[openib-general] Re: [PATCH] mad_rmpp: Fix receive length calculation

[Hal Rosenstock]

Hi Sean,

On Tue, 2005-09-20 at 18:06, Sean Hefty wrote:
> Hal Rosenstock wrote:
> > mad_rmpp: Fix receive length calculation
> 
> Hal, can you explain problem with the current calculation?  Why does pad need to 
> be adjusted by one header size?

Because the data size comes out to be the class specific data size, but
the payload length in the last segment includes the class header as well
as the filled in data in that last segment.

The case that Eitan documented where it wasn't off by 20 (SA class
header size) was due to the fact that the pad went negative because the
payload length was larger than 200 (the SA class specific data).

Make sense now ?

-- Hal