[Openib-windows] RE: HH error codes

[Fab Tillier]

> From: Yossi Leybovich [mailto:sleybo at mellanox.co.il]
> Sent: Monday, August 08, 2005 4:55 AM
> 
> We start to do some basic testing to the AL interface
>
> I notice that when the user give invalid port number the return code is
> IB_ERROR and not IB_INVALID_PORT
>
> After little investigation it looks like all HH_ret_t codes are translated to
> IB_ERROR code regardless to the actual HH code.
>
> In order to support the API all HH error codes need to go through translation
> function that will move them to IBAL codes.

I've been slowly changing those so that the THH functions return ib_api_status_t
instead of HH_ret_t.  This avoids the need for a translation function, and
provides richer error values.  The most recent change was with CQ resize.

Since Mellanox is the maintainer of the HCA code, feel free to put together a
patch for this.

> I also notice that IBAL define cache in mlnx_hob_t structure but when you
> malloc this cache (hca_data.c line 204) you use 2 as hard coded number for the
> number of port
>
> I think its better to obtain this number from the dev_lim query [Mellanox have
> devices with only one port.... and I hope that some day we will have devices
> with more then 2 ports :-) ]

This sounds sane.  The cache was put in place to accelerate local MAD processing
with which I was having some problems (SM requests would timeout on a busy
system).  Go ahead and code up a patch for this.

I'm hoping a lot of these things will get cleaned up as we do the memfree HCA
driver and update the CI API for Windows.

- Fab