[ofa-general] umad agent question?

[Hal Rosenstock]

On Tue, 2007-10-30 at 16:24 -0500, Robert Pearson wrote:
> Sean,
> 
> Stepped out for a while.
> As I mentioned before code below was wrong

What's the change ?

>  although not related to problem.
> If anyone wants that simple mad test code fragment I would be happy to
> submit it fixed up as a coding example but I'm not sure how or where.

The IB diags are examples but use libibmad in addition to libibumad.
There are even some which use vendor class MADs (ibping, ibsysstat,
vendstat).

-- Hal

> Thanks for the help!
> 
> Bob
> 
> -----Original Message-----
> From: Sean Hefty [mailto:mshefty at ichips.intel.com] 
> Sent: Tuesday, October 30, 2007 1:34 PM
> To: Robert Pearson
> Cc: 'Hal Rosenstock'; 'Hal Rosenstock'; general at lists.openfabrics.org
> Subject: Re: [ofa-general] umad agent question?
> 
> > That was one thing I saw. The version of OFED installed on my test machine
> > is 1.1 and that must have been added later. This is not a problem. I just
> > was not expecting the behavior.
> 
> We should make sure that the TID is not an issue.  On the send side, the 
> kernel will set the upper 32-bits of the TID.  This is done to ensure 
> uniqueness among multiple users.  The kernel uses this value to retry 
> requests until it receives a response.
> 
> On the receiving side, the response MAD must set the TID to match what 
> it received.  It looks like the madtest code sets this correctly.  Is 
> this what you see?
> 
> Also, in the following code:
> 
> 	mad_set_field  (mad, 0, IB_MAD_METHOD_F,	method);
> 	mad_set_field  (mad, 0, IB_MAD_RESPONSE_F,	0);
> 
> Does this end up clearing the response bit?
> 
> - Sean