[openib-general] problems on device/ports initialization

[Sean Hefty]

Shirley Ma wrote:
> 
>  > If the MAD code encounters an issue initializing a device,  it will 
> cleanup all resources allocated to that device.  The error handling is 
> per device, rather than per port.
> 
> I am not convinced. First only MAD resouce not all resouces allocated to 
> that device gets cleanup. (Some resources are still allocated for all 

I was only referring to the MAD code.

> ports, and if you continue loading ib_ipoib modules, you can see the ib 
> interface with no QP1, but configuring the interface will fail, and hung 
> in the kernel.)
> Second, the port usability(QP1 & QP0 have both have been created 
> successfully) depends on other port is not a good design.

This is an error handling issue that can be fixed.  Although, I don't think that 
it's necessarily worth doing, since errors should be very rare.

The error wrt to ipoib hanging the kernel, if an earlier error should occur, 
needs to be fixed however.  The problem could be in the MAD layer, SA code, or 
ipoib itself.

- Sean