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

[Thierry]

Hi,

Here is my patch, I have worked on kernel 2.6.18 on centos 5.2
I get some trouble while accessing struc sg_list, but this
implementention allow forwarding information from struct ib_send_wr in
function post_send (in mlx4/qp.c).
The program in userland is also attached but it only print
informations on console.

I will keep looking at this,

Regards,

Thierry



On Wed, Jun 24, 2009 at 7:55 PM, Jason
Gunthorpe<jgunthorpe at obsidianresearch.com> wrote:
> On Wed, Jun 24, 2009 at 06:55:43PM +0300, Moni Shoua wrote:
>
>> I believe that Jason and I still disagree but...
>
>> Jason suggests that I implement this feature with netlink. This
>> approach might have an advantage but if I understand it right this
>> approach requires a patch also to some user application in order to
>> take advantage of this patch.
>
> The kernel devs have made it clear that the preferred way to export
> this information is through netlink and/or a file in /proc/net/..
>
> You never got an answer if /proc/net is truely discouraged or not.
>
> debugfs is completely useless because it cannot be used for actual
> end-user interrogation and the rules are it should not be used as a
> stable kernel-user interface. (ie is NOT a new dumping ground like
> /proc/ was)
>
>> Also, I think that there is a value for a virtual text file under
>> debugfs to monitor rdma_cm connections easily and without any other
>> special app besides 'cat' (just like in IPoIB).
>
> If you have the netlink based program there really is no point in
> including something under debugfs. It is just bloat.
>
>> Finally, this implementation doesn't contradict netlink
>> implementation in the future (and it won't be the first time).
>
> As I said before, the stack has reached a level of maturity that new
> stuff going in should meaningfully 'move the ball forward' toward a
> mature and complete stack. In the context of state reporting that
> means a stable user interface that programs like lsof, netstat, etc,
> can rely upon.
>
> Putting something under debugfs clearly does nothing to advance that
> goal.
>
> Further, there is clearly alot of state information we should be
> exporting to userspace. RDMA-CM stuff is only a tiny portion. netlink
> is the kernel devs answer to all of this.
>
> Jason
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: netlink_trap.patch
Type: text/x-patch
Size: 9335 bytes
Desc: not available
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20090625/30066410/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: receiver.c
Type: text/x-csrc
Size: 3818 bytes
Desc: not available
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20090625/30066410/attachment.c>