[ofa-general] Re: [PATCH v2] rdma_cm: Add debugfs entries to monitor rdma_cm connections

[Jason Gunthorpe]

On Wed, May 06, 2009 at 07:06:52PM +0300, Moni Shoua wrote:
> > Reall the thinking should be 'I want to make lsof work usefully' not
> > 'I want some random and different hack to let me see something'. And
> > yes, that is harder. But the IB stack is now at the point where these
> > small hard things are the sort of work that is needed to get parity
> > with the other stuff in linux..

> debugfs is a common way to export data from kernel to users (IPoIB
> uses it) and it has it's advantages. On the other hand, netlink has
> it's disadvantages so, I don't think that debugfs is the wrong
> way. It's just another way.

Gah, no! Debugfs is NOT ment to be used for users, it is for kernel
debugging. It is specifically not a stable API and commonly used user
space apps should not rely on it. This is why the distros don't mount
it by default.

Viewing the active QPs and RDMA CM connections is not kernel
debugging, it is necessary data for end user app debugging.

> Besides, remember that rdmacm is only aware of part of the opened
> QPs on the host which may lead to a confusion for one who reads the
> output of lsof ("I know that there is an open QP but I don't see it
> in the list")

Sure, but clearly the end user desire is to make lsof work properly
with all the new objects the IB and verbs APIs introduce to the
kernel. I don't think your patch advances that goal at all.

Jason