[ofw] svn commit 2878 has broken IPoIB_NIDS6_CM

[Smith, Stan]

Uri,
  I finally discovered what broke IPoIB....it was not your WorkItem patch; I had forgotten I was running with multiple IPoIB patches and fooled myself into thinking it was the latest IPoIB patch; yours. Sorry for the confusion.

I spent the entire day trying to understand why IPoIB is broken w.r.t. ttcp operation.
Marching backwards thru SVN commits I discovered that svn.2875 ttcp works correctly, while svn.2876 fails.
Seems the patch that Alex delivered (NDIS supplying incorrect length for an IPv4 address) does not set the IPv4 address entirely correctly.
The patch from Alex did not apply to the winOFED SVN tree so I hand crafted a patch.
I compared my hand crafted patched file with the ipoib_port.cpp which Tzachi supplied and found the code of interest to be exactly the same in both files. I'm a bit confused as to why you don't see the ttcp failure in your builds?

Bottom-line: svn.2876 has broken ttcp (tcp) connectivity.

For the time being, until Alex develops a new patch, I will revert svn.2876 next week.

I am on holiday until next Tuesday, perhaps by then Alex will have a tested patch?

I have yet to test the winOFED svn HEAD without svn.2876...next Tuesday.

stan.


Uri Habusha wrote:
> Work do you mean by "no longer works"? Is ping doesn't work? Which
> test did you run
>
> I really surprise since the checkin only move the receive processing
> from dpc to workitem. It shouldn't affect the functionality.
>
> I'll recheck it on Sunday.
>
> Uri
>
> -----Original Message-----
> From: Smith, Stan [mailto:stan.smith at intel.com]
> Sent: Wednesday, September 01, 2010 7:35 PM
> To: Uri Habusha; Tzachi Dar
> Cc: ofw at lists.openfabrics.org
> Subject: svn commit 2878 has broken IPoIB_NIDS6_CM
>
>
> ttcp using an IPoIB interface no longer works with svn.2878.
>
> Remove the IO work queue code (revert to ipoib_ndis6_cm @ svn 2877)
> and IPoIB (ttcp) works fine.
>
> Please review your commit and fix ASAP.
>
> Testing code '_after_' checkin is 'very' desirable behavior!!
>
> Why did the code work in your internal version of IPoIB and not in
> the OpenIB tree?
>
> stan.