP/IP) which =n.

NET_9P_RDMA depends on NET && NET_9P && INFINIBAND && EXPERIMENTAL, but this problem
config has NET=y and INET=n.  And NET != INET.

So INFINIBAND_ADDR_TRANS could depend on NET instead of INET (maybe; I don't
know what interfaces it really needs) or NET_9P_RDMA could depend on INET &&
<rest of list>.  But I don't know which change makes the most sense, or if
some other change does.


HTH.  config attached.
---
~Randy

Attachment: config-rdma
Description: Binary data